Developing a class of MSMs to estimate the effect of time-varying treatments on different types of recurrent time-to-event outcomes

Pragmatic trials and comparative effectiveness research often require causal inference methods to establish cause-effect relationships between interventions and outcomes. With treatments that can change over time, it is crucial to account for time-dependent confounders to accurately estimate treatment effects. Statistical methods to do so have not been developed for recurrent time-to-event outcomes, although such outcomes are often observed in children with chronic kidney disease. This project aims to fill this gap by developing a novel class of statistical methods to estimate the effects of time-varying treatments on recurrent event outcomes. A marginal structural model approach will be applied to the proportional rates model, conditional gap time model, and conditional frailty model to estimate both recurrent event rates and risks of subsequent outcomes after the first outcome event. This project will develop theoretical properties of each model and test its performance using Monte Carlo simulation studies. The new models will then be applied to real data from observational cohort studies and electronic health record databases to answer important clinical questions for children with kidney diseases for the first time. Specifically, this project will enable estimation of the effect of time-varying renin-angiotensin-aldosterone system inhibitor dose on the rate of proteinuria remissions among children enrolled in the Chronic Kidney Disease in Children (CKiD) study; the effect of time-varying corticosteroid use on time from one infection-related acute care event to the next event among patients with glomerular disease in the Cure Glomerulonephropathy (CureGN) study; and the effect of time-varying calcium-based and non- calcium-based phosphate binder use on the time from one skeletal fracture to the next fracture among a heterogeneous population of children with chronic kidney disease in PEDSnet. User-friendly statistical software and associated documentation for implementation of the models will be developed to facilitate their use in a wide range of applications. The tools established by this project will open many new avenues of study for analyzing longitudinal data from observational studies, electronic health record databases, and pragmatic trials. The accurate and precise estimation of time-varying treatment effects on recurrent outcomes will inform improvements in clinical care for children with kidney diseases.