Cardiovascular risk-prediction tools fall short in type 2 diabetes
A new study, reported in Diabetologia in January, tested the performance of 22 cardiovascular risk scores in primary care data from people with type 2 diabetes. The findings suggest these tools do not provide an accurate guide to prevention and treatment decisions. Dr Susan Aldridge reports.
Cardiovascular disease (CVD) remains the main cause of morbidity and mortality among people with type 2 diabetes, despite advances in treatment. Therefore, prevention and treatment of CVD should play an important part in type 2 diabetes management.
Initiation and intensification of CVD treatment in clinical practice is generally guided by risk-prediction algorithms or tools. In the UK, the National Institute for Health and Care Excellence recommends the use of QRISK2 for people with, and without, diabetes. Meanwhile, the European Society of Cardiology does not recommend a specific CVD risk tool. Instead, it stratifies people into three categories based on: presence of target organ damage, number of risk factors, and diabetes duration and age.
In diabetes, CVD differs from that in the general population. There is more heart failure (HF) and peripheral artery disease (PAD), while haemorrhagic stroke is less common. Moreover, the extent of CVD risk among people with type 2 diabetes varies considerably, which points to a need for risk-stratified management.
There are over 300 published CVD risk-prediction tools, many of which have not been validated in people with type 2 diabetes or in people with established CVD. Nor have they even been compared with one another for performance within the same population. So, it is not clear which of these tools should be used when planning CVD risk management in the individual with type 2 diabetes.
Poor performance in type 2 diabetes
Researchers at University College London and Utrecht University have now looked at how well these risk tools actually perform in predicting ‘standard’ CVD – namely coronary heart disease (CHD), stroke and PAD. They also looked at their performance in a broader definition of CVD outcomes, ‘CVD+’, that included HF and atrial fibrillation (AF), as these are more common in diabetes. The study further explored the tools’ performance in individual disease types: stroke, CHD, AF and HF. The study covered tools designed for people with diabetes and those designed for the general population. The latter are often used in clinical practice, as it is just easier to use a single tool.
The performance analysis was applied to the primary care electronic health records of 168,871 individuals with type 2 diabetes without CVD prior to or just after diagnosis. These individuals were followed up until their first CVD event, end of the study or the 10-year follow-up landmark. During this time, 38,335 (22.70%) individuals suffered a CVD, AF or HF event.
The researchers studied 22 different risk-score models predicting the 10-year risk of CVD, nine of which specifically applied to individuals with type 2 diabetes alone. Statistical analysis showed that all the scores performed poorly in predicting outcomes, particularly among those with diabetes and CVD at baseline. The researchers also externally evaluated two risk-prediction tools that are widely used in the UK, namely QRISK2 and QRISK3, which have good predictive ability in the general population. They found these are weaker when applied to people with type 2 diabetes.
The need for a diabetes-specific CVD-prediction tool has been discussed previously. This would need to account for the excess risk in diabetes that can’t be explained by conventional risk factors and would have to include factors like HbA1c and duration of diabetes. In this study, diabetes-specific CVD risk tools didn’t perform any better, so clearly more work is needed on this in order to create such a tool.
However, there was some positive news, for the researchers found that performance of the tools could be dramatically improved by applying a recalibration process using an independent dataset of just a few hundred cases. Given the increased availability of electronic health record data, it should be possible for healthcare commissioners to apply this process at a local level. This could be an attractive and cost-effective alternative to creating an entirely new model and the paper provides a straightforward computer application to support those who want to try this. In the meantime, clinicians and people with type 2 diabetes should be aware that current risk-prediction models may not be providing sufficiently accurate information to guide decisions on CVD prevention and treatment.
Read this paper at:
Dziopa K, Asselbergs F, Gratton J, Chaturvedi N, Schmidt AF. Cardiovascular risk prediction in type 2 diabetes: a comparison of 22 risk scores in primary care settings. Diabetologia online 15 January 2022. https://doi.org/10.1007/s00125-021-05640-y
Any opinions expressed in this article are the responsibility of the EASD e-Learning Programme Director, Dr Eleanor D Kennedy.