Prediction of anorectal malformations with bladder‐neck/urethral‐prostatic fistula using classification and regression tree analysis

• Published in: Journal of clinical ultrasound Vol. 51; no. 3; pp. 436 - 444
• Main Authors: Yuan, Peng, Feng, Wei, Huang, Yao, Wang, Yi
• Format: Journal Article
• Language: English
• Published: Hoboken, USA John Wiley & Sons, Inc 01.03.2023; Wiley Subscription Services, Inc
• Subjects: Anal Canal; Anorectal; anorectal malformations; Anorectal Malformations - diagnostic imaging; Bladder; Classification; classification and regression tree; Fistula; Fistulae; Humans; Infant, Newborn; invertogram; Male; Model accuracy; Neck; Neonates; Perineum; Prediction models; Rectum - surgery; Regression analysis; Regression models; Retrospective Studies; Sensitivity analysis; surgery; Ultrasonic imaging; ultrasound; Urethral Diseases - diagnostic imaging; Urinary Bladder - diagnostic imaging; anorectal malformations; classification and regression tree; invertogram; ultrasound; surgery
• DOI: 10.1002/jcu.23355
• ISSN: 0091-2751; 1097-0096

Background The anorectal malformations (ARMs) with bladder‐neck/urethral‐prostatic fistula in males are the most complex groups in ARMs. It is essential to diagnose them effectively before the primary operation for both surgical modalities and future functional prognosis can be quite different from other types. Our study aimed to develop a classification and regression tree (CART) model for the prediction of ARMs with bladder‐neck/urethral‐prostatic fistula. Methods A total of 132 newborns with ARMs were included retrospectively and randomly assigned to the training sample and test sample in a 3:1 ratio. The pouch‐perineum distance of ultrasound (PPDU), the pouch‐perineum distance of invertogram (PPDI), and the rectum gas above the coccyx (RGAC) on the invertogram were hypothesized can serve as individual predictors. The CART analysis was used to determine the best combination of candidate predictors. The model's performance was assessed by the area under the receiver operating characteristic curve (AUC) and validated in the test sample. Results All three individual predictors were included in the CART model to predict the ARMs with bladder‐neck/urethral‐prostatic fistula in the derivation cohort with the following test characteristics (95% CI): sensitivity 75.6% (60.1 to 86.6); specificity 88.9% (76.7 to 95.4); AUC 0.909 (0.854 to 0.965). The model's predicted accuracy was validated in the test cohort (AUC = 0.883). In all 132 subjects, the AUC of the tree model was significantly superior to that of the best individual index: PPDU (0.901 vs. 0.819; p = 0.005). Conclusions A predictive model that consists of PPDU, PPDI, and RGAC may be useful in predicting ARMs with bladder‐neck/urethral‐prostatic fistula. We propose a predictive model based on CART analysis by the combination of information from ultrasound and invertogram to predict the ARMs with bladder‐neck/urethral‐prostatic fistula, the most complex groups in ARMs. The model can help pediatric surgeons to make an appropriate surgical choice during the neonatal period.