在無尿路出口阻塞之女性病患，群集分析可以用來準確診斷膀胱功能低下

陳人傑¹, 林志杰^1,2, 范玉華^1,2, 盧星華^1,2, 林登龍^1,2, 黃志賢^1,2

臺北榮民總醫院 泌尿部¹; 國立陽明大學醫學院 泌尿學科 書田泌尿科學研究中心²

Use Cluster Analysis to Develop an Accurate and Practical Diagnostic Method for Detrusor Underactivity in Female Patient without Bladder Outlet Obstruction

Jen-Chieh Chen¹, Chih-Chieh Lin^1,2, Yu-Hua Fan^1,2, Shing-Hwa Lu^1,2, Alex T.L. Lin^1,2, William J. Huang^1,2

Department of Urology, Taipei Veterans General Hospital¹; Department of Urology, College of Medicine and Shu-Tien Urological Research Center, National Yang Ming Chiao Tung University²

Purpose:

The International Continence Society (ICS) definition of the detrusor underactivity (DU) lacks of standardized urodynamic (UDS) parameters and thresholds. Standardized and accepted nomograms for the DUA diagnosis have been reported only for men, while there are no definitive standardized UDS criteria for the definition of female DU. We propose to use the cluster analysis to identify DU among female patient without bladder outlet obstruction (BOO) on the daily practice.

Materials and Methods:

We retrospectively reviewed our video-urodynamic study (VUDS) database from 2015 to 2020. All female patients with “weak stream” subscore of International Prostate Symptom Score equal or greater than 3 points were included. The exclusion criteria were patients who had radiological evidence of BOO and situational inability to void on VUDS. The VUDS results were reviewed by two independent researchers, and DU was diagnosed based on the ICS criteria. A total of 80 patients were enrolled, and we divided them into discovery (from 2018 to 2020) and validation cohort (from 2015 to 2018). Hierarchical cluster analysis and K-means clustering were performed with UDS parameters which included detrusor pressure at maximum flow (PdetQmax), maximal urine flow rate (UFR), and voiding time.

Results:

Three groups could be identified by cluster analysis in discovery cohort. Group 1 (17 patients) had low PdetQmax (14 cmH₂O), high maximal UFR (23.5 ml/s), and short voiding time (10.5 seconds). Group 2 (13 patients) had high PdetQmax (27 cmH₂O), medial maximal UFR (14.4 ml/s), and short voiding time (10.6 seconds). Group 3 (8 patients) had low PdetQmax (7 cmH₂O), low maximal UFR (11.4 ml/s), and long voiding time (26.1 seconds). We found that all DU (4 patients) and acontratile detrusor (AcD) (4 patients) patients were clustered in Group 3. In validation cohort, we assigned each data point to the nearest centroid (by Euclidean distance after standardization) of the groups that we obtained from the discovery cohort. We found that 7 of 8 DU or AcD patients in validation cohort were assigned to the Group 3. The positive predictive value and the negative predictive value of this approach for diagnosis of DU or AcD were 87.5% and 97.1%. Besides, we also used the validation cohort to repeat the cluster analysis. We discovered that we were able to replicate the division into three similar groups as the discovery cohort, and the centroids were almost the same.

Conclusions:

Among woman who had weak stream without BOO on VUDS, we detected three urodynamic characters, similarly in two independent patient cohorts, by using cluster analysis. The DU and AcD could be clustered into one of the groups with high sensitivity and specificity. We propose to use this cluster classification to identify DU on the daily practice.