CIMTx: An R Package for Causal Inference with Multiple Treatments using Observational Data

Area of Science:

• Biostatistics
• Epidemiology
• Computational Statistics

Background:

• Causal inference with multiple treatments from observational data presents significant methodological challenges.
• Existing methods often lack unified implementation and comprehensive tools for addressing core causal assumptions.

Purpose of the Study:

• To introduce CIMTx, a software package designed for efficient and unified causal inference with multiple treatments.
• To provide tools for simulating complex multiple treatment data structures.
• To facilitate the assessment of positivity and ignorability assumptions in causal analyses.

Main Methods:

• Implementation of diverse causal inference methods: regression adjustment, inverse probability of treatment weighting (IPTW), Bayesian additive regression trees (BART), generalized propensity score (GPS) with multivariate splines, vector matching, and targeted maximum likelihood estimation (TMLE).
• Techniques for assessing the positivity assumption, including common support identification using IPTW, BART, and vector matching.
• A Monte Carlo sensitivity analysis framework to evaluate departures from the ignorability assumption.

Main Results:

• CIMTx integrates multiple modern causal inference methods into a single, efficient package.
• The package provides practical tools for data simulation in multiple treatment settings.
• CIMTx offers robust methods for evaluating key causal assumptions, enhancing the validity of inference.

Conclusions:

• CIMTx serves as a valuable resource for researchers conducting causal inference with multiple treatments using observational data.
• The package's features for addressing positivity and ignorability improve the reliability and transparency of causal effect estimation.
• CIMTx promotes the adoption of advanced causal inference methodologies in the analysis of binary outcomes.