Pitfalls in interpreting calibration in comparative evaluations of risk models for precision lung cancer screening

Abstract

Lung cancer screening by low-dose computed tomography reduces lung cancer mortality, but reliable risk-based selection of participants is crucial to maximize benefits and minimize harms. Multiple risk models have been developed for this purpose, and their discrimination and calibration performance is commonly evaluated based on large-scale cohort studies. Using a recent comparative evaluation of 10 risk models as an example, we illustrate the merits, limitations and pitfalls of such evaluations.