Impact of Sleep-Related Factors in Alzheimer’s Disease
Name
harrod-jharrod-phd-hst-2026-thesis.pdf
Size
2.16 MB
Format
Adobe PDF
Checksum (MD5)
386680dc1a7da898c9d3369ed70b805c
Author(s)
Harrod, Jordan B.
Advisor(s)
Brown, Emery N.
Boyden, Ed
Date Issued
May 2026
Publisher
Massachusetts Institute of Technology
Abstract
Alzheimer’s disease affects over seven million Americans, erodes individual autonomy, burdens families with difficult decisions about care, and costs the US healthcare system an estimated $781 billion annually. Identifying modifiable risk factors that could prevent or delay onset has become a key priority, and sleep-related exposures such as sleep apnea and sedative medications have emerged as promising candidates based on large observational studies. However, unlike randomized controlled trials, observational studies of dementia risk factors suffer from a particular form of confounding: the outcome of interest is a clinical diagnosis, which is jointly determined by disease incidence and diagnostic effort. Any exposure that increases a person’s contact with the healthcare system can appear to influence dementia risk, regardless of whether a true biological relationship exists. This dissertation leverages two independent cohorts and multiple analytic frameworks to demonstrate that healthcare utilization is a structural confounder in observational dementia research and to develop methodological approaches for addressing it. Chapters 3 and 4 use longitudinal data from the National Alzheimer’s Coordinating Center and electronic health record data from Mass General Brigham to examine associations between sleep-related exposures and incident AD diagnosis. In the NACC cohort, we employ three complementary Cox regression frameworks to examine the sleep apnea-AD association. A lifetime pattern model finds that developing a clinician-recognized sleep apnea diagnosis during follow-up is associated with a reduced hazard of AD conversion, while time-varying models find that current sleep apnea status elevates risk among males with MCI, with attenuation over follow-up. These seemingly contradictory results reflect the distinction between who gets diagnosed and the instantaneous effect of sleep-disordered breathing. A co-participant analysis finds that informant type moderates the association between nighttime behavioral symptoms and AD risk, consistent with the broader finding that measurement is mediated by the clinical context in which data are generated. In the MGB cohort, we use a target trial emulation framework to estimate the association between benzodiazepine exposure and AD diagnosis. The apparent protective association attenuates substantially upon consideration of healthcare utilization. Having an established primary care provider nearly doubles the hazard of receiving an AD diagnosis, and a prescriber specialty analysis reveals that prescriptions from neurologists and psychiatrists are associated with elevated AD hazard relative to controls, consistent with diagnostic pipelines in conjunction than pharmacological effects driving the association. Collectively, these findings converge on a single methodological argument: the clinical infrastructure through which we observe dementia systematically shapes the associations we find. The methodological contributions of this work are applicable to any observational study in which the outcome is a clinical diagnosis and the exposure is correlated with healthcare engagement.
MIT Department
Harvard-MIT Program in Health Sciences and Technology
Terms of Use
In Copyright - Educational Use Permitted
Copyright retained by author(s)
Persistent DSpace Link
