Enhancing Stroke Severity Evaluation: A Machine Learning Approach to Mortality Prediction Versus Traditional Scales

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Informatics

Background:

• The National Institutes of Health Stroke Scale (NIHSS) is a standard but limited tool for assessing acute ischemic stroke severity.
• A need exists for more comprehensive and holistic stroke assessment tools.
• Limitations of current scales necessitate the development of novel indicators for patient outcome prediction.

Purpose of the Study:

• To develop and evaluate a novel machine learning-based severity scale for acute ischemic stroke.
• To compare the predictive performance of the novel scale against the NIHSS.
• To utilize the probability of mortality as a key metric for stroke severity and outcome prediction.

Main Methods:

• A machine learning model was trained on 5983 stroke cases from Vall d'Hebron Hospital (2018-2023).
• The model classifies patient outcome (alive/deceased) using diverse variables.
• The probability of belonging to the 'deceased' class was used as the stroke severity metric.

Main Results:

• The machine learning model achieved an AUC-ROC of 87% for outcome prediction.
• The NIHSS scale at 24h yielded an AUC-ROC of 53%, significantly underperforming the model.
• The novel metric demonstrated superior capability in assessing stroke severity and predicting mortality risk.

Conclusions:

• The proposed machine learning approach offers a more holistic and accurate assessment of stroke severity compared to NIHSS.
• This novel metric can aid in identifying high-risk patients for intensive monitoring and resource allocation.
• Quantifying mortality risk enables more targeted interventions, potentially improving patient outcomes in acute ischemic stroke.