Reliability maps for longitudinal monitoring: a probabilistic framework

INTRODUCTION: In longitudinal monitoring, the interpretation of intra-individual variation traditionally relies on the reference change value (RCV), which integrates within-subject biological variability (CVi) and analytical imprecision (CVa). However, the RCV’s static, binary approach does not allow the explicit estimation of the reliability with which a persistent change is detected over time, nor does it allow the assessment of the operational impact of analytical quality on monitoring.METHODS: Longitudinal monitoring was reformulated in terms of signal-to-noise discrimination. A Monte Carlo simulation based on time-series data from virtual patients was developed, using HbA1c as the model analyte. A sequential observer (Kalman filter) was used to estimate the probability of detecting a change under controlled conditions and to compare the system’s performance with that of the asymmetric RCV model.RESULTS: The system’s performance was summarized in reliability maps, where detection probability depends on the interaction between change magnitude and analytical imprecision. For HbA1c, the Minimum Detectable Delta (MDD), i.e. the change detectable with ≥90% probability under ideal conditions, was about 5.6% in patients with low variability and up to 13.5% with high variability. The framework showed higher sensitivity than the asymmetric RCV model in the present simulations, reflecting the benefit of integrating multiple sequential observations. It also defines a critical CVa, the maximum analytical imprecision compatible with a specific monitoring reliability.DISCUSSION: Reliability maps offer a probabilistic view of detectability in longitudinal monitoring, linking analytical quality to detecting clinical changes. They extend RCV’s role, aiding in monitoring reliability assessment and defining analytical imprecision aligned with clinical goals.