Evaluating the effects of pharmaceutical interventions, social policies, and exogeneous shocks on people’s health and behavior

ABSTRACT

Aging individuals tend to suffer from chronic conditions, some of which manifest in midlife (e.g., type 2 diabetes and hypertension) and some in late life (e.g., neurodegenerative disorders). As the global population increases and as people live longer, finding strategies to prevent or delay these diseases has become a key priority. Concurrent advances in public health and biomedicine offer an array of pharmaceutical (e.g., oral drugs, vaccines) and non-pharmaceutical (e.g., preventative and behavioral health measures) solutions. Meanwhile, exogenous shocks such as pandemics also affect the health and well-being of aging and elsewise vulnerable populations (e.g., immunocompromised individuals, multigenerational households). In such circumstances, pharmaceutical interventions may not be readily available, forcing governments to implement socio-behavioral policies (e.g., lockdowns, mask-wearing mandates) and companies to adopt remote and hybrid work practices.

Natural experiments, such as the social isolation induced by the COVID-19 pandemic or incentive-based vaccine distribution programs aimed to bolster vaccine uptake during this time, provide an opportunity to assess retrospectively the effect of federal, state, or local government policies. Similarly, new drug approvals and clinical equipoise in drug prescription guidelines constitute natural experiments; paired with electronic health records (EHR), they offer the possibility to learn which existing treatments could be repurposed to delay neurodegeneration and/or increase longevity—and for whom these repurposed treatments would work best. However, unlike randomized controlled trials, natural experiments suffer from multiple sources of confounding. The use of appropriate causal inference methods can help mitigate confounding bias, including via weighting and regression discontinuity designs.

This thesis illustrates the use of existing causal inference approaches in population health and proposes new methods to evaluate the effects of pharmaceutical interventions (Chapters 1 and 2), exogenous shocks (Chapters 3, 4, and 5), and socio-behavioral policies (Chapters 3 and 5) on the health and well-being of aging and other vulnerable populations. Specifically, Chapters 1 and 2 leverage the target trial emulation framework to study the comparative effectiveness of antidiabetic and antihypertensive drugs towards preventing dementia or delaying its onset, using EHR data from Mass General Brigham healthcare system. Our target trial emulations suggest the diabetes drug metformin and the antihypertensive drug class of angiotensin receptor blockers as potential repurposing candidates for dementia, especially if initiated before age 70. Chapter 3 uses regression discontinuity designs to quantify the benefits of a local vaccine companion program in Massachusetts during the COVID-19 pandemic. We estimate that this initiative may have bolstered vaccine uptake among older adults aged 75+ by up to 22 percentage points. Chapter 4 implements counterfactual time series modeling to estimate pandemic-period excess mortality associated with overdoses in the US, by substance and geography. We find ~25,650 excess deaths nationally (March 2020-August 2021), disproportionately affecting Southern and Western regions of the country and attributable mainly to synthetic opioids, methamphetamines, and alcohol as well as polysubstance use. Chapter 5 characterizes changes in team coordination among knowledge workers at a large global tech company to better understand the rise of hybrid work practices and their potential implications for well-being. Using two-way fixed effect regression models, we find evidence of voluntary alignment of work schedules with managers and greater co-attendance among employees who were recently hired or work in shared office spaces.

Collectively, these five studies demonstrate how we can effectively learn from data about past events, medical records, and office attendance logs, to provide insights that inform the design of future public health strategies to protect aging individuals and otherwise vulnerable populations.

COMMITTEE

Munther A. Dahleh (chair, advisor), Caroline Uhler, Leo Anthony Celi, Maimuna S. Majumder (Harvard), Mark W. Albers (Harvard)

EVENT INFORMATION

Hybrid event. To attend virtually, please contact the IDSS Academic Office (idss_academic_office@mit.edu) for connection information.