Validation of Survey Questions to Distinguish Type 1 and Type 2 Diabetes Among Adults With Diabetes

Westat, Inc., Rockville MD

Investigators

Linked publications & trials

Abstract

PROJECT SUMMARY/ABSTRACT Type 1 and Type 2 diabetes are distinct clinical conditions with different etiologies, ages of onset, management strategies, risk factors, and outcomes. Currently, the data sources that the Centers for Disease Control and Prevention (CDC) relies upon to monitor trends in diabetes prevalence and incidence are unable to reliably distinguish between types of diabetes. Most of the large federal surveys used for diabetes surveillance have not included questions on diabetes type, and few studies have reported survey-based algorithms for identifying diabetes type. None have compared survey-based identification algorithms with a gold standard case ascertainment in order to validate survey-based assignment of diabetes type. Methods for distinguishing between diabetes types in electronic health records (EHR) data have been tested for children and adults, but additional validation work is needed. Our approach to improving diabetes surveillance in these two areas is based on an integrated study design whereby survey data for diabetes patients are linked with data from their EHR and a gold standard case ascertainment derived from chart review. We will select a diverse sample of diabetes patients that is designed and powered to assess algorithm validity for subpopulations defined by age, sex, and race/ethnicity. We will use rigorous questionnaire development methods to build on items used in previous surveys, cognitively test the new survey module to optimize wording and question order, field the survey using data collection methods similar to established CDC surveys, and analyze the responses relative to a gold standard classification of diabetes type. In addition, we will develop and validate against the gold standard EHR-based algorithms, including a modeling approach that produces a parsimonious rule-based algorithm for determining diabetes type based on the most important clinical variables, and a machine learning approach that uses the gold standard dataset as a starting point to identify implicit patterns that distinguish T1DM and T2DM.