Evaluation of an electronic health record adaptation of a 4-year mortality risk index

Abstract

This thesis describes two studies evaluating the utility, discrimination and validity of an outpatient mortality risk score using an electronic adaptation of the Lee 4-year mortality risk model. In our initial study we evaluated the accuracy of the Lee EHR-adaptation (eLee) to predict 2-year mortality. We implemented an EHR-adaptation of the Lee in an outpatient primary care population for fourteen study providers. We calculated patient mortality risk from the date of enrollment and followed patients for up to two years for the outcome of death. Discrimination of this model was determined by calculating the area under the receiver operator characteristic curve (AUC) by Lee risk points. We also applied a Cox proportional hazards model to three condensed stratums of mortality risk denoted as Low, Intermediate and High to represent clinically relevant and potentially actionable levels of risk. Finally, we compared the aggregate percentage of high and intermediate risk patients in each physician’s panel to contrast differences in physician panels. In the second study, we compared the eLee representing our EHR-adaptation to the original Lee survey for biases of information gathering and retrieval, and overall accuracy of survey and electronic data. Our study included an expansion of the functional status questions to emulate the four original specific questions of functional status from the survey. Both the survey and the electronic EHR-adaptation showed excellent correlation with the adjudicated standard.