Clinical Features Outperform Diabetes Subgroups for Outcomes

Becky McCall

May 02, 2019

Simple clinical features at diabetes diagnosis — such as body mass index (BMI), age, and kidney function — better predict clinical outcomes and are more useful in stratifying patients' treatment than classification of diabetes into subgroups, show the results of a new large study.

Essentially, the current research was a direct response to a Scandinavian study by Ahlqvist and colleagues, reported last year by Medscape Medical News, which defined five subgroups of diabetes with different physiological and genetic profiles (one was type 1 diabetes and the other four were related to type 2 diabetes).

This latest work tests the clinical utility of the subgroup approach by reproducing the analysis and comparing the results to an alternative one based on simple clinical features.

The study was published online April 29 in Lancet Diabetes and Endocrinology by John M. Dennis, PhD, and colleagues from the University of Exeter, UK.

Senior author Andrew Hattersley, MD, professor of molecular medicine at the University of Exeter, told Medscape Medical News that the motivation for this latest work came from the fact that "We know...diabetes is heterogeneous, so we'd like to individualize care by mapping the treatment to the patient, but how do we [most easily] do this?"

"Do we divide diabetes into subgroups, or do we use clinical features to best predict treatment choice?

"In models we actually found that using simple clinical features, in comparison to the subgroup strategy, is a better way" of optimizing therapy choice and "predicting outcomes, including renal failure and glycemic deterioration," said Hattersley, highlighting the key findings.

He added it is known that age at diagnosis predicts glycemic deterioration and estimated glomerular filtration rate (eGFR) predicts renal function, and as such, subgroups are unnecessary.

In an accompanying editorial, Hiddo Heerspink, PhD, clinical pharmacologist, University of Groningen, the Netherlands, sees merit in this approach. "The study adds to emerging data that illustrate that routinely measured patient characteristics can be used to explain, at least in part, the variation in risk of diabetic complications and treatment response among individual patients," he writes.

And commenting on the new findings, Emily Burns, PhD, head of research communications at Diabetes UK, said: "Type 2 diabetes is a complex condition and we need to move beyond a one-size-fits-all approach to treatment."

"This research suggests that healthcare professionals can use simple measurements readily available to them now, including BMI and age of diagnosis, to determine the best treatments for each individual person."

Subgroups of Diabetes or Clinical Features How Useful Are They?

Hattersley said the key to adoption of any predictive tool is how useful it is in real practice.

In their work published a year ago, the Swedish team identified five clusters of patients with diabetes. These divided into three severe and two mild forms of disease: one corresponding to type 1 diabetes and the remaining four representing subtypes of type 2 diabetes (severe insulin-resistant diabetes, severe insulin-deficient diabetes, mild obesity-related diabetes, and mild age-related diabetes).

"We wanted to find out how useful the subgroups were at predicting patient outcomes [HbA1c progression and risk of renal failure] and to compare these findings to using simple clinical characteristics [age, BMI, and eGFR]," Hattersley said.

The researchers also looked at whether a model using simple clinical features performed better than the subgroups in predicting treatment response.

Two large randomized controlled trials, ADOPT and RECORD, were used to provide data on patients with newly diagnosed diabetes (ADOPT, n = 4351) or established diabetes (RECORD, n = 4447), all of whom were randomly assigned to metformin, sulfonylurea, and/or thiazolidinedione therapy.

The Scandinavian cluster analysis was repeated in the ADOPT cohort, and differences in disease progression by the clusters were evaluated. The latter were then contrasted with stratification using simple continuous clinical features comprised of age at diagnosis for glycemic progression and baseline kidney function for renal progression.

"Simpler clinical measures were as or more useful than the clusters for stratifying each outcome assessed," explained Hattersley.

"When we looked at glycemic deterioration, this differed between the subgroups, but when we compared this approach with using simple clinical characteristics, notably age at diagnosis, they performed just as well."

Renal Failure Prediction  Better Using Baseline eGFR

A major point of interest in the Scandinavian cluster analysis was the fact that the insulin resistant subgroup had rapidly deteriorating renal function. 

Hattersley explained that this, in part, reflected a failure to remove patients with pre-existing renal failure.

"We asked whether initial renal function better explained renal failure than subgroups. We found it did. In fact, baseline eGFR markedly outperformed the clusters," he added.

So essentially, clinicians already have a good measure of renal function in the clinic — the patient's creatinine level — and they don't need a subgroup approach to predict renal failure, he argues.

"We need to continue using this measure; don't start thinking there's a magic subgroup approach to it. By using categories rather than numbers in predicting outcomes, you risk losing a lot of information and power to predict."

Clinical Features Versus Subgroups for Predicting Therapy Response

Selection of glucose-lowering therapy was then assessed using the subgroup strategy and compared with an approach combining simple clinical features in the RECORD cohort. 

"The cluster subgroups suggest that age-related mild diabetes patients would get better glucose control on a sulfonylurea than metformin or thiazolidinediones," noted Hattersley.

"But patients in the severe insulin resistance cluster should be on thiazolidinediones. This cluster-based information suggests subgroups can be used in treatment selection."

But the UK group's model found simple clinical characteristics better predicted glycemic control, rather than a subgroup approach.

"We believe the future for optimizing therapy lies in developing models based on clinical features," Hattersley stressed.

"These models will use data from patients' clinical records to inform likely clinical outcome, and then suggest the optimal treatment choice."

But HbA1c and eGFR Are Only Surrogate Outcomes...

Heerspink nevertheless points out in his editorial that further work is needed to look at whether the treatment response in terms of renal or cardiovascular outcomes varies across clusters or clinical variables.

"The [UK] authors analyzed glycemic responses and progression to new onset chronic kidney disease but these responses are surrogate outcomes."

"Further studies are needed to assess whether clinical variables will also outperform a cluster-based approach when clinical renal or cardiovascular outcomes are analyzed."

The study was supported by the UK Medical Research Council. Hattersley has reported no relevant financial relationships. Heerspink has reported being a member of the Study of Diabetic Nephropathy With Atrasentan (SONAR) steering committee. He has served as a consultant for AbbVie, Astellas, AstraZeneca, Boehringer Ingelheim, Fresenius, Gilead, Janssen, Merck, Mundi Pharma, and Mitsubishi Tanabe.

Lancet Diabetes Endocrinol. Published online April 29, 2019. Full text, Editorial

For more diabetes and endocrinology news, follow us on Twitter and on  Facebook.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.