Newswise —
Cerebrovascular accident is one of the most perilous and frequently misdiagnosed health ailments. Individuals of Black and Hispanic ethnicity, females, senior citizens receiving Medicare, and individuals residing in rural regions are less inclined to receive a timely diagnosis to facilitate effective treatment. A recent investigation utilized machine learning techniques and information accessible at the time of admission to the hospital to fashion a prototype that foretells strokes more precisely than existing models.
The study, by researchers at Carnegie Mellon University (CMU), Florida International University (FIU), and Santa Clara University (SCU), appears in the Journal of Medical Internet Research.
Mistakes in diagnosis pose a significant challenge to public health, and preventable fatalities caused by diagnostic errors in stroke are over 30 times more frequent than those caused by myocardial infarction. Identifying stroke is a complicated task due to the presence of multiple ailments that imitate stroke, such as seizures, migraines, and alcohol inebriation. These complexities can result in postponements, which may exacerbate health complications.
The development of an automated screening tool that scrutinizes obtainable data and proposes a diagnosis for stroke carries substantial potential in addressing this challenge. Researchers have turned to artificial intelligence and machine learning methodologies to discern concealed insights from a vast amount of information and produce prognoses for novel patients.
Rema Padman, a Trustees Professor of Management Science and Healthcare Informatics at CMU’s Heinz College, and co-author of the study, explains that "machine learning techniques have been utilized to aid in identifying stroke by interpreting intricate data such as clinical notes and diagnostic imaging results. However, such information may not be promptly accessible when patients are first evaluated in hospital emergency departments, particularly in remote and underserved regions."
Padman and her associates endeavored to devise a stroke forecasting algorithm founded on data that are widely obtainable during a patient's admission. They also evaluated the supplementary value of social determinants of health (SDoH) in prognosticating strokes. SDoH encompasses the conditions individuals are born into, grow up in, reside in, and age in, as well as the catalysts behind these conditions.
The investigation scrutinized over 143,000 hospital admissions of distinct patients in acute care hospitals in Florida between 2012 to 2014. The scientists also examined SDoH data from the American Community Survey by the U.S. Census. Their prototype included factors that healthcare providers and payers typically collect during hospital admission, such as primary demographics (age, gender, race, ethnicity), quantity of chronic illnesses, and primary payer (e.g., Medicare, Medicaid or private insurance).
The researchers' prototype exhibited a high degree of accuracy (with 84% precision in predicting strokes) and sensitivity, surpassing existing measures (which generally overlook up to 30% of strokes). According to the authors, employing the model indicates that it is plausible to anticipate the probability of a patient's condition being a stroke at the moment of hospital admission founded on patients' demographics and social determinants of health that are accessible at the time of admittance, prior to receiving diagnostic imaging or laboratory test outcomes.
Min Chen, co-author of the study and an Associate Professor of Information Systems and Business Analytics in FIU's College of Business, states that "the moderate sensitivity of existing models raises concerns that they miss a significant portion of individuals with a stroke. In hospitals facing a scarcity of medical resources and clinical personnel, our algorithm can complement current models to facilitate promptly prioritizing patients for appropriate intervention."
Xuan Tan, a co-author of the study and a Lecturer in Information Systems and Analytics in the Leavey School of Business at SCU, suggests that "because our model doesn't necessitate clinical notes or diagnostic test results, it might be particularly valuable in confronting the challenges of misdiagnosis when managing walk-in patients with stroke who exhibit milder and atypical symptoms." Tan further suggests that the model could be advantageous in emergency departments of low-volume or non-stroke centers, where providers have limited exposure to stroke on a daily basis, as well as in rural regions where sensitive diagnostic tools are scarce.
The authors of the study acknowledge some limitations of their research. Firstly, since it was a retrospective study, stroke cases were confirmed based on International Classification of Diseases codes, and patient records were not reviewed. Additionally, the authors advise that their algorithm should not be regarded as the gold standard for stroke diagnosis but rather as a tool that complements existing stroke scoring systems employed in hospitals. Lastly, the findings of their study are constrained by the social determinants of health variables that are available in administrative data.
RSS