Skip to Content

Discovering Patterns in Large Datasets

Timely detection of severe patient conditions or concerning events and their mitigation remains an important problem in clinical practice. This is especially true in the critically ill patient. Typical computer-based detection methods developed for this purpose rely on the use of clinical knowledge, such as expert-derived rules, that are incorporated into monitoring and alerting systems. However, it is often time-consuming, costly, and difficult to extract and implement such knowledge in existing monitoring systems.

Our work in this project offers computational, rather than expert-based, solutions that build alert systems from data stored in patient data repositories, such as electronic medical records.

Our approach uses advanced machine learning algorithms to identify unusual clinical management patterns in individual patients, relative to patterns associated with comparable patients, and raises an alert signaling this discrepancy.

Our preliminary studies provide support that such deviations often indicate clinically important events for which it is worthwhile to raise an alert. We propose an evaluation based on physician assessment of alerts that are generated from a retrospective set of intensive-care unit (ICU) patient cases. The project investigators comprise a multidisciplinary team with expertise in critical care medicine, computer science, biomedical informatics, statistical machine learning, knowledge based systems, and clinical data repositories.

This research is supported by National Institute of General Medical Sciences - 5R01GM088224-03



Milos Hauskrecht