Skip to main content

10-09-2018 | Diagnosis | EASD 2018 | News

Predictive model discriminates between type 1 and type 2 diabetes


medwireNews: Researchers have developed a prediction model combining clinical features, autoantibodies, and genetics to help distinguish patients with type 1 diabetes from those with type 2 diabetes.

Presenting the model at the 54th EASD Annual Meeting in Berlin, Germany, Beverley Shields (National Institute for Health Research Exeter Clinical Research Facility, UK) said that clinicians normally rely on clinical features to decide whether a patient has type 1 or type 2 diabetes, but for patients who have some type 1 and some type 2 features, it can be difficult to determine which subtype of diabetes they have.

The team developed the model using data from a cohort of 943 diabetes patients in Exeter. A total of 14% had type 1 diabetes, defined as the presence of severe endogenous insulin deficiency (C-peptide levels <200 pmol/L) and time to insulin requirement of less than 3 years, while the remainder had non-insulin requiring type 2 diabetes, defined as no insulin treatment, or C-peptide levels of more than 600 pmol/L for more than 3 years postdiagnosis.

Shield explained that compared with type 2 diabetes patients, those with type 1 diabetes had a younger age at diagnosis, a lower BMI, a higher type 1 diabetes genetic risk score (GRS), and were more likely to have glutamic acid decarboxylase (GAD) and islet autoantibodies.

When considered individually, these features correctly distinguished between the two patient groups on 63.0–84.6% of occasions on area under the receiver operating characteristic curve analysis, but none of the features entirely separated type 1 and type 2 diabetes, said Shield.

However, she reported that when age at diagnosis and BMI were combined, “you get a much better area under the curve” than either feature alone, with the combination correctly distinguishing between patients with type 1 and type 2 diabetes on 90.4% of occasions. Adding GAD to the model improved the discriminatory power further, to 95.9%.

And combining all the variables resulted in “near perfect discrimination,” she said, identifying patients with the correct diagnosis 97.2% of the time.

The model was validated in an independent cohort of 549 patients, said Shield, and demonstrated similar predictive ability to that in the development dataset. The validation results for the full model including all five variables were not reported, but the model including age, BMI, and GAD correctly discriminated between patients with type 1 and type 2 diabetes on 93.0% of occasions in the validation cohort.

Together, these results indicate that the addition of autoantibodies and GRS to clinical features “improves diagnostic accuracy”, said Shield.

She announced that the model will be made available as an online calculator, and as part of the Diabetes Diagnostics app.

By Claire Barnard

medwireNews is an independent medical news service provided by Springer Healthcare. © 2018 Springer Healthcare part of the Springer Nature group


Novel clinical evidence in continuous glucose monitoring

Novel clinical evidence in continuous glucose monitoring

How real-world studies complement randomized controlled trials

Jean-Pierre Riveline uses data from real-life continuous glucose monitoring studies to illustrate how these can uncover critical information about clinical outcomes that are hard to assess in randomized controlled trials.

This video has been developed through unrestricted educational funding from Abbott Diabetes Care.

Watch the video