Common analysis pitfalls in longitudinal ctDNA studies: lead time, sensitivity and immortal time bias.

Abstract

Circulating tumour DNA (ctDNA) has emerged as one of the most promising biomarkers in oncology, with potential applications from detection of minimal residual disease (MRD) and diagnosis to prognosis, treatment guidance, and response monitoring. This personal view critically examines recurring methodological and statistical pitfalls in ctDNA studies with longitudinal measurements. We highlight important issues with lead time definition and demonstrate that performance metrics like sensitivity and specificity are inherently time-dependent. Furthermore, we show why immortal time bias arises when classifying patients incorrectly as always positive, based on becoming positive during surveillance in MRD related studies. With a combination of toy data examples, simulations, and re-analysis of published data we provide practice-oriented guidance to support the design, analysis, and reporting of ctDNA trials. This viewpoint paper aims to sensitise the community and raise awareness of pitfalls in longitudinal ctDNA analysis, rather than offer definitive solutions-more evidence is needed first.