Causal analysis with longitudinal data and time-dependent exposure using R: A package review

• Traced the evolution of causal inference from Wright's path diagrams in 1920 to the introduction of robust methods like TMLE by Van Der Laan and Rubin in 2006.
• Assessed R packages gfoRmula, ltmle, and Weightit, each implementing G-computation, TMLE, and MSM, for causal effect estimation.
• Identified barriers in biostatistics knowledge translation, such as lack of expertise and user-friendly software.
• Conducted simulations to compare methods across 1,000 iterations, summarizing performance in terms of bias, standard error, and confidence interval coverage.
• Found TMLE to exhibit statistical power and robustness, despite its higher standard error in binary outcome data.
• Recommended the ltmle package for longitudinal causal analysis due to its documentation and robust performance.

Future steps:

• Additional simulation for Bayesian Marginal Structural Model
• Finishing manuscript for publication