TOPLINE:

Researchers developed and validated a machine learning model based on logistic regression that used routine clinical and treatment data to accurately predict in-hospital deaths among patients with acute liver failure and identified major risk factors, such as body temperature and use of certain treatments.

METHODOLOGY:

• Existing prognostic scores for acute liver failure have limited sensitivity and clinical applicability. Researchers conducted a retrospective study to develop and test a machine learning model to predict in-hospital mortality in patients with this condition.
• They identified adult patients with acute liver failure from a database (2008-2022) and used an independent cohort of patients from a Chinese hospital to externally validate the model.
• Two statistical filters (LASSO regression and the Boruta algorithm) were applied to a pool of variables to eventually identify predictors of mortality.
• Researchers trained and compared seven common machine learning approaches — ranging from classic logistic regression and decision trees to advanced ensemble methods and neural networks — to identify the most reliable model. The primary outcome was in-hospital mortality.
• Each model was judged on its ability to discriminate survivors from nonsurvivors (area under the receiver operating characteristic curve [AUC], sensitivity, and specificity), how closely the risk estimates matched real outcomes, potential clinical benefit, and which factors drove its predictions (interpretability via SHAP analysis).

TAKEAWAY:

• The development cohort comprised 1228 patients (median age, 67 years; 41.94% women), of whom 537 (43.73%) died during hospitalization. The external validation cohort included 108 patients, of whom 29 (26.85%) died.
• Logistic regression achieved the most robust performance, with an AUC of 0.802 in internal validation, a sensitivity of 0.671, and a specificity of 0.787. In the external validation cohort, it achieved an AUC of 0.774, a sensitivity of 0.552, and a specificity of 0.873.
• Researchers identified 11 clinically meaningful predictors of mortality, including age, oxygen saturation, and use of sedatives or analgesics, vasopressors, steroids, antibiotics, and others.
• The interpretability analysis identified that the top five predictors of mortality were low body temperature, vasopressor use, older age, use of continuous renal replacement therapy, and use of sedatives or analgesics.

IN PRACTICE:

“Our findings indicate that, for clinical outcome prediction in acute liver failure, simpler and more interpretable models may offer superior robustness and generalizability compared with more complex algorithms,” the authors wrote. “This model is reliable, interpretable, and clinically accessible, offering valuable support for early in-hospital risk stratification and decision-making, potentially complementing existing prognostic tools in ALF [acute liver failure].”

SOURCE:

The study was jointly led by Xuanlin Wu, Qingzhou Song, and Delin Li — all from Guangxi Medical University Cancer Hospital in Nanning, China. It was published online in Digestive and Liver Disease.

LIMITATIONS:

Owing to the retrospective study design, key markers such as imaging or immune tests were not captured. Researchers tested the model in one small hospital group without comparing it directly with standard scores, thus limiting generalizability. Some predictors reflected treatments rather than initial illness.

DISCLOSURES:

The study received support from the Guangxi Natural Science Foundation, the National Natural Science Foundation of China, and the Wu Jieping Medical Foundation. The authors declared having no potential conflicts of interest.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.