Machine-learning-based prediction models in patients with azoospermia caused by azoospermia factor microdeletions

Owing to the lack of specific indicators, azoospermia factor microdeletions are underdiagnosed. A portable prediction tool that can screen for azoospermia factor microdeletions is highly desirable.

Clinical characteristics and laboratory data are collected from 756 patients with or without azoospermia factor microdeletions and subsequently used to develop an online model to predict azoospermia factor microdeletions. A random forest algorithm is used to build the prediction model and five independent risk factors are identified as azoospermia factor microdeletions predictors. Sensitivity, specificity, and accuracy are used to evaluate the performance of the model. The prediction model achieves an area under the curve value of 0.93 and the highest sensitivity value of 0.86.

The online prediction model can be effectively used for azoospermia factor microdeletions diagnosis as it is reliable and portable. It can be widely used in the hospital to help clinicians improve the diagnostic efficiency of azoospermia factor microdeletions.