Use of Machine Learning Techniques to Predict Bladder Outlet Obstruction

Use of Machine Learning Techniques to Predict Bladder Outlet Obstruction in Men with Lower Urinary Tract Symptoms from Noninvasive Home-Uroflowmetry Data 

Lower urinary tract symptoms (LUTS) are a common reason for urological evaluation, and accurate diagnosis is essential for selecting appropriate treatment. The gold standard for diagnosing bladder outlet obstruction (BOO), a common cause of LUTS, is urodynamics, an invasive and time-consuming test. Thus, there is growing interest in using noninvasive tests that can be used more comfortably and more frequently.  

This project explored whether data from CarePath, a handheld home-uroflowmetry device, could be used to help predict BOO. CarePath collects data across multiple voids, providing a more accurate picture of the true voiding pattern and addressing the disadvantages of in-office point-of-care testing. 

Using uroflow-derived metrics along with other clinical measures, I evaluated several classification algorithms, including logistic regression, random forest, and gradient boosting, to assess their ability to diagnose BOO. The models showed high sensitivity but low specificity. Although the models did not achieve the accuracy necessary for clinical use, the predictive signals identified in the flow-derived metrics indicate potential for improved diagnostic accuracy with larger, well-defined cohorts.