Adherence to insulin treatment in participants with type 2 diabetes: Comparison of logistic regression and conditional tree and forests to determine the effective factors

Abstract

Objective: Type 2 diabetes is the most prevalent chronic disease in the world. Timely and appropriate control can significantly reduce the burdens and costs of this disease. Although insulin injection is the most efficient method to control type 2 diabetes, patients avoid this method for unknown reasons. The main aim of the present study is to determine the factors influential in non-adherence to insulin using tools and models that have not been applied in this field so far. Methods: The tendency to insulin injection in 457 patients with type 2 diabetes was investigated in this cross-sectional study using the classic logistic regression and new learning algorithms, including conditional tree, conditional forest, and random forest. Different fits were compared so that the best model can be determined to identify the factors in non-adherence to insulin. Results: Although random forest had the highest accuracy among the fitted models, all the methods had a relative consensus that having life insurance, academic education, and insulin injection experience in immediate family members increase the tendency to accept insulin therapy. Our results also showed that younger patients and those who were committed to a specific diet better approved insulin therapy. Conclusions: The reasons for non-adherence to insulin can be summarized in economic and psychological aspects. Therefore, the health system policies are recommended to address economic issues and also raise public awareness about this treatment method.

Keywords

• type 2 diabetes
• insulin
• decision trees
• forests