Exploring the relationship between psychosocial aspects and glucose control in individuals with type 1 diabetes using standardized questionnaires and continuous glucose monitoring data

Type I diabetes (T1D) is a chronic disease that requires constant self-management by the patient, including monitoring of blood glucose, multiple daily insulin injections, and appropriate lifestyle modifications such as diet and exercise. This strict regimen poses a significant burden on people with T1D, coupled with the psychological challenges of living with the disease. These challenges may contribute to the onset of stress and depression, underscoring the need to explore the interplay between psychosocial and glycemic control. This work investigates the correlation and causal relationships between psychosocial variables and glucose management in T1D. Our analysis is based on a dataset comprising data from 205 patients over three months, including continuous glucose monitoring measurement and self-assessment questionnaires. The final aim is to analyze how psychosocial variables and features extracted from glycemic data interact in people with T1D, providing information to guide improved support strategies. After an initial pre-processing phase to ensure the suitability of the dataset for subsequent analyses, we characterized the study population through graphical represen- tations and statistical tests, such as the Kruskal-Wallis and Mann-Whitney U tests. The study then explored the correlations between the variables and evaluated their characteristics and impact on the outcomes. Both static and dynamic analyses focused on various variables at different time points during the study. Linear models (e.g., Linear Regression) and non-linear models (e.g., Random Forest) were developed to identify key predictors and evaluate model performance. This phase considered multiple targets, such as scores from the depression and diabetes distress questionnaire and clinical variables such as HbA1c, with the corresponding input characteristics customized for each target. Of note, linear models tended to emphasize demographic and questionnaire-based variables as key predictors, while non-linear models successfully highlighted glycemic variables among the most influential factors driving the outcomes. However, none of the tested models provides a satisfactory explanation of the data, as indicated by their low R2 values, suggesting more complex relationships among the variables analyzed. Although this exploratory analysis aimed primarily to understand the data set and investigate how psychosocial aspects influence glycemic control, future work will focus on training predictive models to estimate glycemic outcomes based on psychosocial factors and comprehensively modeling the impact of stress and depression on glycemic outcomes.