Related Experiment Videos
Development and internal validation of a diagnostic prediction model for advanced liver fibrosis based on routine
Junjie Deng1,2, Wen Song1, Yuyu Ma2
1Clinical Laboratory Center, Xinjiang Medical University Affiliated Traditional Chinese Medicine Hospital, Urumqi, Xinjiang, People's Republic of China.
Introduction:
Advanced liver fibrosis is a clinically important stage of chronic liver disease. We developed and internally validated an interpretable model based on routine laboratory tests and benchmarked it against the aspartate aminotransferase to platelet ratio index (APRI) and 4-factor fibrosis index (FIB-4).
Methods:
This single-center, retrospective, diagnostic prediction study included 1853 inpatients, of whom 988 had report-defined F3 to F4 fibrosis. Primary validation used 10 repeated, stratified 7:3 outer resamples, with preprocessing, least absolute shrinkage and selection operator selection, and model refitting restricted to each development subset. Discrimination, calibration, decision curves, precision recall performance, and linear SHapley Additive exPlanations values were evaluated.
Results:
The leakage-aware pipeline achieved a mean (SD) validation area under the curve of 0.800 (0.015), accuracy of 0.753, sensitivity of 0.775, and specificity of 0.729. The final logistic model included platelet count, total bile acids, C3 and C4, red blood cell count, lymphocyte count, and albumin level. In the representative test set, its area under the curve was 0.809 (95% CI, 0.772-0.845) compared with 0.815 for APRI and 0.821 for FIB-4; neither paired difference was statistically significant after Holm adjustment.
Discussion:
The model showed moderate discrimination and interpretable output but was not superior to APRI or FIB-4. External validation is required before clinical implementation.
