Use of quantitative bias analysis to evaluate single-arm trials with real-world data external controls

Area of Science:

• Clinical Trials
• Health Informatics
• Biostatistics

Background:

• Real-world data (RWD) is increasingly used as external controls in single-arm trials (SAT) for regulatory submissions.
• Differences in data generation between RWD and SAT can introduce significant biases.
• Quantitative bias analysis (QBA) offers a method to address these challenges.

Purpose of the Study:

• To illustrate the application of quantitative bias analysis (QBA) for evaluating biases in real-world data (RWD) external controls.
• To demonstrate how QBA can be used in the context of advanced non-small cell lung cancer (NSCLC) with molecular subtypes.
• To assess the impact of specific biases on survival analysis.

Main Methods:

• Described sources of bias in oncology comparing RWD to SAT.
• Simulated a dataset for advanced NSCLC survival analysis using a hypothetical immunotherapy agent.
• Illustrated the impact of three biases: missing confounder, exposure misclassification, and outcome evaluation.

Main Results:

• Hazard ratios (HRs) were estimated using conventional analyses for each simulated scenario.
• Bias-adjusted treatment effects and uncertainty were estimated by selecting appropriate bias models and factors.
• Moderate magnitudes of bias were observed, with variable directions across hypothetical scenarios.

Conclusions:

• QBA provides an intuitive framework for bias analysis, enabling critical evaluation of evidence.
• The accuracy of QBA depends on the correct specification of bias models and factors.
• While study design should minimize bias, QBA is crucial for evaluating the impact of unavoidable biases and assessing evidence quality.