Abstract
This dissertation studied how adherence to metformin therapy, as a first-line medication for patients withtype 2 diabetes mellitus (T2DM) diagnosis, affects glycemic control and racial disparities in the United
States. It adopted a three-paper design and integrated two nationally representative survey datasets,
conducted a systematic review, and performed simulation modeling. Employing Andersen’s Behavioral
Model and the WHO adherence framework, these three papers conceptualized adherence as behavior that
can be changed in a broader socioeconomic and structural context and is not just an individual choice.
The first paper combined 10 years of data from 2010 to 2019 from the Medical Expenditure Panel Survey
(MEPS) to estimate national adherence to metformin monotherapy among U.S. diabetic adults. Adherence
was measured using the Proportion of Days Covered (PDC >= 80). Survey-weighted estimates and
logistic regression models examined racial disparities and their predictors. Including more than 6,000
adults (representing over six million patients), adherence was not significantly different by race after
adjustment. Income, insurance, age, comorbidities, and the number of medications were predictors. When
patients with no medication included, racial differences became more obvious. These findings suggest
that racial gaps in metformin monotherapy and adherence should mainly be explained by socioeconomic
and access to care and not by race alone. It supports the importance of addressing structural predisposing
factors in policy.
The second paper is a systematic review and random-effects meta-analysis of published studies. It
quantified the effect sizes of adherence to metformin monotherapy, as measured by PDC or Medication
vii
Possession Ratio (MPR), and glycated hemoglobin (HbA1C) among diabetic adults. From 1,997 screened
records, eight studies met qualitative criteria. Four estimates pooled in the meta-analysis showed that
adherence (PDC >= 80) is associated with a 0.4 percentage-point decrease in mean HbA1C. The
heterogeneity was moderate with no evidence of publication bias or effect modification by age or sex.
This effect size is clinically meaningful and serves as a parameter for translating adherence improvements
into glycemic control improvements at the population level.
The third paper combines adherence distribution from MEPS (2011-2018), HbA1C distribution from the
National Health and Nutrition Examination Survey (NHANES) (2011–2018), and the pooled effect size
from the meta-analysis. Counterfactual scenarios with higher metformin adherence (i.e., 60%, 70%, and
the highest observed quantile) were defined, and a Monte Carlo simulation was conducted to generate
estimates and uncertainty. The results suggest that improving adherence might have modest gains. Mean
HbA1C was reduced by about 0.10-0.12 percentage points. The prevalence of HbA1C>=7% was reduced
by about 1-3 percent. The younger, uninsured patients had a better result. However, racial disparities in
HbA1C narrowed slightly even under the highest adherence scenarios. One might conclude that
adherence interventions are necessary but insufficient to close racial gaps, while significant gaps remain
in the social and structural determinants of health.