OBJECTIVES: The estimation of dynamic treatment effects is crucial in understanding the real-world impact of interventions that vary based on patient characteristics and clinical history. Traditional methods often focus on static treatment regimes, such as "always treat" or "never treat." However, in practice, clinicians frequently adjust treatments dynamically based on evolving patient conditions, such as A1C levels in diabetes management.

This study explores the use of two advanced causal inference approaches—Longitudinal Targeted Maximum Likelihood Estimation (LTMLE) and Longitudinal Modified Treatment Policy (LMTP)—to estimate dynamic treatment effects. We demonstrate these methods with a synthetic dataset (introducing known (dynamic) treatment effect) and an observational real-world dataset. We also introduce a dashboard that facilitates these analyses for researchers/clinicians.

METHODS: We generated causal static and dynamic treatment effects and estimated mean outcomes under dynamic treatment policies using LTMLE and LMTP methodologies in R. We applied both methods to real-world data to estimate the effect of modifying a treatment based on A1C levels and other clinical events. The data included time-varying covariates and treatment assignments, allowing for a robust analysis of dynamic interventions.

RESULTS: The LTMLE and LMTP approaches consistently recovered the correct static and dynamic treatment effects, demonstrating the feasibility and reliability of these methods in real-world settings. The results showcase the potential of evaluating treatment based on patient-specific (and time-varying) conditions and intercurrent events, opening an avenue of effect estimation beyond only static policies.

CONCLUSIONS: This study highlights the potential of evaluating dynamic treatment policies in real-world settings and the effectiveness of modern causal inference methods. The open-source dashboard developed for this study enables researchers/clinicians to apply these methodologies easily, enhancing their understanding of the impact of dynamic interventions. Future work will focus on expanding the dashboard's capabilities and applying these methods to a broader range of clinical scenarios.