2025 Theses Doctoral

Data-Driven Discovery of Heterogeneous Treatment Effects of Statin Use on Dementia Risk

Jawadekar, Neal

Alzheimer’s disease and related dementias (ADRD) currently afflicts millions of individuals worldwide, and yet there remain no effective treatments. Meanwhile, emerging evidence suggests maintaining cardiovascular health may be one of the most promising prevention strategies for ADRD. For example, hypertension and hyperlipidemia have both been shown to increase one’s risk of dementia, and as such, a variety of studies have been conducted on the link between antihyperlipidemic agents (i.e., statins) and incident dementia risk. However, studies conducted on the relationship between statins, cognitive function, and ADRD have yielded mixed findings thus far. Although randomized controlled trials (RCTs) have found no relationship between statins and cognition, multiple observational studies suggest statin use may help protect against incident dementia risk.

While there are multiple possible reasons for these mixed findings, prior evidence suggests the effect of statins on dementia may vary across certain patient characteristics (e.g., age and baseline co-morbidities). Yet, subgroup-specific investigations have been largely limited and restricted to only one or a few covariates at a time. Improving knowledge of how multiple covariates combine to influence treatment effectiveness could help support more precision prevention initiatives for dementia, where statins are targeted to those who would most benefit. To address this research opportunity, this dissertation investigates how the effects of statins on ADRD may be heterogeneous at the intersection of multiple covariates.

This dissertation is comprised of three aims: a narrative literature review of prior research studies on statins and dementia, with a particular focus on subgroup-specific effects; an epidemiologic study of the United Kingdom Biobank (UK Biobank) observational cohort to estimate the overall average treatment effect (ATE) and conditional average treatment effects (CATEs) of statin use on ADRD risk, and a classification tree analysis to generate interpretable treatment rules as a function of multiple potential effect modifiers; and a statistical analysis to estimate the expected dementia risk from implementing the current standard of care (i.e., assigning treatment based on a 10-year cardiovascular risk score) as compared to alternative scenarios (e.g., assigning treatment based on the interpretable treatment rules; or assigning treatment based on what was received in the real-world).

The narrative literature review suggested that although prior investigations of heterogeneous effects for statins and ADRD are limited, age and several co-morbidities may serve as important effect modifiers of this relationship. Yet, no prior studies explored how multiple covariates in combination (rather than in isolation) influence the varying effectiveness of statins on incident ADRD. The empirical study, using data from the UK Biobank, found that the controlled direct effect of statins on ADRD was close to the null, although there were also notable heterogeneities observed across multiple covariates. While the simple tree-based treatment rule recommended statins to individuals with high triglyceride levels as well as to those without a college education, there were more than a dozen meaningful interaction terms identified overall, with heterogeneities being observed across additional variables including diabetes, cardiovascular disease, LDL cholesterol, and HDL cholesterol.

Finally, the statistical analysis of the public health impact found that implementation of this data-driven treatment rule would result in fewer dementia cases as compared to an existing treatment rule which is nationally implemented in the United Kingdom (i.e., a rule based on whether an individual’s 10-year CVD risk exceeds 10%). Overall, this dissertation investigated how multiple covariates contribute to the heterogeneity in effects of statins on ADRD in a large observational cohort. Furthermore, the analysis of heterogeneous treatment effects helped identify specific subgroups that, if targeted for statin treatment, could help reduce the overall population burden of dementia.