Invited commentary: quantitative bias analysis for time-to-event data needs good validation data

Quantitative bias analysis offers an approach to estimating and adjusting for the impact of residual systematic error, yet to date, most methods have focused on binary outcomes, leaving methods for time-to-event outcomes underdeveloped. Misclassification in time-to-event analyses is challenging because both the events and person-time may contain errors, which requires additional bias parameters. Recent work by Ackerman et al. (Am J Epidemiol. 0000;000(00):0000-0000) illustrates how methods for addressing misclassification of person-time outcomes can be used even when validation data are limited by using expert-informed ranges and simulation to define bias parameters. Although such approaches are better than qualitative assessments, they remain limited by lack of validation data, particularly data capturing person-time measurement error. Improving quantitative bias analysis will require emphasis on designing, conducting, and publishing validation studies that can generate generalizable estimates of misclassification parameters. It also requires methodological innovation and accessible software for person-time-based outcomes, including extensions of regression calibration and risk-based adjustment. Expanding quantitative bias analysis methods for time-to-event analyses will improve the validity, transparency, and interpretability of epidemiologic research and help ensure quantitative bias analysis becomes a routine component of study design and analysis.