부산대학교 도서관

Predicting the future risks of stroke for patients is in high demands. In this paper, we proposed a model predictive of risks of stroke in future 1 year's period for patients across all age, all payor, and all disease groups in Maine, using demographics and clinical histories extracted from Electronic Medical Record (EMR) and clinical notes provided by Health Information Exchange (HIE). A retrospective cohort of 180,196 patients and a prospective cohort of 347,504 patients were constructed for model development and validation, respectively. A logistic regression model based on multivariate analysis was built for risk prediction. The model had a c-statistic of 0.887 in prospective testing, resulting in a sensitivity of 0.410 at a positive predictive value (PPV) of 0.262. Integration of this early-warning system into online patient monitoring platforms enables better management of population with chronic conditions.