PATIENT ADHERENCE MODELING IN DIABETES PATIENTS USING MULTIVARIATE REGRESSION

Open Access
Author:
Xu, Claire
Area of Honors:
Industrial Engineering
Degree:
Bachelor of Science
Document Type:
Thesis
Thesis Supervisors:
  • Harriet Black Nembhard, Thesis Supervisor
  • Paul Griffin, Honors Advisor
Keywords:
  • industrial engineering
  • multivariate regression
  • regression
  • diabetes
Abstract:
Diabetes affects 25.8 million children and adults in the United States and was the seventh leading cause of death in 2007. Evidence shows that higher adherence to treatment is associated with better diabetes control and improved long-term health outcomes for patients. Improving patient adherence will reduce costs to healthcare service and economies as well as reduce complications, increase life expectancy, reduce morbidity, and improve patient quality of life in people living with diabetes. This study focuses on analyzing the effects of physiological factors, including health vitals, comorbidities, and treatment and other factors, on SMBG adherence in diabetes patients using multivariate regression. The findings of this study are that health vitals, comorbidities, and treatment factors are all essential in comprehensively understanding and modeling the factors that impact adherence. Significant variables include A1c, diastolic blood pressure, LDL, peripheral neuropathy, hypothyroidism, CAD/MI, nephropathy, depression, average glucose level, Type 1/Type 2 diabetes, and the number of uncontrolled health vitals. A1c, diastolic blood pressure, nephropathy, and Type 1 diabetes are positively related to non-adherence, while LDL, peripheral neuropathy, hypothyroidism, CAD/MI, depression, average blood glucose, and number of uncontrolled health vitals are negatively related to non-adherence. The regression model created results with an adjusted R2 value of 52.73%. Future work includes improving the descriptive power of the regression models examined. To do so, more data will be collected and analyzed with a focus on increasing the sample size and variability. The physiological data can then be modeled with behavioral risk factors for diabetes patient adherence to form a more comprehensive picture of the treatment process. Furthermore, this research was limited to creating only descriptive models of patient non-adherence. Future work will focus on the development of predictive model that may be used to develop a treatment decision-making model to better mitigate non-adherence risks.