A random forest machine learning model achieved an area under the receiver operating characteristic curve of 0.99 in predicting prognosis of intracerebral hemorrhage patients after hyperbaric oxygen therapy, with activities of daily living score, CT hemorrhage volume, and Glasgow coma scale score identified as key predictors.
Key Findings
Results
The random forest algorithm yielded the highest predictive performance among all models tested for prognosis in intracerebral hemorrhage patients after hyperbaric oxygen therapy.
The area under the receiver operating characteristic curve (AUC) for the random forest model was 0.99.
Model performance was assessed via 10-fold cross-validation.
Multiple models were compared, including multivariate logistic regression and various machine learning approaches.
The study analyzed data from 401 patients with intracerebral hemorrhage who received hyperbaric oxygen therapy.
Methods
Patients were stratified into a good prognosis group and a poor prognosis group based on the modified Rankin scale, with a majority falling in the good prognosis category.
Good prognosis was defined as modified Rankin scale score < 3 (n = 265).
Poor prognosis was defined as modified Rankin scale score ≥ 3 (n = 136).
The total cohort consisted of 401 patients treated at the First Affiliated Hospital of Jiaxing, China.
Significant differences in various factors were observed between the two prognostic groups.
Results
Six independent risk factors for adverse outcomes were identified in intracerebral hemorrhage patients after hyperbaric oxygen therapy.
The independent risk factors identified were: Glasgow coma scale score, computed tomography hemorrhage volume, activities of daily living score, treatment modality, intraventricular bleeding, and rebleeding.
These factors were identified through multivariate logistic regression analysis.
SHAP (SHapley Additive exPlanations) values were used to aid in model interpretation and identify individual risk factor contributions.
These factors were used to develop a logistic regression-based nomogram for clinical use.
Results
Activities of daily living score, computed tomography hemorrhage volume, and Glasgow coma scale score were identified as the key predictors influencing hyperbaric oxygen therapy outcomes.
These three variables were highlighted specifically as the most influential predictors among all identified risk factors.
SHAP analysis provided insights into how each of these individual risk factors contributed to predictions.
The identification was performed using interpretable machine learning methods to support clinical decision-making.
A logistic regression-based nomogram incorporating these predictors was developed to improve prognostic accuracy.
Results
A logistic regression-based nomogram was developed using the identified independent risk factors to support clinical decision-making for intracerebral hemorrhage patients undergoing hyperbaric oxygen therapy.
SHAP values aided in both model interpretation and nomogram development.
The nomogram was based on the six independent risk factors identified through multivariate logistic regression.
The study aimed to address the challenge of inconsistent results of hyperbaric oxygen therapy by providing a reliable predictive tool.
The overall goal was to improve personalized treatment planning for intracerebral hemorrhage patients.
What This Means
This research suggests that machine learning can accurately predict outcomes for patients who have experienced a brain bleed (intracerebral hemorrhage) and are treated with hyperbaric oxygen therapy — a treatment where patients breathe pure oxygen in a pressurized chamber. The researchers analyzed records from 401 such patients and found that a type of machine learning model called a 'random forest' was nearly perfect at predicting whether a patient would have a good or poor recovery, achieving a predictive accuracy score (AUC) of 0.99 out of a possible 1.0. Recovery was measured using the modified Rankin scale, with about two-thirds of patients (265 out of 401) achieving a good outcome.
The study identified six key factors that independently predicted worse outcomes: the patient's level of consciousness (Glasgow coma scale score), the size of the brain bleed as seen on CT scan, the patient's ability to perform daily activities before treatment, the specific type of treatment received, whether blood entered the brain's ventricles, and whether the patient experienced re-bleeding. Among these, daily living ability scores, CT-measured bleed volume, and the consciousness level score were highlighted as the most important predictors. The researchers also used an interpretability tool called SHAP analysis to explain how each factor contributed to individual predictions, and they created a visual chart (nomogram) that clinicians could use at the bedside.
This research matters because hyperbaric oxygen therapy outcomes for brain hemorrhage patients have been inconsistent and hard to predict, making treatment planning difficult. This research suggests that machine learning tools could help doctors identify which patients are at higher risk for poor recovery, potentially allowing for more personalized and targeted treatment decisions. The nomogram in particular could offer a practical, accessible way for clinicians to estimate a patient's prognosis without requiring specialized computing expertise.
Ding C, Wu J, Zhang G, Feng X, Zhu J. (2026). Development and validation of a machine learning-based predictive model for prognosis in cerebral hemorrhage patients after hyperbaric oxygen therapy.. Medical gas research. https://doi.org/10.4103/mgr.MEDGASRES-D-25-00168