Analysis of Alzheimer's disease studies that feature truncated or interval-censored covariates

点击翻译按钮获取中文摘要

SUMMARY More than 6.5 million Americans suffer from Alzheimer’s Disease (AD), and by 2050 this number is expected to double. Yet the development of effective therapies remains an urgent unmet need. In a scenario of highly complex AD pathophysiology and costly and long drug development process, repurposing of drugs approved for other indications is an attractive complementary approach. The rationale for repurposing a drug initially relies on observational studies demonstrating that the cumulative drug exposure is correlated with either a reduction in the risk of developing AD dementia, or with a slowing of the rate of cognitive decline based on serial cognitive evaluations, or with milder AD neuropathological changes at autopsy examination, after adjusting for covariates including age, sex, education, family history of dementia and APOE genotype, and vascular and other modifiable risk factors. However, the vast majority of such published longitudinal studies have ignored the truncation or interval censoring associated with the covariate of interest (e.g., cumulative drug exposure), which is due to either termination of observation by death or non-continuous observation visits in longitudinal studies. Building upon our extensive prior work in the areas of truncation and censoring as well as AD, here we propose to develop methods to more appropriately treat these sampling and measurement problems to avoid bias. We will apply them in two case studies of drugs with opposite purported associations with AD risk -- statins (protective) and proton-pump inhibitors (PPIs, deleterious) -- but mixed findings from longitudinal studies. To this end, we will leverage the strengths of two high-quality publicly available longitudinal datasets: the National Alzheimer’s Coordinating Center (NACC) cohort study and the Harvard Aging Brain Study (HABS). In Aim 1, we propose analytic methods that remove biases arising due to covariate measurements that are truncated in AD studies, such as cumulative statin exposure, including inverse probability weighting, pseudo-observations and reverse regression approaches. In Aim 2, we develop pseudo-observation methods for time-to-event regression with interval-censored covariates. In Aim 3, we conduct and report analyses of NACC and HABS datasets using proposed methods, and develop publicly available R packages for implementation of our proposed methods. Successful completion of these specific aims will produce new statistical methodology that will eliminate the bias that may arise with truncated and interval-censored covariates, which are inherent to longitudinal cohort studies in AD and related dementias. Our proposed research is responsive to the NIA Notice of Special Interest (NOSI): Maximizing the Scientific Value of Secondary Analyses of Existing Cohorts and Datasets in Order to Address Research Gaps and Foster Additional Opportunities in Aging Research (NOT-AG-21-020).