Online Program

333275
Developing a Predictive Risk Model for Type 2 Diabetes


Wednesday, November 4, 2015 : 10:50 a.m. - 11:10 a.m.

Ben Strauss, MS, Children's Environmental Health Initiative, Durham, NC
Marie Lynn Miranda, PhD, School of Natural Resources & Environment, University of Michigan, Ann Arbor, MI
Pamela Maxson, PhD, Duke Center for Community and Population Health Improvement, Duke University, Durham, NC
Background

Diabetes is a growing health risk in the United States, affecting the well-being of our population and costing the nation’s health systems an estimated $218 billion annually. According to the American Diabetes Association, 11.3% of the population over 20 years of age had diabetes in 2010. 

Objective

The Durham Diabetes Coalition is working to reduce the burden of diabetes and understand disparities in Durham County, NC. This project aims to assess the feasibility of constructing risk models from electronic health record data (EHR) and demonstrate the utility of stratifying patients by their risk of adverse outcomes.

Methods

Patient records for patients with type 2 diabetes in 2010-11 from the Duke University Decision Support Repository (N=9,701) were used as a training dataset for a penalized logistic regression model. This model stratifies patients by their risk of a serious adverse diabetes-related outcome in the following calendar year.

Results

The risk algorithm successfully captured those patients at highest risk for a significant event. Patients with highest risk scores in the EHR model tend to be older, be more frequent users of the health system, and have comorbid diagnoses. The risk algorithm is being used as the basis for clinical and community interventions currently taking place in Durham County.

Discussion

Our goal is to create a spatially-informed model for diabetes management that can be replicated nationwide. Risk models based on EHR data provide a valuable clinical tool to efficiently target health care resources to improve health. In future research, incorporating the social and environmental context of a patient will enable the development of a more comprehensive risk model. Additionally, we are on incorporating social data based on neighborhood-level sociodemographic constructs.  In particular, we are considering racial residential segregation, neighborhood deprivation, social instability, health literacy, physical activity outlets, and the food environment.

Learning Areas:

Chronic disease management and prevention
Communication and informatics
Implementation of health education strategies, interventions and programs

Learning Objectives:
Explain the utility of a predictive risk algorithm to stratify patients. Discuss various factors that contribute to risk of serious outcome. Formulate understanding of how risk algorithms based on combined patient and environmental data can be employed to good effect.

Keyword(s): Risk Factors/Assesment, Diabetes

Presenting author's disclosure statement:

Qualified on the content I am responsible for because: As the lead GIS Analyst for the Durham Diabetes Coalition, I have overseen the development and implementation of this risk algorithm and have worked extensively with statisticians and clinicians to make it practicable. My research interests include translation of complex spatial methodologies.
Any relevant financial relationships? No

I agree to comply with the American Public Health Association Conflict of Interest and Commercial Support Guidelines, and to disclose to the participants any off-label or experimental uses of a commercial product or service discussed in my presentation.