Transportability of model-based estimands in evidence synthesis

Area of Science:

• Biostatistics
• Health Technology Assessment
• Evidence Synthesis

Background:

• Effect modifiers typically describe individual-level treatment effect heterogeneity.
• Marginal effect estimates are crucial for population-level health technology assessment decisions.
• Noncollapsible measures can be influenced by prognostic variables, impacting marginal effects.

Purpose of the Study:

• To explore the implications of effect modification and measure collapsibility in evidence synthesis.
• To address the challenges in obtaining reliable marginal effect estimates for population-level decision-making.
• To provide recommendations for improving the transportability of effect estimates across studies.

Main Methods:

• The study discusses theoretical concepts of effect modification and measure collapsibility.
• It analyzes the impact of individual-level versus population-level effect measures.
• The implications for statistical modeling and covariate adjustment in evidence synthesis are examined.

Main Results:

• Purely prognostic variables can modify marginal effects even with individual-level treatment effect homogeneity for noncollapsible measures.
• Unadjusted indirect comparisons may be biased due to imbalances in covariate distributions.
• Covariate adjustment has limitations for noncollapsible measures without individual patient data.

Conclusions:

• Directly collapsible measures enhance the transportability of marginal effects between studies.
• Using directly collapsible measures reduces reliance on covariate adjustment when treatment effects are homogeneous or covariates are balanced.
• Directly collapsible measures aid in selecting appropriate baseline covariates for adjustment when treatment effects are heterogeneous.