A cautionary note concerning the use of stabilized weights in marginal structural models

Area of Science:

• Causal inference
• Biostatistics
• Epidemiology

Background:

• Marginal structural models (MSMs) are vital for estimating causal effects of time-varying treatments.
• Time-dependent confounding is a common challenge in these analyses.
• Inverse-probability-of-treatment weighting (IPTW) is a key method within MSMs.

Purpose of the Study:

• To investigate the bias introduced by common stabilized weights in MSMs.
• To compare the performance of common stabilized weights versus standard weights.
• To evaluate the utility of basic stabilized weights for partial treatment history models.

Main Methods:

• Simulation studies to assess bias under different weighting schemes.
• Analysis of real-world data from the Honolulu Heart Program.
• Comparison of causal effect estimates derived from standard, common stabilized, and basic stabilized weights.

Main Results:

• Common stabilized weights can yield biased causal effect estimates when structural models focus on partial treatment history.
• Standard weights did not exhibit the same bias in the investigated scenarios.
• Basic stabilized weights demonstrated protection against bias in models estimating current or most recent treatment effects.

Conclusions:

• The choice of weights in MSMs is critical and can impact causal inference validity.
• Common stabilized weights may not be appropriate for all MSM applications, particularly those with partial treatment history.
• Basic stabilized weights present a viable alternative for mitigating bias in specific causal effect estimation contexts.