Pragmatic trials: ignoring a mediator and adjusting for confounding

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Health Services Research

Background:

• Pragmatic trials compare new treatments to heterogeneous usual care, complicating direct comparisons.
• Standard statistical assumptions may not fully capture complex relationships, including mediators and heterogeneity, in control arms.

Purpose of the Study:

• To assess statistical methods for confounding adjustment in pragmatic trials when underlying relationships are misspecified.
• To evaluate the impact of unacknowledged mediators and heterogeneity on comparative effectiveness estimates.

Main Methods:

• Simulation studies were employed to test various statistical methods.
• Methods assessed include logistic regression, propensity scores, disease risk scores, inverse probability weighting, doubly robust methods, and standardization.
• Simulations manipulated the presence of mediators, confounding, and heterogeneity.

Main Results:

• Misrepresenting the absence of a mediator leads to misleading comparative effectiveness when using conditional causal effect estimation.
• Estimating the marginal causal effect is generally a more robust approach but may not be suitable for all scenarios.

Conclusions:

• Careful consideration of potential mediators and heterogeneity is crucial when analyzing pragmatic trial data.
• The choice of statistical method (conditional vs. marginal causal effects) significantly impacts the validity of comparative effectiveness findings.