A precision-constrained framework for evaluating noninvasive biomarkers in MASLD and beyond

Area of Science:

• Hepatology
• Artificial Intelligence in Medicine
• Biomarker Discovery

Background:

• Accurate, non-invasive risk stratification for Metabolic Dysfunction-associated Steatotic Liver Disease (MASLD) is crucial.
• Current methods like FIB-4 have limitations in precision and clinical utility.
• There is a significant unmet need for cost-effective diagnostic tools.

Purpose of the Study:

• To develop and validate a novel neural network for non-invasive risk stratification of MASLD.
• To compare the performance of the neural network against the established FIB-4 index.
• To introduce and apply a new metric, RP-AUC0.5-0.7, for evaluating biomarker performance.

Main Methods:

• Development of a clinical records-based neural network integrating patient history, lab tests, and ultrasound data.
• Internal validation on 209 patients comparing the neural network with FIB-4 using ROC-AUC and RP-AUC0.5-0.7.
• External validation on 194 patients to assess the impact on liver biopsy failure rates for MASH prediction.

Main Results:

• The neural network demonstrated superior performance over FIB-4 across various fibrosis stages (F≥2, F≥3, F=4) in internal validation.
• RP-AUC0.5-0.7 analysis confirmed improved performance of the neural network in a clinically relevant precision range.
• External validation showed a significant reduction in liver biopsy failure rate for predicting at-risk MASH from 86.6% to 50.0%.

Conclusions:

• The developed neural network offers a low-cost, accurate, and non-invasive method for MASLD risk stratification.
• The study introduces RP-AUC0.5-0.7 as a valuable metric for biomarker evaluation.
• This approach provides a generalizable framework for assessing screening biomarkers in clinical practice and research.