Building Dual AI Models and Nomograms Using Noninvasive Parameters for Aiding Male Bladder Outlet Obstruction...

Area of Science:

• Urology
• Medical Informatics
• Machine Learning

Background:

• Diagnosing male lower urinary tract symptoms (LUTS) due to bladder outlet obstruction (BOO) is challenging.
• Invasive tests like video-urodynamic studies (VUDS) and pressure-flow studies (PFS) are often required.
• VUDS offers more precise etiological differentiation than PFS for male BOO.

Purpose of the Study:

• To develop noninvasive predictive models for BOO diagnosis.
• To reduce the necessity for invasive diagnostic procedures in men with LUTS.
• To create a decision-support tool for clinicians.

Main Methods:

• Retrospective study of men with suspected BOO undergoing VUDS.
• Development of two predictive models: International Continence Society-defined BOO (ICS-BOO) and video-urodynamic studies-diagnosed BOO (VBOO).
• Utilized 6 machine learning algorithms and validated using cross-validation and an independent test set.

Main Results:

• The ICS-BOO model achieved an AUC of 0.86 in test validation; the VBOO model achieved an AUC of 0.72 externally.
• A dual-model approach could potentially avoid VUDS in 32.6% of patients.
• Combined model predictions enhance clinical decision-making and diagnostic pathway simplification.

Conclusions:

• Two machine learning models effectively predict ICS-BOO and VBOO using noninvasive parameters.
• A dual-model prediction strategy can help avoid VUDS when both models predict BOO positively.
• This approach aids in male BOO diagnosis and reduces reliance on invasive testing.