UC Berkeley

Targeted Maximum Likelihood Estimation (TMLE) is a semiparametric framework that combines machine learning with efficient influence-function theory to produce doubly-robust, asymptotically efficient estimators while providing valid, data-adaptive confidence intervals. Its plug-in structure allows flexible initial fits (via Super Learner or Highly Adaptive Lasso), followed by a targeted update that corrects bias without sacrificing efficiency. These features make TMLE especially attractive for electronic-health-record (EHR) analyses, where high dimensionality, complex missingness, and near-positivity violations routinely undermine traditional regression approaches. This dissertation advances causal inference with electronic health‑record (EHR) data through three tightly connected chapters that move from methodological innovation, to practical guidance, to a large‑scale clinical application on COVID‑19 and Long COVID.Chapter 1 develops a one-step targeted variance estimator for the causal risk ratio (CRR) in scenarios involving treatment, outcome, and baseline covariates, which requires deriving the efficient influence function for the variance of the IF as the basis for constructing the estimator. Simulation results demonstrate that our proposed variance estimator offers improved performance, achieving coverage closer to the nominal level of 0.95 and a lower Type-I error rate.Chapter 2 moves from methodology to practice, focusing on guidance for real-data applications. Observational analyses of electronic health record data using databases such as the National COVID Cohort Collaborative include unique challenges for researchers seeking causal inferences. We explore several challenges and describe potential solutions. This chapter provides an overview of these challenges and solutions, using both simulated data and real-world data, with the outcome of Long COVID as the running example.Chapter 3 applies the longitudinal targeted learning pipeline to evaluate whether vaccination timing affects protection against Long COVID. In a binary analysis of 519,980 patients with acute COVID-19, those who had received ≥ 2 vaccine doses had a 16% lower 12-month risk of Long COVID than unvaccinated patients (adjusted marginal risk ratio = 0.84; 95% CI 0.81–0.88). Second, in a longitudinal analysis of 1,085,291 patients, we compared Long-COVID risk among individuals infected 1-3, 3-5, or 5-7 months after vaccination while adjusting for high-dimensional baseline and time-varying covariates. Relative to infection 1-3 months post-vaccination, the adjusted marginal risk ratios were 0.93 (95% CI 0.62-1.41) for infection at 3-5 months and 1.06 (95% CI 0.72-1.56) for infection at 5-7 months. Overall, COVID-19 vaccination before infection substantially reduced the risk of Long COVID, and this benefit persisted for at least seven months.