The goals of this project are to develop, implement, and evaluate computer-based methods that model usual clinical care and then apply those models to detect individual patient care that is anomalous. In the future, such a system may serve as an “safety net” that continuously monitors patient care, as documented in an EMR, and raises an alert when such care appears to be anomalous. An hypothesis of the project is that such anomalies correspond to medical errors often enough to make such alerting worthwhile. Within the ICU domain the project is investigating the extent to which this hypothesis is supported.
Research Projects and Collaborations
This project is developing and evaluating a probabilistic approach to disease surveillance. The goal of the research is to improve the ability of public health officials and physicians to estimate the current incidence of influenza and other infectious diseases and to predict the future course of epidemics of those diseases. This improved information is expected to better support decisions made by health departments to control epidemics, which is expected to reduce morbidity and mortality from epidemic diseases.
Much of science consists of discovering and modeling causal relationships that occur in nature. There is a pressing need for methods that can efficiently infer causal networks from large and diverse types of biomedical data and background knowledge. This project is developing, implementing, and evaluating an integrated set of tools that support the discovery and sharing of causal knowledge from very large and complex biomedical data, including both observational and experimental data. Areas of investigation include the discovery of the genomic drivers of cancers and of the cell-signaling pathways in those cancers.
This project isinvestigating a novel approach to the problem of detecting multivariate statistical differences across groups of data, which arises in a wide variety of settings. Such circumstances occur naturally in observational studies, where, for example, a clinical researcher may observe a difference in the prevalence of a condition between two groups of patients and would like to explore the reasons behind the difference. Another example is comparative effectiveness research, where it is of interest to understand an observed difference between two clinical treatment approaches.
This project is applying existing Bayesian rule learning methods to high-throughput molecular data (e.g., proteomic and genomic data) to perform disease prediction and biomarker discovery. Datasets being analyzed include those in the domains of lung cancer, breast cancer, amyotrophic lateral sclerosis, and other diseases.