Reviewed by Rupak Datta, MD PhD, Section of Infectious Diseases, Yale School of Medicine
According to two recent studies of hospitalized patients in the United States and the United Kingdom, electronic antimicrobial stewardship tools may optimize antibiotic prescribing among inpatients.
Graber and colleagues developed a visualization tool that allowed for comparison of antimicrobial use among 8 Veterans Affairs (VA) facilities of similar complexity. The visualization tool consisted of interactive web-based antimicrobial dashboard and standardized antimicrobial usage report. This visualization tool was coupled with a monthly learning collaborative. The authors compared the percent change in average monthly antimicrobial use with a pre-post (2014-2016 vs 2016-2018) design with segmented regression and external comparison with 118 uninvolved control facilities. Intervention sites showed a 2.1% decrease (95% CI: -5.7% to 1.6%) in total antimicrobial use pre-post intervention vs 2.5% increase (95% CI, 0.8% to 4.1%) in control facilities. Reductions were observed in anti-MRSA (absolute difference = 4.7% [11.3% decrease versus 6.6% decrease], p=0.92) and antipseudomonal (absolute difference 7.0% [3.4% decrease vs 3.6% increase, p=0.18) antimicrobial use at intervention versus control facilities. Limitations of this work include the non-random selection of intervention sites, inability to analyze individual components of the visualization tool, and inability to attribute causality to the intervention. Nevertheless, this study underscores the importance of electronic antimicrobial stewardship tools with peer comparison.
Moran and colleagues evaluated the accuracy of an open-source machine learning algorithm to predict antibiotic resistance within 48 hours of hospital admission. Between 2010 and 2016, patients with E. coli, K. pneumonaie, or P. aeruginosa isolated from blood or urine specimens in Birmingham, UK were identified. Their demographic, microbiological, and prescription data were used to train an open-source machine learning algorithm called XGBoost to predict resistance to co-amoxiclav and piperacillin/tazobactam. Predictors of resistance and a point-scoring tool were generated using multivariate analysis and compared to the performance of the original prescribers. Overall, there were 15,695 admissions during the study period. The machine-learning system performed statistically but marginally better (AUC 0.70) than medical staff in the selection of appropriate antibiotics and could have reduced the use of unnecessary broad-spectrum antibiotics by as much as 40% among those given co-amoxiclav, piperacillin/tazobactam or carbapenems. In contrast, the AUC for the point-scoring tools ranged from 0.61 to 0.67 and performed no better than medical staff in the selection of antibiotics. Although validation studies are required, these findings hold promise in the use of machine learning to promote antimicrobial stewardship in hospital settings. Machine learning algorithms may help clinicians predict antimicrobial resistance among hospitalized patients presenting with Gram-negative infections associated with bloodstream or urinary sources.
References
Graber CJ, Jones MM, Goetz MB, et al. Decreases in Antimicrobial Use Associated With Multihospital Implementation of Electronic Antimicrobial Stewardship Tools. Clinical infectious diseases: an official publication of the Infectious Diseases Society of America. 2020; 71: 1168-76.
Moran E, Robinson E, Green C, Keeling M and Collyer B. Towards personalized guidelines: using machine-learning algorithms to guide antimicrobial selection. Journal of Antimicrobial Chemotherapy. 2020; 75: 2677-80.