Predicting burn patient mortality with electronic medical records

Although there exists robust literature on mortality-associated factors in burn patients, it is not known how electronic medical records affect outcomes. Using burn injury as a surgical care model of information and communication, we hypothesized that functionality and interoperability of the electronic medical record could serve as determinants of outcome.

We used the state inpatient databases for New York, Washington, California, and Florida for the years 2009 and 2010 for all states, with the additional years of 2012 and 2013 for New York (n = 6,002), and the respective data from the American Hospital Association Information Technology survey. Using International Classification of Diseases, Ninth Revision, codes, we included burn patients and characterized total body surface area burned. We summed the binary answers to questions 1 and 2 and question 3 from the American Hospital Association Information Technology survey to make continuous functionality and interoperability scores. Mortality was predicted using extreme gradient boosting in Python.

In each state in which our models had an accuracy and area under the curve of more than 0.90, electronic medical record functionality but not interoperability was a significant predictor in New York, California, and Florida. Important predictors in each state were, age, duration of stay, total body surface area burned/severity, and total charges. Electronic medical record functionality was more important than all comorbidities except for coagulopathies and electrolyte disorders. Higher functionality scores were associated with mortality (P < .01).

Our data support our hypothesis that electronic medical records may be associated with mortality in burn patients; however, electronic medical records are not having the intended impact on outcomes, and further research needs to elucidate exactly how electronic medical records are being used in clinical settings.