Missingness-aware machine learning using routine laboratory data for distinguishing hepatitis from cirrhosis.

Missingness-aware machine learning models using routine lab data achieve significantly higher discrimination of cirrhosis among hepatitis patients, with AUC up to 0.841.

• Why it matters: Accurate early identification of cirrhosis is crucial for timely management, but conventional scores like FIB-4 and APRI have limited discrimination, and the predictive value of missing data patterns remains underexplored.
• What they did: Researchers developed machine-learning models, including Extra Trees and support vector machines, using 672 patients divided into training and testing sets, incorporating routine lab parameters and missingness indicators, with median imputation for missing values.
• The result: Models with missingness indicators outperformed traditional scores, with the highest AUC of 0.841 and improved sensitivity, suggesting that missing data patterns can enhance hepatitis-to-cirrhosis discrimination in clinical settings.