Urodynamic prognostic factors for large post-void residual urine volume after intravesical injection of onabotulinumtoxinA for overactive bladder

The aim of this study was to identify factors predicting large post-void residual (PVR) (defined as ≥200 mL), an important unsolved problem, after an intravesical injection of onabotulinumtoxinA in patients with overactive bladder syndrome. The data showed that 133 of 290 patients had a large PVR after treatment. Multivariate analysis found that the baseline 3-day daytime frequency episodes and voiding efficiency were independent predictors for postoperative large PVR. A receiver operating characteristic (ROC) curve analysis showed the following optimum cut-off values: (1) 3-day daytime frequency episodes = 25, which has a ROC area of 0.72; and (2) voiding efficiency = 89%, which has a ROC area being 0.66. The predicted logit transformation of probability of large PVR, logit(p), for a given 3-day daytime frequency episodes (a) and voiding efficiency (b%) can be denoted by logit(p) = −5.18 + 0.07 × a + 0.04 × b, with a cutoff value of logit(p) = 0.34 and a ROC area of 0.79. The median value of the persistent large PVR interval was 5 months. In conclusion, low 3-day daytime frequency episodes (<25) and low voiding efficiency (<89%) are associated with large PVR. Besides, logit(p) <0.34 can be used to predict large PVR for its higher ROC area.

retention in a multivariate logistic regression 14 . However, age and diabetes were not associated with postoperative urine retention 14 . These conflicting results led us to identify the independent factors that can predict large PVR after intravesical onabotulinum toxin type A injection for patients with OAB. In addition, the secondary objective was to estimate the recovery time after the presence of postoperative large PVR and identify factors responsible for recovery.
Univariate analysis revealed the following baseline characteristics that were predictors for post-treatment large PVR: age, daytime frequency episodes, PVR, voiding efficiency and cystometric bladder capacity (Table 1). However, only daytime frequency episodes and voiding efficiency were independent predictors for post-treatment large PVR by backward stepwise multivariate logistic regression analysis ( Table 1).
By multivariate logistic regression analysis, baseline 3-day daytime frequency episodes (coefficient = 0.07, P = 0.005) and baseline voiding efficiency (coefficient = 0.04, P = 0.03) were determined to be the significant independent factors to predict the probability of the presence of large PVR with a constant of − 5.18 (95% confidence interval [CI] = − 8.66 to − 1.69, P = 0.004). Thus, the predicted logit(p) for a given baseline 3-day daytime frequency episodes (a) and baseline voiding efficiency (b%) can be denoted by logit(p) = − 5.18 + 0.07 × a + 0.04 × b. The mean and standard deviation of logit(p) was 0.14 and 1.31, respectively. Based on the ROC analysis, the optimum cut-off values of logit(p) = 0.34 were determined with the area under the ROC curve being 0.79 (95% confidence interval [CI] = 0.68 to 0.90; sensitivity = 67.6%, specificity = 84.4%, Fig. 1C).
The median value of the persistent large PVR interval was 5 months (95% CI = 2.5 to 5.5 months, Fig. 2). The baseline variables including age, gender, comorbidities, overactive bladder symptoms score (OABSS), bladder diary and video-urodynamic variables were not significant factors associated with predicting the persistent large PVR interval.

Discussion
We successfully identified low voiding efficiency (< 89%) as an independent factor for predicting post-treatment large PVR. However, variables such as age, comorbidities (including diabetes mellitus), baseline Qmax, PVR, Pdet.Qmax or bladder contractility index were not predictive factors. Previous studies have shown that age, comorbidities, baseline Qmax, PVR, Pdet.Qmax and bladder contractility index are associated with large PVR after intravesical onabotulinum toxin type A injection [10][11][12][13][14] . Comorbidities such as diabetes mellitus and Pdet. Qmax were not statistically significant in our univariate analysis (Table 1). Although age, baseline Qmax, PVR and bladder contractility index were statistically significant in our univariate analysis (Table 1), these variables failed to remain significant in multivariate analysis ( Table 1). The finding of low voiding efficiency as a predictor is similar to the data reported by Jiang et al. 10 and Hsiao et al. 15 .
Osborn et al. reported that preoperative PVR and bladder capacity were associated with postoperative urine retention in a multivariate logistic regression 14 . It is notable that if we excluded voiding efficiency as a variable in our multivariate backward stepwise logistic regression model, then baseline PVR becomes an independent predictor (odds ratio = 1.01, P = 0.04). This finding would be consistent with the data reported by Osborn et al. 14 . However, cystometric bladder capacity remains insignificant as a predictor. PVR and bladder capacity are components of voiding efficiency. Thus, voiding efficiency should be a better predictor than PVR and bladder capacity 14 .
We also identified a novel predictor (daytime frequency episodes) of post-treatment large PVR (Table 1). Low daytime frequency episodes (< 25 episodes during 3 days) can also predict a high incidence of large PVR (Table 1, Fig. 1A). We excluded patients with detrusor underactivity for intravesical onabotulinum toxin type A injection; thus our OAB patients in this study should have intact bladder contractility 16 . From our finding (Table 1, Fig. 1A), the bladder contractility of OAB patients with high daytime frequency episodes would be less affected by onabotulinum toxin type A injection.
In a previous study 13 , comorbidities (including diabetes mellitus, chronic kidney disease, chronic obstructive pulmonary disease and congestive heart failure in previous study) predicted large PVR (odds ratio = 2.2, P = 0.01) 13 . However, all comorbidities are not a predictor in the current study (odds ratio = 1.20, P = 0.45, Table 1). If we use 4 comorbidities (including variables of diabetes mellitus, chronic kidney disease, chronic obstructive pulmonary disease and congestive heart failure as previous study 13 ), instead of all cormorbidities (Table 1), the variable remains insignificant as a predictor of large PVR (odds ratio = 1.23, P = 0.38). Thus, the impact of comorbidities on the post-treatment PVR requires further clarification.
We also found that the median bladder recovery time for patients experiencing large PVR is 5 months (Fig. 2), despite the intravesical onabotulinum toxin type A injection may remain therapeutically effective in patients for up to 1 year 13 . Besides, from the finding of Fig. 3B, male patients had a faster recovery from large PVR than female patients. These data should be valuable for patient consultations.
A limitation of this study is the retrospective design. However, the large sample size and multivariate backward stepwise logistic regression analysis performed in this study should make our analysis reliable and valuable. In addition, because the ROC areas of daytime frequency episodes and voiding efficiency were low, the proposed cut-off values might be less useful. However, logit(p) had a higher ROC area, thus the proposed cut-off value (0.34) of logit(p) can provide a valuable reference for physicians to select appropriate OAB patients for intravesical onabotulinum toxin type A injection.
In conclusion, low 3-day daytime frequency episodes (< 25) and low voiding efficiency (< 89%) are associated with large PVR. Besides, logit(p) < 0.34 can be used to predict large PVR for its higher ROC area. The median recovery time for large PVR was 5 months. Our findings can be used as an initial guide for intravesical onabotulinum toxin type A injection in patients with refractory OAB.  Table 1. Baseline characteristics and logistic analysis for predicting large post-void residual after intravesical onabotulinum toxin type A injection for overactive bladder (n = 290). Values were expressed using the mean ± standard deviation or number (percentage). OAB = overactive bladder syndrome, CKD = chronic kidney disease, COPD = chronic obstructive pulmonary disease, CAD = coronary arterial disease, CHF = congestive heart failure, CVA = cerebral vascular disease, OABSS = total score of overactive bladder symptoms scores questionnaire, USS = total scores of urgency severity scales questionnaire, PPBC = patient perception of bladder condition, Pdet.Qmax = detrusor pressure at maximum flow rate. †Stepwise backward logistic regression analysis was performed using those variables with P < 0.25 at univariate analysis. R 2 = 0.19.

Methods
This study was a retrospective investigation of 290 OAB patients who were treated with intravesical injection of 100 U of onabotulinum toxin type A (Allergan, Irvine, CA, USA) from 2005 to 2014. All patients received their first onabotulinum toxin type A injection.
The indications of intravesical onabotulinum toxin type A injection were OAB patients who were refractory to antimuscarinic treatment. The duration of previous antimuscarinic treatment was at least 3 months. All patients had been treated with at least two different antimuscarinic agents and were still bothered by severe urgency or urgency incontinence of at least one episode per day. All patients were free of urinary tract infections, intrinsic sphincter deficiency, and neurogenic bladder at the time of study enrolment.
The Research Ethics Committee of Hualien Tzu Chi Hospital approved the study. The corresponding author confirmed that all methods were performed in accordance with relevant guidelines and regulations. This study is a post hoc analysis of a previous study 10 . All patients were informed about the possible adverse events after onabotulinum toxin type A injection, and written informed consent was obtained from all patients before treatment. All patients received intravesical suburethelial injections totalling 100 U of onabotulinum toxin type A in the bladder body and excluding the trigone. The video-urodynamic study was routinely performed for the diagnosis of detrusor over activity, bladder outlet obstruction, and intrinsic sphincter deficiency using Life-Tech urodynamics equipment (Stafford, Texas, USA). Women with bladder outlet obstruction and detrusor underactivity were also excluded from this study. A bladder outlet obstruction was defined as the radiologic evidence of bladder outlet narrowing, a voiding detrusor pressure greater than 35 cmH2O and a Qmax less than 15 mL/s or a voiding detrusor pressure greater than 40 cmH2O 17 . If patients did not have a voiding detrusor contractility of more than 10 cmH2O and needed to void by abdominal straining or were unable to void, then detrusor underactivity was diagnosed 18 . The urodynamic parameters of the first sensation of filling, full sensation, cystometric bladder capacity, bladder compliance, Qmax, PVR, voiding detrusor pressure at Qmax (Pdet.Qmax), bladder contractility index (defined as Pdet.Qmax + 5 × Qmax) 19 , and voiding efficiency (defined as voided volume/ bladder capacity × 100%) 19 were measured and recorded in detail.
A three-day voiding diary 20 , uroflowmetry for Qmax, voided volume and PVR was performed at each visit. PVR was measured by transabdominal ultrasound at each visit to the outpatient clinic. The bladder capacity was derived from the sum of the voided volume and PVR.  The urgency severity scale (USS) was measured using a modified version of the validated Indevus Urgency Severity Scale. The USS rates urgency severity as 0, 1, 2, or 3 and is defined as none, mild, moderate, and severe urgency, respectively 21 . We defined urgency incontinence as an urgency incontinence score of 4. The overactive bladder symptom score (OABSS) 22 was also measured at each visit. OAB-wet was diagnosed by the presence of at least one episode of urgency incontinence in the 3-day voiding diary. The remaining patients were considered OAB-dry.
The therapeutic efficacy was graded using the Global Response Assessment (GRA) and categorized as − 3, − 2, − 1, 0, 1, 2, and 3, indicating markedly worse, moderately worse, mildly worse, no change, mildly improved, moderately improved, and markedly improved bladder symptoms, respectively. The patients were closely monitored by personal interview and reported their subjective perception of the bladder condition at 2 weeks, 1, 3, and 6 months after the onabotulinum toxin type A injection. These results were combined with the improvement of OAB symptoms and any adverse effects after onabotulinum toxin type A treatment. A GRA score ≥ 2 at 3 months after onabotulinum toxin type A injection was defined as a successful treatment 23 .
STATA software (Version 11.0; Stata Corp, College Station, TX, USA) was used for statistical analyses. The variables in the univariate regression analysis included age, gender, OAB-wet or OAB-dry, comorbidities, OABSS, USS scores, 3-day voiding diary and urodynamic variables. Multivariate backward stepwise logistic regression analysis was performed using the variables with P < 0.25 from univariate analysis 24 . A P value of less than 0.05 was considered statistically significant. Receiver operating characteristic (ROC) curve analysis was performed to identify the optimum cut-off value for predicting post-treatment large PVR (defined as ≥ 200 mL). The optimal cut-off value was determined by the point on the ROC curve closest to the upper left-hand corner. In addition, multivariate logistic regression analysis including all significant factors was performed to predict the probability of large PVR (= p) and obtain logit transformation logit(p) 25,26 . With all patients' values derived from the equation of logit(p), ROC curve analysis was performed to identify optimum cutoff value.
The persistent large PVR interval was measured from the date of first documented large PVR during the follow-up visits to the date of documented absence of large PVR and afterwards or last follow-up. The survival curve of the persistent large PVR interval was estimated using the Kaplan-Meier method. The P values of between-group differences were calculated from the log-rank test 27 . A Cox proportional-hazards model was used to assess the factors affecting the persistent large PVR interval.