Brandon Fornwalt from Geisinger and his colleagues have revealed groundbreaking research findings, proving that their Artificial Intelligence Model can predict patient death within a year better than practicing cardiologists, relying only on electrocardiogram (ECG) results – a common medical procedure to measure electrical heart activity to detect cardiological conditions.

Scientists admit that they do not yet fully understand how the AI Model is achieving high accuracy rates. Yet the testing phase showed that AI algorytm overperformed “any model you could build out of things that we already measure from an ECG”, shared Fornwalt.

The AI system, tested on historical data from 400.000 patients and their 1.77 million ECG results, scored 0.85 out of 1 on AUC metric, which is used to evaluate how well the predictions correlate with actual death and survival data within a given year. In comparison, the risk scoring model widely used by doctors in US scores from 0.65 to 0.8. 

Moreover, AI Algorithm was also able to accurately predict death, when practicing cardiologists did not spot any risk patterns on ECG results. Providing additional data as patient age or sex did not result in any improvements, that suggests that AI system can derive necessary data from ECG results only.

Fornwalt believes that AI Model is able to spot something that current cardiologies are not able to detect yet, therefore doctors will have an opportunity to learn from it. While some physicians might be uncomfortable with using an algorithm that scientists do not fully understand, research team believes that it is important to proceed to clinical trials for a chance to improve “patient outcomes”.