Variability in non-invasive tests: Challenges and implications for monitoring liver disease

Background: Diagnostic and prognostic non-invasive fibrosis tests (NITs) are increasingly repurposed to monitor liver disease. However, their reproducibility and the minimal detectable change (MDC, ie, the smallest change that exceeds expected measurement variability) are poorly characterized. Methods: We retrospectively assessed variability of blood-based NITs and liver stiffness measurement (LSM) by vibration-controlled transient elastography on the same day (repeatability) and across use-cases when metabolic dysfunction–associated steatotic liver disease, fibrosis, or portal hypertension is suspected. Measurements were done within 90 days. Agreement and variability were quantified using intraclass correlation coefficients (ICC), Bland–Altman analyses, and MDC on the original (absolute) and relative (MDC%) scales. Results: Across 6 cohorts, ICCs were generally good to excellent, but within-subject variability remained substantial. Same-day LSM had a pooled MDC% of 24% on the m/s scale, corresponding to an asymmetric kPa-equivalent window of −43% to +54%. For longitudinal lab-based NITs, pooled MDC% was 47% for FIB-4 and 56% for APRI, while absolute MDC was 13–17 percentage points for LiverPRO and 3 percentage points for CORE. Optimistic estimates remained substantial after excluding the 10% most extreme paired differences. Variability increased with biomarker level for LSM, FIB-4, and APRI, whereas MDC windows for probability-based scores were strongly baseline-dependent and followed an inverse U-shaped pattern. When repeat testing was triggered by an elevated initial LSM, regression to the mean accounted for most or all of the observed decrease. Conclusions: Good agreement in NITs does not guarantee high precision for disease monitoring. MDC depends on biomarker level, clinical setting, measurement scale, and selection for repeat testing. Universal thresholds should not be applied uncritically, and repeated measurements, baseline-dependent interpretation, and regression to the mean should guide monitoring decisions.