A CROSS-SECTIONAL ANALYSIS OF ALCOHOL CONSUMPTION, SEX, AND THE ODDS OF ALZHEIMER’S DISEASE AND OTHER DEMENTIAS IN THE COMMUNITY-DWELLING CANADIAN POPULATION by Jennifer Nicole Tippett B.Sc., University of Victoria, 2013 THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE IN HEALTH SCIENCES UNIVERSITY OF NORTHERN BRITISH COLUMBIA April 2020 Ó Jennifer Nicole Tippett, 2020 ii Abstract Background: The prevalence of Alzheimer’s disease (AD) and other dementias is perpetually increasing in Canada and worldwide with the aging baby boomer population. It is, therefore, important to identify risk factors for these major neurocognitive disorders, such as alcohol consumption, to mitigate the future burden on caregivers and the economy. The purpose of this study was to replicate previous research regarding the dose-response relationship between alcohol consumption and the odds of currently having AD or another dementia. The possibility of a sex effect moderating this relationship was also explored. Participants: Data were obtained for respondents to the combined 2015/16 cycles of the Canadian Community Health Survey who were aged 41 years or older at the time of the survey’s conduction (nweighted = 16,715,618). Methods: Logistic regression was used to crosssectionally assess the relationship between various time- and frequency-related alcohol consumption exposures to outcome dementia status, while controlling for a number of demographic and risk factor variables. Results: A sex effect was identified for drinking at an average frequency of four to six times per week over the past year (p = 0.019, 95% CI: 0.03, 0.73) where women (ORw = 0.13) were more protected against currently having AD or dementia than men (ORm = 0.89) when compared to alcohol abstainers. Binge drinking two to three times per month (OR = 0.19, p = 0.015, 95% CI: 0.05, 0.73) and more than once per week over the past year (OR = 0.16, p = 0.007, 95% CI: 0.04, 0.61) significantly lessened the odds of currently having AD or dementia when compared against alcohol abstainers. A sex effect was present for those who were classified as very heavy drinkers (♂: 6+ drinks/day, ♀: 4+ drinks/day) over the past week (p = 0.018, 95% CI: 1.14, 39.41) where alcohol was protective against currently having AD or dementia in men (ORm = 0.29) and alcohol was a risk factor for currently having AD or dementia in women (ORw = 2.15) when both were iii referenced with alcohol abstainers. Conclusions: With the exception of very heavy drinker women, drinking alcohol was associated with a reduced likelihood of currently having AD or dementia and sex effects were identified for drinking at a moderate frequency over the past year and very heavy drinkers. However, these results should be interpreted with caution due to the possibility of selection, sparse data, and abstainer biases as well as misclassification error. The primary implication of this research is to inform future studies that a more thorough exploration of a sex effect influencing the relationship between alcohol consumption and having AD or dementia is warranted. iv TABLE OF CONTENTS ABSTRACT.......................................................................................................................................................... II TABLE OF CONTENTS ................................................................................................................................... IV LIST OF TABLES ................................................................................................................................................ V LIST OF FIGURES ............................................................................................................................................ VI LIST OF IMPORTANT ABBREVIATIONS .................................................................................................VII ACKNOWLEDGEMENTS ............................................................................................................................ VIII DEDICATION .................................................................................................................................................... IX CHAPTER 1: INTRODUCTION.........................................................................................................................1 CHAPTER 2: LITERATURE REVIEW ............................................................................................................6 DIFFERING TYPICAL AGING FROM DEMENTIA.......................................................................................6 ALCOHOL CONSUMPTION, COGNITIVE DECLINE, AND DEMENTIA ................................................14 SEX DIFFERENCES: DEMENTIA, ALCOHOL CONSUMPTION, AND ALCOHOL’S IMPACT ON DEMENTIA ......................................................................................................................................................23 LITERATURE SUMMARY, RESEARCH QUESTIONS, AND HYPOTHESES ..........................................28 CHAPTER 3: METHODS ..................................................................................................................................30 RESEARCH QUESTIONS AND HYPOTHESES ...........................................................................................30 DATA SOURCE ...............................................................................................................................................31 DATA ANALYSIS ...........................................................................................................................................39 SUITABILITY OF STATISTICAL METHODOLOGY ..................................................................................46 DATA SOURCE AND ANALYSIS: LIMITATIONS AND JUSTIFICATIONS............................................49 CHAPTER 4: RESULTS ....................................................................................................................................52 PREVALENCE OF AD AND DEMENTIA .....................................................................................................52 DESCRIPTIVE STATISTICS ..........................................................................................................................54 MULTIVARIATE MODELS ...........................................................................................................................59 DIFFERENCES BETWEEN INCLUDED AND EXCLUDED SAMPLES ....................................................82 CHAPTER 5: DISCUSSION AND CONCLUSIONS ......................................................................................86 GENERALIZABILITY ....................................................................................................................................94 LIMITATIONS .................................................................................................................................................98 FUTURE RESEARCH ...................................................................................................................................103 CONCLUSIONS.............................................................................................................................................105 REFERENCES...................................................................................................................................................108 APPENDIX A: CALCULATIONS FOR INTERACTION ODDS RATIOS ...............................................121 MODEL B SEX EFFECT INTERACTION CALCULATIONS ....................................................................121 MODEL D SEX EFFECT INTERACTION CALCULATIONS ....................................................................121 v LIST OF TABLES Table 1. Table 2. Table 3. Table 4. Table 5. Table 6. Table 7. Table 8. Variable category, status (predictor, outcome, covariate), Canadian Community Health Survey 2015/16 code names, and descriptions of the variables that were included in the study. Final list of variables included in the analysis with the number of dummy variables indicated in parentheses. Prevalence of Alzheimer’s disease and other dementias by sex. Proxy use, alcohol consumption, demographics, and dementia risk factors among people with and without Alzheimer’s disease or other dementias. Model A: Observed odds ratios relating lifetime alcohol consumption, interview proxy use, smoking status, life stress, highest level of attained education, and age to Alzheimer’s disease or other dementias. Model B: Observed odds ratios relating alcohol consumption frequency in the past 12 months, interview proxy use, smoking status, life stress, highest level of attained education, age, and sex to Alzheimer’s disease or other dementias. Model C: Observed odds ratios relating binge drinking frequency in the past 12 months, interview proxy use, smoking status, life stress, highest level of attained education, age, and sex to Alzheimer’s disease or other dementias. Model D: Observed odds ratios relating Canadian Centre on Substance Use and Addiction drinking risk classification based on drinking frequency in the past week, interview proxy use, smoking status, life stress, highest level of attained education, age, and sex to Alzheimer’s disease or other dementias. p. 36 p. 38-39 p. 52 p. 54-56 p. 60-61 p. 64-66 p. 71-73 p. 76-78 vi LIST OF FIGURES Figure 1. Figure 2. Figure 3. Figure 4. Figure 5. Figure 6. Prevalence of Alzheimer’s disease and other dementias expressed as a percentage by age categorized in five-year increments. Model B: Observed odds ratios demonstrating the likelihood of having Alzheimer’s disease or other dementias associated with various frequencies of alcohol consumption over the past 12 months. Model B: The interaction between consuming alcohol four to six times per week over the past 12 months versus the reference group of no alcohol in the past 12 months and the percentage of those with Alzheimer’s disease and other dementias for (a) men and (b) women. Model C: Observed odds ratios demonstrating the likelihood of having Alzheimer’s disease or other dementias associated with various frequencies of binge drinking over the past 12 months. Model D: Observed odds ratios demonstrating the likelihood of having Alzheimer’s disease or other dementias associated with Canadian Centre for Substance Use and Addiction drinking risk categories based on average drinking frequency over the past seven days in the community-dwelling Canadian population aged 41 and older in 2015 and 2016. CCSA drinking risk categories are as follows: alcohol abstainer (♂ and ♀: 0 drinks/day), light drinker (♂: 1-2 drinks/day, ♀: 1 drink/day), moderate drinker (♂: 3 drinks/day, ♀: 2 drinks/day), heavy drinker (♂: 4-5.9 drinks/day, ♀: 3-3.9 drinks/day), and very heavy drinker (♂: 6+ drinks/day, ♀: 4+ drinks/day). Model D: The interaction between consuming being classified as a very heavy drinker ((♂: 6+ drinks/day average, ♀: 4+ drinks/day average) according to Canadian Centre on Substance Use and Addiction (CCSA) guidelines over the past seven days and the percentage of those with Alzheimer’s disease (AD) and other dementias for (a) men and (b) women in the community-dwelling Canadian population aged 41 years and older in 2015 and 2016. p. 53 p. 67 p. 69 p. 74 p. 79 p. 81 vii LIST OF IMPORTANT ABBREVIATIONS AD Alzheimer’s disease CCSA Canadian Centre on Substance Use and Addiction (formerly known as the Canadian Centre on Substance Abuse) CCHS Canadian Community Health Survey OR Odds ratio RR Relative risk viii ACKNOWLEDGEMENTS First, I would like to sincerely thank my supervisory committee: Drs. Henry Harder, Shannon Wagner, and Sandra Allison. Your thoughtful editing feedback and unwavering support throughout this project kept me motivated in the face of the many roadblocks that I experienced. A special thank-you must also be extended to my external examiner, Dr. Davina Banner-Lukaris for providing me with excellent insights into the importance of distinguishing between gender and sex in research. I also would like acknowledge all of you for your flexibility regarding rescheduling my thesis defence with short notice due to the cycling accident that I had days prior to my original defence date. I must express my endless gratitude to Larine Sluggett of the UNBC Statistics Canada Research Data Centre. Larine went beyond her job description in helping me determine which Statistics Canada surveys could be used in my research, reviewing my Statistics Canada data use application, and always being available to answer any questions that emerged as I conducted my data analysis. Larine was a beacon of hope when I felt like my Thesis research was in jeopardy — thank you! Additional thanks must be offered to the UVic Research Data Centre staff where I performed portions of my data analysis for helping me through the orientation and security clearance processes. Finally, I must express gratitude to my family and friends for being there to celebrate my successes and for offering an ear to listen or a shoulder to cry on when problems arose during this journey. A special shout out must be given to my M.Sc. Health Sciences cohort members and UNBC office mates — working in such close proximity to others going through the same experience was invaluable. I will cherish the time we spent together studying, writing, procrastinating, commiserating, and exploring Prince George and its surrounding areas. ix DEDICATION For my grandparents — Eric and Jean Tippett, and Brian and Eleanor (Cherry) Moss — whose memories of spending time with and life stories I cherish. Thank-you for spurring my passion for research involving older adults. 1 CHAPTER 1: INTRODUCTION The prevalence of major neurocognitive disorder, previously and colloquially known as dementia, is expected to increase dramatically in the near future due to the aging baby boomer population. Globally, it is estimated that 47 million people are living with dementia and 10 million new dementia diagnoses are made every year (World Health Organization [WHO], 2017). The worldwide prevalence of dementia is projected to rise to 75 million in 2030 and 132 million in 2050 (WHO, 2017). In 2016, approximately 564,000 Canadians were assumed to be living with major neurocognitive disorder, totaling $10.4 billion in federal health care and out-of-pocket costs (Alzheimer Society of Canada [ASC], 2016). By the year 2031, 937,000 Canadians are anticipated to be living with this disease and its associated costs are predicted to increase to $16.6 billion (ASC, 2016). Major neurocognitive disorder presents with significant cognitive decline apparent in at least one cognitive domain (complex attention, executive function, learning and memory, language, perceptual-motor, or social cognition); a change in cognition that is objectively recognized by the individual, an informant, or a clinician; and cognitive dysfunction as evidenced by performance on neuropsychological and other clinical tests (American Psychiatric Association [APA], 2013). The perceived impairment in cognition must not be due to delirium or another psychological disorder. Major neurocognitive disorder may be due to Alzheimer’s disease (AD), frontotemporal lobar degeneration, Lewy body disease, vascular dementia, traumatic brain injury, substance or medication use, Parkinson’s disease, Huntington’s disease, other medical conditions, multiple etiologies, or an unspecified reason (APA, 2013). Despite the many varieties of major neurocognitive disorder, AD comprises 60-70% of cases (WHO, 2017). The severity of the disorder may be mild, where individuals have difficulties with minor activities of daily living (e.g., housework); moderate, where individuals have difficulties with 2 major activities of daily living (e.g., feeding); or severe, where individuals are fully dependent on others. The increasing prevalence of major neurocognitive disorder and the perpetually rising costs associated with this disease highlight the importance of identifying risk factors that increase the likelihood of major neurocognitive disorder-related cognitive decline in later life. Major neurocognitive disorder will be referred to by the more prevalent term dementia throughout the remainder of this work. Cognitive decline is an inherent part of normal aging; however, it is typically limited to cognitive functions mediated by the frontal lobes (Deary et al., 2009). Cognitive domains that are dependent on other brain regions remain relatively intact throughout the aging process. Dementia and its preceding states, subjective cognitive decline and mild cognitive impairment, exacerbate the effects of normal aging and can manifest in brain regions other than the frontal lobes, impeding activities of daily living. Cognitive decline, whether pathological or not, is influenced by modifiable and non-modifiable risk factors (Deary et al., 2009; Klimova, Valis, & Kuka, 2017). Higher levels of cognitive reserve, cigarette smoking, and psychological distress are three factors that have been well-documented to influence dementia risk. Elevated cognitive reserve, differences in the way that information is processed that improves the brain’s compensatory capacity, is protective against dementia (Stern, 2009). Cognitive reserve can be estimated from measures of occupational prestige and level of achieved education (Dekhtyar et al., 2015). Cigarette smoking has also been demonstrated to increase dementia risk. Specifically, dementia risk has been found to be increased by 34% per 20 cigarettes smoked daily (Zhong et al., 2015). Finally, both pathological and chronic stress as well as stress in middle age have been observed to increase the incidence of dementia (Doyle, Dunt, & Morris, 2014; Peavy et al., 2012; Sindi et al., 3 2014). Thus, these three factors are important to consider when investigating how other aspects of lifestyle, such as alcohol consumption, influence dementia risk. Alcohol use is extremely prevalent in Canadian society (Canadian Centre on Substance Use and Addiction [CCSA], 2014). In fact, alcohol is the most commonly used psychoactive drug amongst Canadians, with 70% of those aged 15 to 24 and 80% of those aged 25 and older using this drug (CCSA, 2014). Of those who do consume alcohol, 25% engage in risky drinking behaviours and 3.2% abuse alcohol or are alcohol dependent (CCSA, 2014). The risks of consuming alcohol are most often thought to manifest as increased impulsivity, risk-taking behaviour, violence, injury, crime, victimization, and deadly motor vehicle collisions (Moss, 2013; Public Health Agency of Canada [PHAC], 2016); however, research has demonstrated that drinking alcohol can have a direct effect on one’s health in the long term. Alcohol’s relationship with health benefits and risks is complicated — it has a dichotomous relationship with the incidence of illness- and diseaserelated morbidity and mortality (CCSA, 2014). Specifically, mild to moderate drinking can reduce diabetes and cardiovascular disease risk; whereas, heavy drinking is associated with many illnesses such as nutritional and dietary deficiencies, liver cirrhosis, and some cancers (CCSA, 2014; Ross, Wilson, Banks, Rezannah, & Daglish, 2012). The likelihood of dementia in old age has a complex, curvilinear relationship with the amount of alcohol consumed. Specifically, those who abstain from consuming alcohol have a slightly higher risk of cognitive decline than those who drink within the recommended amounts, and heavy drinkers have an elevated risk of dementia when compared to both previously mentioned groups (Lopes et al., 2010; Xu et al., 2017). The negative impact that alcohol has on cognitive function is not limited to old age. Alcohol has been demonstrated to negatively impact cognitive performance in short term memory and executive function tasks as early as 4 middle age (Sabia et al., 2014). It has been suggested that the relationship between alcohol consumption and dementia is influenced by sex; however, to the best of my knowledge, most of this research has focused on the type of alcoholic beverage consumed rather than the volume of alcohol consumed. Furthermore, the results of the little research that has focused on the influence of sex on the alcohol-dementia relationship have been inconsistent. It is thus evident that heavy alcohol consumption has a well-documented negative impact on cognition; however, the potential differences of the effects of light and moderate drinking have not been as thoroughly assessed. Most research has consolidated social drinkers into one group, potentially eliminating a dose-response relationship between the amount of alcohol consumed and dementia. The literature has not been clear regarding whether this affiliation demonstrates a sex effect, therefore, the differential strength of this relationship in women and men will be assessed. The community-dwelling Canadian population with dementia has been underrepresented in the literature. To the best of my knowledge, this will be the first study exploring the relationship between alcohol consumption tendencies, sex, and dementia in this specific population. The purpose of this study is to replicate research (Lopes et al., 2010) regarding the dose-response relationship between the full spectrum of alcohol consumption and self- or proxy-reported dementia status in a community-dwelling Canadian population. The presence of a sex effect will also be investigated as the literature has been unclear about whether the strength of this relationship differs for men and women. Two research questions will be addressed: 1) What is the nature and shape of the relationship between alcohol consumption and the likelihood of dementia? 5 2) Does the association between alcohol consumption and dementia odds vary according to sex? These two research questions will be applied to four varieties of alcohol consumption: lifetime alcohol consumption, alcohol consumption frequency in the past 12 months, binge drinking frequency in the past 12 months, and CCSA drinking risk categorized average alcohol consumption over the past seven days. To isolate the impact of alcohol consumption on dementia risk, other lifestyle-related risk factors for dementia — cigarette smoking-, psychological distress-, and cognitive reserve-related variables — will be controlled for. This research will use data from the combined 2015 and 2016 cycles of the Canadian Community Health Survey (CCHS) accessed from Statistics Canada through the Canadian Research Data Centre Network. 6 CHAPTER 2: LITERATURE REVIEW Differing Typical Aging from Dementia Cognitive decline is a hallmark of the typical aging process and is rooted in brain cell senescence. Typically, cerebral cellular atrophy occurs in an anterior to posterior fashion (i.e., frontal to occipital lobes). Thus, the frontal lobes are afflicted early on, leading to dysfunction in executive function, processing speed, reasoning, and episodic memory (Deary et al., 2009; Nilsson, 2003). Other forms of memory, such as semantic memory, short-term memory, and procedural memory are more robust and retain function throughout aging (Nilsson, 2003), likely because they are less reliant on the frontal lobes. Modifiable and nonmodifiable risk factors can influence the magnitude of brain cell atrophy and cognitive dysfunction (Klimova et al., 2017). Age, ethnicity, sex, and genetic predisposition are examples of non-modifiable risk factors that can influence the likelihood of experiencing cognitive decline. Out of these unchangeable factors, genetics has the most influence over cognitive abilities (Klimova et al., 2017). The presence of lifestyle-related diseases (e.g., cardiovascular disease), substance use (e.g., smoking tobacco, drinking excessively), previous head injuries, sedentary lifestyle, and poor sleep habits are examples of modifiable risk factors that can affect cognitive decline in later life (Deary et al., 2009; Ikeda et al., 2008; Klimova et al., 2017). Eating a diet rich in B vitamins, antioxidants, and omega-3 fatty acids as well as leading a physically active lifestyle are perhaps the most easily accessible intervention points to reduce the chances of cognitive dysfunction (Klimova et al., 2017). Thus, it is apparent that lifestyle is a substantial modifiable risk factor for the cognitive decline associated with natural aging and is imperative to consider in health promotion activities for reducing both an exacerbation of normal cognitive dysfunction and perhaps even dementia risk in the population. 7 Reisberg et al. (2008) proposed a model of cognitive decline associated with dementia, specifically Alzheimer’s disease (AD), which has been recently corroborated by Baker et al. (2017). In this model, pathological cognitive dysfunction progresses through the following stages: subjective cognitive decline, mild cognitive impairment, mild AD, moderate AD, moderate severe AD, and severe AD. Normal cognition is followed by subjective cognitive decline, a phase of increasing dysfunction where the individual has subjective memory concerns but does not yet fulfill diagnostic criteria on objective memory tests. Reisberg et al. (2008) demonstrated that subjective cognitive decline can endure for a maximum of 15 years, resulting in either the individual’s cognitive function reverting to normal or further deterioration to the subsequent stage of decline, mild cognitive impairment. As such, subjective cognitive decline is a risk factor for further cognitive decline: those with subjective cognitive decline are over four times as likely as those with normal cognition to deteriorate to mild cognitive impairment (Reisberg, Shulman, Torossian, Leng, & Zhu, 2010). Mild cognitive impairment is characterized by abnormal changes in memory, language, thinking, or awareness that can be corroborated by informants and objectively measured on cognitive tests (Alzheimer Society of Canada, 2018). There are two forms of mild cognitive impairment: amnestic and non-amnestic mild cognitive impairment (Alzheimer Society of Canada, 2018). With those suffering from the former, memory is the defining domain affected and it is associated with an increased conversion risk to AD. Those with the latter are affected in cognitive domains other than memory and have an increased conversion rate to other forms of dementia, such as Lewy body or frontotemporal dementias. Either amnestic or non-amnestic mild cognitive impairment is associated with increased risk of vascular dementia. Non-amnestic mild cognitive impairment can affect a single domain or multiple domains. The final stage of cognitive decline is dementia, which can vary in severity 8 from mild to severe. While living, those with significant cognition problems can be diagnosed with suspected dementia; however, this suspected diagnosis cannot be verified until a post-mortem autopsy is conducted, and brain physiology is thoroughly examined. Oftentimes, the form of dementia that an individual is suffering from cannot be definitively determined until this point (Johnson, 2011). Genetic allocation, and life experiences and exposures can influence an individual’s ability to retain psychological function despite experiencing neural infarcts, including the cognitive decline experienced with dementia. The way in which this is achieved can be explained with the concepts of brain and cognitive reserve. Brain reserve is an individual’s ability to compensate for brain pathology due to physiological aspects of the brain, such as brain size, and neural and synapse count (Stern, 2009) — i.e., it is physical aspects of the brain that are responsible for the improved ability of an individual to compensate for cognitive deterioration. Alternatively, cognitive reserve explains individual differences in the way that information is processed within the brain that improves its compensatory capacity when faced with neural infarct, injury, or senescence, including AD (Stern, 2012). With cognitive reserve, it is the effectiveness of the synaptic connections within the brain that leads to its ability to cope with degeneration. Cognitive reserve is reliant on neural plasticity, which refers to changes in the secretion of neuromodulators or neurohormones, structure of neuronal axons and dendrites, formation of new synapses, and increased efficacy of preexisting synapses that supports brain development and repair, learning and memory, and the malleability of human behaviour (Salé, Berardi, & Maffei, 2014). Cognitive reserve has two components: neural reserve and neural compensation (Stern, 2012). Neural reserve refers to how cognitively healthy individuals differ with respect to how information is processed by the brain, and how these differences may give some people an advantage over others when 9 experiencing brain pathology. Neural compensation, however, indicates how a neural infarct can change the way in which brain processing occurs to preserve function. As such, an individual with high levels of cognitive reserve will be able to retain psychological function longer than one with low levels of cognitive reserve, despite having the same brain size (Stern, 2012). A combination of genetic endowment and environmental exposure, as well as an interaction between these two factors is thought to promote the accumulation of cognitive reserve throughout the lifespan and hinder brain senescence in aging (Salé et al., 2014). Cognitive reserve can be influenced at any developmental stage by educational and occupational attainment, cognitively complex and stimulating leisure activities, social contact, and even physical activity (Salé et al., 2014). Dekhtyar et al. (2015) conducted a study to explore how cognitive reserve in influenced throughout the lifespan by early childhood education, educational achievement, and data-, people-, and thing-based occupational complexity. The results indicated that higher childhood school achievement and increased data-based complexity in adulthood occupations protected against cognitive decline. Level of education achieved was protective against such decline only if it led to a data-complex occupation. The authors postulated that performance in early schooling supports neural connectivity and thus cognitive flexibility, processing speed, and capacity, which all help to strengthen cognitive reserve. As technology advances, Dekhtyar et al. (2015) suspect that the population will increasingly be exposed to data-enriched careers and will reap the benefits of the protective effects of occupational complexity against cognitive decline. However, it is important to note that exposure to technology, particularly in young people who use social media, has been associated with psychological distress, including increased jealousy towards others, stress levels, and suicidal ideation, as well as decreased sleep (Perry & Singh, 2016). Furthermore, there is evidence to support that cognitive reserve 10 is plastic well into middle age. Cognitive reserve has been observed using electroencephalogram, where high cognitively-functioning older adults were found to allocate more cognitive resources than their younger counterparts to respond with the same level of competence to a stimulus (Riis et al., 2008) — i.e., older adults with higher levels of cognitive reserve are better able to compensate for the reduced cognitive abilities that are normal with aging. In fact, Riis et al. (2008) found that high cognitively-functioning older adults performed similarly to younger adults when presented with novel electroencephalogram stimuli. The average-performing middle-aged adults, alternatively, performed comparably to older adults. The results of this study suggest that experiences in life before middle age set an individual up for either cognitive success or failure in old age (Riis et al., 2008). Thus, interventions may be most effective when implemented, at the latest, during middle-age. Nonetheless, positive life changes at any stage to better one’s health should not be perceived as being fruitless. Dementia Risk Factors. Tobacco smoking, psychological distress, and cognitive reserve — as previously discussed — are three factors that have been well-documented to influence dementia risk. The first factor, tobacco smoking, has been overwhelmingly demonstrated to raise the likelihood of experiencing dementia in old age. A study of Japanese smokers found a dose-response relationship between the number of years spent smoking and the likelihood of dementia; specifically, the authors discovered that smoking for more than 45 years was associated with a twofold increase in the odds of developing dementia in old age (Ikeda et al., 2008). A meta-analytical study by Zhong, Wang, Zhang, Guo, and Zhao (2015) also determined a dose-response relationship between the number of cigarettes smoked and dementia odds. The relative risk of dementia among smoking participants was found to be 1.34, indicating that dementia risk increased by 34% per 20 cigarettes smoked 11 per day (Zhong et al., 2015). Interestingly, smoking increases dementia risk in those lacking the APOE e4 allele more than in those who do carry this allele (Zhong et al., 2015). Smoking cessation was discovered to reduce dementia risk to that of non-smokers (Zhong et al., 2015), demonstrating that smoking only increases dementia risk as long as the individual is regularly smoking cigarettes. Nicotine is known to have a neuroprotective effect within the brain; however, the other toxic chemicals in cigarettes make this protection negligible (Ikeda et al., 2008). It is thought that tobacco smoking increases dementia risk by reducing blood flow to as well as increasing oxidative stress within the brain (Ikeda et al., 2008). Smoking cigarettes has also been associated with reduced cortical volume in brain regions associated with AD and other dementias. Durazzo, Insel, Weiner, and Alzheimer Disease Neuroimaging Initiative (2012) found that cigarette smokers, over a two-year period, experienced higher rates of localized rather than generalized atrophy in the brain in posterior and anterior areas of the brain than non-smokers over the same length of time. Specifically, the affected cortical areas were involved in learning, memory, and the processing of complex, visual, social, and emotional cues, which are all implicated in AD (Durazzo et al., 2012). Evidently, cigarette smoking is an important risk factor for dementia. The second factor, pathological and chronic psychological distress, as well as stress occurring during middle-age have been associated with an increase in dementia likelihood. Stress in its most severe form, post-traumatic stress disorder — a stress-related disorder where an individual, after experiencing trauma, repeatedly re-lives the traumatic event, and displays avoidance, emotional numbness, and increased arousal — is related to an increased likelihood of dementia later in life (Doyle, Dunt, & Morris, 2014). Dementia risk has also been shown to be amplified with increased exposure to chronic stress (Greenberg, Tanev, 12 Marin, & Pitman, 2014; Peavy et al., 2012). Research conducted by Peavy et al. (2012) identified that there is a positive relationship between conversion from pre-dementia states to dementia with increased exposure to stressful life events, such as poor health and financial trouble. The authors also addressed the role of cortisol in this process, as this “fight-or-flight” hormone has been suggested to play an integral role in the association between stress and cognitive decline. Cortisol levels were not found to be associated with deterioration from predementia to dementia; however, a lower cortisol awakening response, which is indicative of exposure to prolonged periods of stress, was related to an increased risk of developing mild cognitive impairment in those who were cognitively healthy at baseline (Peavy et al., 2012). Finally, exposure to stress in middle-age has been connected with increased odds of dementia. Both self-reported stress and work-related stress during middle age have been identified as risk factors for dementia (Sindi et al., 2014; White, 2010). Mechanisms explaining the impact of stress on cognition include changes in the vascular structure of the brain, increased inflammation and oxidative stress, and cortisol-induced damage to the hypothalamus that is hypothesized to spread to other brain regions after surpassing a threshold point (Peavy et al., 2012; Simard, Hudon, & van Reekum, 2009; White, 2010). Nonetheless, stress in and of itself does not cause dementia — it, along with other risk factors, contribute to increased dementia risk (Greenberg et al., 2014). Although one might suspect stress to be related to decreased cognitive reserve, which may explain the increased dementia likelihood in this population, there is evidence to suggest that stress and cognitive reserve influence dementia risk independently (Cabral, Veleda, Mazzoleni, Colarez, NeivaSilva, & Neves, 2016). Unmistakably, exposure to a wide range of stress severities can contribute to increased dementia risk. 13 The third and final well-known factor that can influence dementia risk is cognitive reserve, which was described in the previous section with regards to how it contributes to normal aging. As with normal cognitive decline in aging, there is evidence to suggest that higher levels of cognitive reserve have a protective effect against dementia (Ojagbemi, Bello, & Gureje, 2016; Prince et al., 2012). However, those with extensive cognitive reserve are often diagnosed when the disease is more advanced because more brain infarcts are required before symptoms appear and these individuals are better able to compensate for their cognitive losses (Stern, 2009). In addition to being diagnosed later in the disease process, those with high levels of cognitive reserve have been found to have an increased rate of cognitive decline than those with lower levels of cognitive reserve (Stern, Albert, Tang, & Tsai, 1999). Evidently, cognitive reserve has the ability to significantly influence the likelihood of pathological cognitive decline in old age. Modifiable risk factors like exposure to stress, cigarette smoking, and experiences promoting cognitive reserve are important intervention points for policymakers looking to address the increasing incidence and prevalence of dementia. These three factors have been well-documented to influence dementia risk, thus, they are important to consider when investigating how other aspects of lifestyle may contribute to a dementia diagnosis in old age. Conclusion. Cognitive decline is an inherent part of typical aging; however, it is usually limited to cognitive functions mediated by the frontal lobes. Cognitive domains that are dependent on other brain regions remain relatively untouched. Dementia and its preceding states exacerbate the effects of normal aging and can manifest in brain regions other than the frontal lobes, impeding activities of daily living. Cognitive decline associated with dementia follows a path of increasing dysfunction: subjective cognitive decline, where subjective cognitive deficits occur; mild cognitive impairment, where objective cognitive 14 deficits occur; and finally, dementia. Cognitive decline is a spectrum and is influenced by modifiable and nonmodifiable risk factors. There are several factors which have been welldocumented to affect the likelihood of dementia. Cigarette smoking and various intensities of psychological distress throughout the lifespan have a positive relationship with the odds of being diagnosed with dementia in later life (i.e., more cigarettes smoked and higher levels stress leads to an increased dementia likelihood). Alternatively, cognitive reserve has a negative relationship with the likelihood of dementia diagnosis in old age (i.e., higher cognitive reserve is associated with a lower odds of dementia). Evidently, these factors are important to consider when investigating how other aspects of lifestyle, such as alcohol consumption, influence dementia risk. Alcohol Consumption, Cognitive Decline, and Dementia Alcohol has been viewed in a positive light throughout the majority of human history. During times of inadequate food and water safety, alcohol was an important source of dietary calories and often the only harmless liquid for humans to drink (Cook, 2018). Alcohol was even an important component of medications that physicians prescribed to their patients (Cook, 2018). Alcohol consumption only began to be viewed more critically in the 1980s when the Royal College of Physicians of London put forth what they thought to be harmless limits for the consumption of alcohol — consumption beyond these limits was thought to be hazardous to one’s health. Soon after, the World Health Organization’s International Agency for Research on Cancer declared alcohol to be a group 1 carcinogen, indicating that alcohol is definitively cancer causing in humans, alongside substances such as formaldehyde, asbestos, and plutonium (Cook, 2018; International Agency for Research on Cancer, 2018). That same decade, however, French researchers described a concept called the French Paradox. This concept essentially attributed the cardiac health of French population relative to their intake 15 of saturated fatty acids and cholesterol to their high consumption of alcohol, particularly red wine (Ferrières, 2004). The French Paradox contradicted the perceived risk of drinking in excess of the safe drinking limits recommended by the Royal College of Physicians of London and the declaration that alcohol was a group 1 carcinogen by the World Health Organization. To this day, the French Paradox is pervasive throughout global society, with the general population believing that drinking alcohol can benefit one’s health. In Canada, alcohol is the most commonly used psychoactive substance by Canadians (CCSA, 2014). Like the Royal College of Physicians of London in the 1980s, the Canadian Centre on Substance Use and Addiction (CCSA) has put forth recommendations regarding how much alcohol can be consumed to maximize health benefits and minimize health risks. Long-term health risks can be reduced by limiting daily and weekly alcohol consumption to a maximum of two and 10 drinks, respectively, for women; hazards for men can be minimized by restricting daily and weekly drinking to an upper limit of three and 15 standard drinks, respectively (Butt, Bierness, Gliksman, Paradis, & Stockwell, 2011). Butt et al. (2011) recommend that it is important that days spent drinking are separated by days of alcohol abstinence to prevent habit formation and tolerance effects from occurring. Despite the Canadian government agency’s recommendations, a quarter of Canadians engage in hazardous drinking behaviour, and 3.2% abuse alcohol or are alcohol dependent (CCSA, 2014), putting their long-term health at risk. In fact, more people are killed by alcohol than by motor vehicle collisions and more people are hospitalized because of alcohol than heart attacks (Canadian Institute for Health Information [CIHI], 2018). Alcohol use is extremely pervasive in Canadian culture and is a problem when it comes to morbidity and mortality amongst Canadians. 16 Alcohol consumption has been associated with health benefits and risks. The French Paradox is one of the primary reasons that alcohol is viewed as having a positive influence on health. The researchers who discovered the apparent French Paradox discovered that the French population has a proportionally low risk of coronary heart disease compared to other countries with similar fat and cholesterol consumption (Evans, 2011; Ferrières, 2004). For example, the people of Finland and France have comparable saturated fat intakes, but Finland’s coronary heart disease mortality risk is over five times higher than that of France (Ferrières, 2004). There are two major lifestyle differences between France and other countries that may contribute to this paradoxical discrepancy: the regular consumption of red wine and the adherence to a Mediterranean diet — one rich in fresh fruit and vegetables — among the French. The proposed mechanism behind alcohol’s health benefits has to do with the presence of polyphenols, particularly flavonoids, in both red wine and a Mediterranean diet (Ferrières, 2004). Polyphenols are naturally occurring antioxidants that are found in plants and consumed as a component of the human diet in the form of fruit, vegetables, and their derivatives (e.g., coffee, tea, beer, wine) (Ferrières, 2004; Scalbert, Johnson, & Saltmarsh, 2005). The consumption of polyphenols has been associated with the prevention of cardiovascular disease, some cancers, osteoporosis, diabetes mellitus, and neurodegenerative diseases, such as AD and other dementias (Scalbert et al., 2005). Regarding coronary heart disease, cardiac lesions are thought to result in response to high concentrations of oxidized lipids which are generated via low density lipoprotein oxidation (Ferrières, 2004). Polyphenols, particularly flavonoids, a class of polyphenols, are thought to inhibit low density lipoprotein oxidation, thus preventing cardiac lesions from forming (Ferrières, 2004). This mechanism explains how the French can have good cardiac health despite consuming an abundance of dietary fat. With respect to neurodegenerative diseases, 17 alcohol’s impact can be positive or negative depending on what variety of alcoholic beverage is primarily consumed. Ethanol, the dominant form of alcohol found in alcoholic beverages, is known to increase the production of free radicals and thus levels of oxidative stress in various systems throughout the body (Sun, A., Simonyi, & Sun, G., 2002). For chronic alcohol consumers, this oxidative stress results in neuronal atrophy within the brain (Sun A. et al., 2002). However, the polyphenols found in red wine resulting from grape skins and seeds can mitigate this neuronal atrophy caused by ethanol as well as other oxidative agents (Sun A. et al., 2002). This research suggests that the neuroprotective effects of consuming alcohol with high concentrations of polyphenols can trump the neurodegenerative effects of chronic alcohol consumption. With regards specifically to AD, polyphenols have been found to inhibit the formation of amyloid-β plaques in the brain, which are a defining characteristic of this disease (Loureiro et al., 2017). As evidenced by the French Paradox, polyphenols are the driving force behind the apparent protective effect of alcohol consumption on cardiac and neuronal health. However, it is possible that alcohol has received too much recognition in this association. The Mediterranean Diet also contains high concentrations of polyphenols, and there is no negative effect associated with fresh fruit and vegetable consumption and oxidative stress as there is with ethanol consumption. Alcohol consumption, particularly binge drinking, has many well-documented shortand long-term health risks. Personal injury resulting from vehicle collisions, falls, burns, and drownings; violence; alcohol poisoning; and miscarriage resulting from fetal alcohol spectrum disorder are examples of the short-term health risks of excessive drinking (Centers for Disease Control and Prevention, 2018). Short-term health risks are what the general population usually considers when thinking about the health risks of alcohol consumption. However, alcohol consumption can have a far more insidious, compounding effect on one’s 18 health throughout the lifespan resulting in high blood pressure, stroke, heart disease, liver disease; breast, mouth, throat, esophagus, liver, and colon cancers; mental health and social issues; alcohol dependence; and learning and memory problems, including dementias (Centers for Disease Control and Prevention, 2018). Alcohol consumption has been demonstrated to affect the biochemistry of the brain and body on a microscopic level. Regarding alcohol’s impact on the brain, consuming alcohol within low risk limits has not been associated with structural changes to the gray and white matter of the brain (Preti et al., 2014), suggesting that alcohol’s mechanism of action on cognition occurs at a level of organization that cannot be viewed by neuroimaging technology. Homocysteine, a homologue of the amino acid cysteine found in the blood, has been suggested to provide an explanation for alcohol’s influence on cognitive decline. High blood concentrations of homocysteine have been associated with an increased risk of brain cell atrophy, cognitive decline, and dementia, particularly AD (Turkington & Mitchell, 2010). It has been suggested that high levels of homocysteine can exacerbate the typical characteristics of AD: cognitive dysfunction, and the elevated prevalence of β-amyloid and tau proteins in the brain (Li, Chu, Barrero, Merali, & Praticò, 2014). Although this research was conducted in mice, the authors’ findings support the idea that lifestyle can influence the progression of AD (Li et al., 2014). AD patients tend to present with elevated plasma homocysteine as well as with deficiencies in folate and vitamin B6 (Turkington & Mitchell, 2010). Levels of homocysteine in the blood are moderated by dietary intake of folic acid and vitamins B6 and B12 (Gibson et al., 2008; Turkington & Mitchell, 2010). The literature has not reached a consensus regarding how homocysteine affects cognition or whether modifying one’s diet to include foods rich in folic acid and B vitamins can protect against cognitive decline. Nevertheless, there is 19 evidence to suggest that brain atrophy in those with mild cognitive impairment can be reduced through dietary B vitamin supplementation (Smith et al., 2010). Raised plasma homocysteine may also explain alcohol’s negative impact on cognition. Robinson, Narasimhan, Weatherall, and Beasley (2005) conducted a study assessing blood concentrations of homocysteine in those participating in an in-patient alcohol detox program. Blood samples were taken when the patients were admitted and discharged from the hospital. Upon admission, over half of the sample had plasma homocysteine concentrations exceeding the reference range (Robinson et al., 2005). After a few days when the detox program was completed, plasma homocysteine levels dropped compared to their hospital admission levels; however, roughly a quarter had readings that remained higher than the normal range (Robinson et al., 2005). The authors propose that there is a dose-response relationship between alcohol consumed and plasma homocysteine levels, as even those who drink moderate amounts of alcohol demonstrate increased plasma homocysteine (Robinson et al., 2005). Perhaps those whose plasma homocysteine levels remained high upon completion of the program drank much more alcohol than the other study participants, resulting in a longer length of time for the body to return to homeostatic plasma homocysteine levels. Alcohol consumption has also been linked to decreased blood concentrations of the B vitamins and folic acid which, as previously mentioned, moderate plasma homocysteine concentrations (Gibson et al., 2008). Thus, alcohol not only increases blood concentrations of homocysteine, but decreases the availability of homocysteine’s regulators. Homocysteine and its regulators appear to play a significant role in alcohol’s effect on cognitive decline in those who drink alcohol, whether moderately or heavily. The previous few paragraphs demonstrate that polyphenols are largely responsible for the supposed health benefits of alcohol 20 consumption and that elevated plasma homocysteine levels are responsible for the negative impact of alcohol consumption on the brain and body. As previously mentioned, alcohol consumption has been associated with exacerbated cognitive decline. Regulations have been proposed by both researchers and government agencies as to what are considered safe alcohol consumption limits that maximize the supposed health benefits and minimize the many health risks. These recommended doses of alcohol are often described in terms of standard drinks. A standard drink is one containing approximately 13.6g of pure alcohol and is the equivalent to a 341mL bottle of 5% beer, cider, or cooler; a 43mL shot of 40% hard liquor; or a 142mL glass of 12% wine (Rethink Your Drinking, 2016). Xu et al. (2017) suggest that a daily alcohol dose of less than or equal to 1.25 standard drinks, regardless of sex, optimized protection against diseases including heart disease, some cancers, stroke, and hypertension in addition to dementia; and alcohol consumption in excess of 3.8 standard drinks per day was associated with significantly elevated risks of the aforementioned afflictions. Alternatively, Sabia et al. (2014) found that, for men only, consistently drinking 3.5 standard drinks per day led to an increased risk of cognitive decline in all domains. In women, regularly drinking more than 1.8 standard drinks led to dysfunction in the cognitive domain of executive function only. There is evidence to suggest that the relationship between alcohol consumption and dementia likelihood is curvilinear and complex. A study by Lopes et al. (2010) identified a J-shaped association between alcohol consumption and risk of both cognitive decline and dementia for women only, such that women who abstain from consuming alcohol have a slightly elevated risk of cognitive decline when compared to light drinkers, but a much lower risk than women who are heavy drinkers. However, the increased risk of dementia for those who abstain from drinking could be explained by the abstainer group being generally unhealthier than the other 21 alcohol consuming groups (Cook, 2018; Hassing, 2018). Xu et al. (2017) have demonstrated a U-shaped relationship between alcohol consumption and risk of all-cause dementia and AD, but not vascular dementia. Nevertheless, the latter study did not include those who did not consume alcohol, which could explain the discrepancy between the shape of the association between alcohol consumption and cognitive decline. Other research has suggested that the effect of light to moderate drinking on dementia risk is dependent on dementia type. Specifically, the likelihood of developing unspecified dementia and AD has been suggested to be reduced by light or moderate drinking (Panza et al., 2012). The type of alcohol consumed may have a variable impact on cognition. Drinking alcoholic beverages produced from fruit and vegetables, particularly wine, has offered protection against cognitive decline due to their polyphenol content; however, drinking beer was correlated with elevated dementia risk (Xu et al., 2017). Furthermore, those who consume alcohol are more likely to use tobacco products (Ames et al., 2010), which, as mentioned previously, has been associated with a higher dementia risk (Ikeda et al., 2008). Evidently there is a link between alcohol consumption (including type of alcohol consumed), cognitive decline, and dementia. Unfortunately, the impacts of severe drinking on dementia risk have been studied more extensively than drinking within government-issued guidelines. Additional factors influencing alcohol consumption. Smoking tobacco products and psychological distress have been demonstrated to increase the likelihood of drinking alcohol. Although it is difficult to find peer-reviewed sources discussing the simple association between alcohol consumption and tobacco use, two studies were identified that explored this relationship in adolescents and young adults. Reed, Wang, Shillington, Clapp, and Lange (2007) found a positive relationship between both the amount of alcohol consumed and drinking frequency, and cigarette smoking in a population of undergraduate 22 students. Another study by Grucza and Bierut (2006) found that smokers drink more than those who have never smoked, and that those who smoke were over four times more likely to suffer from alcohol use disorders than their non-smoker counterparts who drank the same amount of alcohol. Unmistakably, there is an existing correlation between alcohol consumption and tobacco smoking, such that those who smoke tobacco drink alcohol more than those who do not and are more likely to exhibit problematic drinking behaviour. There are four factors that drive individuals to consume alcohol: coping, conforming, inducing a positive mood, and socializing; however, problematic drinking most often results when alcohol is used as a coping mechanism (PHAC, 2016). Psychological distress may drive an individual to use alcohol as a coping tool. Psychological distress, indeed, has a positive relationship with the prevalence of alcohol use (Balogun, Koyanagi, Stickley, Gilmour, & Shibuya, 2014). Additionally, Barnes (2013) reported that perceived stress was more reliably able to predict drinking problems than amount of alcohol consumed, and that sex influences stress’ impact on alcohol consumption. Specifically, males were more susceptible to alcohol use problems than females when experiencing stress (Barnes, 2013). It is noticeable that there is a strong connection between perceived life distress and alcohol consumption. The previously presented evidence demonstrates the importance of accounting for psychological distress and tobacco smoking when assessing the relationship between alcohol consumption and dementia. Conclusion. Alcohol use has been extremely prevalent throughout human history and this trend endures in modern Canadian society despite warnings from public health officials about its negative health effects. Alcohol’s effect on health is dualistic in nature, with mild to moderate drinking, particularly of beverages containing an abundance of polyphenols, being 23 linked to positive long-term health outcomes and heavy drinking being associated with increased morbidity and mortality. The likelihood of experiencing cognitive dysfunction such as dementia in old age has a curvilinear relationship with the amount of alcohol consumed, as those who abstain from consuming alcohol appear to have a slightly higher risk of cognitive decline than those who drink within the recommended amounts but those who drink heavily have a very high dementia risk. However, alcohol abstainer groups in research are often composed of individuals with poorer overall health than those in other alcohol consumption groups. The presence of high levels of homocysteine in the blood, as well as deficiencies in its regulators, has been suggested to provide a biological link between alcohol consumption and increased incidence of cognitive decline. Heavy alcohol consumption has a welldocumented negative impact on cognition; however, the effects of light and moderate drinking have not been as thoroughly assessed. Most research has consolidated social drinkers into one group, potentially eliminating a dose-response relationship between the amount of alcohol consumed and cognitive dysfunction. Finally, tobacco smoking and psychological distress influence the likelihood of consuming alcohol; therefore, it is important to consider their roles as confounding variables when addressing the impact of alcohol consumption on dementia risk. Sex Differences: Dementia, Alcohol Consumption, and Alcohol’s Impact on Dementia Women have been significantly underrepresented as research subjects. Historically, the typical research participant has been Caucasian and male, leaving potential sex differences between men and women underexplored or unexplored in academic literature (Liu & Manger, 2016). It should be noted that there are also inequalities in the ethnicities represented in research, with those other than Caucasian being underrepresented (Liu & Manger, 2016); however, the focus here will be on sex differences. Women have been 24 excluded as study participants due to the assumption that men are an adequate proxy for women, women being perceived as more confounding and expensive subjects due to fluctuating hormone levels, and reproductive concerns for women of child-bearing age or for those who were pregnant (Liu & Manger, 2016). Women and men differ with regards to their lifestyle, environment, behaviour, and biology (Liu & Manger, 2016). These differences result in variations in disease prevalence, presentation, diagnosis, severity, and outcomes, in addition to differences in the efficacy of pharmaceutical drugs as well as in how these drugs are metabolized (Liu & Manger, 2016). However, the historical exclusion of women from research has adversely effected women’s health outcomes. For example, cardiovascular disease is the leading cause of death for both men and women; however, the majority of the research on the diagnosis and treatment of this disease has been conducted on middle-aged men (Harvard Medical School, 2014). Men often have atherosclerotic plaques that form in the large coronary arties of the heart while women tend to have small vessel disease, where plaques form in the very small blood vessels of the heart (Harvard Medical School, 2014). Resultantly, women are often told that their heart is healthy because imaging techniques are used to check for blockages in the coronary arteries, not the smaller heart vessels and, if left untreated, this condition may result in a heart attack (Harvard Medical School, 2014). Women also have double the likelihood of experiencing unexpected side effects to pharmaceutical drugs than men, with 80% of drugs in the United States being withdrawn from the market for this reason (Bruinvels, Burden, McGregor, Ackerman, Dooley, Richards, and Pedlar, 2017). Thus, due to the inherent differences between the sexes, it is important first, to actively include women in research studies and, second, to consider how sex differences may affect relationships being explored in research. Regarding the present study, 25 it is important to consider how the prevalence of dementia, alcohol consumption habits, and the impact of alcohol consumption on the development of dementia vary by sex. Dementia disproportionately affects men and women, with 65% of those who are diagnosed with dementia, including AD, after the age of 65 being women (Alzheimer Society of Canada, 2019). A study by Chêne et al. (2015) found that one in five women would be diagnosed with dementia in their lifetime while the same lifetime risk for men was one in 10. However, despite women being more likely to have dementia in their lifetime, the preceding state to dementia, mild cognitive impairment, is more prevalent in men. Men are also more likely to stall along the progression to dementia in this state and, therefore, less likely to be diagnosed with dementia (Petersen et al., 2010). One explanation for this paradox has to do with men having a higher risk of cardiac health issues earlier in life than women. As a result, the men who survive to the age where dementia is a concern have better cardiovascular health, which reduces their dementia risk (Chêne et al., 2015). Furthermore, it has been suggested that women transition more abruptly between normal cognition and dementia than men, but at a later age (Petersen et al., 2010). Evidently, dementia rates inherently differ by sex, emphasizing the importance of taking sex differences into consideration when investigating this disease. Men and women vary with regards to their alcohol consumption tendencies. The prevalence of drinking, including heavy drinking, is higher in men than in women (Wilsnack, Wilsnack, Kristjanson, Vogeltanz-Holm, & Gmel, 2009). Women are more likely than men to both abstain from alcohol during their life and stop drinking if they have consumed alcohol at one point (Wilsnack et al., 2009). In Canada, 24% of males aged 12 and older reported to be heavy drinkers; whereas, 15% of women identify as heavy drinkers (Tam, 2018). Canadian women are of particular concern with regards to their risky drinking behaviours. 26 Although the proportion of women who are heavy drinkers is comparatively low, they are trending towards catching up with men with regards to alcohol-related deaths. According to a report from the Canadian Institute for Health Information (2018), the number of deaths in women that can be attributed to alcohol consumption has grown by 25% since 2001; deaths in men has only risen 5% over the same period of time. Furthermore, the number of women who are hospitalized because of drinking alcohol and related complications rose by 3% between 2015/16 and 2016/17; the hospitalization rate for men only increased 0.6% in the same period (CIHI, 2018). High-volume alcohol consumption rates also vary between Canadian women of varying sexual identities. Women who are bisexual or lesbian have heavy drinking rates of 30% and 25%, respectively; whereas, the same rate for heterosexual women is 15% (Tam, 2018). It is clear that alcohol consumption habits differ between men and women; therefore, the inherent differences in drinking tendencies between men and women might have a variable impact on dementia likelihood. This evidence highlights the importance of considering sex differences when examining the relationship between alcohol consumption and dementia. To the best of my knowledge, there is little research specifically addressing the differential relationship between alcohol consumption and dementia for men and women, and the studies that have investigated this association have reached conflicting conclusions. One study by Lyu and Lee (2014) looked at alcohol consumption and cognitive impairment and whether sex impacted this relationship. The authors found that the influence of alcohol consumption on cognitive impairment, measured by effect size, did not differ significantly between the sexes (Lyu & Lee, 2014). This suggests that sex does not impact the alcohol/cognition relationship and, surprisingly, that it may not be necessary to consider sex differences in this case. Conversely, Wardzala et al. (2018) found evidence of sex differences 27 in the alcohol consumption and cognitive decline relationship despite the effects of alcohol being negligible when sex was not stratified. When the sexes were separated, male and female moderate drinkers — in this study defined as no more than four drinks per day or 14 drinks per week for men and no more than three drinks per day or seven per week for women — were found to be protected against cognitive decline and brain pathology, but to varying degrees (Wardzala et al., 2018). The authors identified a significant protective effect of moderate drinking against decline in men; the protective effect of moderate drinking on cognitive decline in women was trivial (Wardzala et al., 2018). The authors explained that this may be due to women being more susceptible to the negative effects of alcohol than men (Wardzala et al., 2018). It is clear that the little research that exists regarding the relationship between sex, alcohol consumption, and dementia has reached conflicting results, emphasizing the importance of conducting further research in this area. Despite women being inherently underrepresented as research participants, there are clear sex differences in dementia prevalence and the way that alcohol tends to be consumed. More women than men are diagnosed with dementia; however, the sex differences in the onset of cardiovascular issues and the impact of these issues on survival to the age where dementia tends to become more common may help explain this phenomenon. Men consistently out-drink women, but the gap between the sexes has been narrowing in recent years. The results of the little research addressing the influence of sex on the relationship between alcohol consumption and dementia likelihood have been conflicting. Some studies have shown that sex does not influence this relationship; whereas, others have demonstrated that alcohol’s apparent protective effect against dementia is amplified in men but is negligible in women. The evidence presented here highlights the importance of stratifying by 28 sex as well as the academic necessity of this stratification due to the inconsistency of the results of previous research. Literature Summary, Research Questions, and Hypotheses With the aging baby boomer population, it is expected that both the incidence of dementia and its economic burden on families and governments will increase. Thus, it is important to identify modifiable risk factors that may inflate this and subsequent populations’ likelihood of developing dementia. One modifiable risk factor that has been proposed as being hazardous to cognitive health is alcohol consumption, which is extremely prevalent in Canadian society. The aforementioned literature has provided evidence that alcoholic drinking is linked to an increased dementia risk; however, much of this research has focused on extreme alcohol consumption and has failed to investigate sex differences. Those that have investigated a sex effect have led to conflicting results. The purpose of this study was to replicate previous research (Lopes et al., 2010) regarding the dose-response relationship between the full spectrum of alcohol consumption and self-or proxy-reported dementia status in a community-dwelling Canadian population. The presence of a sex effect was also investigated as the literature has been unclear about whether the strength of this relationship between alcohol and dementia differs for men and women. Two research questions were addressed: 1) What is the nature and shape of the relationship between alcohol consumption and the likelihood of dementia? 2) Does the association between alcohol consumption and dementia odds vary in strength according to sex? These two research questions will be used to explore the relationship between AD and dementia odds and four timelines/frequencies of alcohol consumption: lifetime alcohol 29 consumption, drinking frequency in the past 12 months, binge drinking frequency in the past 12 months, and CCSA drinking risk categorization based on the average number of drinks consumed in the past week. To avoid the relationship between alcohol consumption, sex, and dementia being obscured by confounding factors, other lifestyle-related risk factors for dementia — specifically, cigarette smoking, psychological distress, and cognitive reserverelated variables — were controlled for. This research used data from the combined 2015 and 2016 cycles of the Canadian Community Health Survey available from Statistics Canada and accessed through the Canadian Research Data Centre Network. 30 CHAPTER 3: METHODS Research Questions and Hypotheses The methodologies described hereafter were designed to build four logistic regression models, each regarding the relationship between various timelines of alcohol exposure and differing frequencies of alcohol consumption and the likelihood of having dementia or AD. The first model had to do with whether survey respondents had been exposed to alcohol in their lifetime. The second regarded survey respondents’ frequency of alcohol consumption in the past 12 months. The third concerned the frequency with which survey respondents engaged in binge drinking within the past 12 months. Finally, the fourth model pertained to average number of drinks consumed over the past week, categorized according to the Canadian Centre on Substance Use and Addiction’s (CCSA) drinking guidelines. Each model was run first without and second with a sex interaction term to determine whether sex played a significant role in moderating the relationship between alcohol consumption and AD or dementia. All of these logistic regression models controlled for the confounding variables cigarette smoking, psychological distress, and cognitive reserve, as the literature has suggested that these variables may influence the likelihood of one developing AD or dementia. The goals of this research were twofold. First, what was the nature and shape of the relationship between the four previously mentioned categorizations of alcohol consumption and the odds of AD or any other dementia? Second, did the association between the four categorizations of alcohol consumption and AD or any other dementia vary in strength according to sex? Based on the review of the literature, four hypotheses were posed to answer the four previously mentioned research questions. First, it was anticipated that those who had consumed alcohol in their lifetime would have a higher AD or dementia likelihood than those 31 who had not consumed alcohol in their lifetime. Second, it was predicted that the relationship between alcohol consumption frequency in the past 12 months and AD or dementia likelihood would be J-shaped, such that those who did not consume alcohol, consumed alcohol in moderation, and frequently consumed alcohol have a moderate, reduced, and elevated dementia risk, respectively. Third, it was expected that the relationship between binge drinking frequency in the past 12 months and AD or dementia odds would be J-shaped, with those who did not consume alcohol in the past year, those who drank but did not binge drink, and those who engaged in binge drinking having a moderate, reduced, and elevated AD or dementia likelihood, respectively. Fourth, it was assumed that the relationship between CCSA drinking risk categorization based on alcohol consumption in the past seven days and AD or dementia risk would be J-shaped, such that alcohol abstinence, light, moderate, heavy, and very heavy alcohol consumption yield moderate, moderate-low, low, high, and very high dementia risk, respectively. Regarding the sex effect component of this research, it was expected that sex would moderate the relationship between the four categorizations of alcohol consumption and risk of AD or dementia in men and women; however, it was unknown how these rates would differ. As such, this analysis was exploratory in nature. Data Source The Canadian Community Health Survey (CCHS) is a cross-sectional survey that was first conducted in the year 2000 with the goal of streamlining the health data that was being collected from Canadians and improving the translation of knowledge gained from the analysis of this data to the Canadian population (Statistics Canada, 2017a). The CCHS consists of data regarding the health status, health care utilization, and determinants of health for Canadians (Statistics Canada, 2017a). The survey itself consists of four sections: core 32 content, theme content, optional content, and rapid response content (Statistics Canada, 2017a). The core content is posed to all survey respondents and will be unchanged until the year 2021. The theme content is also presented to all survey respondents, but these questions relate to a specific topic or theme which changes annually or biannually. The optional content provides the provinces and territories with the ability to collect data from their residents regarding topics of public health importance at the provincial or territorial level and changes biannually. Finally, the rapid response content is only asked of those residing in the provinces and provides an opportunity for organizations, on a cost-recovery basis, to ask questions on topics of interest to them. The specific version of the CCHS that was utilized in this research was the combined 2015 and 2016 cycles of the CCHS. The following sections — Participants, Sampling, and Data Collection — were paraphrased from the CCHS 2016 Microdata File User Guide (Statistics Canada, 2017a). Participants. The target population of the combined 2015 and 2016 cycles of the CCHS was Canadians aged 12 and older at the time the survey was conducted who resided in any of the ten provinces or three territories. Prospective interviewees were excluded if they were: living on Aboriginal reserves or settlements; full-time members of the Canadian Forces; foster care youth aged 12-17; institutionalized; or living in two remote health regions in the province of Québec (Région du Nunavik and Région des Terres-Cries-de-la-BaieJames). Sampling. A geographical frame applied to the sampling procedures for the CCHS. The country was divided into health regions with the territories comprising their own and the provinces being subdivided into several health regions. In British Columbia, for example, health regions were created based on the provincial health service delivery areas served by the health authorities throughout the province. In total, the provinces were divided into over 33 100 health regions in addition to the three territorial health regions. A minimum sample size of 500 respondents per health region was applied to ensure that the collected data was of acceptable quality; however, a maximum sampling fraction of one per 20 households was adhered to so as to reduce the likelihood that dwellings in less populous areas would be repeatedly sampled between this and other Statistics Canada surveys. A national sample size goal of 120,000 adults was set with 117,000 and 3,000 respondents being sought in the provinces and territories, respectively. In the ten provinces, the ideal sample size was allocated first within each province and second to the provincial health regions based on their respective population sizes. First, dwellings within each province were stratified based on geographical (health region) and socioeconomic criteria. Second, within each stratum, between 150 and 250 dwellings or households were clustered together. Third, individual dwellings were systematically sampled for inclusion from within the clusters. This process differed in the territories. Larger and smaller territorial communities were distinguished with the former comprising its own stratum and the latter being grouped into strata with other similar communities based on characteristics such as population, median household income, geographical information, and proportion of Inuit and/or Aboriginal persons. There were six strata in the Yukon, ten in the Northwest Territories, and ten in Nunavut. For strata composed of groups of communities, one community within each stratum was selected for inclusion based on a probability that was proportional to the communities’ population sizes. Dwellings were then selected using the same procedure as for the provinces. 34 The sample size goal for youth respondents was 10,000. Youth were selected from a list and then contacted for interviews over the phone. The total desired sample size for both youth and adults was 130,000. Data collection. Interviews, whether conducted in person or over the phone, consisted of three parts: entry, health content, and exit components. The entry and exit components consisted of questions designed to aid the interviewer in contact initiation, sample information collection, respondent selection, and determination of case status. The health content portion contained the majority of survey questions asked of the interviewees and consisted of the health modules. Since youth were individually selected using a list, they were specifically contacted and asked whether or not they wished to participate in the survey. If they desired to be included, the interview would begin. This process was slightly more complicated for adult respondents. In this case, dwellings were selected and contacted. A knowledgeable household member was asked to provide basic demographic information for all the residents of the dwelling and then one member of the dwelling was asked to participate in the more thorough health content survey component. For the 2015 and 2016 combined cycles of the CCHS, the national response rates for youth and adults were 54.7% and 59.9%, respectively; and the total national response rate was 59.5%. Several efforts were made to reduce the non-response rate. Brochures and letters were sent to the sampled households prior to the collection period to introduce the dwelling residents to the survey and emphasize the importance of participating. If first contact was made at an inopportune time, interviewers were instructed to ask when a more convenient time would be to conduct the interview. If the household could not be reached over the phone, an in-person visit was scheduled. If these efforts were not successful, an additional 35 information package was left on the doorstep and follow-up phone calls were placed at varying times of the day. In the case that participation was refused, a Statistics Canada staff member would phone or visit the dwelling to discuss the importance of taking part. To prevent language barriers from being an issue, every Statistics Canada office employed interviewers with skills in many languages. In 2016, youth interviews required parent/guardian consent. Every effort was made to ensure the parents/guardians were comfortable with the interview taking place, including providing the parents/guardians with information and a copy of the questionnaire. Finally, proxy interviews were conducted in cases where the primary respondent was physically or mentally unable to participate in the survey themselves. Measures were taken to ensure the data collected was of sufficient quality including verifying and monitoring interviewer performance and utilizing internal reports to monitor data collection targets and data quality. Population of interest. The specific population of interest in this study was composed of combined 2015 and 2016 CCHS cycle respondents who were 41 years of age or older at the time the survey was conducted and who responded yes or no to the question “Do you have Alzheimer’s disease or any other dementia?”. Participants were included regardless of whether they completed the survey independently or using a proxy. Prospective participants were excluded from the study if they were missing data for any of the variables of interest, with the exception of variables ALW_010 through ALW_040, which were included despite missing data. The rationale for including these variables will be discussed below. The final unweighted and weighted samples consisted of 67,629 and 16,715,618 respondents, respectively. 36 Variables. Names and descriptions for several variables of interest that were extracted from the 2015/16 CCHS are listed below in Table 1. Table 1. Variable category, status (predictor, outcome, covariate), CCHS 2015/16 code names, and descriptions of the variables that were included in the study. Variable Category Status Code Name Description Alzheimer’s Outcome CCC_145 Indicates whether each individual has or disease/dementia variable does not have dementia status Weekly volume of Predictor ALW_010 Number of drinks consumed on the first alcohol consumed variables day of the week ALW_015 Number of drinks consumed on the second day of the week ALW_020 Number of drinks consumed on the third day of the week ALW_025 Number of drinks consumed on the fourth day of the week ALW_030 Number of drinks consumed on the fifth day of the week ALW_035 Number of drinks consumed on the sixth day of the week ALW_040 Number of drinks consumed on the seventh day of the week Alcohol Predictor ALC_005 Ever had a drink in lifetime consumption variables frequency ALC_010 Had a drink in the past 12 months ALC_015 Frequency of alcohol consumption in the past 12 months ALC_020 How often more than 5 or 4 alcoholic drinks have been consumed at one time in the past 12 months for men and women, respectively (binge drinking) Sex Predictor DHH_SEX Respondent sex variable Cigarette smoking Covariate SMKDVSTY Smoking status Psychological Covariate GEN_020 Life-related psychological distress distress Cognitive reserve Covariate LBFF14 Occupation status EHG2DVR9 Highest level of attained education Age Covariate DHH_AGE Respondent age Proxy use Covariate ADM_PRX Health component completed by proxy 37 All variables were taken from the database verbatim for analysis, with the exception of a number of variables which were recoded so that the “no” responses possessed the lowest numeric value (i.e. zero) and variables ALW_010 through ALW_040 which described weekly alcohol consumption by day of the week. Regarding the former, some variables were coded in the CCHS such that the “no” responses (e.g., did not consume any alcohol in the past year, lifetime cigarette abstainer) were ones. To keep variable coding consistent and to aid in logistic regression, these variables were recoded so that these responses were recorded as zeros. These variables can be identified by their letter B suffix. Regarding the latter, missing data for variables ALW_010 through ALW_040 was controlled for by creating a composite variable called Mean_DailyAlcohol which was the sum of the number of alcoholic drinks that were disclosed to have been consumed within the past week divided by the number of days in the past seven days where alcohol consumption totals were disclosed. This variable and its predecessors were included despite missing data being present because it was a concern that excluding potential participants for refusing to respond or not remembering how much alcohol they consumed on any day out of the past seven days might significantly diminish the sample size for the study. Respondents’ mean daily alcohol consumption was then categorized based on alcoholic drinking recommendations put forth by the CCSA. The recommendations put forth by the CCSA dictate that women and men should consume no more than two or three standard drinks per day and 10 or 15 standard drinks per week, respectively (Butt et al., 2011). The specific categories that were used were as follows: alcohol abstainers, who did not consume alcohol; light drinkers, who consumed less alcohol than the low-risk guidelines put forth by the CCSA; moderate drinkers, who consumed alcohol within the limits suggested by the CCSA; heavy drinkers, who consumed alcohol in excess of the CCSA’s advised limits but no more 38 than double these limits; and very heavy drinkers, who consumed alcohol in excess of double the CCSA’s advised limits (Butt et al., 2011). Mean daily alcohol consumption as recorded by the variable Mean_DailyAlcohol was categorized into a new variable called CCSA_Risk_Categories_Daily according to these guidelines. An alpha level of p < 0.05 was used to determine statistical significance. Upon review of the occupation data provided by variable LBFF14, it was concluded that it would be impossible to include cognitive reserve as a composite variable comprised of occupation and education data. Occupation was not recorded for respondents who were retired at the time the survey was conducted. As a result, it was not possible to include cognitive reserve status as a covariate. Education alone was used in its place. Additionally, it was decided that the variable ALC_010B (drank alcohol in the past 12 months) should be omitted due to its similarity with ALC_015 (frequency of alcohol consumption in the past 12 months). The final list of variables included in the analysis, including those taken verbatim from the database and those that were recoded or computed, is displayed below in Table 2. Table 2. Final list of variables included in the analysis with the number of dummy variables (k) indicated in parentheses. Variable Name Variable Description Dummy Coded? CCC_145B Has Alzheimer’s disease or any other No dementia ADM_PRXB Health component completed by proxy No DHH_AGE Age No ALC_005B Had a drink - lifetime No ALC_015 Drank alcohol - frequency - past 12 months Yes (k = 7) ALC_020 Drank 5 (male) / 4 (female) or more drinks - Yes (k = 6) frequency - past 2 months ALW_TOTAL Total weekly alcohol consumption with No missing values replaced with 0 ALW_DaysResponded Days in week where alcohol consumption No was reported 39 Mean_DailyAlcohol Mean daily alcohol consumption (total alcohol consumption divided the number of days that alcohol consumption was reported) CCSA_Risk_Categories CCSA drinking categorization according to _Daily Mean_DailyAlcohol Categorized_AgeB Categorized age DHH_SEX Sex DHH_AGE Age GEN_020B Perceived life stress SMKDVSTYB Smoking status DHG2DVR9 Highest level of education achieved No Yes (k = 4) Yes (k = 10) No No Yes (k = 4) Yes (k = 5) Yes (k = 8) Data Analysis Methodological approach. This study employed a quantitative approach, a research method designed to evaluate objective theories and hypotheses through the statistical evaluation of the relationships between numerical, measurable values (Cresswell, 2009). This systematic procedure was used to investigate the aforementioned research questions and hypotheses. Specifically, this quantitative research assumed a nonexperimental, case-control design. A case-control design is a commonly used observational epidemiological research approach using data that has been collected in the past (El-Masri, 2014). This design assesses the relationship between an exposure and an outcome of interest; specifically, individuals who have the outcome of interest are compared to those who do not have the outcome of interest with regards to their status on a number of exposure variables (El-Masri, 2014). The case-control approach was the most relevant design for this research as those with dementia or AD were compared to those without dementia or AD with respect to a number of potential dementia or AD risk factors. The fact that this study used retroactively collected CCHS data as well as that disease outcomes were already known eliminated many other research designs from contention (e.g., cohort study, randomized controlled trial). 40 Case-control studies are advantageous because they allow investigators to address multiple risk factors for a disease at one time, are more time efficient than other research approaches as the outcome of interest has already occurred, and are useful when studying uncommon diseases (Himmelfarb Health Sciences Library, 2011). Although dementia is not a rare disease, within the context of the CCHS data, there were proportionally few participants who had been diagnosed with AD or dementia. Specifically, a weighted sample of 147,911 participants were diagnosed with AD or dementia at the time the surveys were conducted out of the total weighted sample of 16,715,618 participants. Using a case-control design allows for inferences to be made about whether the predictor, or exposure, variables of interest make an individual more or less likely to have dementia. One significant disadvantage of using this approach is that those who have the outcome of interest are oftentimes more likely to exaggerate the influence of potential risk factors that they think may have contributed to the maladies that they are facing (Himmelfarb Health Sciences Library, 2011). However, this research used data from the CCHS, which is an extremely large survey containing an incredible number of variables having to do with multiple diseases and numerous risk factors. Because of this, it is less likely that participants were able to exaggerate the contributions of alcohol consumption to their present AD or dementia status as they were unaware when the data was collected that this research would address the alcohol/dementia relationship. Furthermore, this project was realized and conducted well after the data was collected. In conclusion, a quantitative, case-control study design was the most appropriate and relevant choice to address whether alcohol consumption can predict AD or dementia status in a community-dwelling Canadian population. Like with any research approach, there are 41 advantages and disadvantages to using this design; however, in this case it is believed that the advantages outweigh the disadvantages. Statistical methodology. This research employed a cross-sectional design to analyze data from the combined 2015 and 2016 cycles of the CCHS. As previously mentioned, respondents’ data was included if they were at least 41 years of age when the survey was conducted and did not have any missing data in their responses to the survey questions included as variables, with the exception of variables ALW_010 through ALW_040, as previously discussed. Two observational groups were created, distinguished by the respondents’ status regarding the outcome variable, AD or dementia status: one consisting of those who self- or proxy-reported as having AD or dementia and another made up of those who did not self- or proxy-report as having AD or dementia (i.e., the control group). Microdata files were requested from the Statistics Canada Research Data Centre Network for the combined 2015 and 2016 cycles of the CCHS. The statistical software IBM SPSS Statistics 24 was used to apply the inclusion and exclusion criteria and to prepare the data for analysis. Data analysis preparation, or cleaning and screening of the data, was conducted as outlined in Tabachnick and Fidell (2013) and included inspecting the data for accuracy of data input, univariate outliers, assessing normality of the distribution, and checking linearity and homoscedasticity. The data file was then merged with the bootstrapping and master weights files. The IBM SPSS data file was next exported into Stata/SE 15.1 (64 bit), hereafter referred to as Stata, so that subsequent data analysis could be conducted with bootstrapping weights applied. Bootstrapping procedures were employed so as to allow probability-based estimation of the population parameter from this survey data, estimating the sampling distribution and generating the resampled confidence intervals (Mooney, 2008). 42 Statistical analyses were carried out in Stata in three stages. First, the data were analyzed using multiple logistic regression with AD or dementia status representing the outcome variable and a number of alcohol consumption-related variables acting as predictors. Covariates were included as dictated by the review of the literature. Multiple logistic regression analysis was used to independently investigate two research questions as described below and according to the logistic regression modelling procedure outlined in Chapter 4 of Hosmer, Lemeshow, and Sturdivant (2013). Second, post-hoc tests were conducted to explore the differences between participants in the AD or dementia and control groups on each predictor variable (see Table 2 above) using bivariate logistic regression. A significance level of p < 0.05 was used for these analyses. The prevalence of AD or dementia in this population was also determined for the entire sample, and for men and women. Third, descriptive statistics in the form of cross-tabulations were generated to describe the sample. Bivariate logistic regressions between all variables and both the outcome variable of interest, AD or dementia status, as well as between all variables and the included and excluded samples were run. 1. What is the nature and shape of the relationship between alcohol consumption and the likelihood of dementia? Based on the review of the literature, it was hypothesized that the relationship between alcohol consumption and dementia would be positive or J-shaped, depending on the alcohol-related variable in question. For the hypothesis regarding lifetime alcohol consumption — a binary alcohol variable — it was assumed that the relationship between dementia odds and alcohol consumption would be positive, such that having consumed alcohol in the lifetime would yield an increased dementia risk than having not consumed alcohol in the lifetime. For the remaining hypotheses regarding frequency of alcohol 43 consumption in the past 12 months, binge drinking in the past 12 months, and CCSA drinking categorization based on mean daily alcohol consumption in the past week — categorical alcohol variables — J-shaped relationships were expected, such that those who drank moderate amounts of alcohol or consumed alcohol at a moderate frequency had a reduced dementia likelihood and those who abstained from alcohol consumption or drank heavily with regard to amount or frequency had a slightly elevated and markedly elevated dementia risk, respectively. Any variables that were revealed to share a statistically significant relationship with the outcome variable, dementia status, during the bivariate logistic regression analysis were included as covariates in this portion of the analysis. The sex variable was excluded at this stage of analysis. To investigate the relationship between various forms of alcohol consumption on dementia likelihood, this portion of analysis was broken down into four independent models, each with an alcohol consumption variable acting as a lone predictor. These models assessed the relationship between (a) lifetime alcohol consumption (variable ALC_005B), (b) alcohol consumption frequency within the past 12 months (variable ALC_015), (c) binge drinking frequency within the past 12 months (variable ALC_020), and (d) CCSA drinking risk classification within the past seven days (variable CCSA_Risk_Categories_Daily) and the likelihood of having AD or another dementia. This stage of analysis was segregated into four models due to concerns over collinearity between the alcohol consumption variables if they were all introduced into a single logistic regression model. Additionally, these four models will illuminate how alcohol consumption in three different time-related scopes, lifetime, past year, and past week, as well as differences between “normal” alcohol consumption and binge drinking affects dementia odds. Since the relationship between alcohol consumption and AD or dementia was anticipated to be curvilinear for categorical alcohol consumption variables 44 (ALC_015, ALC_020, and CCSA_Risk_Categories_Daily), dummy coding was used, allowing for the slopes to vary between categories (Szklo & Nieto, 2014). Dummy coding was also utilized for categorical covariates. The predictor variables and covariates that were dummy coded can be viewed above in Table 2. All variables were entered into their respective models at the same time. The four models were run with bootstrapping and master weights applied, model fit was assessed using " goodness of fit and link tests, and influential observation analysis was conducted to determine the presence of multivariate outliers in the models. A significance level of p < 0.05 was used to determine statistical significance. The final portion of data analysis in Stata involved producing descriptive statistics for the final sample. Cross-tabulations were produced between the AD or dementia status variable (CCC_145B) and all binary or categorical predictor variables. For the noncategorized age variable (DHH_AGE), mean age and mean age by dementia status were determined and differences between groups were assessed via regression. Categories in two variables, ALC_020 and CCSA_Risk_Categories_Daily, had to be collapsed due to low cell counts and to adhere to the data release requirements of the CCHS. This only applied to the descriptive statistics portion of the analysis. The bivariate logistic regression analyses had to be redone with these collapsed categories as the CCHS considers bivariate logistic regression to be akin to descriptive statistics. Finally, cross-tabulations and regressions were produced between the inclusion/exclusion variable Final_Sample and all variables in order to determine whether those who were included in the analysis differed significantly from those who were excluded. For the non-categorized age variable (DHH_AGE), mean age and mean age by inclusion/exclusion were determined and differences between groups were assessed via regression. A significance level of p < 0.05 was used to determine statistical significance. 45 2. Does the association between alcohol consumption and dementia odds vary according to sex? It was hypothesized that a sex effect would be present, such that the relationship between alcohol consumption and AD or dementia would be different for men and women, but it was unknown how the rates would differ between the sexes based on the review of the literature. As such, this analysis was exploratory in nature and the hypothesis was nondirectional. This portion of the analysis was identical to that previously described; however, the sex predictor variable was introduced into the multiple logistic regression equation. Interaction terms for the alcohol consumption variables ALC_005B, ALC_015, ALC_020, and CCSA_Risk_Categories_Daily and sex were introduced to each of the four alcohol consumption models — (a) lifetime alcohol consumption, (b) alcohol consumption frequency within the past 12 months, (c) binge drinking frequency within the past 12 months, and (d) CCSA drinking risk classification within the past seven days — to analyze this complex relationship. Bootstrapping and master weights were applied prior to running the models. Initially, only the interaction variables and the interaction term were included in the model. If this model was statistically significant, the model was run again including the significant covariates from the initial bivariate logistic regression analysis. Model fit was assessed using " goodness of fit and link tests, and influential observation analysis was conducted to identify multivariate outliers. A significance level of p < 0.05 was used to determine statistical significance. For each of the four models — (a) lifetime alcohol consumption, (b) alcohol consumption frequency within the past 12 months, (c) binge drinking frequency within the past 12 months, and (d) CCSA drinking risk classification within the past seven days — a 46 decision was made as to whether the final model should include the sex interaction term or not. This was based on whether the model including the interaction term was statistically significant. Suitability of Statistical Methodology Logistic regression is the best statistical tool to analyze this data as the outcome variable is binary (dementia/no dementia) and the predictor variables were both continuous and discrete (Tabachnick & Fidell, 2013). Furthermore, this technique is often used in doseresponse risk assessment, which is the ultimate goal of the present study, and the output of the analysis is easily interpretable as it can be readily converted into odds ratios (Szklo & Nieto, 2014). Possible alternative analytical approaches include discriminant function analysis, multiway frequency analysis, and multiple regression. These techniques are more powerful in some ways than logistic regression, but they have strict assumptions that must be met regarding the type of outcome and predictor variables included in the analysis. The data in this case is not appropriate for any of these alternative methods as each of them requires exclusively discrete or continuous outcomes and/or predictors and, as previously mentioned, this study included combination of discrete and continuous predictors (Tabachnick & Fidell, 2013). Logistic regression is generally viewed as a large sample technique. Problems can occur if the ratio of cases to predictor variables is too small, such as inflated parameter estimates and standard errors (Tabachnick & Fidell, 2013). The issue of sample size selection for this statistical technique has been a well-debated topic in the literature. Peduzzi, Concato, Kemper, Holford, and Feinstein (1996) suggest a minimum of 10 cases (or events) per predictor variable in logistic regression. Using this rule and the fact that this study included 12 independent variables (or predictors), the minimum sample size would be 120 47 dementia/AD cases. The unweighted rounded counts of survey respondents with dementia/AD in the combined 2015/16 cycles of the CCHS exceeded the minimum sample size recommendation by over five times, leading to the conclusion that logistic regression is an acceptable analytical choice in this case. Furthermore, logistic regression was utilized by Nabalamba and Patten (2010) on a weighted sample of 79,971 people with dementia/AD, which is slightly over half of the present weighted sample size for the combined 2015/16 CCHS cycles at 147,911. Nabalamba and Patten (2010) were thus able to successfully utilize this technique with fewer weighted participants and more independent/predictor variables, further supporting the use of this technique in this case. Odds Ratio, Relative Risk, and Bias. As the present study was a case-control study that utilized multivariate logistic regression, the measure of association between exposure and outcome was the odds ratio (OR). As previously mentioned, one of the primary advantages to logistic regression is that the output of this methodology, the beta coefficient, can be readily converted into easily-interpretable ORs. Mathematically, the OR is the ratio of the probability of developing the disease in exposed individuals divided by the ratio of the probability of developing the disease in nonexposed individuals, and is thus a ratio of ratios (Szklo & Nieto, 2014). ORs are often used to approximate another measure of association, the relative risk (RR), which is the ratio of the incidence or risk in those exposed to the disease divided by the same ratio is unexposed individuals (Szklo & Nieto, 2014). RRs are often expressed as proportions or percentages (Davies, Crombie, & Tavakoli, 1998). The OR is equivalent to the RR plus built in bias (Szklo & Nieto, 2014). If the bias term is close to one, then the OR is an appropriate approximation of the true RR. However, if the bias deviates from one, the result is an inflated OR that is biased away from the null hypothesis, amplifying the perceived association between exposure and outcome (Szklo & Nieto, 2014). 48 When approximating the RR from the OR, the goal is to minimize the bias term. This can be done by adhering to the rarity assumption, ensuring that the true risk does not dramatically differ between the comparison groups, or maximizing the sample size. First, the rarity assumption assumes that the disease in question has a low incidence, resulting in a bias term that is approximately equal to one and, therefore, negligible (Szklo & Nieto, 2014). If this assumption is upheld, the OR is a good estimator of the RR; however, if the assumption is not upheld, as is the case with very common conditions, the built-in bias term may be inflated and the resultant OR may dramatically overestimate the true RR (Szklo & Nieto, 2014). Second, bias can be increased when the risk of the event in question differs significantly between the comparison groups. When risks in both groups being compared are low (around 20%), the odds will acceptably approximate the risks and the OR will be an acceptable estimator of the RR (Davies et al., 1998). However, if the risk in either group rises above the threshold of 20%, the bias is inflated and the OR is an increasingly poor predictor of the RR (Davies et al., 1998). Third, bias can be minimized by ensuring a large sample size. Nemes, Jonasson, Genell, and Steineck (2009) conducted a study to determine how sample size influenced the bias present in ORs. The authors found that as the sample size, irrespective of ratio of cases to controls, increased, the bias approached zero (Nemes et al., 2009). Bias correction measures — such as the bias corrected estimate, jack-knife, and bootstrapping — may be employed during data analysis to adjust for small sample size and reduce bias (Nemes et al., 2009). Although ORs are a credible measure of association and more readily interpretable than beta coefficients, RRs are often sought as the most easily interpretable form of statistical output (Davies et al., 1998). ORs can be used to approximate the true RR if the built-in bias term is small, thus the rarity assumption is upheld, the risk does not significantly differ 49 between groups, and the sample size is large. In the present study, the ORs resulting from logistic regression analysis will not be interpreted as RRs. Data Source and Analysis: Limitations and Justifications Two obvious limitations of this research were the cross-sectional nature of the study and the necessary use of self-reported data for AD or dementia diagnosis and alcohol consumption. Cross sectional studies, by definition, are unable to determine causation. This study attempted to determine whether lifelong alcohol consumption was a risk factor for AD or dementia in old age (i.e., whether alcohol consumption causes AD or dementia). As the alcohol consumption data was collected at the same time as that for AD or dementia status, it is important to relate present alcohol consumption to that in the past. Shaw, Krause, Liang, and McGeever (2011) found that alcohol consumption tends to decline with increasing age with the exception of those who are heavy drinkers. Heavy drinkers’ alcohol consumption tendencies remained stable throughout the aging process (Shaw et al., 2011). Assuming that this relationship is true of the present study’s population, there is justification for the assumption that self-reported alcohol consumption was lesser or equal to that in the past. Alcohol consumption for heavy drinkers at present time should have been an accurate representation of alcohol consumption in the past and that for those who were not heavy drinkers should be underestimated. Assuming that the decline in alcohol consumption with age is proportional (i.e., those who were moderate drinkers are light drinkers presently and those who were light drinkers are now alcohol abstainers), it may be possible to differentiate effects between groups at the lower end of the alcohol consumption spectrum. Thus, despite this study being cross-sectional in nature, it may still be possible to extrapolate a doseresponse effect from the data. 50 Using self-reported disease status for a disease such as dementia is clearly problematic, as those who have the disease may not have been able to accurately classify themselves as having or not having the disease. This population may additionally have misclassified themselves with regards to the independent variables and covariates included in this study. In the present sample from the combined 2015/16 cycles of the CCHS, 59.97% and 2.69% of those with and without dementia, respectively, used proxies to answer the survey questions. The high rate of proxy usage in those with dementia has the potential to reduce the chance of misclassification and bias in the study. A study by Farias, Mungas, and Jagust (2005) found that those in the earlier stages of dementia tend to more accurately selfreport their symptoms than informants. The inverse was true for those with dementia — dementia patients tended to underestimate their symptoms and informants gave a more accurate account of information regarding the severity of symptoms (Farias et al., 2005). The results of this study suggest that both self-reported dementia status for those with less severe dementia and proxy-reported dementia status for those in more advanced stages of the disease were likely to be valid. For those who did not use proxies, there is still a chance of misclassification if they are in moderate or advanced stages of the disease. According to Nabalamba and Patten (2010) who published a study using self-reported dementia status from cycle 3.1 of the CCHS, there is no evidence that self-reported dementia is a valid method of classifying those with dementia. Unfortunately, as far as I am aware, there has been no published research comparing the validity of self-reported dementia to health professional-diagnosed dementia. However, the previously mentioned journal article by Farias et al. (2005) provides evidence that, as long as those who had more severe dementia used proxies to respond to the questionnaire, the validity of this study should not have been compromised. 51 The validity of self-reported alcohol consumption data may also be questionable, particularly for participants who suffer from alcoholism or engage in binge drinking, as they may not want to admit to problematic drinking. The survey questions asked by the combined 2015/16 cycles of the CCHS can be classified as graduated-frequency and short-term recall assessments of alcohol consumption (McKenna, Treanor, O’Reilly, & Donnelly, 2018). McKenna et al. (2018) found that graduated-frequency measures of alcohol consumption in surveys (i.e., those presenting pre-determined categories of alcohol consumption) were the most valid and reliable form of assessing alcohol drinking compared to short term recall and quantity-frequency measures. Short-term recall survey structure achieved good or mixed reliability and validity, as opposed to the quantity-frequency method, which did not score as highly. This study provides evidence that, despite the combined 2015/16 cycles of the CCHS relying on self-reported data, the format of the questions in the survey were adequately valid and reliable measures of alcohol consumption. Despite there being significant limitations to using a cross-sectional study design with self-reported data, I believe that its use in this case was adequately justified by the review of the literature. The results of this research will be valuable to inform further investigations of the alcohol consumption and dementia/AD relationship in a community-dwelling Canadian population. 52 CHAPTER 4: RESULTS Prevalence of AD and Dementia The total and by sex weighted estimates and proportions of community-dwelling Canadians aged 41 and older who self- or proxy-identified as having Alzheimer’s disease (AD) or another form of dementia in the combined 2015 and 2016 cycles of the Canadian Community Health Survey (CCHS) are displayed below in Table 3. Table 3. Prevalence of Alzheimer’s Disease (AD) and other dementias by sex in the communitydwelling Canadian population aged 41 and older in 2015 and 2016. Included are the weighted estimates of the number of individuals with AD/dementia, the observed proportions of those with AD/dementia, and the 95% confidence intervals (CI) of the observed proportion for women, men, and both sexes. Weighted Observed 95% CI of Observed Proportion Estimate Proportion (%) Women 70,352 0.817 0.658 0.975 Men 77,559 0.957 0.765 1.149 Both Sexes 147,911 0.885 0.760 1.010 Approximately 148,000 individuals indicated that they had AD or another form of dementia. When the sexes were considered separately, a higher proportion of men had positively selfidentified as having AD or another dementia than women. When proportions of those with AD or another form of dementia were segregated into five-year age range cohorts (Figure 1), the general trend was proportional such that increasing age led to a higher prevalence of having AD or dementia. 26 24 22 20 18 16 14 12 10 8 6 4 2 91+ 86.0 to 90.9 81.0 to 85.9 76.0 to 80.9 71.0 to 75.9 66.0 to 70.9 61.0 to 65.9 56.0 to 60.9 51.0 to 55.9 46.0 to 50.9 0 41.0 to 45.9 Prevalence of Alzheimer's Disease or other dementias (%) 53 Age (years) Figure 1. Prevalence of Alzheimer’s disease (AD) and other dementias expressed as a percentage by age categorized in five-year increments for community-dwelling Canadian adults aged 41 years and older surveyed in the combined 2015 and 2016 cycles of the Canadian Community Health Survey. However, there were some discrepancies from the general positive trend between age and prevalence of AD and dementia. The prevalence increased steadily for the first three age cohorts, after which the prevalence decreased by over 3% for the following two cohorts. From age cohort 66-70.9 years to 81-85.9 years, the trend was once again positive and increased by approximately 15%. A decrease in the prevalence of AD or dementia was identified for the final two age cohorts, with a dramatic depreciation for the 91 and older age cohort, where the prevalence of AD or dementia was comparable to those 25 years their junior. 54 Descriptive Statistics Statistics describing the weighted study population in the form of two by two contingency tables were generated for every variable against AD or dementia status. The results are displayed below in Table 4. Table 4. Proxy use, alcohol consumption, demographics, and dementia risk factors among people with and without Alzheimer’s Disease (AD) or other dementias in the community-dwelling Canadian population aged 41 years and older in 2015 and 2016. Has AD/Dementia Does Not Have AD/Dementia Proportion 95% CI Proportion 95% CI (%) (%) Alcohol Use Drank in lifetime Yes 88.49 83.80 91.95 93.33 92.93 93.72 No 11.51 8.05 16.20 6.67 6.28 7.07 Drank in past year Yes 54.07 47.31 60.68 78.24 77.69 78.78 No 45.93 39.32 52.69 21.76 21.22 22.31 Frequency of drinking in past year No alcohol 45.93 39.32 52.69 21.76 21.22 22.31 Less than once 13.76 10.51 17.81 15.97 15.48 16.46 per month Once per month 8.12 3.92 16.09 7.39 7.07 7.73 2-3 times per 6.67 4.19 10.44 9.27 8.88 9.67 month Once per week 7.27 4.84 11.59 12.02 11.60 12.46 2-3 times per 6.74 4.55 9.88 17.71 17.21 18.22 week 4-6 times per 2.65 1.11 6.16 6.88 6.56 7.21 week Every day 8.86 5.97 12.95 9.00 8.66 9.35 Binge drinking frequency in past year No alcohol 45.93 39.32 52.69 21.76 21.22 22.31 Never 44.44 38.00 51.08 43.36 42.73 43.99 Less than once 5.61 3.31 9.37 19.31 18.81 19.82 per month 1-3 times per 2.36 0.94 5.80 9.49 9.10 9.89 month 55 Once per week More than once per week Drinking frequency in past week* Alcohol abstainer (♂ and ♀: 0 drinks/day) Light drinker (♂: 1-2 drinks/day, ♀: 1 drink/day) Moderate drinker (♂: 3 drinks/day, ♀: 2 drinks/day) Heavy and very heavy drinker (♂: 4+ drinks/day, ♀: 3+ drinks/day) Demographics Proxy use Yes No Sex Female Male Risk factors Perceived life stress Not at all stressful Not very stressful A bit stressful Quite a bit stressful Extremely stressful Smoking status Lifetime abstainer (never smoked whole cigarette) Experimental smoker (at least 1 cigarette, current non-smoker) Former occasional 1.08 0.57 0.32 0.27 3.59 1.19 3.47 2.61 3.24 2.42 3.72 2.81 69.46 62.33 75.76 49.06 48.40 49.72 28.82 22.57 36.00 47.30 46.65 47.95 0.74 0.31 1.72 2.05 1.89 2.21 0.99 0.37 2.58 1.59 1.44 1.77 59.97 40.03 52.94 33.39 66.61 47.06 2.69 97.31 2.44 97.04 2.96 97.56 47.56 52.44 40.64 45.42 54.58 59.36 51.57 48.43 51.30 48.16 51.84 48.70 25.12 19.97 31.08 16.24 15.79 16.70 19.22 30.65 20.70 14.79 24.85 15.18 24.59 37.15 27.58 23.67 38.66 18.03 23.15 38.00 17.52 24.19 39.32 18.55 4.31 2.45 7.46 3.41 3.18 3.65 40.47 33.81 47.49 35.74 35.09 36.40 10.29 6.53 15.96 13.05 12.63 13.48 4.61 2.46 8.46 3.78 3.53 4.04 56 smoker (current non-smoker) Former daily 35.17 29.12 41.74 30.82 30.24 31.41 smoker (current non-smoker) Current 0.68 0.23 1.98 3.38 3.15 3.62 occasional smoker Current daily 8.78 5.73 13.23 13.23 12.81 13.66 smoker Highest level of attained education Grade 8 or lower 18.90 13.37 26.02 5.66 5.38 5.95 Grade 9-10 7.15 5.08 9.98 6.10 5.82 6.93 Grade 11-13 4.62 2.55 8.23 3.50 3.27 3.75 Secondary school 22.56 17.25 28.93 21.99 21.47 22.52 graduation, no post-secondary Trade certificate 9.28 6.66 12.77 9.85 9.51 10.21 or diploma College 13.19 9.88 17.41 22.90 22.33 23.48 certificate or diploma (nontrades) University 2.26 1.16 4.34 3.71 3.48 3.96 certificate or diploma below bachelor’s level Bachelor’s 15.25 9.76 23.05 17.32 16.80 17.85 degree Certificate, 6.79 4.34 10.48 8.97 8.58 9.37 diploma, or degree above bachelor’s level *Note. Drinking frequency in the past week was determined using Canadian Centre on Substance Use and Addiction (CCSA) drinking risk categorization by mean daily alcohol consumption in the past week. The following trends were identified for the five alcohol consumption variables. Regarding lifetime alcohol consumption, participants were more likely to have consumed an alcohol beverage in their lifetime regardless of their AD or dementia status. However, approximately 5% more of the population who had never consumed a drink in their lifetime had AD or 57 dementia than did not have AD or dementia. Concerning alcohol consumption in the past year, more people drank who did not have AD or dementia than who did have AD or dementia. The ratio of those who drank to those who did not drink in the past year was greater for those without AD or dementia (i.e., the number of those who did drink was much higher than those who did not drink), while the ratio for those who did have AD or dementia was more or less equal (i.e., the number of those who did drink was only slightly higher than those who did not drink). In relation to the frequency of alcohol consumption over the past year, the most drastic difference between those who did and did not have AD or dementia was that over twice as many people with AD or dementia had not drank at all on the past 12 months. For those who had consumed alcohol in the past year either infrequently (less than once per month once per month, or two to 3 times per month) or very frequently (every day), proportions of those with or without AD or dementia were similar; however, the proportion was always higher for those without AD or dementia. Group differences in proportions were more significant in the middle alcohol consumption frequency groups, albeit once again always higher for those without AD or dementia: almost twice as many people drank once per week who did not have AD than who did, almost three times as many people consumed alcohol two to three times per week who did not have AD than who did, and almost three times as many people drank alcohol four to six times per week who did not have AD than who did. In relation to binge drinking frequency in the past year, once again, there were over twice as many people who did not consume alcohol at all in the past year who had AD or dementia when compared to those who did not have these diseases. Proportions of those who had consumed alcohol in the past year but had not engaged in binge drinking behaviour were approximately equal and greater than 40% for both those who did and did not have AD or dementia, although the proportion was slightly higher for those with AD or dementia. For 58 those who had engaged in binge drinking, proportions were between 3 to 4.5 times greater for those who did not have AD or dementia than did. Finally, with reference to Canadian Centre on Substance Use and Addiction (CCSA) drinking risk classification over the past week, once again there were more individuals (approximately 1.5 times more) with AD or dementia who abstained from consuming alcohol in the last seven days. More people who did not have AD or dementia were classified as light, moderate, and heavy/very heavy drinkers than those who had AD or dementia. Specifically, there were 1.6 times more light drinkers, 2.8 times more moderate drinkers, and 1.6 times more heavy and very heavy drinkers without than with AD or dementia. Demographic information for the weighted study population will be described hereafter. The mean age for the entire sample was 59.04 years (95% CI: 58.95, 59.14). When separated by AD or dementia status, the mean age for those without AD or dementia was 58.91 years (95% CI: 58.81, 59.01) and that for those with AD or dementia was 74.55 years (95% CI: 72.67, 76.43). The difference in the uncategorized mean ages for those with and without AD or dementia was statistically significant, p = 0.000. The proportions of men and women who had dementia in the sample were approximately equal; however, the proportion was 5% higher for men than women. There were 22.3 times more people with AD or dementia who used proxies to respond to the survey than who used proxies but did not have AD or dementia. Just over 40% of those with AD or dementia responded to the survey independently (i.e., without using a proxy) compared to over 97% of those who neither had AD or dementia nor used a proxy to respond to the survey. Finally, trends identified in the risk factor variables will be discussed. Regarding selfreports of life-related distress, there were greater proportions of those with AD or dementia who reported that they found their life not at all stressful, quite a bit stressful, or extremely 59 stressful than those without AD or dementia. There were more people without AD or dementia who stated that their life was not very or a bit stressful than those without AD or dementia. The most drastic difference between the AD/dementia and no AD/dementia groups was seen in those who reported their life to be not at all stressful, where one quarter of those with AD or dementia stated that they experienced very little stress. With respect to smoking status, there was a greater proportion of those without AD or dementia who reported being experimental smokers and current occasional or daily smokers. There was a greater proportion of those with AD or another form of dementia who reported being lifetime smoking abstainers and former occasional or former daily smokers. The greatest difference was observed in those who reported they were current occasional smokers, where there were almost five times as many individuals without AD or dementia identifying as this type of smoker than those with AD or dementia. Referencing education as a risk factor for dementia, there was a greater proportion of individuals with AD or dementia who had a highest level of attained education between below grade 8 to grade 12 or 13 (depending on province of residence) than those without AD or dementia. This relationship was inversed for those with some form of post-secondary education, whether trade school to education beyond a Bachelor’s degree, where those without AD or dementia were more prominent. The most significant difference was observed in those with a grade 8 or lower level of education, where there were over three times as many people with AD or dementia than without. Multivariate Models Four models were assessed for the presence of a sex effect in order to determine which model to select as the final model. Each addressed a different frequency or timeline of alcohol consumption: Model A assessed lifetime alcohol consumption, Model B looked at alcohol consumption frequency in the past year, Model C addressed binge drinking frequency 60 over the past year, and Model D detailed risky drinking as classified by the CCSA safe drinking guidelines over the past 7 days. The model selection process and final models will be discussed hereafter. Model A: Lifetime alcohol consumption. Model A was run a total of three times: once including the main predictor variable — had a drink in lifetime — and all covariates, yet excluding sex; once including just the main predictor variable, sex, and drinking in the past year/sex interaction; and once including the main predictor variable, all covariates, sex, and drinking in the past year/sex interaction. The first iteration of this model — which excluded the drinking in the past year and sex interaction term — was selected as the final model because the interaction term was not statistically significant at the p < 0.05 level for either of the models including the interaction term. The results of the final model assessment are displayed below in Table 5. Table 5. Model A: Observed odds ratios (ORs) relating lifetime alcohol consumption, interview proxy use, smoking status, life stress, highest level of attained education, and age to Alzheimer’s disease (AD) or other dementias in the community-dwelling Canadian population aged 41 years and older in 2015 and 2016. 95% confidence intervals (CIs) and p-value data is also presented. Statistical significance at the p < 0.05 level is indicated by an asterisk (*). Correlates of AD and other OR 95% CI p-value dementias Lifetime alcohol consumption No† 1.00 … … Yes 1.04 0.64 1.69 0.866 Proxy use No† 1.00 … … Yes 26.32 19.50 35.51 0.000* Age 41 to 45.9 years† 1.00 … … 46 to 50.9 years 3.04 0.73 12.74 0.128 51 to 55.9 years 7.45 2.08 26.71 0.002* 56 to 60.9 years 4.92 1.14 21.24 0.033* 61 to 65.9 years 5.49 1.49 20.27 0.011* 61 66 to 70.9 years 71 to 75.9 years 76 to 80.9 years 81 to 85.9 years 86 to 90.9 years 91+ years Perceived life stress Not at all stressful† Not very stressful A bit stressful Quite a bit stressful Extremely stressful Smoking status Lifetime abstainer (never smoked whole cigarette) † Experimental smoker (at least 1 cigarette, current non-smoker) Former occasional smoker (current nonsmoker) Former daily smoker (current non-smoker) Current occasional smoker Current daily smoker Highest level of attained education Grade 8 or lower † Grade 9-10 Grade 11-13 Secondary school graduation, no postsecondary Trade certificate or diploma College certificate or diploma (non-trades) University certificate or diploma below bachelor’s level Bachelor’s degree Certificate, diploma, or degree above bachelor’s level Note. † Reference category. 8.40 13.86 17.09 40.80 43.52 20.58 2.43 4.08 5.02 12.32 12.46 5.13 29.08 47.05 58.21 135.11 152.00 82.52 0.001* 0.000* 0.000* 0.000* 0.000* 0.000* 1.00 0.77 1.06 1.40 1.04 … 0.51 0.72 0.86 0.56 … 1.16 1.56 2.30 1.93 0.205 0.772 0.179 0.913 1.00 … … 1.07 0.65 1.77 0.781 1.34 0.58 3.06 0.493 1.15 0.81 1.63 0.434 0.37 0.09 1.55 0.174 1.44 0.85 2.44 0.177 1.00 0.67 1.42 1.12 … 0.38 0.64 0.66 … 1.18 3.17 1.88 0.165 0.391 0.677 1.13 0.66 1.93 0.655 0.90 0.52 1.56 0.719 0.84 0.36 1.98 0.694 1.26 1.25 0.67 0.64 2.37 2.44 0.479 0.511 62 Overall, the model, which was bootstrap replicated 991 of the desired 1000 replications, was significant with Wald " (29, n = 16,715,618) = 995.83 and p = 0.000. In this model, only proxy use and nine of the ten age categories were statistically significant at predicting outcome membership when all other variables were held at their reference values. The primary variable of interest, lifetime alcohol consumption, was not an accurate predictor of AD or dementia group membership at the p < 0.05 level. With regards to proxy use, those who used a proxy to respond to the survey were over 26 times more likely to have AD or another form of dementia than those who did not use a proxy to respond to the survey when all other covariates were held constant. With reference to categorized age, the following statistically significant results were found. All age categories, with the exception of those aged 46 to 50.9 years, significantly increased the odds of having AD or another form of dementia when compared to the reference group consisting of those aged 41 to 45.9 years of age. Those aged 51 years or older had roughly a 4 to 43 times the odds of having AD or another form of dementia in comparison to the reference group. Generally, the odds of having AD or dementia increased with increasing age with the exception of those aged 91 years or older. This age group saw a dramatic reduction in AD or dementia odds when compared to the preceding age category. The fit of the model was assessed using " goodness of fit and link tests. Results of the former indicated that the observed sample distribution of the model was not significantly different from the distribution that was expected, " (9, n = 16,715,618) = 9.52, p = 0.390. The latter test indicated that link, or misspecification, error was present in the model, Bhat = 1.35, phat = 0.000, Bhatsq= 0.05, phatsq = 0.015 — i.e., the squared dependent variable had explanatory power. 63 Finally, influential observation analysis was conducted on the model. Multivariate outliers were identified ( $%&'()*+ = 567) and the model was run again excluding these individuals. The results were assessed for changes in the sign or direction of the odds ratios or changes in significance. The p value for current occasional smokers changed from being nonsignificant (p = 0.174) to significant (p = 0.002); however, this singular change was not enough to warrant excluding these 567 individuals from the final model and the original model was accepted. Model B: Alcohol consumption frequency in the past 12 months. Model B was run a total of three times: once including drinking frequency in the past year and all covariates yet excluding sex and drinking frequency in the past year/sex interaction variables; once only including alcohol consumption frequency in the past year, sex, and drinking frequency in the past year/sex interaction variables; and once including the primary predictor variable alcohol consumption frequency in the past year, all covariates, and the drinking frequency in the past year/sex interaction term. The third iteration of this model was selected as the final model as the second model containing only the alcohol consumption frequency in the past year, sex, and drinking in the past year/sex interaction variables indicated a statistically significant sex effect. Once the covariates were added back into the model in the third iteration, the statistically significant sex effect was still present. The results of the final model are display below in Table 6. 64 Table 6. Model B: Observed odds ratios (ORs) relating alcohol consumption frequency in the past 12 months, interview proxy use, smoking status, life stress, highest level of attained education, age, and sex to Alzheimer’s disease (AD) or other dementias in the communitydwelling Canadian population aged 41 years and older in 2015 and 2016. 95% confidence intervals (CIs) and p-value data is also presented. Statistical significance at the p < 0.05 level is indicated by an asterisk (*). Correlates of AD and other OR 95% CI p-value dementias Alcohol consumption frequency in the past 12 months No alcohol in past 12 1.00 … … months† Less than once per 0.91 0.51 1.62 0.746 month Once per month 1.27 0.27 5.91 0.762 2-3 times per month 0.91 0.40 2.07 0.822 Once per week 1.00 0.35 2.83 0.999 2-3 times per week 0.56 0.29 1.07 0.080 4-6 times per week 0.89 0.30 2.68 0.835 Every day 0.69 0.37 1.29 0.247 Proxy use No† 1.00 … … Yes 24.58 17.80 33.95 0.000* Age 41 to 45.9 years† 1.00 … … 46 to 50.9 years 2.99 0.71 12.69 0.137 51 to 55.9 years 7.72 2.14 27.81 0.002* 56 to 60.9 years 4.87 1.11 21.38 0.036* 61 to 65.9 years 5.43 1.47 20.01 0.011* 66 to 70.9 years 8.42 2.40 29.48 0.001* 71 to 75.9 years 13.51 3.96 46.10 0.000* 76 to 80.9 years 16.76 4.89 57.41 0.000* 81 to 85.9 years 38.58 11.55 128.86 0.000* 86 to 90.9 years 42.41 12.01 148.54 0.000* 91+ years 19.24 4.69 78.93 0.000* Perceived life stress Not at all stressful† 1.00 … … Not very stressful 0.79 0.52 1.20 0.269 A bit stressful 1.04 0.71 1.52 0.830 Quite a bit stressful 1.36 0.83 2.22 0.219 Extremely stressful 1.00 0.53 1.88 0.992 Smoking status 65 Lifetime abstainer (never smoked whole cigarette) † Experimental smoker (at least 1 cigarette, current non-smoker) Former occasional smoker (current nonsmoker) Former daily smoker (current non-smoker) Current occasional smoker Current daily smoker Highest level of attained education Grade 8 or lower † Grade 9-10 Grade 11-13 Secondary school graduation, no postsecondary Trade certificate or diploma College certificate or diploma (non-trades) University certificate or diploma below bachelor’s level Bachelor’s degree Certificate, diploma, or degree above bachelor’s level Sex Male† Female Interaction Term drinking frequency in past year x sex No alcohol in the past 12 months x female† Less than once per month x female Once per month x female 2-3 times per month x female Once per week x female 1.00 … … 1.18 0.72 1.93 0.522 1.42 0.61 3.29 0.412 1.24 0.87 1.77 0.225 0.37 0.09 1.57 0.177 1.48 0.87 2.52 0.146 1.00 0.66 1.41 1.16 … 0.37 0.63 0.68 … 1.16 3.14 1.98 0.144 0.401 0.586 1.17 0.68 2.00 0.571 0.95 0.55 1.63 0.848 0.90 0.38 2.16 0.818 1.31 1.31 0.69 0.67 2.49 2.57 0.415 0.427 1.00 1.17 … 0.73 … 1.87 0.504 1.00 … … … 0.53 0.24 1.17 0.117 0.40 1.07 0.08 0.32 2.09 3.52 0.278 0.912 0.65 0.19 2.21 0.488 66 2-3 times per week x female 4-6 times per week x female Every day x female Note. † Reference category. 0.89 0.33 2.43 0.821 0.14 0.03 0.73 0.019* 0.80 0.33 1.90 0.608 Overall, the model, which was bootstrap replicated 976 out of the desired 1000 times, was statistically significant with Wald " (43, n = 16,715,618) = 1058.95 and p = 0.000. The main predictor variable of interest in this model, alcohol consumption frequency in the past year, was not a good predictor of AD or dementia group membership at the p = 0.05 level for any of the variable categories; however, that for drinking two to three times per week approached statistical significance. Despite not being statistically significant, the distribution of the alcohol consumption frequency in the past year and odds ratio for AD or dementia status relationship is shown below in Figure 2. Alcohol consumption frequency in the past 12 months 67 No alcohol† Less than once per month Once per month 2-3 times per month Once per week 2-3 times per week 4-6 times per week Every day 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 Observed OR Figure 2. Model B: Observed odds ratios (ORs) demonstrating the likelihood of having Alzheimer’s disease (AD) or other dementias associated with various frequencies of alcohol consumption over the past 12 months in the community-dwelling Canadian population aged 41 and older in 2015 and 2016. The dashed line indicates an OR of 1, or no association between the exposure (frequency of alcohol consumption) and outcome (AD or other dementias). Statistical significance at the p < 0.05 level, if present, is indicated by an asterisk (*). The reference category is marked as †. The general trend observed was that of an inverted parabola, such that AD or dementia risk increased from no alcohol in the past year to consuming alcohol once per month, and then generally decreased for those who drank multiple days out of the week or every day. However, it was indicated that alcohol consumption was protective or had no effect on AD or dementia risk with the exception of those who drank once per month, who had a greater odds of having AD or dementia. The only components of this model that were statistically significant at predicting AD or dementia group membership were proxy use, nine of the ten age categories, and the interaction between drinking alcohol four to six times per week and sex. Those who used 68 proxies were over 24 times more likely to have AD or another form of dementia than those who did not use proxies when all other covariates were held constant. All age categories, with the exception of those aged 46 to 50.9 years, significantly increased the odds of having AD or another form of dementia when compared to the reference group, which was composed of those aged 41 to 45.9 years. Those aged 51 years of age and older were between approximately 5 and 42 times more likely to have AD or dementia than those in the reference group when all other covariates were held constant. Generally, the trend was that AD or dementia risk increased with increasing age; however, those in the highest age category, 91 years of age or older, had less than half the odds of having AD or dementia than those in the preceding age category. The statistically significant interaction for sex and drinking alcohol four to six times per week was further analyzed to determine the nature of the interaction. Odds ratios for those who drank four to six times per week, referenced against alcohol abstainers, were calculated for both men and women (calculations can be found in Appendix A). The odds ratio for men who drank four to six times per week compared to alcohol abstainers was calculated to be 0.89 and that for women was found to be 0.13. The odds ratio for women was further away from 1 (i.e., no effect) for women than for men. Proportions were also generated for both men and women for drinking alcohol four to six times per week over the past year and AD or dementia status variables to visualize the interaction between sex and alcohol consumption. These proportions were then converted into bar graphs and are displayed below in Figure 3. 69 Percentage of men with Alxheimer's disease or other dementias (%) a) 0.5 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 b) 0.5 Percentage of women with Alzheimer's disease or other dementias (%) 0 0.45 No alcohol† 4-6 times per week No alcohol† 4-6 times per week 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 Figure 3. Model B: The interaction between consuming alcohol four to six times per week over the past 12 months versus the reference group of no alcohol in the past 12 months and the percentage of those with Alzheimer’s disease (AD) and other dementias for (a) men and (b) women in the community-dwelling Canadian population aged 41 years and older in 2015 and 2016. The reference category is marked as †. The dashed line shows how the trends vary between the sexes. The slope of the trend line connecting the those who abstained from alcohol in the past year and those who drank four to six times per week in the past year was more negative for 70 women than men. There were more women who had AD or dementia who were alcohol abstainers than who drank four to six times per week over the past 12 months. The trend was also the same for men. There were more men with AD or dementia who were alcohol abstainers than who drank four to six times per week over the past 12 months. However, when the proportions of those with AD or dementia who drank four to six times per week were compared between men and women, there were slightly more men (0.044%) in this category than women (0.0042%). The fit of the model was assessed using " goodness of fit and link tests. Results of the former indicated that the observed sample distribution of the model was not significantly different from the distribution that was expected, " (9, n = 16,715,618) = 6.27, p = 0.713. The latter test indicated that link, or misspecification, error was not present in the model, Bhat = 1.19, phat = 0.000, Bhatsq= 0.03, phatsq = 0.144 — i.e., the squared dependent variable did not have explanatory power. Finally, influential observation analysis was conducted. Multivariate outliers were identified ( $%&'()*+ = 827) and the model was run again excluding these individuals. The results were assessed for changes in the sign or direction of the odds ratios or changes in significance. No changes were observed; therefore, it was concluded that the multivariate outliers did not influence the outcome significantly and the original model was retained. Model C: Binge drinking frequency in the past 12 months. Three iterations of Model C were run: one including binge drinking frequency in the past 12 months and all covariates, yet excluding sex and binge drinking/sex interaction variables; once including only binge drinking frequency in the past 12 months, sex, and binge drinking/sex interaction variables; and once including all covariates and interaction terms. All model repetitions were 71 statistically significant at a p value of 0.000; therefore, the decision regarding whether or not to include the interaction term in the final model was based on statistical significance of the interactions. The second time that the model was run, including only the interaction variables and interaction term, two interaction permutations were significant: never binge drinking and female (ORobs = 0.51, p = .016) and binge drinking two to three times per month and female (ORobs = 0.15, p = .047). Due to the significant interactions, the third iteration of the model was selected as the final model. In the final model, the interactions were no longer significant, implying that the addition of the covariates to the model explained some of the relationship between the interaction term and the outcome. The results of the final model for binge drinking frequency in the past 12 months are displayed below in Table 7. Table 7. Model C: Observed odds ratios (ORs) relating binge drinking frequency in the past 12 months, interview proxy use, smoking status, life stress, highest level of attained education, age, and sex to Alzheimer’s disease (AD) or other dementias in the community-dwelling Canadian population aged 41 years and older in 2015 and 2016. 95% confidence intervals (CIs) and p-value data is also presented. Statistical significance at the p < 0.05 level is indicated by an asterisk (*). Correlates of AD and other OR 95% CI p-value dementias Binge drinking frequency in the past 12 months No alcohol in past 12 1.00 … … months† Never 0.98 0.61 1.59 0.943 Less than once per 0.49 0.22 1.10 0.082 month Once per month 0.96 0.20 4.61 0.961 2-3 times per month 0.19 0.05 0.73 0.015* Once per week 0.44 0.07 2.69 0.375 More than once per week 0.16 0.04 0.61 0.007* Proxy use No† 1.00 … … Yes 24.30 17.76 33.25 0.000* Age 41 to 45.9 years† 1.00 … … 72 46 to 50.9 years 51 to 55.9 years 56 to 60.9 years 61 to 65.9 years 66 to 70.9 years 71 to 75.9 years 76 to 80.9 years 81 to 85.9 years 86 to 90.9 years 91+ years Perceived life stress Not at all stressful† Not very stressful A bit stressful Quite a bit stressful Extremely stressful Smoking status Lifetime abstainer (never smoked whole cigarette) † Experimental smoker (at least 1 cigarette, current non-smoker) Former occasional smoker (current nonsmoker) Former daily smoker (current non-smoker) Current occasional smoker Current daily smoker Highest level of attained education Grade 8 or lower † Grade 9-10 Grade 11-13 Secondary school graduation, no postsecondary Trade certificate or diploma College certificate or diploma (non-trades) University certificate or diploma below bachelor’s level Bachelor’s degree 3.09 7.35 4.75 5.06 7.69 12.12 14.79 34.01 38.04 16.78 0.65 1.83 1.01 1.24 1.98 3.22 3.86 9.33 9.77 3.76 14.61 29.48 22.36 20.55 29.90 45.65 56.68 123.91 148.15 74.93 0.154 0.005* 0.048* 0.024* 0.003* 0.000* 0.000* 0.000* 0.000* 0.000* 1.00 0.78 1.06 1.40 1.02 … 0.52 0.72 0.85 0.55 … 1.17 1.56 2.31 1.92 0.232 0.778 0.187 0.939 1.00 … … 1.18 0.71 1.94 0.524 1.44 0.62 3.32 0.396 1.25 0.88 1.78 0.215 0.37 0.08 1.68 0.198 1.57 0.95 2.59 0.076 1.00 0.69 1.43 1.19 … 0.39 0.61 0.69 … 1.21 3.33 2.05 0.195 0.412 0.520 1.20 0.69 2.10 0.517 0.95 0.55 1.65 0.869 0.91 0.37 2.23 0.839 1.35 0.71 2.57 0.352 73 Certificate, diploma, or degree above bachelor’s level Sex Male† Female Interaction term - binge drinking in the past 12 months x sex No alcohol in past 12 months x female† Never x female Less than once per month x female Once per month x female 2-3 times per month x female Once per week x female More than once per week x female Note. † Reference category. 1.31 0.65 2.63 0.447 1.00 1.17 … 0.75 … 1.83 0.478 1.00 … … … 0.56 0.91 0.31 0.22 1.02 3.75 0.059 0.891 0.39 0.18 0.04 0.03 3.49 1.13 0.398 0.068 2.35 4.82 0.14 0.90 39.71 25.92 0.555 0.067 The final model was bootstrap replicated 394 of the desired 1000 times and was statistically significant with Wald " (41, n = 16,715,618) = 1632.65 and p = 0.000. The main variable of interest in this model, binge drinking frequency in the past 12 months, was a good predictor of AD/dementia group membership for those who binge drank two to three times per month and for those who binge drank more than once per week. Specifically, the former and latter were associated with an 81% and 84% reduction in the likelihood of having AD or another form of dementia, respectfully. The full distribution of the relationship between binge drinking in the past 12 months and the odds of having AD or dementia is displayed below in Figure 4. 74 Binge drinking frequency in past 12 months No alcohol† Never Less than once per month Once per month 2-3 times per month * Once per week More than once per week * 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 Observed OR Figure 4. Model C: Observed odds ratios (ORs) demonstrating the likelihood of having Alzheimer’s disease (AD) or other dementias associated with various frequencies of binge drinking over the past 12 months in the community-dwelling Canadian population aged 41 and older in 2015 and 2016. The dashed line indicates an OR of 1, or no association between the exposure (frequency of binge drinking) and outcome (AD or other dementias). Statistical significance at the p < 0.05 level, if present, is indicated by an asterisk (*). The general observed trend was negative such that increased frequency of binge drinking was associated with a decreased risk of AD or dementia. As in the previous models, the only two covariates that were statistically significant at the p < 0.05 level in predicting AD/dementia group membership were proxy use and all age categories with the exception of those aged 46 to 50.9 years. Proxy use was associated with over 24 times the odds of having AD or dementia compared to those who did not use proxies when all other covariates were held constant. Belonging to one of the age categories beyond and including 51 years of age was associated with an increase of between approximately 4 and 38 times the odds of having AD or dementia compared to those aged 41 to 45.9 years 75 when all other covariates were held constant. The odds of having AD or dementia generally increased with increasing age, with the exception of those who were 91 years of age or older who had less than half the likelihood of having AD or dementia than those in the preceding age category. The fit of the model was assessed using " goodness of fit and link tests. The results of the former indicated that the observed sample distribution was not significantly different from the expected distribution, " (9, n = 16,715,618) = 5.57, p = 0.782. The results of the latter demonstrated that link, or misspecification, error was not present in the model, Bhat = 1.18, phat = 0.000, Bhatsq = 0.03, phatsq = 0.15) — i.e., the squared dependent variable did not have any explanatory power in the model. Finally, influential observation analysis was conducted to determine the influence of multivariate outliers. Multivariate outliers were identified (noutliers = 44), excluded from the model, and the model was re-run. The results were screened for changes in sign or direction of odds ratios and changes in significance. No such changes were found; therefore, the decision was made to not exclude multivariate outliers from the model. Model D: CCSA drinking risk classification based on drinking frequency in the seven days. Three variations of Model D were run: one including all covariates but excluding the sex variable and interactions between sex and drinking frequency in the past week categorization; once including only the interaction variables, drinking frequency in the past week and sex, and the interaction term; and once including all covariates and the interaction between drinking frequency classification over the past week and sex. All three models were statistically significant at a p value of 0.000. In the second iteration of the model, both light drinkers and very heavy drinkers were statistically significant with regards to their ability to predict AD/dementia group membership at p < 0.05. There was also a statistically significant 76 interaction between sex and light drinkers and sex and very heavy drinkers at p < 0.05. In the third iteration, however, statistical significance for light and very heavy drinkers was eliminated and only the interaction between very heavy drinkers and sex remained at p < 0.05, suggesting that the covariates are important and explain some of the relationship between drinking and AD/dementia status seen in the second iteration. As a result, the third iteration of the model, including all covariates and the interaction between drinking frequency in the past seven days, was selected as the final model. The results of this model can be viewed below in Table 8. Table 8. Model D: Observed odds ratios (ORs) relating Canadian Centre on Substance Use and Addiction (CCSA) drinking risk classification based on drinking frequency in the past week, interview proxy use, smoking status, life stress, highest level of attained education, age, and sex to Alzheimer’s disease (AD) or other dementias in the community-dwelling Canadian population aged 41 years and older in 2015 and 2016. 95% confidence intervals (CIs) and p-value data is also presented. Statistical significance at the p < 0.05 level is indicated by an asterisk (*). Correlates of AD and other OR 95% CI p-value dementias Drinking frequency in past week (average) Alcohol abstainer (♂ and 1.00 … … ♀: 0 drinks/day)† Light drinker (♂: 1-2 0.95 0.59 1.53 0.844 drinks/day, ♀: 1 drink/day) Moderate drinker (♂: 3 0.57 0.20 1.66 0.302 drinks/day, ♀: 2 drinks/day) Heavy drinker (♂: 4-5.9 1.22 0.16 9.06 0.847 drinks/day, ♀: 3-3.9 drinks/day) Very heavy drinker (♂: 0.29 0.07 1.13 0.074 6+ drinks/day, ♀: 4+ drinks/day) Proxy use No† 1.00 … … Yes 24.83 18.12 34.03 0.000* Age 77 41 to 45.9 years† 46 to 50.9 years 51 to 55.9 years 56 to 60.9 years 61 to 65.9 years 66 to 70.9 years 71 to 75.9 years 76 to 80.9 years 81 to 85.9 years 86 to 90.9 years 91+ years Perceived life stress Not at all stressful† Not very stressful A bit stressful Quite a bit stressful Extremely stressful Smoking status Lifetime abstainer (never smoked whole cigarette) † Experimental smoker (at least 1 cigarette, current non-smoker) Former occasional smoker (current nonsmoker) Former daily smoker (current non-smoker) Current occasional smoker Current daily smoker Highest level of attained education Grade 8 or lower † Grade 9-10 Grade 11-13 Secondary school graduation, no postsecondary Trade certificate or diploma College certificate or diploma (non-trades) University certificate or diploma below bachelor’s level 1.00 3.11 7.61 4.99 5.55 8.39 13.92 16.93 40.20 42.81 20.45 … 0.76 1.91 1.07 1.42 2.36 3.82 4.69 11.20 11.30 4.79 … 12.72 30.32 23.27 21.71 29.79 50.77 61.16 144.33 162.12 87.22 0.115 0.004* 0.041* 0.014* 0.001* 0.000* 0.000* 0.000* 0.000* 0.000* 1.00 0.78 1.06 1.39 1.02 … 0.51 0.72 0.83 0.57 … 1.18 1.57 2.32 1.83 0.239 0.758 0.206 0.952 1.00 … … 1.14 0.70 1.83 0.600 1.41 0.62 3.19 0.415 1.20 0.85 1.69 0.309 0.38 0.09 1.63 0.191 1.48 0.86 2.53 0.153 1.00 0.68 1.46 1.17 … 0.38 0.65 0.71 … 1.21 3.23 1.93 0.193 0.357 0.540 1.18 0.72 1.93 0.518 0.94 0.54 1.64 0.838 0.88 0.39 1.98 0.761 78 Bachelor’s degree Certificate, diploma, or degree above bachelor’s level Sex Male† Female Interaction - CCSA categorized drinking risk x sex Alcohol abstainer x female† Light drinker x female Moderate drinker x female Heavy drinker x female Very heavy drinker x female Note. † Reference category. 1.34 1.33 0.73 0.68 2.46 2.60 0.347 0.402 1.00 1.08 … 0.76 … 1.53 0.684 1.00 … … … 0.55 0.56 0.29 0.12 1.03 2.75 0.061 0.479 0.35 7.46 0.02 1.41 5.00 39.41 0.438 0.018* The final model, which was successfully bootstrap replicated 345 of the desired 1000 times, was statistically significant, Wald " (37, n = 16,715,618) = 1121.38, p = 0.000. The categories of the main variable of interest in this model, CCSA drinking risk classification, was not a good predictor of AD/dementia group membership. However, very heavy drinkers approached statistical significance and were 71% less likely to have AD or dementia than alcohol abstainers when all other covariates were held constant. The trend of the relationship between CCSA drinking risk classification risk for alcohol consumption over the past seven days and the likelihood of having AD or dementia is displayed below in Figure 5. 79 CCSA drinking risk categorization Alcohol abstainer† Light drinker Moderate drinker Heavy drinker Very heavy drinker 0 1 2 3 4 5 6 7 8 9 Observed OR Figure 5. Model D: Observed odds ratios (ORs) demonstrating the likelihood of having Alzheimer’s disease (AD) or other dementias associated with Canadian Centre for Substance Use and Addiction (CCSA) drinking risk categories based on averages drinking frequency over the past seven days in the community-dwelling Canadian population aged 41 and older in 2015 and 2016. CCSA drinking risk categories are as follows: alcohol abstainer (♂ and ♀: 0 drinks/day), light drinker (♂: 1-2 drinks/day, ♀: 1 drink/day), moderate drinker (♂: 3 drinks/day, ♀: 2 drinks/day), heavy drinker (♂: 4-5.9 drinks/day, ♀: 3-3.9 drinks/day), and very heavy drinker (♂: 6+ drinks/day, ♀: 4+ drinks/day). The dashed line indicates an OR of 1, or no association between the exposure (frequency of alcohol consumption) and outcome (AD or other dementias). Statistical significance at the p < 0.05 level, if present, is indicated by an asterisk (*). The reference category is marked as †. The general trend for this relationship was negative, such that increased alcohol consumption throughout the week yielded a reduced likelihood of having AD or another form of dementia. As with Models A through C, proxy use and categorized age, with the exception of those aged 46 to 50.9 years, were the only covariates that demonstrated good predictability of whether an individual would or would not have AD or dementia. The former indicated that those who used proxies were over 24 times more likely to have AD or another form of dementia than those who did not use proxies to respond to the survey when all other 80 covariates were held constant. The latter suggested that those who belonged to age categories over and including 51 years were between approximately 4 and 42 times more likely to have AD or dementia than those in the reference category who were between the ages of 41 and 45.9 years. Interestingly, the odds of having AD or dementia increased rather steadily with increasing age until the 91 years or older group where the odds were cut by half. This model resulted in a statistically significant interaction between the very heavy drinker CCSA drinking risk class, referenced against alcohol abstainers, and sex. Odds ratios were calculated comparing the two sexes with regards to those who were classified as very heavy drinkers compared to alcohol abstainers (calculations can be found in Appendix A). The odds ratio for very heavy drinker men compared to alcohol abstainers was calculated to be 0.29 and that for women was found to be 2.15. The odds ratio for men implies that being a very heavy drinker is protective against dementia in men but is associated with a two-fold increase in the odds of dementia in women. Proportions were also generated for both men and women for those who reported being very heavy drinkers over the past week and AD or dementia status to visualize the interaction between sex and alcohol consumption. These proportions were then converted into bar graphs and are displayed in Figure 6 below. 81 Percentage of men with AD or other dementias (%) a) 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Alcohol abstainer† Percentage of women with AD or other dementias (%) b) Very heavy drinker 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Alcohol abstainer† Very heavy drinker Figure 6. Model D: The interaction between consuming being classified as a very heavy drinker ((♂: 6+ drinks/day average, ♀: 4+ drinks/day average) according to Canadian Centre on Substance Use and Addiction (CCSA) guidelines over the past seven days and the percentage of those with Alzheimer’s disease (AD) and other dementias for (a) men and (b) women in the community-dwelling Canadian population aged 41 years and older in 2015 and 2016. The reference category is marked as †. The dashed line shows how the trends vary between the sexes. The slope of the trend line connecting those who abstained from alcohol in the past week and those who were heavy drinkers in the past week was more negative for women than men. 82 There were more women who had AD or dementia who were alcohol abstainers than very heavy drinkers — i.e., those who drink four or more drinks per day average for women. The trend was also the same for men. There were more men with AD or dementia who were alcohol abstainers than very heavy drinkers — i.e., six or more drinks per day average for men. However, when the proportions of those with AD or dementia who were very heavy drinkers were compared between men and women, there were slightly more women (0.0025%) in this category than men (0.0016%). Fit of the model was assessed using two approaches: " goodness of fit and link tests. The former test indicated that the observed sample distribution did not differ significantly from the expected distribution, " (9, n = 16,715,618) = 11.05, p = 0.273. The latter suggested that link, or misspecification error, was present in the model, Bhat = 1.28, phat = 0.000, Bhatsq = 0.04, phatsq = 0.041 — i.e., the squared dependent variable did have some explanatory power in the model. Finally, influential observation analysis was attempted; however, no multivariate outliers were identified. It was concluded that the model should remain as is. Differences Between Included and Excluded Samples The included and excluded samples were then analyzed using simple linear regression to determine whether the samples differed significantly from one another. The included and excluded groups differed significantly on the following variables: AD or dementia status, proxy use, binge drinking habits, CCSA drinking risk classification, age — both categorized and continuous, perceived life stress, smoking habits, and highest level of attained education. The two groups did not differ with regards to their lifetime alcohol use, both variables addressing alcohol use in the past year (binary and frequency variables), and respondent sex. 83 A larger proportion of those with dementia were excluded from the study sample than were included, Wald " (1, n = 17,461,161) = 23.16, p < 0.000, R2 = 0.002. Of the excluded sample, 3.14% had AD or dementia versus only 0.88% in the included sample. Proxy use also saw a larger proportion of those with dementia present in the excluded than included sample, Wald " (1, n = 17,461,161) = 80.86, p < 0.000, R2 = 0.009. Eleven percent of the excluded sample used proxies to respond to the survey; whereas, only 3.20% of the included sample used proxies. Binge drinking habits differed significantly between the two groups, Wald " (1, n = 17,461,161) = 11.93, p < 0.001, R2 = 0.001. Percentages for the sub-categories of the binge drinking variable differed between the two groups by more than 5% for those who had not consumed alcohol at all in the past 12 months and for those who binge drank less than once per month in the past year. For the former, 26.88% of those excluded and 21.97% of those included drank no alcohol in the past year — i.e., there were fewer people included in the final sample who had not consumed alcohol in the past year. For the latter, 13.98% of those excluded and 19.19% of those included binge drank less than once per month in the past 12 months — i.e., a larger proportion of those included engaged in monthly binge drinking. CCSA drinking risk classification for alcohol consumption over the past seven days differed significantly between those who were included and excluded from the study, Wald " (1, n = 17,461,161) = 5.13, p < 0.024, R2 = 0.000. The sub-categories of the CCSA drinking risk variable differed by more than 5% between the excluded and included samples for those who were classified as alcohol abstainers and light drinkers. For the former, 54.69% of those excluded and 49.24% of those included were classified as alcohol abstainers — i.e., more people were excluded who were classified as alcohol abstainers. For the latter, 41.23% 84 of those excluded and 47.14% of those included were classified as light drinkers — i.e., more light drinkers were included in the study than were excluded. Categorized age differed significantly between the included and excluded samples, Wald " (1, n = 17,461,161) = 97.57, p < 0.000, R2 = 0.005. There was a greater proportion of younger adults between the ages of 41 and 65.9 years who were included in the study and, alternatively, a larger percentage of older adults who were 66 years of age or older who were excluded from the study. The continuous age variable also significantly differed between the included and excluded groups, Wald " (1, n = 17,461,161) = 100.36, p < 0.000, R2 = 0.005. The mean age for those excluded was 63.28 years and that for those included was 59.05 years. The included sample was therefore younger than the excluded sample. The included and excluded samples differed significantly with regards to their ratings of perceived life stress, Wald " (1, n = 17,461,161) = 7.65, p < 0.006, R2 = 0.000. The perceived life stress sub-categories that differed the most dramatically (i.e., more than 3%) were those who reported their life as not being stressful at all, a bit stressful, and quite a bit stressful. For the first sub-category, 20.38% of those excluded versus 16.31% of those included reported their life to be not at all stressful — i.e., more people were excluded who did not experience stress at all. For the second sub-category, 35.93% of those excluded and 23.63% of those included reported their life to be not very stressful — i.e., more people were excluded who did not experience much stress. For the third sub-category, 14.30% of those excluded versus 18.05% of those included reported their life to be quite a bit stressful — i.e., more people were included who found their life to be quite stressful. Cigarette smoking habits differed significantly between the two groups, Wald " (1, n = 17,461,161) = 7.38, p < 0.007, R2 = 0.000. The smoking habit sub-categories that differed 85 the most significantly (i.e., more than 3%) were those who reported being experimental and former daily smokers. For the former, 10.85% of those excluded and 13.03% of those included identified as experimental smokers — i.e., more people were included who were experimental smokers. For the latter, 34.31% of those excluded and 30.86% of those included reported being former daily smokers — i.e., more people were excluded who used to smoke cigarettes daily. Finally, highest level of attained education differed significantly between groups, Wald " (1, n = 17,461,161) = 7.66, p < 0.006, R2 = 0.000. Education level sub-categories that differed significantly (i.e., more than 3%) were secondary school graduation, certificate/diploma, and Bachelor’s degree. For the first sub-category, 28.14% of those excluded and 21.99% of those included had achieved secondary school graduation — i.e., more people were excluded from the study who had graduated from secondary school. For the second sub-category, 18.83% were excluded and 22.81% were included who had a certificate or diploma — i.e., more people were included in the study who had a certificate or diploma. For the third sub-category, 13.64% were excluded and 17.3% were included who had received a Bachelor’s degree — i.e., more people were included in the study who had a Bachelor’s degree than were excluded. 86 CHAPTER 5: DISCUSSION AND CONCLUSIONS This research — which used data from the Canadian Community Health Survey (CCHS), a large nationwide government health survey — had two primary goals. First was to determine the nature and shape of the relationships between lifetime alcohol consumption, alcohol consumption frequency over the past 12 months, binge drinking frequency over the past 12 months, and Canadian Centre for Substance Use and Addiction (CCSA) drinking risk classification based on alcohol consumption over the past seven days and the odds of AD and other dementias. Second was to explore the presence of a sex effect in the moderation of the four previously described relationships. As a result, four hypotheses were constructed. First, it was expected that those who consumed alcohol in their lifetime would have a higher likelihood of AD or other dementias than those who had abstained from alcohol in their lifetime. Second, the relationship between alcohol consumption frequency in the past 12 months and the odds of AD or other dementias was anticipated to be J-shaped, such that alcohol abstinence, moderation, and heavy consumption would be associated with moderate, low, and high odds of AD, respectively, Third, the relationship between binge drinking frequency in the past 12 months and the odds of AD or other dementias was predicted to be J-shaped, such that alcohol abstinence, binge drinking in moderation, and frequent binge drinking would be associated with moderate, low, and high odds of AD, respectively, Fourth, the relationship between CCSA drinking risk classification and AD or other dementias would be J-shaped, such that alcohol abstinence, moderation, and heavy consumption would be associated with moderate, low, and high likelihoods of dementia, respectively. Finally, it was predicted that a sex effect would be present in all of these four relationships — i.e., the relationship between various forms of alcohol consumption and the odds of AD or dementia would vary based on sex. 87 The first hypothesis regarding lifetime alcohol consumption was rejected. Those who had consumed alcohol in their lifetime did not have a significantly higher odds of having AD or another form of dementia than those who had abstained from alcohol consumption throughout their life. Furthermore, a sex effect was not found to moderate this relationship. The second hypothesis, which predicted that the relationship between alcohol consumption frequency over the past 12 months would be J-shaped, was also rejected. The observed association between alcohol consumption frequency and the odds of having dementia resembled a somewhat abstract inverted J shape and trended towards higher amounts of alcohol consumption being protective against rather than a risk factor for dementia. A sex effect was identified; however, only for those who drank four to six times per week. The sex effect indicated that consuming alcohol four to six times per week was more protective against dementia in women than in men. The third hypothesis, which anticipated that the relationship between binge drinking in the past 12 months and the odds of AD or dementia would be J-shaped, was also rejected. The observed relationship was generally negative, such that increased frequency of binge drinking was associated with a decreased likelihood of having dementia, and this protective effect of alcohol consumption was only present at a statistically significant level for those who binge drank two to three times per month and more than once per week. Sex also did not moderate the relationship between binge drinking frequency and AD or dementia likelihood. Finally, the fourth hypothesis, which expected that the relationship between CCSA drinking risk categorization and AD and other forms of dementia would be J-shaped, was rejected. The observed shape of the relationship was approximately what was anticipated; however, the results were not statistically significant. A sex effect was identified in this model for only those who were classified as very heavy 88 drinkers, where men were protected against AD and women had a two-fold increased dementia risk. The only statistically significant results pertaining directly to the hypotheses were threefold. First, a sex effect was present for those who, on average, consumed between four and six alcoholic beverages per week in the past year. Both men and women had an odds ratio of less than one (ORmen = 0.89, ORwomen = 0.13), indicating that drinking an average of four to six times per week in the past year made it less likely that an individual currently had AD or dementia compared to those who did not consume alcohol in the past year. When men and women were compared, women were seemingly more protected against AD than men (i.e., their odds ratio was further from one than that of men) and this difference was statistically significant. Second, binge drinking an average of two to three times per month and more than once per week in the past 12 months were good predictors of AD or dementia status. Binge drinking two to three times per month and more than once per week were associated with an 81% and 84% reduction in the odds of having AD or another dementia when compared to those who abstained from alcohol in the past year. Third, a sex effect was present for those who were classified as very heavy drinkers based on the CCSA drinking risk guidelines and alcohol consumption habits over the past week. Men who were classified as very heavy drinkers (consuming six or more alcoholic drinks per day on average in the past week) had an odds ratio of less than one (ORmen = 0.29), indicating that being a very heavy drinker was associated with a decreased likelihood of currently having AD or dementia when compared to alcohol abstainers. Women who were classified as being very heavy drinkers (consuming four or more alcoholic drinks per day on average in the past week) had an odds ratio greater than one (ORwomen = 2.15), indicating that being a very heavy drinker was associated with an increased likelihood of currently having AD or dementia when 89 compared to alcohol abstainers. This difference between the sexes was statistically significant. These results were as expected because a sex effect was observed and that drinking an average of four to six times per week is protective against dementia; however, it was not anticipated that alcohol consumption would be protective against AD or dementia for those who were classified as very heavy drinkers, as they are consuming alcohol beyond what is considered to be “in moderation” or “healthy”. Furthermore, for those classified as being very heavy drinkers, alcohol consumption in men reduced the likelihood of having AD; whereas, the opposite effect was observed in women. Alcohol has long been touted as being protective against many health issues in the long term, including dementia. As mentioned previously, most studies suggest that the association between alcohol consumption and AD is J- or U-shaped, such that those who abstain from alcohol have a higher risk of AD than those who drink in moderation, and those who drink heavily have a higher AD risk than both alcohol abstainers and those who drink in moderation. The finding that women who were very heavy drinkers (i.e., consumed four or more drinks on average per day) were more at risk of presently have AD or another form of dementia than very heavy drinker men, who in fact were protected against having AD or another form of dementia, is consistent with this prediction. Furthermore, this finding corroborates the results of a study conducted by Lopes et al. (2010) in which a J-shaped relationship between alcohol consumption and cognitive dysfunction was identified for women only, at least with regards to very heavy drinking women. Furthermore, the finding that women are more negatively affected by heavy drinking than men is consistent with Wardzala et al. (2018). It is important to note that the current study was cross-sectional; therefore, it can only be concluded that women who were very heavy drinkers in seven days 90 before the survey was conducted have a significantly higher odds of having AD or dementia than their male counterparts. For those who consumed an average of four and six alcoholic beverages per week for the last year, both women and men were protected against a current diagnosis of AD or another form of dementia. Other studies have reported on the health benefits of drinking alcohol. Ferrères (2004) and Evans (2011) researched the French Paradox and found that the French who tend to drink red wine, which is rich in polyphenols, and adhere to a Mediterranean diet were less likely to have coronary heart disease than those in other countries who consumed different food. The polyphenols found in red wine have also been demonstrated to reduce the formation of amyloid- plaques (Loureiro et al., 2017). Research conducted by Goldwater, Karlamangla, Merkin, and Seeman (2019) found that those who consumed any amount of alcohol in the past month had lower scores of multisystem physiologic dysfunction than those who abstained from drinking. However, a concept called abstainer bias could help to explain the observed preventative effect of alcohol for males and females who drank between four to six alcoholic drinks on average per week over the past year and for males who were classified as very heavy drinkers based on their alcohol consumption over the last week. Abstainer bias results when the vast majority of those who do not consume alcohol are in poorer general health than those who do consume alcohol, and its effect is exacerbated in studies involving older participants (Cook, 2018; Hassing 2018). Oftentimes, abstainer bias is not controlled for in scientific research. These generally unhealthy individuals are, thus, combined with true alcohol abstainers which biases this control or reference group towards being less healthy than those who do drink alcohol. A study by Hassing (2018) actively controlled for abstainer bias when investigating the effect of alcohol use on cognitive performance. The results, unsurprisingly, demonstrated that there 91 was a negative dose-response relationship between alcohol consumption and cognitive performance, rather than the J-shaped relationship that may other studies were reporting. A significant negative effect of alcohol consumption was even observed at low doses, those that would fall within low risk drinking guidelines, such as those put forth by the CCSA. In the present study, however, women who consumed an average of four to six alcoholic drinks per week over the past year were more protected against the adverse effects of alcohol than men who drank the same number of beverages containing alcohol. Research addressing a potential sex effect with regards to the effect of alcohol on cognitive function, particularly AD or other dementias, is relatively rare. However, one study by Wardzala et al. (2018) found that men were more likely to cognitively benefit from consuming alcohol than women. The results of this study, specifically pertaining to women who consume an average of four to six drinks per week, were inconsistent with that by Wardzala et al (2018). The discrepancy between the present study and that by Wardzala et al (2018) could be explained by low cell counts for men and women in this alcohol consumption category. Unfortunately, the cell counts could not be released from the Research Data Centre due to confidentiality concerns. However, it can be reported that only 6.84% of the total sample (i.e., those with and without AD or dementia combined) consumed alcohol between four and six times per week in the past 12 months, the lowest proportion for any level of the alcohol consumption frequency in the past year variable. As a result, it is possible that sparse data bias is present and inflated the significance of this relationship. When cases are rare — here, sex, alcohol consumption, and dementia status were being looked at — the coefficients or, in this case odds ratios, tend to be biased away from the null (Greenland, Mansournia, & Altman, 2016). The implication of this is that when cell counts are low, one is more likely to find statistical significance. This is possibly also true of the significant sex effect for those who were 92 classified as very heavy drinkers in the past week. In fact, the cell counts for those who were very heavy drinkers did not meet the CCHS data release criteria for descriptive statistics, meaning that the cell counts were fewer than five. As a result, for the descriptive data to be released, categories had to be collapsed. In this case, heavy and very heavy drinking categories were combined. Even when combined, this group constituted of a total of 1.6% of the entire sample. Three other interesting findings of this research were the higher prevalence of AD or dementia for men than women, the sudden drop in the prevalence of AD or dementia for those aged 91 or older compared to their immediately younger counterparts, and that those who used proxies to respond to the CCHS survey were, in every model, over 24 times more likely to have AD or dementia than those who did not respond to the survey using proxies. Historically, women have been considered to be more susceptible to developing dementias, particularly AD (Derreberry & Holroyd, 2017; Welsh, 2019); therefore it was anticipated that the proportion of those with AD or dementia would be higher for women than men as female sex is viewed as a risk factor for developing these detrimental cognitive diseases. However, more men than women were found to have AD or another form of dementia in the present study. Recently an alternative explanation for the increased prevalence of AD and other dementias in women has been proposed. A recent paper by Mielke (2018) suggests that age is the primary risk factor for AD, not sex. The results of studies are often skewed to show an increased proportion of women affected by dementias than men because women live longer than men, on average, and are therefore more at risk of developing AD. Mielke (2018) states that studies assessing the incidence (i.e., new cases) of AD in the United States show that women and men of the same age are equally likely to develop dementia, even after surpassing the age of 85. Similar studies conducted in the 93 European Union and the United Kingdom offer conflicting results. Incidence of AD is higher in women in the former and higher in men in the latter, suggesting that new cases of AD is influenced by both temporal and geographical factors (Mielke, 2018). Although the present study did not look at incidence, but rather prevalence or proportion, the fact that the most recent view is that age rather than sex is the predominant risk factor for developing AD could explain the inflated prevalence of AD and other dementias in men in our sample. It is also possible that the sample of the Canadian population interviewed in these cycles of the CCHS was disproportionately afflicted with forms of dementia other than AD (e.g., Lewy body, vascular, and Parkinson’s Disease dementias). It has been suggested that these other forms of dementia are more commonly found in men than women. Unfortunately, the CCHS does not record what type of dementia a respondent is suffering from so this was unable to be assessed. The previously mentioned paper by Mielke (2018) only looked at sex and AD. Whether age is also the prevailing risk factor for these other forms of dementia is currently unknown. As previously discussed, the most important risk factor for developing AD or dementia is age. Therefore, it was anticipated that there would be a positive, proportional relationship between age and the prevalence of AD and other forms of dementia. This study identified a mostly positive relationship between the two factors; however, after at the age of 91 or older, the prevalence of AD and dementia dropped to 4.71%, the lowest proportion seen since those in the 46 to 50.9 years of age category. One possible explanation of this result is that women tend to live longer and healthier lives than men (Peck et al., 2017). There are three prevailing hypotheses as to why this is the case (Peck et al., 2017). The heterogametic hypothesis states that women have two X chromosomes, which provides genetic redundancy when mutations occur. The second hypothesis suggests that estrogen has 94 a positive effect on inflammation, lipid profiles in the blood, the immune system, and reduces oxidative stress. The third hypothesis attributes elevated estrogen levels in women to their decreased incidence of cardiovascular disease and other chronic diseases when compared to men, at least in premenopausal women. Men therefore have a higher risk of mortality at a younger age than women due to their lack of genetic redundancy as well as lower inherent levels of the health-protecting hormone estrogen. Those who reach the age of 91 or older are likely men and women who have positive genetic and lifestyle factors and are, therefore, less likely to develop AD or dementia. Furthermore, the average age of diagnosis of AD and dementia is 75 years (Barnes, J. et al, 2018) and most individuals, once diagnosed, survive for eight years (Alzheimer’s Association, 2020). Those who have reached the age of 91 or older have, therefore, surpassed both the average age of dementia diagnosis and mortality. Finally, regardless of model, respondents who used proxies reliably had a vastly increased odds of currently having AD or dementia than those who did not use proxies. Specifically, the former were reliably 24 or more times more likely to currently have AD or dementia than the latter. This result was not surprising. Proxies were used in cases where the intended respondent was not able to provide answers to the survey questions, either due to physical or mental ailments (Statistics Canada, 2017a). AD and dementia would fall into this category. Using proxies for these individuals likely increased the reliability of responses as Faria et al. (2005) suggest that those with more advanced dementias tend to lack the ability to provide accurate information about themselves in research settings. Generalizability The generalizability of this research to the Canadian population is maximized by the use of survey weighting and minimized by group differences that resulted from the application of the study’s inclusion and exclusion criteria. First, weighting is used in survey 95 research to make the results generalizable to the population from which the sample was taken. In this instance, each respondent to the CCHS was assigned a survey weight, which indicates the number of people in the target population that the respondent represents — in this case, the majority of the Canadian population (Statistics Canada, 2017a). Weighting allows the results of such research to be generalizable to the entire target population rather than only those sampled (Statistics Canada, 2017a). However, there are some limitations to the generalizability of results that are obtained using CCHS data, even with weighting applied. As stated previously in the methodology section, there were a number of Canadians who were excluded from being considered as a respondent based on factors such as location of residence, occupation, or living situation. Specifically, those living on Aboriginal reserves or in two remote health regions of Québec (Région du Nunavik and Région des Terres-Criesde-la-Baie-James), who are full-time members of the Canadian Forces, youth in foster care between the ages of 12 and 17 years, or institutionalized Canadians were excluded from being considered as respondents (Statistics Canada, 2017a). Therefore, the results of this study cannot be generalized to these populations, although youth were not considered for inclusion in this study due to them being too young to be asked about their AD or dementia status. The generalizability of the results is also limited temporally. The derived weights were designed to be representative of the Canadian population in the years 2015 and 2016 (Statistics Canada, 2017a); therefore, it would be incorrect to assume that the results would generalize to the Canadian population today, as the demographics of our country are constantly changing. Second, observed differences between the included and excluded respondents limited the generalizability of the results of this research. The decision regarding whether or not to include an individual in the final sample was based on the inclusion and exclusion criteria 96 stated in the methodology section. Respondents were required to be 41 years of age or older and to have responded either yes or no to the CCHS survey question regarding AD or dementia status to be included in the final sample. Respondents were excluded from the final sample if they were missing data on any of the variables, with the exception of a number of variables that were used to calculate the CCSA risk classification variable. The final sample, based on these inclusion and exclusion criteria, were found to differ significantly on a number of factors: AD/dementia status, proxy use, binge drinking habits, CCSA drinking risk classification, age, perceived life stress, cigarette smoking, and highest level of attained education. As a result, the final sample from which the results will be generalized to the majority of the Canadian population was significantly different from that which the sampling weights were derived from. Those who were excluded from the final sample were more likely to have dementia, have used proxies to complete the survey, be alcohol abstainers, be over the age of 66 years, have reported experiencing no life stress, be former daily smokers, and to have attained a secondary school level of education. Those who were included in the final sample were more likely to binge drink at least once per month, be classified under the CCSA drinking guidelines as a light drinkers, be between the ages of 41 and 65.9 years, report experiencing a bit or quite a bit of life stress, be an experimental cigarette smoker, and have a level of education equivalent to a post-secondary certificate or diploma, or a Bachelor’s degree. Many of the significant differences between the included and excluded groups identified are logical given that those with missing data were excluded from the final sample. As previously stated, a larger proportion of those with AD or dementia were excluded from the study, likely because these respondents had data missing due to their lack of memory. This population is sensibly more likely to use proxies to respond to the survey, as their memory may restrict them from responding accurately; less likely to currently drink, 97 especially if their AD or dementia is advanced to the state of needing a caregiver where the caregiver would be the one supplying the individual with alcohol; be older, as, generally, the incidence and prevalence of AD and dementia increases with age; be less stressed as those with dementia tend to have a reduced level of awareness; be less likely to smoke as, once again, caregivers would likely be the ones supplying them with cigarettes; and have lower attained levels of education, as lower levels of education is a risk factor for developing cognitive impairment. As a result, the sample that was analyzed here was skewed towards being healthier, younger, and to have received more years of formal education. This may negatively impact the generalizability of the results to the target population as the data that was analyzed differed significantly from the data that was collected by the CCHS but not analyzed. Furthermore, the CCHS excluded individuals who were institutionalized, including those living in long-term care or nursing homes. As a result, a significant proportion of Canadians who have been diagnosed with AD or another form of dementia have been excluded here. According to Canadian Institute for Health Information (2020), 61% of those with dementia live outside of long-term care or nursing facilities. From this information, it can be deduced that 39% of the AD and dementia population have not been considered in this study. Although weighting improves the generalizability of the findings to the target population, care should be taken in this case when extrapolating these findings on the Canadian population due to the identified differences between those included and excluded from the study. The results can only confidently be generalized to the portion of the Canadian population that the included weighted sample represents. 98 Limitations There are seven significant limitations to this research: the study’s cross-sectional nature; the possibility of the presence of selection, sparse data, and abstainer biases; the lack of being able to assess the link between cognitive reserve and AD and other dementias; the presence of link error in the multivariate logistic regression models; and the desired number of bootstrap replicates not being met. First, this study was cross-sectional, meaning that the presence of the exposure (i.e., alcohol consumption) and outcome (i.e., AD or dementia status) were assessed at the same point in time (Pandis, 2014). As a result, these types of studies are extremely limited in their ability to identify causality between the exposure and outcome (Pandis, 2014). This inability to determine a temporal link between exposure and outcome is exacerbated here by the fact that many of the alcohol consumption variables assessed drinking frequency, at most, within the past year of the survey being conducted. It is unlikely that alcohol consumption in the past 12 months would influence one’s odds of having AD or dementia in such a short time span. Furthermore, the only longitudinal alcohol consumption variable was a simple yes or no question regarding whether the individual had consumed any alcohol at all within their lifetime. The only conclusions that can be drawn from the present study are that present drinking status may or may not be a risk factor for currently living with a diagnosis of AD or dementia. Unfortunately, this is the reality of designing a study around pre-existing data. Researchers relinquish control over the data that is collected from respondents and therefore must use what is available in the best way possible to meet their research goals. However, cross-sectional studies are useful despite this significant limitation. They can be used to inform hypothesis generation for future, more rigorous studies assessing the same causal relationship (Pardis, 2014). 99 Selection bias in an inherent risk when conducting a cross-sectional study. This form of bias occurs when those who are included in a study differ significantly in their characteristics from those who were excluded from the study; therefore, the obtained estimates differ from what would be obtained if selection was done at random (Pardis, 2014). As previously discussed, there were statistically significant differences between those included and excluded from the present study. Ideally, there would be no significant difference between the two groups on any of the variables, indicating that the selected sample was equivalent to that obtained through randomization and, therefore, representative of the total sample and the target population. Those with data missing on any of the variables of interest, with the exception of those used to calculate CCSA drinking risk, were immediately excluded from the final sample. Through the analysis of group differences, it was found that a larger proportion of those excluded from than included in the final sample had AD or dementia. This likely led to a snowball effect, as other variables linked to AD and dementia were also identified to vary significantly between groups. As a result, it is problematic to extrapolate the results of this study to the target population, as the obtained sample may not be truly representative of the target population. Sparse data bias results from there being too few cases for certain combinations of exposures and outcomes (i.e., when cell sizes are small) and can even occur in very large datasets (Greenland et al., 2016). This issue can be exacerbated in many forms of multivariate analysis where coefficients are exponentiated to yield ratio estimates (Greenland et al., 2016). Here, the majority of the statistically significant results were identified in cells where there was the potential for the case count to be small. For example, one of the statistically significant findings for CCSA drinking risk classification found a statistically significant sex effect for very heavy drinkers. The very heavy drinker category had to be 100 collapsed with the previous category, heavy drinkers, when running the descriptive statistics to meet the CCHS’ minimum cell count requirement of five cases per cell. This indicates that there were fewer than five cases in one of the AD or dementia/no AD or dementia cells for very heavy drinkers. As a result, the findings for the results where there was the potential for data to be sparse should be applied critically. Abstainer bias is a relatively new concept that describes how many studies investigating lifestyle-related exposures (e.g., drinking alcohol, smoking cigarettes) to an outcome often use abstainer groups as the reference or control, and that these groups are often in poorer general health than non-abstainers (Hassing, 2018). The results are therefore skewed to show that it is “healthier” to use said substance than to abstain (Hassing, 2018). Many of the significant findings in this case reflected that consuming alcohol is protective against AD or dementia. In this case, 45 to 69% of those with AD or dementia had abstained from drinking alcohol during the period questioned on the survey, which is a larger proportion than for those without AD or dementia. It is therefore likely that the alcohol abstainer groups were skewed towards being unhealthier and more likely to currently have AD or dementia. As the abstainer group was also the reference group for all of the tested models, it is logical that, since those who drank alcohol were healthier than those who did not, a protective effect for alcohol consumption against dementia was identified. Hassing (2018) attempted to quell abstainer bias by excluding non-drinkers from her study and did not find a protective effect for light alcohol consumption against cognitive decline. Alternatively, abstainer bias could be controlled for by including a general health variable as a covariate in the model. This would ensure that the influence of general health was held constant throughout the model so that the true effect of alcohol consumption on AD or dementia status could be identified. In retrospect, it would have been interesting and likely 101 necessary to control for general health. As previously mentioned, the primary argument for using mild substance users as the reference or control groups in substance exposure studies is that abstainer groups often include true lifetime abstainers and former substance users. However, a study by Park, Ryu, and Cho (2017) investigated the impact of including, in this case, former drinkers amongst the alcohol abstainer group when investigating the relationship between alcohol consumption and cardiovascular disease. The authors found that true alcohol abstainers and former drinkers did not differ in their incidence of cardiovascular disease (Park et al., 2017). Furthermore, an observed protective effect of alcohol against cardiovascular disease was found for occasional drinkers both when the reference group was composed of true abstainers only and true abstainers and former drinkers combined (Park et al., 2017). This indicates that abstainer bias is not necessarily caused by poorer general health of those included in the abstainer group and negates the need to statistically control for general health. Ultimately, this is why I did not control for general health in my logistic regression analyses. Additionally, the decision to use alcohol abstainers as the reference categories for all four of my logistic regression models, although not ideal, was made due to the nature of the primary predictor variable for Model A, lifetime alcohol consumption. Respondents to this variable indicated that they either had or had not consumed alcohol in their lifetime. Given the binary nature of the variable, I thought that it would be most appropriate to use alcohol abstainers as the reference group in Model A as well as in the remainder of my models to increase model consistency. Cognitive reserve was unable to be controlled for due to the CCHS not containing an appropriate measure of occupation for respondents. Cognitive reserve was going to be created as a composite of educational attainment and occupational prestige. As previously mentioned, only those who were currently employed were asked about their field of work; 102 those who were retired were not asked about their former occupation. As a result, only education level was controlled for. Since this study was designed using pre-existing data, this was unable to be rectified. Link or misspecification error was identified in models A (lifetime alcohol consumption) and D (CCSA drinking risk classification). Misspecification error identifies that that the assumptions of the model are not correct, therefore the results can easily be misleading or outright wrong, and often results from nonlinearity, incorrect regressors, an error term being correlated with one of the covariates, or heteroscedasticity of the error term (MacKinnon, 2019). Here, misspecification error was found in models having to do with lifetime alcohol consumption and CCSA drinking risk classification. The former was simply a yes or no question that respondents answered: Have you ever had a drink in your lifetime? (Statistics Canada, 2017b). The latter was a variable that was computed from a number of CCHS variables having to do with how many drinks respondents had consumed in the past seven days (Statistics Canada, 2017b). I suspect that the root of the issue here has to do with the nature of the primary predictors of interest. For example, the primary predictor for Model A was a yes or no question and that for Model D was a computed composite variable of several survey variables. It is possible that these vague variables are not suitable for assessing this relationship or there is an unknown error term that is correlated with one of the covariates in the models. It is unlikely that nonlinearity, incorrect regressors, or heteroscedasticity are issues here as these potential issues were screened for. Ultimately the presence of misspecification or link error is reflective of flaws in these two exposure variables and any conclusions from these models should not be interpreted as adequately representing reality. 103 Finally, the desired number of bootstrap replicates was not attained in any of the multivariate logistic regression models. One or more parameters could not be estimated in 9, 24, 606, and 655 bootstrap replicates in Models A, B, C, and D, respectively out of the desired 1000 replicates. The small number of failed replicates in Models A and B are not overwhelmingly concerning; however, over half of the bootstrap replicates failed in Models C and D. Lumley (2017) suggests that there are eight reasons why bootstrapping may fail: the presence of correlation, constraints, extrema, lack of smoothness, serious outliers, a zero derivative, sparse estimators, or overfitting. I believe that the presence of extreme outliers, sparse estimators, and overfitting led to the bootstrap replicates failing in this case. The presence of sparse data has been discussed previously. Outliers were a known issue in the present study. Initially, all data analysis was conducted with outliers included in and excluded from the final sample; however, the decision was made to continue with outliers included as excluding them eliminated all heavy drinkers from the sample and heavy drinking was the main exposure of interest. This issue was likely compounded by the fact that these outliers who drank heavily often originated in cells with sparse data. Furthermore, logistic regression can be prone to overfitting (Tabachnick & Fidell, 2013). These factors likely contributed to the failing of bootstrapping replicates that was observed. Regarding why more replicates were failed with Models C and D than Models A and B, it is likely that the exposure variables in the former two models possessed more sparse data, outliers, and overfitting than the latter models. Future Research The previously mentioned limitations, despite negatively affecting the validity of the study, serve useful in informing future research on the topic of AD or dementia and alcohol consumption. Future studies, if feasible, should aim to be longitudinal in nature so that the 104 true causal relationship between alcohol consumption and AD or dementia, if present, can be identified. If a longitudinal study is not possible, a retrospective cross-sectional study could be conducted with past alcohol consumption variables that are better tailored to assess lifetime alcohol exposure. In the present study, one of the alcohol variables was a simple yes or no question regarding whether respondents had consumed any alcohol at all in their lifetime, two had to do with alcohol consumption frequency in the past year, and one assessed drinking frequency over the past week. Therefore, three of the four main alcohol variables of interest, at most, looked at alcohol consumption over the past year. Having a variable with more detailed information regarding lifetime alcohol consumption would be beneficial. For example, a categorical variable asking about alcohol consumption frequencies at various age groups or during the period where alcohol was consumed most frequently. This would serve to better causally link alcohol consumption with subsequent AD or dementia diagnosis. Future research should also aim to limit the amount of selection, sparse data, and abstainer bias present in the sample. First, selection bias, in this case referencing the statistically significant differences between those included in and excluded from the analysis, could be reduced by including all respondents in the analysis and imputing missing data. It is possible that there was an underlying reason as to why those not included were eliminated from contention. Making this change would increase the generalizability of the results to a larger portion of the community-dwelling Canadian population ages 41 or older. Sparse data bias could be reduced through the use of matching, a technique used to ensure that cases and controls are as equal as possible. This would ensure equivalency, at least between those with and without AD or dementia. Unfortunately, when looking at extreme levels of exposure to a particular disease risk factor, it is very likely that sample and cell sizes will be small. A 105 priority should be made in future research to control for abstainer bias. This can be done, as proposed by Hassing (2018), via eliminating alcohol abstainers from the analysis completely or through the addition of a general health variable to the model so that this factor can be controlled for statistically. It would also be interesting for future studies to be able to include additional demographic factors in the analysis that were unable to be controlled for here. For example, cognitive reserve could not be controlled for due to the lack of a recorded occupation for retirees in the CCHS. Many of the recommendations here revolve around collecting one’s own data, which results in more control over the variables available so that they are a better measure of the intended exposure. However, collecting data independently comes with a slew of additional problems, most significantly participant recruitment; if a longitudinal study is conducted, participant retention; and small sample sizes. Surveys are an invaluable source of information due to their large sample sizes and their ability to be readily generalized to the population represented. It is possible that in future research, survey datasets could be linked together in a longitudinal manner to provide a more complete view of the alcohol consumption and AD or dementia relationship. For example, a sample of repeatedly surveyed individuals could be consolidated to determine when they were diagnosed with AD or dementia. This would also allow for their alcohol consumption over the years to be recorded. Conclusions The purpose of this study was to replicate previous research in their assessment of the dose-response relationship between the full spectrum of various time- and frequencydependent alcohol consumption variables and self- or proxy-reported AD or dementia status in the community-dwelling Canadian population. An exploratory analysis regarding whether sex influenced an effect on this relationship was also addressed. The main findings were as 106 follows. Lifetime alcohol consumption was not a good predictor of present AD or dementia status, and no significant sex effect was identified. Consuming alcohol at a frequency of four to six times per week over the past 12 months was found to be significantly more protective against currently having AD or dementia in women than men when compared to those who had not consumed alcohol in the past year. Binge drinking at a frequency of two to three times per month and more than once per week over the past year were associated with a reduced likelihood of currently having AD or dementia compared to those who had not consumed alcohol in the past year; a significant sex effect was not found. Finally, men who were classified as very heavy drinkers — who drank six or more drinks per day over the past seven days — according to the derived CCSA drinking risk classification system were less likely to currently have AD or dementia than those who abstained from consuming alcohol; very heavy drinker women, who drank four or more drinks per day over the past seven days, according to the determined CCSA drinking risk guidelines were over two times more likely to currently have AD or dementia than alcohol abstainers. Although statistically significant at a p value of 0.05, these findings should be interpreted with caution due to the limitations of the study, particularly the likely presence of selection, sparse data, and abstainer bias as well as misspecification error. The implications of this research are limited due to the previously mentioned study flaws. However, one of the benefits of a cross-sectional study is that it is useful in informing future research, particularly when it comes to forming hypotheses in more robust studies. Informing future research is, therefore, the primary implication of the present study. This research sparks curiosity regarding whether a sex effect is present for moderate to heavy alcohol consumption and the likelihood of having AD or dementia in the communitydwelling Canadian population aged 41 or older. It would be interesting to see whether this 107 sex effect is sustained when better retrospective and longitudinal alcohol consumption variables are modelled against current AD or dementia status. 108 References Alzheimer Society of Canada. (2016). Report summary: prevalence and monetary costs of dementia in Canada. Retrieved from https://doi.org/10.24095/hpcdp.36.10.04 Alzheimer Society of Canada. (2018). Other dementias: mild cognitive impairment. Retrieved from https://alzheimer.ca/sites/default/files/files/national/otherdementias/other-dementias_mild-cognitive-impairment.pdf Alzheimer Society of Canada. (2019). Dementia numbers in Canada. Retrieved from https://alzheimer.ca/en/Home/About-dementia/What-is-dementia/Dementia-numbers Alzheimer’s Association. (2020). Alzheimer’s and dementia in Canada. Retrieved from https://www.alz.org/ca/dementia-alzheimers-canada.asp American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders: DSM-5. Washington, D.C: American Psychiatric Association. Ames, S. C., Stevens, S. R., Werch, C. E., Carlson, J. M., Schroeder, D. R., Kiros, G., ... Offord, K. P. (2010). The association of alcohol consumption with tobacco use in black and white college students. Substance use & Misuse, 45(7-8), 1230-1244. doi:10.3109/10826080903554192 Baker, E., Iqbal, E., Johnston, C., Broadbent, M., Shetty, H., Stewart, R., ... Dobson, R. J. B. (2017). Trajectories of dementia-related cognitive decline in a large mental health records derived patient cohort. PLoS One, 12(6) doi:10.1371/journal.pone.0178562 Balogun, O., Koyanagi, A., Stickley, A., Gilmour, S., & Shibuya, K. (2014). Alcohol consumption and psychological distress in adolescents: A multi-country study. The Journal of Adolescent Health: Official Publication of the Society for Adolescent Medicine, 54(2), 228. 10.1016/j.jadohealth.2013.07.034 Barnes, G. E. (2013). Personality, life stress, alcohol consumption and alcohol problems in 109 Vancouver couples. Procedia - Social and Behavioral Sciences, 84, 1464-1468. 10.1016/j.sbspro.2013.06.774 Barnes, J., Bartlett, J. W., Wolk, D. A., Van Der Flier, W. M., & Frost, C. (2018). Disease course varies according to age and symptom length in Alzheimer’s disease. Journal of Alzheimer's disease, 64(2), 631-642. Bruinvels, G., Burden, R. J., McGregor, A. J., Ackerman, K. E., Dooley, M., Richards, T., and Pedlar, C. (2017). Sport, exercise and the menstrual cycle: where is the research? British Journal of Sports Medicine, 51(6), 487. Butt, P., Beirness, D., Gliksman, L., Paradis, C., & Stockwell, T. (2011). Alcohol and health in Canada: A summary of evidence and guidelines for low-risk drinking. Ottawa, ON: Canadian Centre on Substance Use and Addiction. Cabral, J. C. C., Veleda, G. W., Mazzoleni, M., Colares, E. P., Neiva-Silva, L., & Neves, V. T. D. (2016). Stress and cognitive reserve as independent factors of neuropsychological performance in healthy elderly. Ciência & Saúde Coletiva, 21(11), 3499-3508. 10.1590/1413-812320152111.17452015 Canadian Centre on Substance Abuse. (2014). Alcohol. Retrieved from http://deslibris.ca.prxy.lib.unbc.ca/ID/248546 Canadian Institute for Health Information. (2020). Dementia in home and community care. Retrieved from https://www.cihi.ca/en/dementia-in-canada/dementia-across-thehealth-system/dementia-in-home-and-community-care Canadian Institute for Health Information. (2018, May 31). Alcohol harm on the rise for Canadian women. Retrieved from https://www.cihi.ca/en/alcohol-harm-on-the-risefor-canadian-women?utm_source=crm&utm_medium=email&utm_campaign= yhsccq&utm_content=mediareleaseEN 110 Centres for Disease Control and Prevention. (2018, January 3). Fact Sheets — Alcohol Use and Your Health. Retrieved from https://www.cdc.gov/alcohol/fact-sheets/alcoholuse.htm Chêne, G., Beiser, A., Au, R., Preis, S. R., Wolf, P. A., Dufouil, C., & Seshadri, S. (2015). Gender and incidence of dementia in the Framingham Heart Study from mid-adult life. Alzheimer's & Dementia: The Journal of the Alzheimer's Association, 11(3), 310320. Cook, A. (2018, March 21). Alcohol: Tonic or Toxin? CBC Radio: Ideas. Episode retrieved from https://www.cbc.ca/radio/ideas/alcohol-tonic-or-toxin-1.4584264 Creswell, J. W. (2009). Research design: Qualitative, quantitative, and mixed methods approaches (3rd ed.). Thousand Oaks, Calif: Sage Publications. Davies, H. T. O., Crombie, I. K., & Tavakoli, M. (1998). When can odds ratios mislead? BMJ, 16, 989-991. Deary, I. J., Corley, J., Gow, A. J., Harris, S. E., Houlihan, L. M., Marioni, R. E., ... Starr, J. M. (2009). Age-associated cognitive decline. British Medical Bulletin, 92(1), 135152. doi:10.1093/bmb/ldp033 Dekhtyar, S., Wang, H., Scott, K., Goodman, A., Koupil, I., & Herlitz, A. (2015). A lifecourse study of cognitive reserve in dementia — from childhood to old age. The American Journal of Geriatric Psychiatry: Official Journal of the American Association for Geriatric Psychiatry, 23(9), 885-896. doi:10.1016/j.jagp.2015.02.002 Derreberry, T. M., & Holroyd, S. (2017). Dementia in women. Psychiatric Clinics, 40(2), 299-307. Doyle, C., Dunt, D., & Morris, P. (2014). Stress and dementia. International Psychogeriatrics / IPA, 26(8), 1235. doi:10.1017/S1041610214001033 111 Durazzo, T. C., Insel, P. S., & Weiner, M. W., & Alzheimer Disease Neuroimaging Initiative. (2012). Greater regional brain atrophy rate in healthy elderly subjects with a history of cigarette smoking. Alzheimer's & Dementia: The Journal of the Alzheimer's Association, 8(6), 513-519. 10.1016/j.jalz.2011.10.006 El-Masri, M. M. (2014). Case-control study design. The Canadian Nurse, 110(6), 10. Evans, A. (2011). The French paradox and other ecological fallacies. International Journal of Epidemiology, 40(6), 1486-1489. doi:10.1093/ije/dyr138 Farias, S. T., Mungas, D., & Jagust, W. (2005). Degree of discrepancy between self and other-reported everyday functioning by cognitive status: Dementia, mild cognitive impairment, and healthy elders. International Journal of Geriatric Psychiatry, 20(9), 827-834. Ferrières J. (2004). The French paradox: lessons for other countries. Heart (British Cardiac Society), 90(1), 107-11. Gibson, A., Woodside, J. V., Young, I. S., Sharpe, P. C., Mercer, C., Patterson, C. C., ... Evans, A. (2008). Alcohol increases homocysteine and reduces B vitamin concentration in healthy male volunteers-a randomized, crossover intervention study. QJM: An International Journal of Medicine, 101(11), 881-887. doi:10.1093/qjmed/hcn112 Goldwater, D., Karlamangla, A., Merkin, S. S., & Seeman, T. (2019). Compared to nondrinkers, individuals who drink alcohol have a more favorable multisystem physiologic risk score as measured by allostatic load. Plos One, 14(9), e0223168. Greenberg, M. S., Tanev, K., Marin, M., & Pitman, R. K. (2014). Stress, PTSD, and dementia. Alzheimer's & Dementia: The Journal of the Alzheimer's Association, 10(3 Suppl), S155-S165. 10.1016/j.jalz.2014.04.008 112 Greenland, S., Mansournia, M. A., & Altman, D. G., 2016. Sparse data bias: A problem hiding in plain sight. BMJ, 352, i.1981. Grucza, R. A., & Bierut, L. J. (2006). Cigarette Smoking and the Risk for Alcohol Use Disorders Among Adolescent Drinkers. Alcoholism, Clinical and Experimental Research, 30(12), 2046–2054. http://doi.org/10.1111/j.1530-0277.2006.00255.x Harvard Medical School. (2014). Heart attack and stroke: men vs. women. Harvard Heart Letter, 24(9), 1, 7. Hassing, L. B. (2018). Light alcohol consumption does not protect cognitive function: A longitudinal prospective study. Frontiers of Aging Neuroscience, 10(March), Article 18. Himmelfarb Health Sciences Library (2011, November). Study design 101: Case control study. Retrieved from https://Himmelfarb.gwu.edu/tutorials/studydesign101 /casecontrols.html Hosmer Jr, D. W., Lemeshow, S., & Sturdivant, R. X. (2013). Applied Logistic Regression (3rd ed.). Hobokin, New Jersey, USA: John Wiley & Sons. Ikeda, A., Yamagishi, K., Tanigawa, T., Cui, R., Yao, M., Noda, H., … Iso, H. (2008). Cigarette smoking and risk of disabling dementia in a Japanese rural community: A nested case-control study. Cerebrovascular Diseases, 25(4), 324–331. https://doi.org/10.1159/000118377 International Agency for Research on Cancer. (2018). Agents Classified by the IARC Monographs, Volumes 1-123. World Health Organization. Retrieved from https://monographs.iarc.fr/agents-classified-by-the-iarc/ Johnson, K. (2011, March 10). Dementia diagnosis often refined at autopsy. Medscape. Retrieved from https://www.medscape.com/viewarticle/738750 113 Klimova, B., Valis, M., & Kuka, K. (2017). Cognitive decline in normal aging and its prevention: A review on non-pharmacological lifestyle strategies. Clinical Interventions in Aging, 12, 903-910. Lopes, M. A., Furtado, E. F., Ferrioli, E., Litvoc, J., & de Campos Bottino, C. M. (2010). Prevalence of alcohol-related problems in an elderly population and their association with cognitive impairment and dementia. Alcoholism Clinical and Experimental Research, 34(4), 726-733. doi:10.1111/j.1530-0277.2009.01142.x Li, J., Chu, J., Barrero, C., Merali, S., & Praticò, D. (2014). Homocysteine exacerbates βamyloid pathology, tau pathology, and cognitive deficit in a mouse model of Alzheimer disease with plaques and tangles. Annals of Neurology, 75(6), 851-863. doi:101002/ana.24145 Liu, K. A., & Mager, N. A. (2016). Women's involvement in clinical trials: historical perspective and future implications. Pharmacy Practice, 14(1), 708. doi:10.18549/PharmPract.2016.01.708 Loureiro, J., Andrade, S., Duarte, A., Neves, A., Queiroz, J., Nunes, C., ... Pereira, M. (2017). Resveratrol and grape extract-loaded solid lipid nanoparticles for the treatment of Alzheimer’s disease. Molecules, 22(2), 277. doi:10.3390/molecules22020277 Lumley, T. (2017, February 1). Biased and Inefficient: When the bootstrap doesn’t work. Retrieved from https://notstatschat.tumblr.com/post/156650638586/when-thebootstrap-doesnt-work Lyu, J., & Lee, S. H. (2014). Alcohol consumption and cognitive impairment among Korean older adults: Does gender matter? International Psychogeriatrics, 26(2), 335-340. MacKinnon, J. G. (2019, December 28). Specification error. Retrieved from 114 https://www.encyclopedia.com/social-sciences/applied-and-social-sciencesmagazines/specification-error McKenna, H., Treanor, C., O’Reilly, D., & Donnelly, M. (2018). Evaluation of the psychometric properties of self-reported measures of alcohol consumption: A COSMIN systematic review. Substance Abuse Treatment, Prevention, and Policy, 13(1), 6. Mielke, M. M. (2018). Sex and gender differences in Alzheimer’s disease dementia. The Psychiatric Times, 35(11), 14. Mooney, C. (2008). Bootstrapping. In P. J. Lavrakas (Ed.), Encyclopedia of Survey Research Methods (pp. 66-67). Thousand Oaks, CA: SAGE Publications, Inc. Moss, H. B. (2013). The impact of alcohol on society: A brief overview. Social Work in Public Health, 28(3-4), 175-177. doi:10.1080/19371918.2013.758987 Nabalamba, A., & Patten, S. B. (2010). Prevalence of mental disorders in a Canadian household population with dementia. Canadian Journal of Neurological Sciences, 37(2), 186-194. Nemes, S., Jonasson, J. M., Genell, A., Steineck, G. (2009). Bias in odds ratios by logistic regression modelling and sample size. BMC Medical Research Methodology, 9(1), 56-56. Nilsson, L. (2003). Memory function in normal aging. Acta Neurologica Scandinavia, 107(s179), 7-13. doi: 10.1034/j.1600-0404.107.s179.5x Ojagbemi, A., Bello, T., & Gureje, O. (2016). Cognitive reserve, incident dementia, and associated mortality in the Ibadan study of ageing. Journal of the American Geriatrics Society, 64(3), 590-595. doi:10.1111/jgs.14015 Pandis, N. (2014). Cross-sectional studies. American Journal of Orthodontics and 115 Dentofacial Orthopedics, 146(1), 127-129. Panza, F., Frisardi, V., Seripa, D., Logroscino, G., Santamato, A., Imbimbo, B. P., ... Solfrizzi, V. (2012). Alcohol consumption in mild cognitive impairment and dementia: Harmful or neuroprotective?: Alcohol and cognitive decline. International Journal of Geriatric Psychiatry, 27(12), 1218-1238. doi:10.1002/gps.3772 Park, J., Ryu, Y., & Cho, S. (2017). The effect of reference group classification and change in alcohol consumption on the association between alcohol consumption and cardiovascular disease. Alcoholism: Clinical and Experimental Research, 41(2), 379387. doi:10.1111/acer.13299 Peavy, G. M., Jacobson, M. W., Salmon, D. P., Gamst, A. C., Patterson, T. L., Goldman, S., ... Galasko, D. (2012;2011;). The influence of chronic stress on dementia-related diagnostic change in older adults. Alzheimer Disease and Associated Disorders, 26(3), 260. 10.1097/WAD.0b013e3182389a9c Peck, M. S., Porter, A. T., Hayes, B. D., Otieno, D., Hargrove, A. J., Mayhand, C. C., ... & Brown, L. M. (2017). Sex and longevity: Why women live longer than men. Journal of Student Research, 6(1), 19-23. Peduzzi, P., Concato, J., Kemper, E., Holford, T. R., & Feinstein, A. R. (1996). A simulation study of the number of events per variable in logistic regression analysis. Journal of Clinical Epidemiology, 49(12), 1373-1379. Perry, B., & Singh, S. (2016). A virtual reality: Technology′s impact on youth mental health. Indian Journal of Social Psychiatry, 32(3), 222-226. 10.4103/0971-9962.193190 Petersen, R. C., Roberts, R. O., Knopman, D. S., Geda, Y. E., Cha, R. H., Pankratz, V. S., ... Rocca, W. A. (2010). Prevalence of mild cognitive impairment is higher in men. the mayo clinic study of aging. Neurology, 75(10), 889-897. 116 Preti, A., Muscio, C., Boccardi, M., Lorenzi, M., de Girolamo, G., & Frisoni, G. (2014). Impact of alcohol consumption in healthy adults: A magnetic resonance imaging investigation. Psychiatry Research: Neuroimaging, 224, 96-103. doi:10.1016/j.pscychresns.2014.06.005 Prince, M., Acosta, D., Ferri, C. P., Guerra, M., Huang, Y., Rodriguez, J. J. L., ... Acosta, I. (2012). Dementia incidence and mortality in middle-income countries, and associations with indicators of cognitive reserve: a 10/66 Dementia Research Group population-based cohort study. The Lancet, 380(9836), 50-58. Public Health Agency of Canada. (2016). The Chief Public Health Officer’s Report on the State of Public Health in Canada 2015: Alcohol Consumption in Canada. Retrieved from http://healthycanadians.gc.ca/publications/department-ministere/state-publichealth-alcohol-2015-etat-sante-publique-alcool/alt/state-phac-alcohol-2015-etat-aspcalcool-eng.pdf Reed, M. B., Wang, R., Shillington, A. M., Clapp, J. D., & Lange, J. E. (2007). The relationship between alcohol use and cigarette smoking in a sample of undergraduate college students. Addictive Behaviors, 32(3), 449-464. 10.1016/j.addbeh.2006.05.016 Reisberg, B., Prichep, L., Mosconi, L., John, E. R., Glodzik-Sobanska, L., Boksay, I., ... de Leon, M. J. (2008). The pre-mild cognitive impairment, subjective cognitive impairment stage of Alzheimer's disease. Alzheimer's & Dementia: The Journal of the Alzheimer's Association, 4(1 Suppl 1), S98. Reisberg, B., Shulman, M. B., Torossian, C., Leng, L., & Zhu, W. (2010). Outcome over seven years of healthy adults with and without subjective cognitive impairment. Alzheimer's & Dementia: The Journal of the Alzheimer's Association, 6(1), 11-24. doi:10.1016/j.jalz.2009.10.002 117 Rethink Your Drinking. (2016). What is a standard drink? Retrieved from https://www.rethinkyourdrinking.ca/what-is-a-standard-drink/ Riis, J. L., Chong, H., Ryan, K. K., Wolk, D. A., Rentz, D. M., Holcomb, P. J., & Daffner, K. R. (2008). Compensatory neural activity distinguishes different patterns of normal cognitive aging. Neuroimage, 39(1), 441-454. doi:10.1016/j.neuroimage.2007.08.034 Robinson, G., Narasimhan, S., Weatherall, M., & Beasley, R. (2005). Raised plasma homocysteine levels in alcoholism: Increasing the risk of heart disease and dementia? The New Zealand Medical Journal, 118(1216), U1490. Ross, L. J., Wilson, M., Banks, M., Rezannah, F., & Daglish, M. (2012). Prevalence of malnutrition and nutritional risk factors in patients undergoing alcohol and drug treatment. Nutrition (Burbank, Los Angeles County, Calif.), 28(7-8), 738-743. 10.1016/j.nut.2011.11.003 Sabia, S., Elbaz, A., Britton, A., Bell, S., Dugravot, A., Shipley, M., ... Singh-Manoux, A. (2014). Alcohol consumption and cognitive decline in early old age. Neurology, 82(4), 332-339. doi:10.1212/WNL.0000000000000063 Salé, A., Berardi, N., & Maffei, L. (2014). Environment and brain plasticity: Towards an endogenous pharmacotherapy. Physiological Reviews, 94(1), 189-234. 10.1152/physrev.00036.2012 Scalbert, A., Johnson, I. T., & Saltmarsh, M. (2005). Polyphenols: Antioxidants and beyond. The American Journal of Clinical Nutrition, 81(1 Suppl), 215S-217S. doi:10.1093/ajcn/81.1.215S Shaw, B. A., Krause, N., Liang, J., & McGeever, K. (2011). Age differences in long-term patterns of change in alcohol consumption among aging adults. Journal of Aging and Health, 23(2), 207-227. 118 Simard, M., Hudon, C., & van Reekum, R. (2009). Psychological distress and risk for dementia. Current Psychiatry Reports, 11(1), 41-47. 10.1007/s11920-009-0007-z Sindi, S., Hakansson, K., Hagman, G., Kulmala, J., Soininen, H., Kareholt, I., ... Kivipelto, M. (2014). Mid-life work-related stress increases dementia risk in late-life: The CAIDE 30-year study. Alzheimer's & Dementia, 10(4), P746. 10.1016/j.jalz.2014.05.1408 Smith, A. D., Smith, S. M., de Jager, C. A., Whitbread, P., Johnston, C., Agacinski, G., ... Refsum, H. (2010). Homocysteine-lowering by B vitamins slows the rate of accelerated brain atrophy in mild cognitive impairment: A randomized controlled trial. PloS One, 5(9), e12244. doi:10.1371/journal.pone.0012244 Statistics Canada. (2017a). Canadian Community Health Survey Annual Component User Guide: 2016 Microdata File. Statistics Canada. (2017b). Canadian Community Health Survey Annual Component: 2016 Master File Data Dictionary. Stern, Y. (2009). Cognitive reserve. Neuropsychologia, 47(10), 2015-2028. doi:10.1016/j.neuropsychologia.2009.03.004 Stern, Y. (2012). Cognitive reserve in ageing and Alzheimer’s disease. The Lancet Neurology, 11(11), 1006–1012. https://doi.org/10.1016/S1474-4422(12)70191-6 Stern, Y., Albert, S., Tang, M. X., & Tsai, W. Y. (1999). Rate of memory decline in AD is related to education and occupation: Cognitive reserve? Neurology, 53(9), 1942. Sun, A. Y., Simonyi, A., & Sun, G. Y. (2002). The "French paradox" and beyond: Neuroprotective effects of polyphenols. Free Radical Biology & Medicine, 32(4), 314. Szklo, M., & Nieto, F. J. (2014). Epidemiology: Beyond the basics (3rd ed.). Burlington, 119 Mass: Jones & Bartlett Learning. Tabachnick, B. G., & Fidell, L. S. (2013). Using multivariate statistics (6th ed.). Boston, Mass: Pearson. Tam, T. (2018, October). The Chief Public Health Officer’s Report on the State of Public Health in Canada 2018. Retrieved from https://www.canada.ca/content/dam/phacaspc/documents/corporate/publications/chief-public-health-officer-reports-statepublic-health-canada/2018-preventing-problematic-substance-use-youth/2018preventing-problematic-substance-use-youth.pdf Turkington, C., & Mitchell, D. (2010). The encyclopedia of Alzheimer’s disease (2nd ed.). New York, NY: Facts on File Inc. University of California, Los Angeles. (2020). Institute for Digital Research and Education Statistical Consulting: Deciphering Interactions in Logistic Regression. Retrieved from https://stats.idre.ucla.edu/stata/seminars/deciphering-interactions-in-logisticregression/ Wardzala, C., Murchison, C., Loftis, J. M., Schenning, K. J., Mattek, N., Woltjer, R., . . . Wilhelm, C. J. (2018). Sex differences in the association of alcohol with cognitive decline and brain pathology in a cohort of octogenarians. Psychopharmacology, 235(3), 761-770. Welsh, T. J. (2019). Multimorbidity in people living with dementia. Case Reports in Women's Health, 23. White, L. (2010). Educational attainment and mid-life stress as risk factors for dementia in late life. Brain: A Journal of Neurology, 133(Pt 8), 2180-2182. 10.1093/brain/awq201 Wilsnack, R. W., Wilsnack, S. C., Kristjanson, A. F., Vogeltanz-Holm, N. D., & Gmel, G. (2009). Gender and alcohol consumption: Patterns from the multinational GENACIS 120 project. Addiction, 104(9), 1487-1500. World Health Organization. (2017). Dementia Fact Sheet. Retrieved from http://www.who.int/mediacentre/factsheets/fs362/en/ Xu, W., Wang, H., Wan, Y., Tan, C., Li, J., Tan, L., & Yu, J. (2017). Alcohol consumption and dementia risk: A dose–response meta-analysis of prospective studies. European Journal of Epidemiology, 32(1), 31-42. doi:10.1007/s10654-017-0225-3 Zhong, G., Wang, Y., Zhang, Y., Guo, J. J., & Zhao, Y. (2015). Smoking is associated with an increased risk of dementia: A meta-analysis of prospective cohort studies with investigation of potential effect modifiers. PloS One, 10(3), e0118333. doi:10.1371/journal.pone.0118333 121 Appendix A: Calculations for Interaction Odds Ratios Model B Sex Effect Interaction Calculations As the logistic regression model indicated that there was a statistically significant interaction between the sexes for those who drank alcohol four to six times per week, odds ratios were calculated, in accordance with University of California, Los Angeles (2020), to demonstrate the influence of this interaction. The data used for the calculations are as follows: Source Drinking 4-6 times per week (h) Female (f) Drinking 4-6 times per week#Female (h#f) Odds Ratio (OR) 0.8894141 1.173245 0.1415279 OR for h1/h0 (i.e., drinking 4-6 times per week over alcohol abstainers) for f0 (i.e., male): ORh = 0.8894141 OR for h1/h0 (i.e., drinking 4-6 times per week over alcohol abstainers) for f1 (i.e., female): ORh x ORh#f = 0.8894141 x 0.1415279 = 0.1258769 Model D Sex Effect Interaction Calculations As the logistic regression model indicated that there was a statistically significant interaction between the sexes for those who were classified as very heavy drinkers based on their alcohol consumption over the past week, odds ratios were calculated, in accordance with University of California, Los Angeles (2020), to demonstrate the influence of this interaction. The data used for the calculations are as follows: Source Very heavy drinkers (h) Female (f) Very heavy drinkers#Female (h#f) Odds Ratio (OR) 0.2876408 1.07625 7.461305 122 OR for h1/h0 (i.e., very heavy drinkers over alcohol abstainers) for f0 (i.e., male): ORh = 0.2876408 OR for h1/h0 (i.e., very heavy drinkers over alcohol abstainers) for f1 (i.e., female): ORh x ORh#f = 0.2876408 x 7.461305 = 2.14618