Affiliation:
1. First Affiliated Hospital of Shantou University Medical College
2. Affiliated Shantou Hospital of Sun Yat-senUniversity
3. Affiliated Jieyang People’s Hospital of Sun Yat-sen University
4. Shantou University Medical College
Abstract
Abstract
Background The prediction of the prognosis of aneurysmal subarachnoid hemorrhage (aSAH) is a complex matter. Although the National Institutes of Health Stroke Scale (NIHSS) has been linked to intracerebral hemorrhage and ischemic stroke, its predictive value with regard to aSAH outcomes is unclear and requires investigation. This study aims to assess whether NIHSS is an independent and valuable predictor of aSAH outcomes and determine whether machine learning model with NIHSS could further enhance the predictive performance.Methods This study involved 1195 patients who experienced ruptured aSAH between 2013 and 2022. Patients from two additional tertiary hospitals were used as external validation. Various statistical learning methods, such as logistic regression (LR), random forest (RF), decision trees, and XGBoost, were utilized to examine the relationship between NIHSS and the modified Rankin Scale (mRS) at 1 month. Machine learning models and logistic regression models were trained to predict functional outcomes using data gathered at the time of admission. Functional outcomes were assessed using mRS for neurologic disability, which was dichotomized into good (mRS ≤ 3) and poor (mRS ≥ 4) outcomes.Results The NIHSS on the first day after aSAH was revealed as an independent predictor of the patient’s 1-month outcome. The NIHSS was an independent predictor of an unfavorable outcome after aSAH (OR, 1.08; 95% CI, 1.04–1.13, P < 0.001). Adding the NIHSS score to the multivariate model significantly improved its discrimination for an unfavorable outcome after aSAH (the receiver operator characteristics curve [AUC], 0.782; 95% CI, [0.746, 0.817] vs AUC, 0.842; 95% CI, [0.805, 0.878]; P < 0.001). Moreover, the machine learning models, including Support vector machine(SVM),XGboost and Random Forest(RF) with AUCs of 0.874, 0.812 and 0.795, respectively, further improved the discrimination for the unfavorable outcome after aSAH.Conclusions The NIHSS is a reliable and straightforward predictor of an unfavorable prognosis for patients with aSAH. Compared to translational LR, the use of machine learning techniques could further improve the performance of the multifactorial model that incorporates the NIHSS for an unfavorable prognosis in patients with aSAH.
Publisher
Research Square Platform LLC