Identifying Type 1 and Type 2 Diabetes Through EHR/EMRs

June 10, 2019
Patrick Campbell

Sue Kirkman, MD, professor of medicine at UNC Chapel Hill, discusses the results of a study she led that attempted to create a method of identifying what type of diabetes through EHR data.

For years, the Centers for Disease Control and Prevention (CDC) has been attempting to identify a method that would allow them to differentiate type 1 and type 2 diabetes patients based upon data from electronic health records.

Sue Kirkman, MD, professor of medicine at UNC Chapel Hill, is the current chair of the National Diabetes Education Program, which is a joint initiative of the CDC and National Institute of Diabetes and Digestive and Kidney Diseases, and she recently led a study into the development of a “gold standard” to identify T1D and T2D from EHR data.

The study, which was presented at the American Diabetes Association (ADA) 2019 Scientific Sessions in San Francisco, CA, used data from more than 50,000 adult patients in the UNC Health Care System. From this data, they dev eloped two models to determine diabetes type. The first model was a decision tree and the second was a weighted model.

The two models exhibited 89% congruency when identifying T1D and 96% congruent for identifying T2D. Overall, 70% of the sample had T2D, 27% has T1D, and 3% had indeterminate or other diabetes.

Despite the results of the study, Kirkman told MD Magazine® in an interview that they do not expect the CDC to utilize the method as it is cost- and labor-intensive, rather they will use the results to further develop a model that is both accurate and cost-effective.

MD Mag: What were the results of your study looking to develop a “gold standard” for identifying T1D and T2D from electronic medical records?

Kirkman: So, we're presenting some research that we're doing with the CDC and it's really related to a larger problem, which is that when the CDC does their surveillance of diabetes in this country they can't determine type of diabetes. So, when they come out with numbers that say 33 million people in the United States have diabetes, we don't really know how many of those people have type 1 diabetes versus type 2 diabetes and one of the reasons that's a problem is that type 1 is a lot less common than type 2 and you know we know that prevalence and incidence of type 2 diabetes has been growing in the last few decades and so the type 1 statistics tend to sort of get lost in the midst of all that.

So, the CDC has a real interest in being able to better determine like how much type 1 is there and what are the outcomes of people with type 1 diabetes. So, to back up a little bit in order to be able to develop survey questions which is really how the CDC does their surveillance you have to have something to validate those survey questions against and so that's really what the research that we're presenting here is all about.

So, the "gold standard" is really just what's going to kind of underlie this work in terms of being able to better determine how many people have type 1 and type 2 diabetes. We went through a very detailed process where we looked you know very closely at 5,000 charts of people with diabetes, went through all kinds of data in their EHR, reviewed chart notes, and tried to get more information from that and fed all this information into two different models and tried to come up with you know who's got type 1 and who's got type 2 and when is it kind of indeterminate. We don't really expect that this is going to be the way we determine how many people have type 1 and type 2 in the future. It's very labor intensive, we really just did it as kind of the background so that we can then develop telephone survey questions that will more easily answer the question of what type of diabetes do people really have.