Deficiencies of Rule-Based Technology-Generated Antibiograms for Specialized Care Units.
David M HillLorraine A TodorPublished in: Antibiotics (Basel, Switzerland) (2023)
The objective of this study was to compare the pathogens and susceptibilities of the current automated, rule-based technology (RBT) antibiogram with one manually collected through chart review with additional rules applied. This study was a two-year, retrospective cohort study and included all bacterial cultures within the first 30 days from patients admitted to a single Burn Center. The current RBT antibiogram served as the control, and new antibiogram versions were created using additional rules and compared to the control. Six-hundred fifty-seven patients were admitted (61% excluded for lack of cultures). 59% had at least one hospital-acquired risk factor, with over one-third having recent illicit drug use and one-third having a recent hospitalization. Of the 410 cultures included, 57% were Gram-negative, and half were from wound infections. Sensitivities were significantly different when comparing the manual and the RBT version after including factors such as days since admission, presence of hospital-acquired risk factors, or previous antibiotic courses. Recommended empiric Gram-negative antibiotics changed from double coverage to a single β-lactam with >90% susceptibility. The susceptibilities between the first and subsequent courses were dramatically different. Before developing an antibiogram or interpreting the output, it is important to consider which automated criteria are utilized, especially for units with extended lengths of stay.
Keyphrases
- gram negative
- multidrug resistant
- risk factors
- healthcare
- end stage renal disease
- palliative care
- machine learning
- chronic kidney disease
- emergency department
- deep learning
- high throughput
- prognostic factors
- newly diagnosed
- peritoneal dialysis
- quality improvement
- electronic health record
- health insurance
- antimicrobial resistance