Predicting Cardiovascular risk from retinal photographs and limited demographic data: An external validation of a New Zealand algorithm
Retinal photographs are routinely taken from people with cardiovascular comorbidities (i.e. diabetes and hypertension). These images have the potential to be rapidly analyzed at low cost and used to improve cardiovascular risk prediction. This information could be made available immediately to the patient and their health care provider to both inform and motivate better risk management, allowing better targeting of expensive drugs to high-risk patients.
The main modifiable factors for cardiovascular risk are diabetes and hypertension. However, to be effective, these factors must be proactively controlled by the patient, via positive change in lifestyle and medication uptake and compliance. Hence, it is essential to assess patient's cardiovascular risk early, and support the patient with education and medication, which will then significantly reduce the long term cost of intervention.
Cardiovascular disease (CVD) is the commonest cause of hospitalisation and premature death for people with diabetes, although their CVD risks are highly variable. Expensive, new drugs are now available for reducing CVD risk in people with diabetes, and their benefits are greatest in those at highest risk. However, current CVD risk prediction equations only have modest accuracy, largely because the available predictors are all indirect measures of CVD, and more accurate CVD risk stratification is needed to better target these new medications to the right people. The retina is the only part of the human vasculature that is directly visible by non-invasive means. Several studies have recently shown that an artificial intelligence (AI) deep learning retinal image algorithm can be used for estimating CVD risk. We hypothesise that retinal vascular image analysis using AI could significantly improve current CVD risk prediction equations.
It has been well substantiated that cardiovascular disease (CVD) and chronic kidney disease (CKD) are intimately linked. Therefore, it is further hypothesized that by further developing AI models that can identify features implicating CKD from fundus images, and then including those additional features into the previously proposed cardiovascular disease model, it might be possible to increase the performance of the CVD model even further. As such, the project will be extended to investigate the application of AI algorithms on the detection of both CKD and CVD.
Recent trials have demonstrated the benefits of several diabetes medications such as SGLT2i and GLP1-RA , as well as new lipid-targeted monoclonal antibodies therapies that inhibit PCSK9. Both SGLT2i and GLP1RA are currently under consideration for restricted access funding in New Zealand based on estimated CVD risk, while the PCSK9 inhibitors are currently unfunded. Each of these medications act to reduce CVD risk further, when added to existing standard therapies for glucose, blood pressure and lipid-lowering. The effectiveness and cost-effectiveness of these medications is directly proportional to patients' pre-treatment CVD risk, and therefore the accurate targeting of the medications requires accurate risk prediction.
Scope extension: Predicting Cardiovascular risk from retinal photographs and limited demographic data: An external validation of a New Zealand algorithm