Developing a Learning Electronic Medical Record (LEMR) system

Electronic medical records (EMRs) are capturing increasing amounts of patient data that can be leveraged by machine-learning methods for computerized decision support. This project focuses on the development of intelligent EMRs that contain adaptive and learning components to provide decision support using the right data, at the right time.

This work is funded by a R01 grant from the NLM, NIH.

Outlier-based monitoring and alerting

Statistical anomalies in patient management actions may correspond to medical errors. In this project machine-learning methods are used to identify patient-management actions that are unusual with respect to actions used to manage comparable patients in the past. And an alert is raised if such an action is deemed a statistical outlier. This method complements existing knowledge-based detection and alerting methods that are clinically precise, but costly to build. PIs of the project are Milos Hauskrecht, Gilles Clermont and Gregory F. Cooper.

This work is funded by a R01 grant from the NIGMS, NIH.

Personalized modeling for precision medicine

In predictive modeling in medicine, the typical paradigm consists of learning a single model from a database of individuals, which is then applied to predict outcomes for any future individual. Such a model is called a population-wide model because it is intended to be applied to an entire population of future individuals. In contrast, personalized modeling focuses on learning models that are tailored to the characteristics of the individual at hand. Personalized models that are optimized to perform well for a specific individual are likely to have better predictive performance than the typical population-wide models that are optimized to have good predictive performance on average on all future individuals. Moreover, personalized models can identify features such as genomic factors that are specific for an individual thus enabling precision medicine.

Predicting catheter salvage in central line-associated bloodstream infections

Central-line associated bloodstream infections (CLABSI) are the most common pediatric healthcare-associated infection, and are associated with significant morbidity, mortality and cost. CLABSI treatment requires a choice between removal of the central venous catheter and attempted central venous catheter salvage through antimicrobial treatment including long-dwell, high-concentration antibiotic lock therapy. This presents a clinical conundrum when continued central venous catheter use is medically important. Unsuccessful salvage exposes patients to the risk of infection recurrence or subsequent line failure and removal, whereas preemptive central venous catheter removal incurs the risk of invasive procedures and loss of potential future central venous catheter sites. The ability to produce individual predictions of clinically-relevant outcomes in attempted central venous catheter salvage can help the clinician in taking the appropriate action.

Realtime identification of brain ischemia during surgery

Brain ischemia and stroke that occurs during surgery is a devastating complication and with increasing number of surgical procedures this complication is likely to increasing. Intraoperative monitoring of brain activity like routine cardiac monitoring, has the potential to identify brain ischemia. In this project, machine learning models are used to build models that will identify and alert brain ischemia in real time.