Systematic review of risk prediction models for catheter-associated urinary tract infection.

OBJECTIVE

To systematically review and critically appraise published risk prediction models for catheter-associated urinary tract infection (CAUTI) in adult inpatients.

METHODS

Eight electronic databases were searched from inception to June 30, 2026. Two reviewers independently screened records and extracted data using the CHARMS checklist; risk of bias and applicability were assessed with PROBAST.

RESULTS

Thirteen reports describing 18 prediction models from 12 independent cohorts were included. All were retrospective, with sample sizes ranging from 140 to 122,386 and CAUTI incidence from 1.3 to 24.7%. Logistic regression predominated (11 of 13 reports). Derivation AUC values ranged from 0.611 to 0.993; nine exceeded 0.800. The most frequently identified predictors were age, catheter dwell time, diabetes mellitus, level of consciousness, and serum albumin. All 13 reports were rated high overall risk of bias by PROBAST, driven principally by temporal leakage - the incorporation of time-varying predictors into models intended for use at catheter insertion - insufficient events-per-variable ratios, data-driven variable selection, and near-universal absence of external validation. Only two reports conducted external validation (AUC = 0.758 and 0.970). Calibration was systematically underreported.

CONCLUSION

Derivation AUC values are prone to optimism owing to pervasive methodological deficiencies, and current evidence is insufficient to support clinical implementation of any existing model. Future research should prioritize multicenter external validation over novel model development, with prospective designs, a priori predictor specification, adequate sample sizing, and full TRIPOD+AI-compliant reporting.