Simple Clinical Features More Effective Than Clustered Subgroups in Stratifying T2D

Doctor measuring blood sugar of patient with diabetes
Doctor measuring blood sugar of patient with diabetes
Simple continuous clinical features were better for stratifying patients with type 2 diabetes than data-driven clusters for predicting treatment response and other clinical patient outcomes.

Models based on simple continuous clinical features might be better for stratifying patients with type 2 diabetes (T2D) than data-driven clusters that differ in disease progression and treatment response, according to study results published in The Lancet Diabetes & Endocrinology.

Researchers conducted this study to compare the clinical utility of a subgroup-based approach for predicting outcomes in T2D vs using outcome-specific models that employ simple patient characteristics such as sex, age at diagnosis, body mass index, and baseline hemoglobin A1c level.

Patients with T2D were identified from 2 previous studies: 3802 people from the ADOPT trial, stratified into 5 therapy groups using data-driven cluster analysis, and 4057 people from the RECORD trial. Clinical outcomes, including glycemic progression and kidney function, and response to different diabetes medications were assessed using the clusters strategy and the clinical features strategy in both study cohorts.

The researchers found that while clusters differed in glycemic progression, age at diagnosis predicted disease progression as well as or better than cluster formation. They also discovered differences between clusters in chronic kidney disease incidence, and that a model of glomerular filtration rate at baseline better predicted time to chronic kidney disease.

There were also useful differences between clusters in stratifying glucose-lowering treatment response, particularly regarding benefit of thiazolidinediones in the severe insulin-resistant diabetes cluster and sulfonylureas in the mild age-related diabetes cluster. However, it was determined that simple clinical features were able to select therapy for individual patients significantly better than clusters.

Related Articles

Limitations to this study included the potential that participants may not be representative of all patients with diabetes because of original trial exclusion criteria.

The researchers said their results suggest “precision medicine in type 2 diabetes is likely to have most clinical utility if it is based on an approach of using specific phenotypic measures to predict specific outcomes, rather than assigning patients to subgroups.”


Dennis JM, Shields BM, Henley WE, Jones AG, Hattersley AT. Disease progression and treatment response in data-driven subgroups of type 2 diabetes compared with models based on simple clinical features: an analysis using clinical trial data [published online April 29, 2019]. Lancet Diabetes Endocrinol. doi:10.1016/S2213-8587(19)30087-7