Detrusor underactivity (DU) has the feature of relatively low maximum flow rate (Qmax) and cannot currently be differentiated from bladder outlet obstruction (BOO) without invasive pressure flow studies (PFS). The aim of this study was to test the hypothesis that a decision tree model employing free flow parameters could non-invasively discriminate DU from BOO in men.

Free flow data from 158 men with BOO and 135 men with DU diagnosed by PFS at a single specialist centre between 2012 and 2018 were selected for analysis in this study. Inclusion criteria for DU were bladder contractility index (BCI)<100, Bladder Outlet Obstruction Index (BOOI)<20 and bladder voiding efficiency (BVE)<90%, while the criteria for BOO were BCI≥100, BOOI≥40 and BVE≥90%.

Flow data were analysed in both the time and frequency domains to derive non-invasive parameters which were determined to be statistically different between the two groups [1, 2]. Those parameters having statistical differences were utilized in a classification and regression tree (CART) model (Figure 1). The data were split randomly into 70% for training and 30% for testing. In order to minimise false positive occurrences, the maximum tree depth was set to three, and the number of minimum cases in parent and child nodes was set to twenty and seven, respectively. Statistical analysis was performed in SPSS and parameters were derived in MATLAB.

The training model and the test results for discriminating DU from BOO are presented in Figure 1. In the training phase, 102 BOO and 92 DU were employed, and yielded a 75.8% overall discrimination accuracy, 89.2% sensitivity and 60.9% specificity (Figure 1, left). The remaining 56 BOO and 43 DU were used in the testing phase, and yielded a 72.7% overall discrimination accuracy, 85.7% sensitivity and 55.8% specificity (Figure 1, right).

It is worth noting that during the training phase of the DU/BOO CART model, node 4 of level 2 identified 20 DU patients and none with BOO. During testing, the same node identified 8 DU patients and 1 with BOO, which still provided a chance to subgroup 20.7% of DU patients with a true positive rate of 96.5%. The CART model predicted BOO or DU with a higher probability, though in a small portion, when the data were classified into a child node.

This study tested the diagnostic utility of a CART model to discriminate DU from BOO non-invasively by employing uroflowmetry parameters. Though the 74% accuracy is not ready for the clinical use, it is a significant improvement compared to any individual parameter in discriminating DU from BOO, and has the potential to improve when tested further with a larger database. 
The adjustability of the threshold values in each decision may hold promise for subgrouping a greater number of severe DU and/or BOO patients. This model was only tested using a single database, and further validation is needed in a larger cohort to improve robustness.

Li, R., Gammie, A., Zhu, Q. & Nibouche, M. 2017 “Mathematical analysis on Urine Flow Traces for Non-invasive Diagnosis of Detrusor Underactivity in Men”, Neurourology and Urodynamics, S3(36), pp. 87-88.
Li, R., Gammie, A., Zhu, Q. & Nibouche, M. 2018 “Median frequency and sum of amplitude changes in rising slope: two potential non-invasive indicators for differentiating DU with BOO in male”, Neurourology and Urodynamics. S5(37), pp. 248-249.