Non-Invasive Clinical Parameters for the Prediction of Urodynamic Bladder Outlet Obstruction: Analysis Using Causal Bayesian Networks

et al. (2014) Non-Invasive Clinical Parameters for the Prediction of Urodynamic Bladder Outlet Obstruction: Analysis Using Causal Bayesian Networks. PLoS ONE 9(11): e113131. doi:10.1371/journal.pone.0113131 Non-Invasive Clinical Parameters for the Prediction of Urodynamic Bladder Outlet Obstruction: Analysis Using Causal Bayesian Networks Myong Kim 0 Abhilash Cheeti 0 Changwon Yoo 0 Minsoo Choo 0 Jae-Seung Paick 0 Seung-June Oh 0 Mauro Gasparini, Politecnico di Torino, Italy 0 1 Department of Urology, Seoul National University Hospital , Seoul , Korea , 2 Department of Computer Science, School of Computing and Information Sciences, Florida International University , Miami, FL , United States of America, 3 Department of Biostatistics, Robert Stempel College of Public health & Social Work, Florida International University , Miami, FL , United States of America Purpose: To identify non-invasive clinical parameters to predict urodynamic bladder outlet obstruction (BOO) in patients with benign prostatic hyperplasia (BPH) using causal Bayesian networks (CBN). Subjects and Methods: From October 2004 to August 2013, 1,381 eligible BPH patients with complete data were selected for analysis. The following clinical variables were considered: age, total prostate volume (TPV), transition zone volume (TZV), prostate specific antigen (PSA), maximum flow rate (Qmax), and post-void residual volume (PVR) on uroflowmetry, and International Prostate Symptom Score (IPSS). Among these variables, the independent predictors of BOO were selected using the CBN model. The predictive performance of the CBN model using the selected variables was verified through a logistic regression (LR) model with the same dataset. Results: Mean age, TPV, and IPSS were 6.2 (67.3, SD) years, 48.5 (625.9) ml, and 17.9 (67.9), respectively. The mean BOO index was 35.1 (625.2) and 477 patients (34.5%) had urodynamic BOO (BOO index $40). By using the CBN model, we identified TPV, Qmax, and PVR as independent predictors of BOO. With these three variables, the BOO prediction accuracy was 73.5%. The LR model showed a similar accuracy (77.0%). However, the area under the receiver operating characteristic curve of the CBN model was statistically smaller than that of the LR model (0.772 vs. 0.798, p = 0.020). Conclusions: Our study demonstrated that TPV, Qmax, and PVR are independent predictors of urodynamic BOO. - Funding: This study was supported by grant No. 04-2010-1080 from the Seoul National University Hospital Research Fund. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing Interests: The authors have declared that no competing interests exist. Urodynamic study (UDS) is considered the gold standard for clinical assessment of bladder outlet obstruction (BOO) in patients with benign prostatic hyperplasia (BPH) [1]. Patients with urodynamic BOO show higher efficacy after transurethral surgery [2]. In this respect, BOO is helpful in stratifying BPH patients eligible for surgical treatment. However, UDS has significant limitations in terms of invasiveness, cost, and morbidity [3]. Numerous attempts have been made to substitute non-invasive clinical parameters for UDS to predict BOO; however, individual variables, including symptom score [4], prostate specific antigen (PSA) level [5], free uroflowmetry (UFM) [6], post-void residual (PVR) urine volume [7], and prostate size [8], have shown a poor to weak correlation with BOO. To improve prediction ability, combinations of non-invasive clinical parameters have been investigated to predict BOO [917]. The statistical methods used for combinations were diverse from the cumulative scoring system [9], to the construction of a formula by logistic regression analysis [1013], to the artificial neural network (ANN) models [1417]. However, these attempts had limited predictive performance. Moreover, the need to use numerous clinical parameters makes clinical application difficult. Furthermore, some predictive models [1417] could not explain which variables are comparatively important for BOO owing to their black box nature [18]. Causal Bayesian networks (CBN) have emerged as an advanced alternative to conventional statistical models in the medical field [19]. The benefit of this model is that it can visualize the interaction of causes and rule out indirect causes of events [20]. Hence, we aimed to identify non-invasive clinical parameters to predict BOO using a CBN model. To the best of our knowledge, this study is the first to test CBN model for BOO prediction. Prostate volume (ml) Total prostate volume Transitional zone volume Prostate specific antigen (ng/ml) International Prostatic Symptom Score (IPSS) IPSS-quality of life Uroflowmetry parameters Maximum flow rate (ml/sec) Post-void residual volume (ml) Urodynamic study parameters Maximal urethral closure pressure (cmH2O) Functional urethral length (mm) First desire (ml) Normal desire (ml) Strong desire (ml) Compliance (ml/cmH2O) PdetQmax (cmH2O) Opening pressure (cmH2O) Bladder outlet obstruction index PdetQmax, detrusor pressure at maximum flow rate. doi:10.1371/journal.pone.0113131.t001 Materials and Methods I. Data collection The Institutional Review Board of Seoul National University Hospital (SNUH) approved the protocol of this study. A database of 2,492 patients that were older than 45 and that had lower urinary tract symptoms (LUTS) was created from records dated between October 2004 and August 2013. The data were retrieved from the urodynamic database registry and Electronic Medical Records System of SNUH. All information was anonymised and de-identified prior to analysis. Patients with a history of previous genitourinary surgery, pelvic radiation therapy, urinary tract infection, urethral stricture, interstitial cystitis, and neuropathy suggesting neurogenic bladder or incomplete evaluations were excluded. Thus, after excluding 1,111 such patients (44.6%), the data from 1,381 patients were analyzed. Clinical parameters of the subjects, including history, physical examination, International Prostatic Symptom Score (IPSS) [21], UFM, PVR, PSA, prostate volume (PV) measured by transrectal ultrasonography, and UDS results were retrieved. UFM (Flowmaster, Medical Measurement System, Enschede, Netherlands) results were obtained as free flow, whenever voided volume was less than 120 ml, and fails were repeated. PVR was measured after UFM using an ultrasound bladder scanner (BladderScan BVI 3000, Verathon Inc., WA, USA). All UDS were performed using a multichannel video system (UD-2000, Medical Measurement System) according to International Continence Society (ICS) recommendations [22]. The BOO index, which is equal to detrusor pressure at maximal flow rate (PdetQmax)226maximal Total subjects (N = 1381) flow rate (Qmax), was used to determine BOO [23]. Patients with BOO Index $40 were considered as obstructed. II. Datab (...truncated)