Distinguishing Selection Bias and Confounding Bias in Comparative Effectiveness Research

Area of Science:

• Health Services Research
• Biostatistics
• Clinical Epidemiology

Background:

• Comparative effectiveness research (CER) guides treatment decisions using evidence.
• CER studies face challenges, notably bias.
• Confounding and selection bias are distinct phenomena with different impacts on validity.

Purpose of the Study:

• To formally distinguish selection bias from confounding bias in CER.
• To clarify scientific, design, and analysis issues related to bias in CER.
• To address the conflation of "treatment-selection bias" with confounding bias in the literature.

Main Methods:

• Formal distinction between selection bias and confounding bias.
• Analysis of their distinct consequences on internal and external validity.
• Consideration of a study on depression treatment and weight change.

Main Results:

• Confounding bias compromises internal validity; selection bias compromises external validity.
• These biases can occur simultaneously, even when confounding is controlled.
• Distinct statistical methods and covariate information are needed to mitigate each bias.

Conclusions:

• Clear distinction between selection and confounding bias is crucial for valid CER.
• Researchers must use appropriate, distinct methods to control for both biases.
• Future CER studies require careful planning to collect relevant data for both confounding and selection bias control.