Disclosures: The researchers report no relevant disclosures.
November 25, 2020
2 min read

Machine learning improves prediction of cerebral ischemia after subarachnoid hemorrhage

Disclosures: The researchers report no relevant disclosures.
You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

We were unable to process your request. Please try again later. If you continue to have this issue please contact

Machine learning models significantly outperformed standard models in predicting delayed cerebral ischemia and functional outcomes at 3 months after a subarachnoid hemorrhage, according to findings published in Neurology.

“After subarachnoid hemorrhage (SAH), delayed cerebral ischemia (DCI) is the biggest contributor to poor functional outcomes,” Jude P.J. Savarraj, PhD, a bioinformatics postdoctoral fellow in the department of neurosurgery at McGovern Medical School, and colleagues wrote. “Previous studies show that several [electronic medical record] parameters, including white blood count panel, measures of coagulation and fibrinolysis, serum glucose and sodium and vital signs (including ECG and BP) are either marginally or strongly associated with DCI and functional outcomes.”

The researchers hypothesized that machine learning models would be able to learn these associations and accurately predict DCI and functional outcomes and outperform standard models.

To test this, Savarraj and colleagues performed a retrospective analysis of outcomes among 451 patients [women, 290; average age, 54 years; median modified Rankin Scale score (mRS) at discharge = 3; median mRS at month 3 = 1] who had a subarachnoid hemorrhage between July 2009 and August 2016. They selected the machine learning model with the best average area under the curve on the training set, using a 10-fold cross-validation approach. The model they used, artificial neural networks, demonstrated a 10-fold cross-validation AUC of 0.78±0.16 on the ”training set,” according to the study results.

The researchers trained machine learning models and standard models to predict DCI and functional outcomes with data collected within 3 days of admission. They compared predictions of standard models with the machine learning model for each outcome measure, including DCI (n = 399), outcome at discharge (n = 393) and outcome at 3 months (n = 240). A clinician prognostication team prospectively predicted the 3-month outcome for 90 patients, which Savarraj and colleagues also compared with the machine learning and standard models.

Machine learning models resulted in predictions with the following AUC curves: DCI = 0.75±0.07 (95% CI, 0.64-0.84), discharge outcome = 0.85±0.05 (95% CI, 0.75-0.92) and 3-month outcome = 0.89±0.03 (95% CI, 0.81-0.94) for 3-month outcomes. Machine learning models outperformed standard models, with improved AUC scores in delayed cerebral ischemia (0.2; 95% CI, –.02 to 0.4), discharge outcomes (0.07±0.03; 95% CI, 0.0018-0.14) and 3-month outcomes (0.14; 95% CI, 0.03 to –0.24). According to the researchers, the physician team’s 3-month outcome prediction performance matched the machine learning model.

“[Machine learning] improves prediction of DCI and functional outcomes compared to standard models. It matches attending physician’s performance in predicting 3-month outcomes,” the researchers wrote. “Their performance must be evaluated in patient cohorts from other centers. In the future, the model can be expanded to include other variables, including imaging and specimen biomarkers to improve performance.”