Clinical Performance Evaluation of Demographic-Based Liver Volumetry

Patient demographic-based standard liver volume (SLV) models are widely used to estimate patient-specific hepatic metabolic demand and support future liver remnant assessment after hepatectomy. These formulae rely on anthropometric variables such as weight, height, and body surface area; however, their accuracy for individual patients remains uncertain. In an exploratory single-center study, we evaluated ten published SLV formulae and one in-house multivariable regression model against CT-derived liver volumes as the anatomic reference standard in 100 patients with normal livers, using Bland-Altman analysis. Mean percentage bias ranged from -20% to +11%, with only 4-32% of cases achieving prediction errors within ±5%, and the 95% limits of agreement were wide for every model, spanning from ±25% to ±36% around the mean bias. Because percentage bias is expressed relative to the measured volume, the width of the limits of agreement was almost entirely determined by calibration (r = 0.99 with signed mean bias). Once this scaling was removed, the ten published formulae were indistinguishable, with calibration-adjusted limits of agreement of ±31.3% to ±32.7% (mean ±32.0%, SD 0.40). Rescaling each formula by a single multiplicative constant (0.924 to 1.282) reduced mean bias to between +1.2% and +3.6% and converged the limits of agreement to ±32.4% to ±33.1%, raising the proportion of patients estimated within ±10% from 13% to 54% for the most poorly calibrated formula; refitting both slope and intercept gave no further improvement. The in-house model, using five predictors fitted within this cohort, reached ±29.4% and was the only specification below this floor. In this exploratory single-center cohort, prediction error was wide across the observed range and showed no dependence on predicted liver volume. The limitations for SLV appear to lie in the demographic predictors rather than in the reference measurements. Therefore, future work should focus on predictors that capture information beyond body size, such as measures of body composition or liver function.