Endocrine Abstracts

Introduction: During Ramadan, the daily lifestyle undergoes changes that influence the glycemic balance of both fasting and non-fasting diabetic patients. The aim of this study is to predict glycemic imbalance during Ramadan in non-fasting diabetic patients using Machine learning models.

Methods: This is a prospective study which included all non-fasting diabetic patients during the month of Ramadan and followed at the endocrinology department of the Sheikh Khalifa University Hospital in Casablanca. We collected anthropometric and metabolic parameters during three consultations (before, during and after Ramadan). To predict imbalance, we trained artificial intelligence models based on Machine Learning using the collected data. Finally, we ran several simulations with the best-performing model, using the variables that proved to be the main predictors of poor glycemic control.

Results: We included 154 patients. The prevalence of poor glycemic control among patients was 52.6%. The Extra Tree Classifier was the best-performing model for predicting imbalance (accuracy = 0.87, AUC = 0.87). The most significant variables for predicting imbalance were: baseline caloric intake and its evolution, female sex, baseline weight, evolution of BMI and waist circumference, and total cholesterol level after Ramadan.

Conclusion: The clinical use of our results will enable us to target risk factors in order to improve glycemic control in non-fasting diabetics.