Machine Learning Algorithm Diagnoses Stroke with 83% Accuracy

Stroke is one of the most frequently misdiagnosed medical conditions, and prompt detection is crucial for effective treatment. Patients treated within an hour of symptom onset have a higher chance of survival and avoiding long-term brain damage. Data reveals that Blacks, Hispanics, women, older adults on Medicare, and rural residents are less likely to be diagnosed within this critical timeframe. Existing pre-hospital stroke scales overlook about 30% of cases. New research has shown that a machine learning (ML) algorithm, utilizing hospital data and social determinants of health data, can diagnose a stroke quickly—before laboratory test results or diagnostic images become available—with 83% accuracy. This finding suggests the possibility of reducing stroke misdiagnosis and enhancing patient monitoring, enabling medical staff to identify stroke patients or those at risk sooner and improving patient outcomes.

Researchers at Florida International University (Miami, FL, USA) developed the ML algorithm for better stroke diagnosis utilizing data from suspected stroke patients, such as age, race, and number of underlying conditions. Social determinants of health (SDoH) are non-medical factors like race, income, and housing stability that influence a wide range of health outcomes. The researchers utilized emergency department and hospitalization records from Florida hospitals between 2012 and 2014, combined with SDoH data from the American Community Survey, to create the ML stroke prediction algorithm. Their analysis included 143,203 unique patient hospital visits. Stroke-diagnosed patients were typically older, had more chronic conditions, and primarily relied on Medicare.

Image: ML algorithm uses hospital data and social determinants of health data to diagnose a stroke (Photo courtesy of Freepik)

With the researchers' ML algorithm, when a patient arrives at a hospital with stroke or stroke-like symptoms, an automated, computer-assisted screening tool quickly analyzes the patient's information. If the algorithm predicts a high risk for stroke, a pop-up alert is triggered for the emergency department team. Current ML methods often focus on interpreting clinical notes and diagnostic imaging results, which may not be available upon patient arrival, especially in rural and underserved communities. This technology is presently undergoing pilot testing in the emergency departments of various prominent healthcare systems.

"As we add more data it's learning data," said Min Chen, associate professor of information systems and business analytics at FIU Business and one of the researchers. "Our algorithm can incorporate a lot of variables to analyze and interpret complex patterns, which will allow emergency department care teams to make better and faster decisions."

Related Links:
Florida International University