# Machine learning for early prediction of the infection in patients with urinary stone after treatment of holmium laser lithotripsy

## TL;DR

This study uses machine learning to predict postoperative infections in patients after a specific urinary stone treatment, helping doctors make better decisions.

## Contribution

A novel machine learning strategy is proposed for early prediction of postoperative infection after holmium laser lithotripsy.

## Key findings

- Logistic Regression achieved the best performance with an AUC of 0.734 in predicting infection.
- Preoperative urine leukocyte count was identified as the most important variable for prediction.
- Key variables were interpreted using SHAP values to support clinical decision-making.

## Abstract

Patients after holmium laser lithotripsy have a certain probability of getting postoperative infection. An early and accurate diagnosis of postoperative infection allows a timely administration of appropriate antibiotic treatment. However, doctors can not accurately determine whether the patient has the infection. Here, a novel strategy is put forward to assist in predicting postoperative infection early by using machine learning methods. We retrospectively collected 1006 cases of patients with urinary stone after treatment of holmium laser lithotripsy from Zhejiang Provincial People’s Hospital. Feature engineering was added to filter the important characteristics and Miceforest multiple imputation method was applied to tackle the missing data problem. We used 5-fold cross-validation to train and validate the six machine learning methods. Besides, we could also find key variables important to postoperative infection by explaining the model. The hyperparameters were constantly adjusted to achieve the best performance of model. The result showed that LR had a best performance in independent datasets with AUC of 0.734. And the SHAP values indicated that preoperative urine leukocyte count was the most important variable to the prediction. Our study enables accurate predictions of infection in urology perioperative periods, the key variables can be interpreted better and more accurately to support clinical decision making.