External validation and updating of a model to predict urinary tract infection after urogynecologic surgery


      To externally validate and update a previously published model predicting postoperative urinary tract infection (UTI) after benign hysterectomy in a geographically distinct cohort of women undergoing urogynecologic surgery.

      Materials and Methods

      The validation cohort included 351 women who underwent urogynecologic surgery at an academic health system between January and August 2019. Individuals who underwent sacral neuromodulation, non-pelvic floor surgeries, and repeat surgery within 6 months were excluded. Clinical and surgical characteristics were abstracted from the medical record. Primary outcome was incidence of UTI up to 6 weeks after surgery, defined as: UTI symptoms with positive urine culture, physician diagnosis, or decision to treat with antibiotics per the National Surgical Quality Improvement Program guidelines. Missing predictor values were imputed. Predicted probabilities of postoperative UTI were generated for the validation cohort using the original model described by El-Nashar et al. This model included 4 predictors: cystocele, menopausal status, hormone therapy (HT), and preoperative postvoid residual volume >150mL. Predicted probabilities were compared to actual events to determine model performance. Accuracy was measured using the concordance index (C-index), calibration curves, and the Brier score. Model coefficients and intercepts were subsequently refit to an updated model that included additional clinically relevant candidate predictors which were selected using redundancy analysis and variable clustering. Internal validation of the updated model was performed using bootstrapping to correct for bias. Decision curves were used to identify the range of threshold probabilities in which each model was of greatest value.


      In the validation cohort, the overall postoperative UTI prevalence was 11% (95% CI 8, 15). The original model had low discriminatory ability in the validation cohort (C-index = 0.53) and was miscalibrated. Recalibration alone did not improve model performance. The model coefficients and intercepts were updated to the validation cohort after adding the new predictors of sling procedure, HT, age, recurrent UTI history, postoperative bladder drainage, ASA score, and hysterectomy. Model discrimination improved (C-index = 0.61, 95% CI 0.52,0.68) with Brier score of 0.1 and moderate calibration (Figure 1). Decision curve analysis demonstrated benefit with treatment thresholds of 5 to 45% (Figure 2).


      The original postoperative UTI model required updating to achieve acceptable discrimination and calibration in a cohort of women undergoing urogynecologic surgery. This updated model can be used to identify individuals at high-risk for postoperative UTI and may facilitate targeted prevention strategies.
      Figure thumbnail fx1
      Figure thumbnail fx2