Data-Driven Modeling of Usual Clinical Care
Clinical care is complex and often fast paced. Preventable medical errors can and do occur, as has been well documented in recent years. Clinical guidelines and rule-based alerts provide clinical decision support that is intended to reduce medical errors. These methods are driven by expert knowledge. As such, they tend to focus on high impact areas in which medical errors are either prevalent, serious, or both. However, the coverage of such methods is relatively narrow.
We are investigating data-driven methods for helping avoid medical errors. Machine-learning methods are applied to electronic health record (EHR) data to derive computer-based probabilistic models of usual care. These models, which can be complex and time-oriented, represent the probability of various types of care being given for different types of medical conditions. The care of a current patient, as revealed by his or her EHR, is automatically compared to the model of usual care that has been constructed from past-patient cases. If some aspect of the care of the current patient has a low probability, then an alert is sent to the patient’s clinician that indicates the care is unusual.
Milos Hauskrecht, Ph.D., Gregory F. Cooper, M.D., Ph.D., Gilles Clermont, M.D., M.P.H., and Shyam Visweswaran, M.D., Ph.D. are faculty on a project that is investigating these methods in the area of critical care medicine. The project is currently developing and evaluating models and alerts in a laboratory setting. Once this approach is validated in the laboratory, the next stage will be to evaluate it in a limited clinical setting.
Sample of Related Publications:
Batal I, Sacchi L, Bellazzi R, Hauskrecht M. Multivariate time series classification with temporal abstractions. In: Proceedings of the International Florida AI Research Society Conference (FLAIRS) (2009).
Hauskrecht M, Valko M, Batal I, Clermont G, Visweswaran S, Cooper GF. Conditional outlier detection for clinical alerting. In: Proceedings of the Fall Symposium of the American Medical Informatics Association (Nov 2010).
Visweswaran S, Mezger J, Clermont G, Hauskrecht M, Cooper GF. Identifying deviations from usual medical care using a statistical approach. In: Proceedings of the Fall Symposium of the American Medical Informatics Association (Nov, 2010).
Batal I, Hauskrecht M. Mining clinical data using minimal predictive rules. In: Proceedings of the Fall Symposium of the American Medical Informatics Association (Nov, 2010).
Valko M, Hauskrecht M. Feature importance analysis for patient management decisions. In: Proceedings of the International Congress on Medical Informatics (MEDINFO), Cape Town, South Africa (2010)