Methods to control for unmeasured confounding in pharmacoepidemiology: an overview

Area of Science:

• Pharmacoepidemiology
• Biostatistics
• Health Research Methods

Background:

• Unmeasured confounding poses significant challenges in pharmacoepidemiologic research.
• Various methods exist to address unmeasured confounding during study design or data analysis.

Purpose of the Study:

• To review common methods for detecting and controlling unmeasured confounding.
• To offer recommendations for their appropriate use in pharmacoepidemiology.

Main Methods:

• Design-phase methods include case-only designs (case-crossover, case-time control, self-controlled case series) and prior event rate ratio adjustment.
• Data-analysis methods encompass negative control, perturbation variable, instrumental variable, sensitivity analysis, and ecological analysis.
• Substudy-based approaches involve external adjustment, propensity score calibration, two-stage sampling, and multiple imputation.

Main Results:

• A comprehensive list of methods for managing unmeasured confounding is presented.
• These methods are categorized by their application in study design, data analysis, or through substudies.
• The review details specific techniques within each category, such as case-crossover and instrumental variable methods.

Conclusions:

• The effectiveness of methods for handling unmeasured confounding varies across studies and databases.
• Integrating statistical evidence with clinical knowledge is crucial for interpreting pharmacoepidemiologic study results accurately.