Causal Multistate Models to Evaluate Treatment Delay.

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Multistate models are valuable for analyzing time-dependent events but are not typically used for causal inference.
• Estimating the causal effects of treatment strategies, especially involving delays, requires robust methodologies.

Purpose of the Study:

• To propose and evaluate a novel estimator combining multistate models with g-computation for causal inference.
• To estimate the causal effect of treatment delay strategies on recovery probabilities.
• To assess the impact of delaying treatment, such as awaiting natural recovery for 3 months.

Main Methods:

• Developed a g-computation-based estimator integrated with an illness-death multistate model.
• Formulated necessary causal and modeling assumptions for identification and estimation.
• Utilized an illness-death model where illness signifies treatment and recovery signifies recovery.

Main Results:

• A simulation study demonstrated that the proposed method offers more efficient data utilization compared to cloning-censoring-reweighting.
• The methodology was applied to real-world data from a cohort of 1896 couples with unexplained subfertility undergoing intrauterine insemination.
• The study estimated the causal effect of treatment delay on recovery in this cohort.

Conclusions:

• The proposed method effectively combines multistate modeling with g-computation for causal inference in treatment delay scenarios.
• The approach provides a more data-efficient alternative to existing methods for analyzing complex event trajectories.
• This methodology has practical applications in reproductive health, specifically in understanding treatment delay effects on subfertility outcomes.