Presented by: Belén Giménez MD
Urology Department, Universidad de Chile, Hospital Clínico San Borja Arriarán

Introduction

Artificial Intelligence (AI) can be used to recognize patterns and make predictions of future events from large amount of data. Due to the increasing incidence of urolithiasis, and its known association with genetic, nutritional, and environmental factors, it is relevant to implement interventions that help prevent the development of urolithiasis in the general population. Our aim was to create an AI model to predict the probability of a patient to develop kidney stones.

Materials

An extensive questionnaire was created to collect information about different risk factors for urolithiasis in individuals with and without history of kidney stones. Demographic, nutritional and exercise habits information, medical and family history, along with data from blood and urine analysis were collected. A supervised Machine Learning (ML) model was developed (Python Software Foundation). The performance of four models (Logistic regression, decision tree classifier, random forest classifier, extra trees classifier) to predict the occurrence of kidney stones was evaluated by determining the weight of each variable.

Results

,

A total of 730 questionnaires answered by Chilean individuals were included. All four models showed a high range of performance (70-83%) and identified as the strongest predictors for kidney stones development, the history of surgery for kidney stones, BMI of 30, beer consumption (3 times per week), burnt orange color of voided urine, history of calcium phosphate stones, weight gain in the past year (>5% of basal weight) and low water intake (<1 liter). Clear or pale straw yellow urine, not feeling thirsty during the day, use of potassium citrate, and regular consumption of calcium, fruits, and vegetables were identified as protective factors.

Conclusion

The use of ML algorithms to predict the risk of urolithiasis is reliable and effective. The models analyzed could be used to identify patients at higher risk of developing urolithiasis and address modifiable factors to reduce the incidence of this disease. The collection of quality clinical information and the application of predictive risk algorithms is an emergent tool that will allow the development of preventive public policies for this chronic disease of growing importance.

Funding

None

Lead Authors

Juan Fulla, MD, MsC
Universidad de Chile

Co-Authors

Antonia Reyes, MD
Universidad de Chile

Camila Cortés, MD
Universidad de Chile

Sofía Astorga, MD
Universidad de Chile

Nicolás Urnía, Medical Intern
Universidad del Desarrollo

Catherine Sanchez, DVM, PhD
Clinica las Condes