Researchers in the study used machine learning techniques and patient data on arrival at the hospital to develop a model that predicts stroke with greater accuracy than current models.
Research published earlier this year Journal of Medical Internet Researchresearchers say, may help address potentially fatal diagnostic errors and delays in stroke diagnosis.
Stroke is one of the most dangerous and commonly misdiagnosed medical conditions. Black and Hispanic people, women, Medicare patients, and rural residents are all less likely to be diagnosed in time for effective treatment. Similar to stroke.
Death from preventable stroke due to diagnostic error occurs approximately 30 times more frequently than death from myocardial infarction. Automated screening tools that analyze available data and suggest stroke diagnoses could help address this issue.
Artificial intelligence and machine learning can identify hidden insights and generate predictions from large amounts of data. Machine learning methods have been used to detect stroke by interpreting clinical notes and imaging results. However, when patients are first triaged in hospital emergency departments, such information may not be readily available, especially in rural and underserved communities.
Researchers in this study sought to develop a stroke prediction algorithm based on data available at the time of patient admission. Social determinants of health (SDoH) in predicting stroke were also assessed. This includes the conditions under which people are born, grow up, live in and age. They examined her more than 143,000 individual patients who visited acute care hospitals in Florida between 2012 and 2014, along with her SDoH data from the US Census’ American Community Survey. We also incorporated variables routinely collected by providers at the time of admission, including age, gender, race, ethnicity, chronic disease, and major insurance companies.
Their model showed 84% accuracy in predicting stroke and greater sensitivity than existing scales that miss up to 30% of strokes. Therefore, it may be possible to predict the possibility of a stroke at the time of admission, even before the results of imaging and laboratory tests are available.
The researchers found their model to be a small emergency department where providers have limited exposure to daily strokes, and in rural areas with limited access to mild and/or atypical strokes. He said it could be particularly helpful in addressing potential misdiagnosis in symptomatic walk-in stroke patients. Sensitive diagnostic tools or incomplete data collection capabilities.
Because their study was retrospective, confirmation of stroke cases relied on the International Classification of Diseases code rather than patient records. The researchers recommend using the algorithm as a model rather than a standard to complement existing stroke scoring systems used in hospitals.
“The modest sensitivity of existing models raises concerns that they will miss a significant proportion of stroke patients,” said co-author Min Chen, an associate professor at Florida International University, in a statement. “In hospitals where medical resources and clinical staff are scarce, our algorithms will complement current models to help quickly prioritize patients for appropriate intervention.”
