Prediction of infective complications after retrograde intra renal surgery using Machine learning

Abstract

Background

To compare the models obtained with classical statistical methods and machine learning (ML) algorithms to predict postoperative infective complications (PICs) after retrograde intrarenal surgery (RIRS).

Material and methods

Patients who underwent RIRS between January 2014 and December 2020 were retrospectively screened. Patients who did not develop PICs were classified as Group 1 and patients who developed as Group 2.

Results

Three-hundred and twenty-two patients were included in the study; 279 patients (86.6%) who did not develop PICs were classified as Group 1, and 43 patients (13.3%) who developed PICs were classified as Group 2. In multivariate analysis, the presence of diabetes mellitus, preoperative nephrostomy, and stone density were determined to be factors that significantly predicted the development of PICs. The area under the curve (AUC) of the model obtained by classical Cox regression analysis was 0.785, and the sensitivity and specificity were 74% and 67%, respectively. With the Random Forest, K- Nearest Neighbour, and Logistic Regression methods, the AUC was calculated as 0.956, 0.903, and 0.849, respectively. RF's sensitivity and specificity were calculated as 87% and 92%, respectively.

Conclusion

With ML, more reliable and predictive models can be created than with classical statistical methods.

Keywords:

• Retrograde intrarenal surgery
• infective complications
• machine learning
• sepsis

Declaration of interest

No potential conflict of interest was reported by the author(s).