Measurement error and bias in real-world oncology endpoints when constructing external control arms.

Area of Science:

• Oncology
• Biostatistics
• Real-World Evidence

Background:

• Randomized controlled trials (RCTs) are the gold standard for treatment evaluation.
• External control arms (ECAs) using real-world data (RWD) are increasingly explored in oncology.
• Accurate measurement of real-world oncology endpoints, like progression-free survival (PFS), is challenging and limits ECA acceptance.

Purpose of the Study:

• To identify and describe key factors contributing to measurement error in real-world PFS (rwPFS) within multiple myeloma (MM).
• To distinguish between misclassification bias and surveillance bias in endpoint derivation and assessment.
• To quantify the impact of these biases on the comparability of rwPFS to trial PFS.

Main Methods:

• Distinguished between misclassification bias (endpoint derivation/ascertainment) and surveillance bias (outcome observation/assessment).
• Described how misclassified progression events and irregular assessment frequencies in MM RWD contribute to these biases.
• Conducted a simulation study to illustrate the behavior of these biases individually and in combination.

Main Results:

• Mismeasured median PFS (mPFS) can be substantially biased by measurement error.
• False positive misclassifications biased mPFS earlier (-6.4 months), while false negatives biased it later (+13 months).
• Irregular assessment frequencies, with correct event classification, showed minimal bias (0.67 months).

Conclusions:

• Simultaneous misclassification and irregular assessments can create bias exceeding the sum of individual biases.
• Understanding endpoint measurement error in RWD is vital for robust ECA construction in oncology.
• Simulations quantifying measurement error impact aid ECA study planning and result interpretation.