Single-arm Trials With External Comparators and Confounder Misclassification: How Adjustment Can Fail.

Area of Science:

• Biostatistics
• Epidemiology
• Clinical Trial Design

Background:

• Single-arm trials with external comparators assess experimental drugs in rare diseases.
• Regulatory agencies are exploring expanded use of these studies.
• Existing guidance focuses on outcome misclassification, neglecting confounder misclassification.

Purpose of the Study:

• To illustrate how misclassified confounders can bias study estimates.
• To provide quantitative examples of bias using smoking misclassification in claims data.
• To offer a tool for calculating bias in adjusted estimates.

Main Methods:

• Utilized causal diagrams to demonstrate bias from misclassified confounders.
• Employed plausible misclassification values for smoking in pharmaceutical claims data.
• Developed a bias calculation tool for adjusted estimates.

Main Results:

• Adjustment for misclassified confounders can increase bias when misclassification differs between data sources.
• Single-arm studies are particularly vulnerable to this bias due to perfect confounder-treatment alignment.
• Bias can occur even with strong confounder-data source associations if misclassification varies.

Conclusions:

• Differential confounder misclassification must be considered in single-arm trial design.
• Subsample validation and bias correction methods are vital for integrating trial and real-world data.