Why heart risk is hard to predict in type 1 diabetes

By Dr. Liji Thomas, MDReviewed by Lauren HardakerApr 20 2026

A large European study uncovers hidden cardiovascular risk patterns in type 1 diabetes, showing how smarter profiling could help doctors catch complications earlier and tailor prevention.

Study: Precision cardiovascular risk prediction in type 1 diabetes: An IMI2 SOPHIA analysis. Image credit: Chinnapong/Shutterstock.com

Cardiovascular disease (CVD) is a leading cause of death in type 1 diabetes (T1D). Risk assessment is complicated by the presence of chronic hyperglycemia alongside lipid abnormalities and hypertension. To bridge this gap, a study in Nature Communications applied an existing phenotype-driven risk prediction tool to T1D patients to refine CVD risk stratification based on discordance between body mass index (BMI) and cardiometabolic biomarkers.

Existing literature shows that CVD risk in T1D patients remains high despite good blood sugar control, making weight gain a major concern. In an earlier study involving one of these authors, researchers distinguished five discordant profiles in the general population to improve the detection of CVD risk.

Discordance profile allocation in T1D

The current study sought to assess whether these profiles apply to T1D by replicating this framework in a T1D population. The authors extended the analysis to include T1D patients. They analyzed cross-sectional data from approximately 44,000 T1D patients in three cohorts (KUL, DPV, and SIDIAP) across multiple centers in Europe. Conventional CVD markers were analyzed, including demographic, anthropometric, lifestyle, blood biomarkers, and blood pressure data.

They calculated discordance scores for each individual based on how well the biomarkers matched the BMI. Each subject was then assigned probabilities for each phenotype, emphasizing the continuous rather than categorical nature of this exercise. Finally, they used the Uniform Manifold Approximation and Projection (UMAP) method to plot all this data and compare the results with those of the original study.

Discordant hyperglycemic profile overrepresented in T1D

This demonstrated that three profiles were predominantly represented in T1D: concordant, hyperglycemic, and inflammatory, although other profiles were present at lower frequencies.

The discordant hyperglycemic phenotype accounted for 2.5 % of people in the original study, but 55 %–76 % in the T1D population. Compared with the concordant profile, glycated hemoglobin (HbA1c) was higher in the hyperglycemic group, while lower HbA1c levels were associated with the concordant (lower-risk, more general population–like) profile.

Current model shows selective improvements over conventional scoring

They then compared two sex-specific survival prediction models: one based on SCORE2, a CVD risk stratification tool recommended by the European Society of Cardiology that incorporates biomarkers and other CVD risk-related variables, and the other that added the assigned profile probabilities. The aim was to identify which one performed better at predicting major adverse cardiac events (MACE).

The results showed that adding profile allocation probabilities improved predictions in specific models, outcomes, and cohorts, but not universally.

Significant likelihood ratio testing, a commonly used gold-standard approach for comparing nested models, demonstrated improved prediction of macrovascular complications in the KUL cohort, although this was limited to males.

Similarly, MACE prediction improved for males in the SIDIAP cohort, while extended MACE prediction improved for females in the same cohort. In addition, retinopathy prediction was enhanced in males in the KUL cohort and in females in the DPV cohort.

These findings are consistent with the improvements in MACE prediction reported by the original study authors, particularly among men in the UK.

Comparison with other tools

In comparison, other risk prediction tools designed for diabetes, such as the United Kingdom Prospective Diabetes Study (UKPDS) (primarily developed for type 2 diabetes) and T1D-specific tools such as STENO-T1D, have been shown to underestimate CVD risk in T1D or do not include BMI, unlike the newer LIFE-T1D, which also considers kidney and retinal complications.

Net benefit analysis favors current model

When the benefit to the population of using these tools is considered, rather than improved predictive performance alone, the original study showed the net benefit of using any models, including those with discordant profile data, across a range of MACE probabilities up to 15 %, compared to either treating nobody or treating everybody (no intervention or universal intervention, respectively).

At a 10 % MACE risk at 10 years, this model identified an additional four people who were correctly treated, while avoiding 37 unnecessary interventions per 10,000 people tested.

If only T1D patients are tested using this model, two additional interventions would be correctly performed at this threshold (for men in the SIDIAP cohort), while avoiding 5,746 unnecessary interventions per 10,000 people tested, according to decision curve analysis estimates.

The specific mechanisms underlying these distinctions appear to trace back to the differences in fasting glucose, for men and women, systolic blood pressure differences in women, and low-density lipoprotein (LDL, ‘bad’ cholesterol) in men. These represent potential preventive targets for reducing CVD risk in this population.

Chronic hyperglycemia in T1D may mask other relevant cardiovascular risk factors and profiles, making risk stratification more difficult. Further, CVD pathways in people with good glucose control may differ from those in people with hyperglycemia.

Advantages of this approach

Notably, this approach relies on routinely collected clinical biomarkers and does not require additional specialized testing beyond the waist-to-hip ratio, an easily obtained metric, and imposes no additional burden on the healthcare system. However, it could help clinicians determine when and how to test for CVD prevention in the at-risk T1D population, provided it is integrated into clinical routine, such as through publicly available digital tools (e.g., https://shiny.gbiomed.kuleuven.be/UMAP_app/), especially if linked to electronic health records.

In addition, even small gains in the ability to correctly identify high- and low-risk patients are important for identifying complications early and preventing them.

Potential limitations

The study used cross-sectional data, whereas real-world data changes over time and could shift risk profiles towards other phenotypes. Longitudinal studies are required to trace changes in these profiles in the general and T1D population.

However, new evidence suggests that cross-sectional data can predict outcomes as well as, or even better than, longitudinal data, and are easier to access in daily clinical practice.

The study used European data, limiting its generalizability.

Implications and future directions

The findings confirm the validity of the original model when extended to T1D patients. It also illustrates an association between better glycemic control and lower-risk profiles, rather than establishing a direct causal reduction in CVD risk, in contrast to earlier reports suggesting glycemic control alone may not influence CVD risk.

Future risk models should include glucose control in risk allocation. Longitudinal studies could help validate BMI-based risk categories and follow the evolution of risk profiles. This would help develop tailored preventive and treatment strategies for type 1 diabetes.

Download your PDF copy by clicking here.