1170. Applying Machine Learning Algorithms to Predict Multi-Drug-Resistant Bacterial Infections From Prior Drug Exposure. (26th November 2018)

• Record Type: Journal Article
• Authors: Lendrum, Elizabeth; Haslam, David; Ambroggio, Lilliam
• Is Part Of: Open forum infectious diseases. Volume 5(2018)Supplement 1
• Page Start: S353
• Page End: S353
• Publication Date: 2018-11-26
• Subjects: Communicable diseases -- Periodicals; Medical microbiology -- Periodicals; Infection -- Periodicals; 616.9
• DOI: 10.1093/ofid/ofy210.1003
• Languages: English
• ISSNs: 2328-8957

Abstract

Abstract: Background: Multi-drug-resistant (MDR) infection in the acute care setting prolongs hospital stay and causes high mortality, especially in the pediatric population. Being able to predict MDR infection risk upon or during admission could help prevent and reduce morbidity and mortality in children requiring acute care in the future. This study aimed to develop and validate a predictive model for MDR infection in the pediatric population using machine learning (ML) analysis. Methods: The study population included hospitalized pediatric patients diagnosed with MDR infection between January 1, 2010 and March 8, 2018. All positive cultures during that period were coded as growing either an MDR or non-MDR organism. ML was performed with random forest (RF) analysis to determine whether hospital drug exposure in the 90 days prior to culture was able to accurately classify cultures as positive for an MDR or non-MDR organism. Results: During the study period, 7, 551 positive cultures were defined as MDR out of a total of 26, 913 cultures (28% of all positive cultures). When all cultures were included in the analysis, RF was modestly successful at classifying MDR vs. non-MDR organisms. Significant improvements in classification accuracy were obtained by subdividing cultures based on growth of individual species. RF was able to classify MDR Enterococcus with accuracy = 0.87, positive predictive value of 0.81, and negative predictive value of 0.88. Surprisingly, exposure to many nonantibiotic drugs were important in predicting antibiotic resistance, indicating either that these drugs altered risk directly, or were correlated with MDR risk indirectly. Conclusion: Drugs without known antimicrobial activity were important predictors of MDR status. Nonantimicrobial drug exposure may be a marker for disease types or therapeutic interventions that place patients at higher risk of MDR infection. Monitoring antimicrobial and nonantimicrobial drug exposure may accurately identify patients at highest risk of MDR infection. Disclosures: All authors: No reported disclosures.