Accurate risk stratification for TAVI
The average age of the European population is rising. This so called double population ageing refers to the increasing proportion of individuals aged 65 years or older is expected to increase 2.5 fold in the upcoming decades. Because aortic stenosis is strongly associated with age, the number of patients requiring aortic valve replacement—either through open surgery or percutaneous approaches—is expected to increase in the coming years.1 Selecting the appropriate technique and determining whether a patient is eligible for intervention at all, depends largely on the estimated risk of postoperative complications. Currently, EUROSCORE II is used for this purpose, yet accurate risk prediction remains challenging and carries significant clinical implications.2
Improving risk stratification by deriving biological age from chest radiographs
Stambollxhiu et al.3 elegantly demonstrate that risk prediction can be improved beyond traditional scores by using deep learning to estimate biological age from chest X rays. Their convolutional neural network, trained on 100,000 chest radiographs, integrates radiographic signs of ageing into a predicted biological age. When tested in 1,269 patients who underwent either TAVI or surgical AVR, the algorithm achieved a higher area under the curve—indicating superior predictive performance—for mortality compared with EUROSCORE II. Although the algorithm was also numerically better than chronological age, this difference was not statistically significant.3
A new trend: deriving biological age from retinal and face images
This study fits within a broader trend in which deep learning applied to subtle markers of ageing outperforms conventional risk scores in predicting adverse events. Another article in the July issue of European Heart Journal – Digital Health reports a model predicting major adverse cardiac events using subclinical signs of atherosclerosis extracted from retinal images.4 A previously published study in the European Heart Journal (2020) showed that coronary artery disease could even be predicted from a selfie.5 It is an interesting field: deep learning applied to classical markers of ageing appears capable of surpassing traditional risk scores. This important line of research may help clinicians refine risk stratification for cardiac surgery and TAVI. It also raises questions about how heart teams will determine treatment decisions for patients with severe symptomatic aortic stenosis in the future. Rather than relying on a score, a retina image, a chest X ray, or even a selfie might one day indicate whether a patient is fit for surgery. Stambollxhiu et al. should be congratulated on their results and publication. Further research is eagerly awaited.
