Constructing predictive models for vaginal surgery in patients with noninvasive gynecological conditions

To develop predictive models for vaginal operative route selection based on clinical variables that can be easily assessed preoperatively in patients with noninvasive gynecological conditions.

Retrospective study.

University Hospital.

Women with routine gynecological surgeries via different approaches.

The medical records of 315 women without prolapse and undergoing hysterectomy, adnexal cystectomy or myomectomy were reviewed. Multiple logistic regression analysis was used to identify factors associated with the vaginal approach for each procedure. Predictive models were generated and optimal cut-off points were identified using the receiver operating characteristic curve.

Predictive models for different vaginal surgical procedures.

For hysterectomy, the patient's body mass index, dysmenorrheal complaints and uterine size were identified as negative predictors for vaginal hysterectomy, whereas previous vaginal delivery was positive. For adnexal cystectomy, adnexal pathology was a negative predictor, whereas previous vaginal delivery and ovarian cyst size were positive. For myomectomy, the body mass index and number of fibroids were negative predictors while previous vaginal delivery was positive. All three models were able to predict the vaginal procedures undergone by women and the areas under the curve were 0.88, 0.95 and 0.92, respectively. Each optimal model cut-off value (logit(p) = 0.53, 0.36, 0.73) resulted in good sensitivity (92.3%, 100% and 87.5%, respectively) and specificity (77.8%, 88.6% and 90.9%, respectively).

These predictive models, which used clinical variables that can be easily assessed preoperatively, may help surgeons to select candidates for different vaginal procedures.