Stacking ensemble learning model to predict 6-month mortality in ischemic stroke patients.

Area of Science:

• Neurology
• Artificial Intelligence
• Biostatistics

Background:

• Acute ischemic stroke requires timely treatment, but guidelines have gray areas regarding reperfusion therapy.
• Predictive models for mortality can aid clinical decision-making in ambiguous cases.

Purpose of the Study:

• To develop a machine learning-based mortality prediction model for acute ischemic stroke patients who are not candidates for reperfusion therapy.
• To evaluate the model's performance in predicting 6-month all-cause mortality.

Main Methods:

• A stacking ensemble learning model was developed, utilizing an artificial neural network as the ensemble classifier.
• Seven base classifiers (K-nearest neighbors, support vector machine, extreme gradient boosting, random forest, naive Bayes, artificial neural network, logistic regression) were employed.
• Clinical data from the International Stroke Trial database was used, selecting variables available at patient presentation.

Main Results:

• The stacking ensemble model demonstrated acceptable performance in predicting 6-month mortality for the target patient group.
• Key performance metrics included an area under the receiver-operating characteristics curve of 0.783, accuracy of 71.6%, sensitivity of 72.3%, and specificity of 70.9% on a validation set.

Conclusions:

• The developed stacking ensemble model provides a valuable tool for predicting mortality in acute ischemic stroke patients not undergoing reperfusion therapy.
• This model can support clinical decision-making in situations not clearly covered by existing practice guidelines.