An XGBoost machine learning model achieved AUC=0.911 for predicting hematoma expansion in hypertensive intracerebral hemorrhage patients with MAFLD, with SHAP analysis revealing non-linear, synergistic statistical associations between hepatic fibrosis, systemic inflammation, and local hemodynamic parameters.
The XGBoost model achieved the best discrimination among five predictive algorithms for hematoma expansion in HICH-MAFLD patients.
SHAP analysis revealed non-linear, synergistic statistical associations between hepatic fibrosis, systemic inflammation, and local hemodynamic parameters in predicting hematoma expansion.
Metabolic dysfunction-associated fatty liver disease (MAFLD) was identified as a comorbidity that exacerbates hematoma expansion risk in hypertensive intracerebral hemorrhage.
The FIB-4 index (a marker of hepatic fibrosis severity) was identified as a key predictor variable with synergistic interactions with other risk factors for hematoma expansion.
The SHAP-integrated XGBoost framework was proposed as a tool for individualized risk stratification of hematoma expansion in HICH patients with MAFLD.
This research suggests that patients who have both a type of brain bleed called hypertensive intracerebral hemorrhage (HICH) and a liver condition called metabolic dysfunction-associated fatty liver disease (MAFLD) face a particularly high risk of their brain bleed getting worse (called hematoma expansion). The researchers used data from 529 such patients to build and compare five different computer-based prediction models. The best-performing model — called XGBoost — was highly accurate, correctly identifying high-risk patients about 91% of the time based on the AUC score. The researchers then used a technique called SHAP to understand which factors the model relied on most and how they interacted with each other. The SHAP analysis revealed something notable: the liver-related factors (specifically a measure of liver scarring called the FIB-4 index and a marker of body-wide inflammation called hs-CRP) did not simply add to the brain bleed risk in a straightforward way. Instead, they appeared to dramatically amplify the risk already created by local factors like the initial size of the brain bleed and blood pressure — in what the researchers describe as a potential 'liver-brain axis.' This suggests that the metabolic and inflammatory changes caused by fatty liver disease may interact with the bleeding process in the brain in complex, compounding ways. This research suggests that doctors managing patients with both a brain bleed and fatty liver disease may benefit from tools that account for liver health and inflammation markers, not just traditional stroke factors like blood pressure and bleed size. The development of a specific, interpretable machine learning model for this combined patient group could support earlier identification of those at greatest risk of worsening, potentially helping guide more tailored monitoring and treatment decisions.
Yang Z, He J, Song Y, Meng Q. (2026). Metabolic dysfunction-associated fatty liver disease exacerbates hematoma expansion in intracerebral hemorrhage: an explainable machine learning approach.. Frontiers in endocrinology. https://doi.org/10.3389/fendo.2026.1912965