Using machine learning and hospital data, scientists have developed an accurate AI model to detect a stroke.
The group from Carnegie Mellon, Florida International, and Santa Clara Universities published their findings earlier this year in the Journal of Medical Web Research.
The AI Model Contained Healthcare Provider and Payer Data
The purpose of this study was to develop a stroke-prediction algorithm by analyzing more than 143,000 hospital visits of patients in Florida acute care hospitals from 2012 to 2014, and social determinants of health data from the US Census Bureau’s American Community Survey.
Basic demographic information, the number of chronic illnesses, and insurance coverage were all added to their model, which is based on data typically provided by healthcare providers and payers.
The scientists claimed that the model had an accuracy of 84% in predicting strokes, making it more accurate than previously used scales which can miss as many as 30% of cases.
Based on demographics and social determinants of health known at the time of entry, the authors say it is possible to predict the likelihood of a patient’s condition being a stroke at the time of hospital presentation, before obtaining any diagnostic imaging or laboratory test results.
Read more: Advocating For Earlier Breast Cancer Screening For Black Women: Why It’s Important?
Limitations of the Study
The retrospective nature of the study, the fact that stroke cases were confirmed solely using ICD codes without consulting patient records, and the lack of social determinants of health variables in the administrative data were all drawbacks.
They also stressed that their algorithm is not meant to replace established stroke grading systems in hospitals but rather serves as an auxiliary tool.
People of African-American and Hispanic descent, females, the elderly, those on Medicare, and those living in rural regions have much lower rates of timely stroke diagnosis compared to the general population.
Read more: IRS Recommends Free Online Paycheck Withholdings Tool To Prevent Potential 2023 Tax Bill