Assessment of Factors Associated with Treatment Adherence Among Elderly Patients with Multimorbid Type 2 Diabetes Mellitus Using Lasso-Logistic Regression.

AIMS AND

OBJECTIVES

This study aimed to investigate treatment adherence among elderly patients with multimorbid type 2 diabetes mellitus (T2DM), and analyze the influencing factors.

DESIGN

A single-centre, cross-sectional study design.

METHODS

In this study, convenience sampling was used to examine elderly patients with multimorbid T2DM seeking treatment at six community health service centers within the Jinqiao Medical Alliance in the Pudong New Area of Shanghai between May and July 2024. Demographic and disease-related data were collected including treatment adherence, self-care activities, social support, cognitive function, and depression. Factors influencing treatment adherence were investigated through three machine learning approaches: random forest algorithm for detecting non-linear patterns, multiple linear regression for linear relationship analysis, and Lasso-Logistic regression with L1 regularization to optimize feature selection while controlling multicollinearity. This tripartite methodology synergistically combines ensemble learning, parametric modeling, and sparse logistic regression to ensure robust predictor identification.

RESULTS

This study found that the average treatment adherence score for elderly patients with multimorbid T2DM was 45.30 (SD = 5.99). Integrated machine learning (random forest, Lasso-Logistic regression, and linear regression) identified four key determinants: elevated HbA1c (β = -4.417, P < 0.01) and depression (β = -1.207, P < 0.01) significantly reduced adherence, whereas improved self-care (β = 0.081, P < 0.01) and higher income (β = 0.589, P < 0.01) enhanced compliance. This multi-method approach validated predictors through both linear and non-linear modeling frameworks.

CONCLUSION

This study quantifies adherence in elderly T2DM patients (Mean=45.30) and identifies four modifiable predictors through advanced modeling. Prioritized interventions should focus on enhancing glycemic control through intensified HbA1c monitoring for upward trends and integrating depression management into diabetes care plans, while leveraging self-care capacity and economic support as foundational enhancers through tailored guidance and support programs to improve treatment adherence, optimize health outcomes, and minimize morbidity in this population.