Causal Inference using Electronic Health Records in Primary Care

Prospective randomized experiments are conducted to evaluate the efficacy (or safety) of a treatment (or an intervention). However, clinical trials are expensive and prone to recruitment challenges especially when detecting rare adverse events. In these circumstances, large observational healthcare repositories, such as those comprised of electronic health record (EHR) data collected in routine primary care, may provide an alternative source to address such causal objectives. The overarching aim of this dissertation is to emulate randomized experiments using a causal framework with methodological considerations for treatment-confounder feedback with informative observation process and subject-specific (time-invariant) unmeasured covariates, high-dimensional propensity score estimation with surrogate covariates, and continuous-time causal inference with marked point processes. First, we applied a longitudinal causal inference framework using calibrated weights with non-overlapping time-intervals to evaluate the effectiveness of glucose-lowering medications using glycemic index of Hemoglobin A1c as the primary outcome. Second, we constructed marginal structural models estimated through calibrated weights obtained via several machine learning algorithms, amalgamated with the SuperLearner method for the estimation of diabetes care provisions. This work was then extended to an article on continuous-time causal inference using optimal weights to assess the long-term adverse effects of glucose lowering medication (e.g., sodium-glucose cotransporter 2 inhibitor) for the recurrent outcome of urinary tract infection. This dissertation follows the idea that events and processes influence each other using the hypothetical (or counterfactual) constructs with some temporal latency between the exposure and the outcome. From a pragmatic perspective, the utility of this dissertation is to describe the estimation of causal effects through specific events and processes available in primary care EHRs.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.