Estimating individual risk of catheter-associated urinary tract infections using explainable artificial intelligence on clinical data

Background: Catheter-associated urinary tract infections (CAUTIs) increase clinical burdens. Identifying the high-risk patients is crucial. We aimed to develop and externally validate an explainable, prognostic prediction model of CAUTIs among hospitalized individuals receiving urinary catheterization. Methods: A retrospective cohort paradigm was applied for model development and validation using data from 2 hospitals and used the third hospital's data for external validation. Machine learning algorithms were applied for predictive modeling. We evaluated the calibration, clinical utility, and discrimination ability to choose the best model by the validation set. The best model was assessed for the explainability. Results: We included 122,417 instances from 20-to-75-year-old subjects. Fourteen predictors were selected from 20 candidates. The best model was the random forest for prediction within 6 days. It detected 97.63% (95% confidence interval [CI]: ± 0.06%) CAUTI positive, and 97.36% (95% CI: ± 0.07%) of individuals that were predicted to be CAUTI negative were true negatives. Among those predicted to be CAUTI positives, we expected 22.85% (95% CI: ± 0.07%) of them to truly be high-risk individuals. We provide a web-based application and a paper-based nomogram for using this model. Conclusions: Our prediction model accurately detected most CAUTI-positive cases, while most predicted negative individuals were correctly ruled out.