Predicting long-term outcomes in intracerebral hemorrhage: a comparative study of machine learning models highlights the prognostic value of hematoma clearance.

Four key predictors of long-term functional outcomes after ICH surgery were identified: age, admission GCS score, ICH grade, and hematoma clearance ratio.

• Predictors were selected using Least Absolute Shrinkage and Selection Operator (LASSO) regression.
• The cohort consisted of 185 ICH patients who underwent stereotactic surgery at a single center.
• This was a retrospective cohort study design.
• All four variables were incorporated into predictive models across multiple algorithms.

Hematoma clearance ratio was associated with both 6-month and 12-month functional outcomes following ICH surgery.

• The hematoma clearance ratio was identified as a prognostically relevant variable across both follow-up time points.
• The Youden index suggested a threshold of 67.3% for the hematoma clearance ratio.
• Restricted cubic spline (RCS) analysis indicated an exploratory cut point near 81%, which the authors state 'should be considered hypothesis-generating.'
• The study highlights hematoma clearance ratio as a key finding in terms of its prognostic value for long-term functional recovery.

Logistic regression demonstrated internally validated performance comparable to the evaluated machine learning models in this small-sample setting.

• ML algorithms compared included Gaussian Naïve Bayes (GNB), Linear Discriminant Analysis (LDA), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost).
• Model performance was evaluated using AUC, accuracy, sensitivity, specificity, and other metrics.
• The sample size was limited to 185 patients, which the authors characterized as a 'limited sample size setting.'
• The authors conclude that LR 'may serve as an interpretable research model' in this context.

Older age, higher ICH grade, and lower GCS score were each associated with a lower probability of favorable long-term outcomes.

• These associations were identified across the prognostic modeling framework applied to the 185-patient cohort.
• ICH grade was one of the four LASSO-selected predictors alongside age, GCS, and hematoma clearance ratio.
• GCS was measured at admission.
• These findings are consistent with established clinical knowledge about ICH prognosis.

The study used a retrospective, single-center design with internal validation only, limiting generalizability.

• The cohort included 185 patients from a single center.
• Only internal validation was performed; external validation was not conducted.
• The authors describe this as a 'single-center retrospective cohort.'
• The limited sample size was explicitly noted as a constraint on the comparative evaluation of ML models.