Predicting delayed cerebral ischemia after subarachnoid hemorrhage using physiological time series data.

Area of Science:

• Neurology
• Medical Informatics
• Machine Learning

Background:

• Delayed cerebral ischemia (DCI) is a major complication following subarachnoid hemorrhage (SAH).
• Current DCI prediction tools often rely on admission imaging, which may have limitations in precision.
• Accurate prediction of DCI is crucial for timely intervention and improved patient outcomes.

Purpose of the Study:

• To develop and validate a prediction model for DCI after SAH.
• To investigate the utility of temporal unsupervised feature engineering from physiological time series data.
• To compare the performance of the novel model against standard clinical grading scales.

Main Methods:

• Utilized a dataset of 488 SAH admissions, with models trained on 80% and validated on 20%.
• Extracted features from physiological time series (blood pressure, heart rate, respiratory rate, oxygen saturation) using an unsupervised random kernel approach.
• Trained and validated classifiers including Partial Least Squares and Support Vector Machines on derived and combined feature sets.

Main Results:

• The prediction model using random kernel-derived physiologic features achieved an AUC of 0.74.
• A combined model incorporating baseline data and physiologic features reached an AUC of 0.77.
• The novel approach demonstrated significantly higher classification accuracy compared to standard grading scales (AUC 0.58-0.60).

Conclusions:

• Unsupervised feature engineering from high-frequency physiological time series data offers a computationally inexpensive and accurate method for DCI prediction.
• This approach surpasses the predictive performance of traditional methods based on demographics and standard grading scales.
• The developed model holds promise for enhancing DCI risk stratification and guiding clinical management after SAH.