The goal of this project is to develop advanced computational, rather than expert-based, solutions to detect anomalous patient-management actions. Our approach works by identifying patient-management actions that are unusual with respect to actions used to manage comparable patients in past and by raising a patient-specific alert when such a patient is prospectively encountered. The main hypothesis investigated in this work is that statistical anomalies in patient management actions correspond to medical errors often enough to justify the alerting. The new approach is designed to complement existing knowledge-based detection and alerting methods that are clinically precise, but costly to build.
Knowledge-based vs. outlier-based error detection. Classical approaches in detection of medical errors utilize a set of expert-defined rules to detect anomalous events. Apart from the fact that these systems often lack well-defined statistical underpinning, adaptation of the systems to a new environment is impossible without direct interaction with a human. On the other hand, statistical methods rely on past data as opposed to carefully extracted expert knowledge. The benefit of the statistical approach is that it is adaptive and can be applied to a wide range of conditions.
Conditional anomaly (outlier) detection. Detection of unusual events becomes an important issue in highly interconnected and computerized environments, mostly because it is demanding both in terms of human labor and capital investment. Anomaly detection provides a set of techniques that are capable of identifying rare (or in other words anomalous) events in in large datasets. However, existing anomaly detection methodology focuses mostly on detection of anomalous data entries in the datasets. This biases anomalies to events that occur with a low prior probability, for example, patients suffering from a rare disease or exhibiting a rare combination of symptoms. In our work we are interested in finding anomalies in outcomes or patient management decisions with respect to patients who suffer from the same or similar condition. To account for the conditional aspect of anomaly detection we have introduced, developed and continue developing a new conditional anomaly detection framework that seeks to detect unusual values for one or a subset of variables (outcomes, decision) given the values of the remaining variables.
Summary of Research Goals:
Collaborators: , Department of Critical Care Medicine,University of Pittsburgh School of Medicine
Publications:
M. Hauskrecht, M. Valko, I.Batal, G. Clermont, S. Visweswaran, G. Cooper.
Conditional Outlier Detection for Clinical Alerting.
Annual American Medical Informatics Association (AMIA) Symposium , November 2010. [Homer Warner Award, AMIA 2010].
I. Batal, and M. Hauskrecht. Mining Clinical Data using Minimal Predictive Rules.
Annual American Medical Informatics Association (AMIA) Symposium , November 2010.
S. Visweswaran, J. Mezger, G. Clermont, M. Hauskrecht, G. Cooper. Identifying Deviations from Usual Medical Care using a Statistical Approach.
Annual American Medical Informatics Association (AMIA) Symposium , November 2010.
I. Batal, M. Hauskrecht. Constructing Classification Features using Minimal Predictive Patterns.
19th ACM Conference on Information and Knowledge Management, November 2010.
I. Batal, M. Hauskrecht. A Concise Representation of Association
Rules using Minimal Predictive Rules. ECML/PKDD, September 2010.
M. Valko, and M. Hauskrecht. Feature importance analysis for patient management
decisions. 13th International Congress on Medical Informatics , Cape Town, South Africa,
September 2010.
I. Batal, L. Sacchi, R. Bellazzi, and M. Hauskrecht. A Temporal Abstraction Framework for Classifying Clinical Temporal Data.
AMIA anual conference , 2009.
I. Batal, L. Sacchi, R. Bellazzi, and M. Hauskrecht. Multivariate Time Series Classification with Temporal Abstractions. In Proceedings of the
Twenty-Second International Florida AI Research Society Conference (FLAIRS 2009), May 2009.
Michal Valko, Gregory Cooper, Amy Seybert, Shyam Visweswaran, Melissa Saul, Milos Hauskrecht. Conditional anomaly detection methods for patient-management alert systems.
Proceedings of the ICML Workshop on Machine Learning in Health Care Applications, 25th International Machine Learning Conference, Helsinki, Finland, 2008.
M. Hauskrecht, M. Valko, B. Kveton, S. Visweswaram, G. Cooper.
Evidence-based anomaly detection in clinical domains.
Proceedings of the Annual American Medical Informatics Association (AMIA) Symposium , 2007.
M. Valko and M. Hauskrecht . Distance metric learning for conditional anomaly detection. In Proceedings of the
Twenty-First International Florida AI Research Society Conference, FLAIRS, 2008.
J. Mezger, G. F. Cooper, M. Hauskrecht, G. Clermont, S Visweswaran.
Detecting Deviations from Usual Medical Care. Annual American Medical Infortmatics Association (AMIA) Conference
The web page is updated by milos.