Bugs, Drugs and Data: Antibiotic Resistance, Prevalence and Prediction of Bug-Drug Mismatch using Electronic Health Records (EHR) Data

Abstract

Bug-Drug Mismatch (BDM) occurrences are an important and modifiable category of inappropriate antibiotic therapy (IAAT) that increases adverse outcomes for patients and drives overall antibiotic resistance (AR). Surveillance of baseline AR, emerging trends in resistance among priority bacterial pathogens and prevalence of BDM with respect to the age of the patients and the type of health care-setting are required due to differences in antimicrobial need and use in these populations. Additionally, very little is known about the risk factors associated with BDM occurrence.

We performed a retrospective study using de-identified, electronic health record (EHR) data in the Cerner Health Facts™ data warehouse. We assessed antibiotic susceptibility data between the years 2012 to 2017 and visualized the slope coefficient from linear regression to compare changes in resistance over time. We examined the prevalence of BDM for critically important antibiotics and clinically relevant pathogens between the year 2009 to 2017 in four groups of patients: adults; children; children treated in freestanding pediatric facilities and children treated in blended facilities (adults and children). We implemented multiple logistic regression as a reference model to identify risk factors for BDM occurrences and compared the predictive performance measure with 4 machine learning models (logistic regression with lasso regularization, random forest, gradient boosted decision tree and deep neural network).

The trends in resistance rates to clinically relevant antibiotics were influenced by age and care setting. BDM prevalence for several critically important antibiotics differed between children and adults as well as within pediatric and blended facilities. Risk factors such as age of the patient, patient comorbidities and size of the facility were significantly associated with BDM occurrence. Additionally, the machine learning models developed in our study has a high predictive ability (C-statistic), higher sensitivity, specificity, positive predictive value and positive likelihood ratio to identify BDM occurrence than the reference model.

This study describes the utility of data visualization to interpret large scale EHR data on the trends of AR, prevalence and risk factors of BDM which are critical in tailoring antibiotic stewardship efforts to improving appropriate antibiotic prescribing and ultimately reduce AR.