The goal of this project to develop and evaluate a learning electronic medical records (L-EMR) system that draws a physician's attention to the right data, at the right time. It learns how to do so by analyzing patterns of patient data access f many physicians in many past cases in the EMR, and learning which EMR data to highlight that are relevant for making clinical decisions in a given patient. The hypothesis underlying this research is that the L-EMR system will have sufficiently high precision and recall in highlighting relevant data, decrease the average time to assess an intensive care unit (ICU) patient case, and be judged by critical care medicine (CCM) physicians to be clinically useful.
The first aim of this project is develop a highly-usable L-EMR user interface. The L-EMR user interface will include zoomable time-series displays of lab-results, med-orders, and vital signs. Usability studies of the L-EMR user interface will guide revisions and enhancements.
The second aim of the project is to train statistical models that can be applied to a patient case to predict relevant lab-results, med-orders, and vital signs. We will enlist CCM physicians to review a set of retrospective ICU patient cases on a focused set of clinical conditions. Participants will review these cases as if they were active patients, identifying relevant lab- results, med-orders, and vital signs. We will train and evaluate statistical models to predict relevant data, and identify the best performing algorithm to include in the L-EMR system.
The third aim of the project is to evaluate the L-EMR system. We will recruit CCM physicians to evaluate an L-EMR system based on user interfaces from Aim 1 and statistical models trained using the best performing algorithm in Aim 2 to highlight relevant data items. We will measure the precision and recall of the data-highlighting functionality for assessing patient cases and making clinical decisions (e.g., lab and medication orders), the time required to assess cases with and without the highlighting, and physicians' assessments of the strengths and weaknesses of the L-EMR system. If the results of these experiments are positive, as anticipated, this project will introduce a computational method that has significant potential to improve future EMR systems and enhance patient care.
The purpose of this research is to develop and evaluate a learning electronic medical records (EMR) system that draws a physician's attention to the right data, at the right time. The system works by analyzing patterns of EMR usage of physicians, and learning which EMR data to highlight that are relevant in a given patient. The main idea underlying the approach is that patterns of past EMR usage patterns can be exploited to selectively highlight clinically useful patient data.
Showing the most recent 10 out of 15 publications
