Following surgery for pelvic organ prolapse (POP) or stress urinary incontinence (SUI), 2.5-43% of women have post-operative urinary retention (PUR).  Known risk factors for PUR include age >50, lower BMI, advanced POP, baseline voiding dysfunction, previous incontinence surgery, spinal anesthesia, intra-operative fluid administration >750 mL, estimated blood loss >100 mL, post-operative opioid use, and post-operative UTI (1).   The objective of this analysis was to build a statistical model to produce individualized risk predictions of post-operative urinary retention (PUR) for women undergoing pelvic floor reconstructive surgery.

Institutional Research Ethics Board approval was obtained. A retrospective chart review of all women who underwent pelvic reconstructive surgery for SUI or POP by a single urogynecologist at a tertiary referral centre between 2011-2014 inclusive was performed. Seven surgical procedures were included: vaginal hysterectomy, anterior vaginal repair, posterior repair, laparoscopic assisted vaginal hysterectomy, laparoscopic sacrocolpopexy, midurethral sling (trans-obturator) and laparoscopic sling. Patients who failed 2 trials of void on post-operative day 1 (post void volume >150cc with voided volume >200cc) or any patient who required re-catheterization in the first two days after surgery were defined as having PUR.   Potential risk factors studied for PUR risk included age, menopause status, body mass index, number of medical co-morbidities, grade of prolapse (vault, anterior or posterior), pre-operative voiding dysfunction, previous SUI/POP procedures, type of surgery (see above), duration of surgery and estimated blood loss >300cc. 

In univariate analysis, the frequency of pre-operative and intra-operative potential risk factors for PUR were compared using Student's t-test or Chi-square test in these 2 groups with P <0.05 considered statistically significant.  

Logistic regression methodology was then used to develop a nomogram to model the probability of PUR. The first set of covariates considered were those hypothesized to be clinically relevant: age, all  7 types of surgery, pre-op history of voiding dysfunction, diabetes, hypertension, number of other comorbidities.  These predictors were treated as the baseline model, and the remaining predictors were explored to determine if they add additional predictive value. The following process was used to obtain the final prediction model: 1) Imputation: Missing values in the dataset were imputed using Multiple Imputed Chained-Equations (MICE) in order to avoid removing entire cases who only have a few missing measures. This allowed the full collection of patients (N=334) to be used in analysis (mice package in R was used).  2)Best-subsets model selection:  Because of the relatively small number of observations to the number of possible variables to use in the dataset, it was of interest to use a model selection procedure to obtain a subset of the predictors that explain the response the best. This protects the final model from overfitting the observed data, which would have accuracy issues when trying to generalize to future patients. Best-subsets variable selection looks at every possible combination of predictor variables and chooses the subset that fits the best according to the criteria defined by Hastie (3) .  In order to reduce bias, the modeling process was bootstrapped to assess the variability it has from sample to sample.

334 women underwent SUI/POP procedures. The incidence of PUR was 30.8% (103/334). On univariate analysis, comparing the PUR and non-retention groups, there was no statistical difference in mean age, menopause status, mean body mass index, pre-operative voiding dysfunction, high-grade vault or posterior prolapse, duration of surgery and estimated blood loss >300cc.  The PUR group was significantly more likely have a higher grade anterior prolapse [67.0% vs. 52.4%, P =0.004], anterior vaginal repair [76.7% vs 54.8%, P=0.00005], and laparoscopic sling [17.5% vs. 6.9%, P=0.003].  The non-retention group was significantly more likely to have fewer medical co-morbidities [24.7% vs. 14.6%, P =0.02], had previous SUI/POP procedures [27.3% vs. 12.6%, P =0.003], and undergo the following procedures: midurethral sling (trans-obturator) [49.8% vs 37.9%, P=0.04] and laparoscopic sacrocolpopexy [25.0% vs 8.74%, P=0.00006].

A total of 10 predictive variables were considered for the nomogram: all 7 surgical procedures, diabetes, hypertension, medical co-morbidities. Of these, 6 were included to the final model after sequential analysis of each variable’s predictive function: diabetes, medical co-morbidities, laparoscopic sling, anterior vaginal repair, laparoscopic sacrocolpopexy, and vaginal hysterectomy (Figure 1).

This nomogram provides a practical way to obtain individualized predictions of the risk of PUR, with each variable accorded its own number of points (as per the top line). After adding up the points acquired from each variable, drawing a vertical line from the ”Total points” line to the “Probability of PUR” then provides the estimated risk for an individual patient.

This is the first nomogram specifically developed to predict the risk of PUR in women undergoing pelvic reconstructive surgery.  The nomogram is easy to use and provides a visual tool when counselling patients pre-operatively regarding their individual risk of PUR.

Geller EJ. Prevention and management of postoperative urinary retention after urogynecologic surgery. Int J Womens Health 2014;28:829–38.
Niël-Weise BS, van den Broek PJ. Urinary catheter policies for short-term bladder drainage in adults. Cochrane Database of Systematic Reviews 2005, Issue 3. Art. No.: CD004203.
Hastie T, Tibshirani R, Friedman J.  Elements of Statistical Learning, 2009, 2nd ed Springer Verlag.