Preoperative prediction of the selection of the NOTES approach for patients with symptomatic simple renal cysts via an interpretable machine learning model: a retrospective study of 264 patients

Background: There are multiple surgical approaches for treating symptomatic simple renal cysts (SSRCs). The natural orifice transluminal endoscopic surgery (NOTES) approach has gradually been applied as an emerging minimally invasive approach for the treatment of SSRCs. However, there are no clear indicators for selecting the NOTES approach for patients with SSRCs. We aimed to investigate the preoperative clinical determinants that influence the selection of the NOTES approach in patients with SSRCs and to construct a prediction model to assist the surgeons in selecting the NOTES approach.

Methods: Clinical data from 264 patients with SSRCs from a single-center medical institution were included. Predictors were analyzed via the least absolute shrinkage and selection operator and multivariable logistic regression. Various machine learning classification algorithms were evaluated to determine the optimal model. An interpretive framework for personalized risk assessment was developed via SHapley Additive exPlanations (SHAP).

Results: Preoperative factors predicting the selection of the NOTES approach included cyst growth, the presence of renal calculus, body mass index, history of diabetes, history of cerebrovascular disease, hemoglobin level, and the platelet (PLT) count. The logistic classification model was identified as the optimal model, with area under the curve of 0.962, an accuracy of 0.868, a sensitivity of 0.889, and a specificity of 1.000 in the test set.

Conclusion: A logistic regression model was constructed and tested via the SHAP method, providing a scientific basis for selecting the NOTES approach for patients with SSRCs. This method offers effective decision support for doctors in choosing the NOTES approach.