Clinical Predictors of Failed Medical Treatment in Patients with Tubo-ovarian Abscess: External Validation of a Recently Published Risk Score.

Study Objective: To assess the external validity of a recently published clinical risk score estimating the risk of failed medical treatment in patients with tubo-ovarian abscess (TOA) based on 4 clinical variables on admission.

Design: The probability of failed medical treatment predicted from the reference risk score was compared with the observed rates in a retrospective cohort of patients with TOA. Results were assessed using rigorous methods for clinical prediction models.

Setting: Safety-net teaching hospital system in Houston, Texas.

Patients: One hundred and sixty nine consecutive patients admitted with TOA between 2011 and 2018 were included. Some were treated conservatively with intravenous antibiotic agents; others required a drainage procedure.

Intervention: Electronic health records were reviewed and the 4 clinical predictors of failed conservative treatment were captured (age, white blood cell count on admission, abscess size, and presence of bilateral abscess). A clinical risk score was calculated for each patient. The prediction model was created using the risk score in a multivariate logistic regression. Then the calibration, discrimination, and accuracy of the model were evaluated to perform the external validation analysis.

Measurements And Main Results: Among 169 eligible patients, 50.2% were successfully treated with intravenous antibiotic agents and 49.8% needed abscess drainage. Patients undergoing drainage were more likely to be older, be diabetic, to present with elevated white blood cell count and fever, and to have a larger abscess size on univariate analysis. Among the 4 known predictors of drainage, abscess size was found to be the strongest. Significant difference in clinical characteristics was noted between our cohort and the reference cohort, and the model needed recalibration to adjust for these differences. The area under the receiver operating characteristic curve was 0.77 (0.71-0.84) indicating good discrimination. The Brier score was favorable (0.19) and the observed and predicted rates were similar ranging across different risk scores.

Conclusions: Our results provide external validation of a simple clinical risk score predicting failed medical treatment in patients with TOA.