Sex and APOE genotype differences related to statin use in the aging population

Abstract Background Significant evidence suggests that the cholesterol‐lowering statins can affect cognitive function and reduce the risk for Alzheimer's disease (AD) and dementia. These potential effects may be constrained by specific combinations of an individual's sex and apolipoprotein E (APOE) genotype. Methods Here we examine data from 252,327 UK Biobank participants, aged 55 or over, and compare the effects of statin use in males and females. We assessed difference in statin treatments taking a matched cohort approach, and identified key stratifiers using regression models and conditional inference trees. Using statistical modeling, we further evaluated the effect of statins on survival, cognitive decline over time, and on AD prevalence. Results We identified that in the selected population, males were older, had a higher level of education, better cognitive scores, higher incidence of cardiovascular and metabolic diseases, and a higher rate of statin use. We observed that males and those participants with an APOE ε4–positive genotype had higher probabilities of being treated with statins; while participants with an AD diagnosis had slightly lower probabilities. We found that use of statins was not significantly associated with overall higher rates of survival. However, when considering the interaction of statin use with sex, the results suggest higher survival rates in males treated with statins. Finally, examination of cognitive function indicates a potential beneficial effect of statins that is selective for APOE ε4–positive genotypes. Discussion Our evaluation of the aging population in a large cohort from the UK Biobank confirms sex and APOE genotype as fundamental risk stratifiers for AD and cognitive function, furthermore it extends them to the specific area of statin use, clarifying their specific interactions with treatments.


BACKGROUND
Population aging has been recognized as a key policy issue worldwide.
The proportion and absolute number of older people are increasing dramatically: by 2040, nearly one in seven people is projected to be aged over 75 years. 1 Projections suggest there will be 66.1 million people aged 80 years and over in the European Union by 2080. 2 These trends will have a major impact on public spending. In the UK, the Office for Budget Responsibility forecasts total spending to increase from 33.6% to 37.8% of gross domestic product (GDP) between 2019 and 2064-equivalent to current £79 billion-due mainly to the aging population. 3 In the United States, Medicare expenditures are projected to rise to 6% to 9% of GDP with a predicted strain on federal budget and the national economy. 4 The burden of these expenditures will This suggests an increasing prevalence of chronic age-related conditions with long-duration preclinical phases such as Alzheimer's disease (AD). 3 Sex differences in longevity are documented and feature in many species in addition to humans. [5][6][7] While it is common for women to live longer than men, the magnitude of the difference in longevity differs across cultures and is modifiable by environmental factors; the difference in life span is declining in developed nations. 8 Cholesterol metabolism has been shown to have an important role in age-related disease such as AD 9,10 and mounting evidence suggests that statins, a class of cholesterol-lowering drugs, may effect cognitive function and risk for older age-associated AD and dementia. [11][12][13][14][15][16][17] Clinical trials evaluating the effects of statins in patients diagnosed with AD have failed to meet primary outcomes, resulting in no significant therapeutic benefit. [18][19][20][21] However, medical bioinformatic analyses conducted over the past 5 years indicate that statin therapies are associated with reduced risk of AD. 11,22 Recent studies investigating the benefits of statins on neurological outcomes suggest that when statins are prescribed for population at risk of age-related diseases, they are associated with decreased incidence of AD, dementia, Parkinson's disease, multiple sclerosis, and amyotrophic lateral sclerosis. 23 The interaction of genotypes of apolipoprotein E (APOE), a risk factor for AD involved in cholesterol metabolism, with statins' pharmacodynamics and pharmacokinetics has been largely investigated, [24][25][26] indicating a significant effect of the genetic polymorphisms on treatment responses in term of plasma lipid profile 27,28 and a strong association with the risk and the course of coronary heart diseases. 29 risk of AD, also slowed cognitive decline. 31 Sex differences, as well as the effects of the APOE genotype, are well documented in statin drug response. [32][33][34] In a recent examination of the association between statin use and the incidence of AD, it was found that reduction in AD risk varied across statin molecules, sex, and race/ethnicity. 35 A major resource to enable investigations in aging populations is the UK Biobank, 36  The aims of our study are (1) to assess differences in treatments in the aging population and identify potential stratifiers for greater beneficial effects of statins; and (2) to evaluate the effect of statin use in the aging population on survival, AD incidence, and cognitive decline.
While previously the cost-effectiveness of a polypill, including simvastatin, to prevent cardiovascular diseases has been assessed in the UK Biobank cohort, 37 to the best of our knowledge our work is the first to report on statin use within the UK Biobank's aging population. Study design and analytical strategy is described in Figure 1. Aim I (orange) is to assess differences in treatments in the aging population and identify potential stratifiers for greater beneficial effects of statins, to achieve the aim we determined drug exposure and assess their differences, focusing on statins treatments. Aim II (green) is to evaluate the potential beneficial effects of statin use in the aging population on survival, Alzheimer's disease (AD) incidence and cognitive decline; to achieve the aim we used the matched cohort and the identified stratifiers, derived from Aim I

Disease diagnoses
Referring to the work in Zissimopoulos et al., 35

Determination of drug exposure
The medications category of the UK Biobank contains data on type and number of regular treatments taken by each individual. Data are obtained through a verbal interview by a trained nurse and coded via Read codes. We built a code set (reported in Appendix A in supporting information) for each of the medication groups of interest previously linked to cognitive impairment and included: statins, nonstatin cholesterol-lowering drugs, AD medications, antidepressants, non-steroidal anti-inflammatory drugs (NSAIDs), estrogens, diabetes medications, vitamin E, omega-3 and derivatives, and medications for long-term asthma management.

Determination of APOE genotype
In the UK Biobank the APOE genotype is directly genotyped via SNPs rs429358 and rs7412. Values for either of the two SNPs were available for 299,627 participants; of these 47,299 participants were missing a value for one of the two SNPs and were therefore excluded. A total of 252,327 participants were included. APOE genotype missingness is due to UK Biobank enrollment procedures (i.e., participants recently enrolled for which the information is not available yet) or technical issues, therefore we assume are missing at random. 43,44 Further consideration regarding missingness mechanisms are reported in Appendix C in supporting information.

Statistical analyses
We compared and contrasted the population stratified by sex and APOE ε4 genotype. To test for significant differences among the four groups (female APOE ε4, female non-APOE ε4, male APOE ε4, male non-APOE ε4) we applied the Kruskal-Wallis test for continuous variables, and chi-square for categorical ones.
We compared APOE ε4 carriers within females and males using t tests and chi-square. The Cochran-Mantel-Haenszel test was used for stratified analyses considering population distributions in ethnicity strata. Analyses of baseline differences in cohort characteristics were corrected for multiple testing using the Bonferroni correction, as indicated in Table 1.
All analyses were computed using R version 3.2.3. Results are presented as the main effect with a 95% confidence interval. A significance level of 5% was used for main inferences.
To study drug exposure, while minimizing the effects of possible confounders and including relevant stratifiers, we applied propensity scoring to assess the comparability of case mix and created matched data sets for each drug category.
Given the definition of propensity scores (PS; i.e., the probability of being treated) this step allows us to compare the score in females and males, thus assessing relevant differences in treatments between sexes. To adjust for different distributions of characteristics across treated groups (age, social-economic status, education level, and rel-evant diagnoses for each drug), patients were stratified based on their propensity of being treated with a specific drug. It is important to note that sex is not included as a potential confounder, as the aim of this analysis was to study its correlation with treatments, and then use it as a stratifier for the following analyses.
For each drug, we derived a sample matched (with a 1:1 ratio) on the PS and compared the probability of being treated (i.e., PS itself) between females and males with t tests. Analyses were performed using the functions "matchit" and "match.data" from the MatchIt R package, 45 using logistic regression to estimate the PS and the nearest neighbor method for matching the cohorts.
To further study statin exposure differences, we applied a logistic regression model and conditional inference tree (to visually illustrate associations between selected covariates and response) on the matched cohort (where the PS is computed based on treatment with statins). In both models we assess the exposure to statins on the basis of covariates not included in the PS analyses (i.e., sex, AD, dementia, and APOE ε4 genotype). Age, social-economic status, education level, and relevant diagnoses for each drug were not included as covariates in the regression models as they were used to match the cohorts. We used the "rpart" and "rpart plot" function of the "rpart" package. 46 After we assessed the probability of being treated with statins and identified key stratifiers, we evaluated the effect of statins on specific outcomes (i.e., survival, AD prevalence, and cognitive decline) in the matched cohort.
To examine the effect of statin use on survival, we used death records captured by the UK Biobank. We used baseline measurements to build a survival model, left-censored at baseline. Right-censoring was applied at the last follow-up date or date of death (if occurred). Survival was studied with a Cox regression model adjusted by sex, APOE genotype, AD diagnosis, and dementia diagnosis. We performed the analysis with the "coxph" function of the "survival" package. 47 For assessing longitudinal cognitive patterns in relation to statin use we included individuals who had at least two measurements including baseline assessment. This selection of participants may have introduced some bias but was essential to determining slope of change in cognitive measures.
To test for differences in the rate of change of the cognitive measures between statin-user and non-user groups over follow-ups we used a linear mixed-effects model (using the "lme4" package 48 To examine potential effects of statins use on AD prevalence, we conducted a cross-sectional analysis on individuals who were

The UK Biobank aging population
From the entire UK Biobank cohort, 252,327 who were aged 55 or over at recruitment (baseline) had a determined APOE genotype and baseline data, and were selected for our investigations (Table 1). Of these, 14,523 (4.717%) had data available from their first follow-up visit and 2,677 (0.87%) had data available from baseline, first, and second follow-up visits (for a full description of selection criteria for each analysis see Figure S1 in supporting information).
We found no differences in the population distribution in the four main classes (defined by sex and APOE genotypes), nor were there any differences when stratified by ethnicity.
A comparison of females (n = 136,665) and males (n = 115,662) revealed that the two groups differ in terms of age, education level, cognitive measures, disease diagnoses, and statin use, but not in Townsend deprivation or AD incidence.
Data illustrate that in the selected population, males are older, have a higher level of education, better cognitive scores, higher incidence of cardiovascular and metabolic diseases, and higher rate of statins use. We further compared males and females stratified by APOE genotypes (carriers vs. non-carriers of the APOE ε4 allele). In both females and males, statistically significant differences were found for disease diagnoses, including AD and dementia, and in use of different statins, excluding pravastatin in males.
Dementia and AD diagnoses do not overlap for the majority of the cases, except in 150 subjects, which represents 10.8% of the total population with a diagnosis of dementia or AD (n = 1390).
As for cognitive measures at baseline, only RT was found statistically significant different in both females and males comparing APOE ε4 carriers and non-carriers.

Drug exposure in the aging population
To assess drug exposure in the aging population, datasets matched via PS were created for each drug ( Table 2). Further results regarding the matching process for statins treatments are reported in Appendix D in supporting information. We observed significant differences in drug exposure between females and males ( Figure 2). Females are less likely to be treated with antidepressants, asthma medication, diabetes drugs, and non-statin lipid lowering drugs; and more likely to be treated with NSAIDs and omega 3s.
To examining statin exposure differences, we applied a logistic regression model and conditional inference tree to assess the exposure in the matched dataset (Table S1 in supporting information) on the basis of features not included in the PS analyses (i.e., sex, AD and dementia diagnoses, and APOE ε4 genotype-indicated in Table S1 as non-matched).
Based on the regression model, males (z-value = 51.2, odds ratio A second approach to visually illustrate statin exposure differences by stratifiers, is based on recursive partitioning and reports the results as logical tree structures (Figure 3). Treatment with statins is stratified by sex, APOE ε4 genotype, and degenerative diseases. However, the model suggests that treatment is stratified by APOE ε4 genotype in males (nodes 14 and 15), but not in females. Tree models also provide lists of rules, which summarize the branch path to each final node and its predicted probability. Within our model, the rule associated with the lowest probability of being treated (0.21) is the one including females

F I G U R E 3
Results of the tree model. Each node shows the predicted class (Yes = treated or No = not treated). Color legend indicates the fitted value. Each tree node reports the predicted class, the predicted probability of the class (i.e., of being treated), and the actual percentage of observations in the node belonging to the class. Branches indicate the value of the variable for which the node was split. For example, the first node includes the whole population, split on the basis of sex; node two indicates the female population, where the probability of being treated is 0.45, the predicted class in "No"; node three indicates the male population, where the probability of being treated is 0.63, the predicted class in "Yes"  Figure S2 in supporting information.

Effects of exposure to statins
To examine the effect of statin use on survival, death records captured by the UK Biobank were used. We performed the following analysis on the dataset matched on statin PS, thus including as covariates sex, APOE genotype, AD, dementia diagnoses, and their interactions with statin treatment.
The matched data set included 6622 death events (3170 in statin users and 3452 in non-users). The multivariate Cox regression analysis (Table 3 and Figure S4 in supporting information) revealed that use of statins was not associated with overall higher rates of survival (Pvalue = .206). On the other hand, considering the interaction of statin use with sex, the results suggest higher survival rates in males treated with statins.
As suggested by our analyses, individuals differ in probability of statin use on the basis of strata defined by sex and APOE genotype.
Here we examined whether differences in use of statins have an effect on RT changes.
To assess changes in cognitive patterns measured by RT, we included individuals who had at least two measurements (from two visits) after  (Table 4). Changes in RT were significantly associated with time from baseline (scores worsened in time, as already described in Lyall et al. 39 ) as well as sex; males had worse performance over time. Significant differences (P = .03) were found in RT changes between statin users and non-users when stratified by APOE genotype, as can be seen in Figure 4. We tested the differences in RT variations in time (Slope.yrs) between statin users and non-users in strata (see Figure 4B). Larger slopes indicate faster cognitive deterioration. No significant differences were seen. However, as suggested by Figure 4B, different behav-   Table S4 in supporting information.

DISCUSSION
Statins have greater beneficial effects on cognitive function in APOE ε4 homozygotes, 11 and it has been demonstrated 35  or Alzheimer's Disease Assessment Scale-cognitive subscale (ADAScog). 59 We could not conduct longitudinal analysis of change in cognition in the AD cohort as data were not available. This is most likely due to dropout, as individuals diagnosed with AD are less likely to followup with a study such as the UK Biobank. Nevertheless, our analyses revealed that statin use in APOE ε4 carriers decreases the risk for AD, in alignment with findings from previous studies. 11 Another valuable addition to the analyses presented here would be a further stratification of the patients according to biochemical markers such as cholesterol or triglyceride levels in plasma. Future analyses should include this information.
The UK Biobank has several potential biases: general ones such as the enrollment of a mostly White population, with higher socioeconomic status, and specific for study, including possible selection biases, such as higher rates of depression in females. While the UK Biobank's cohort contains a mixture of prevalent conditions, including dementia and AD, it is important to note that ICD-10-CM codes might not always be accurate, particularly for these types of diagnoses. Furthermore, given the observational nature of the UK Biobank data, our results showing that statin use was associated with lower risk of cognitive impairment in AD among APOE ε4 carriers, needs to be further validated by a randomized clinical trial.
Our evaluation of the aging population in a large-scale cohort from the UK Biobank identified important sex differences related to statin use. Our results suggest that patient stratification that includes APOE genotype and consciousness of sex bias could significantly reduce risk of AD in both men and women.

ACKNOWLEDGMENTS
We acknowledge funders and data providers. Funding for this work was enabled by the National Institute on Aging grants R34 AG049652 "Systems Pharmacology for Predictive Alzheimer's Therapeutics: SysPharmRx-AD" and by P01 AG026572 (to RDB), and by Centre.

CONFLICTS OF INTEREST
The authors have no financial conflicts of interest.

ETHICAL APPROVAL AND CONSENT TO PARTICIPATE
This research has been conducted using the UK Biobank Resource under Application Number 19923 "A precision medicine approach for treatment and prevention of Alzheimer's disease using statins." UK Biobank has approval from the North West Multi-centre Research Ethics Committee (MREC), which covers the UK. All participants have previously provided consent for UK Biobank data and samples to be used for research.

AVAILABILITY OF DATA AND MATERIALS
The data that support the findings of this study are available from UK Biobank. Restrictions apply to the availability of these data, which were used under the application 19923 license for the current study, and so are not publicly available.

AUTHOR CONTRIBUTIONS
Arianna Dagliati performed experiments, analyzed data, and wrote the manuscript draft. Arianna Dagliati, Nophar Geifman, Niels Peek, and Roberta Diaz Brinton designed and developed the study. All authors revised and edited the manuscript.