Recently, according to the latest news, British researchers have launched an AI tool that can predict a patient’s risk of death within 10 years and the probability of developing heart failure and other diseases in the future based on electrocardiograms, with an accuracy rate of over 70%.
The NHS in England is to trial a “superhuman” artificial intelligence tool that predicts a patient’s risk of disease and dying early. The new technology, known as AI-ECG risk estimation, or Aire, is trained to read the results of electrocardiogram (ECG) tests, which record the electrical activity of the heart and are used to check for problems.
It can detect problems in the structure of the heart that doctors would not be able to see, and flag patients who may benefit from further monitoring, tests or treatment.
In a world first, it will initially be trialled at Imperial College Healthcare NHS trust and Chelsea and Westminster hospital NHS foundation trust, before being tested in other hospitals. It is understood hundreds of patients will be recruited in the first instance, with numbers then scaled up for further studies.
Research published in the Lancet Digital Health journal found Aire could correctly identify a patient’s risk of death in the 10 years after the ECG in 78% of cases. Researchers trained Aire using a dataset of 1.16m ECG test results from 189,539 patients. The platform could also predict future heart failure in 79% of cases, future serious heart rhythm problems in 76% of cases, and future atherosclerotic cardiovascular disease – where the arteries narrow, making blood flow difficult – in 70% of cases.
Dr Fu Siong Ng, a reader in cardiac electrophysiology at Imperial College London and a consultant cardiologist at Imperial College Healthcare NHS trust, said: “The vision is every ECG that will be done in hospital will be put through the model. So anyone who has an ECG anywhere in the NHS in 10 years’ time, or five years’ time, would be put through the models and the clinicians will be informed, not just about what the diagnosis is, but a prediction of a whole range of health risks, which means that we can then intervene early and prevent disease.
“If, for example, it says you’re at high risk of a specific heart rhythm problem, you could be more aggressive in preventative treatment to prevent it from happening. There are some linked to weight, so you can put them through weight-loss programmes. You might even think about earlier medical treatments to prevent things from progressing, but that will be the subject of the clinical studies that we plan to do.”
Dr Arunashis Sau, a British Heart Foundation (BHF) clinical research fellow at Imperial College London’s National Heart and Lung Institute and a cardiology registrar at Imperial College Healthcare NHS trust, said the goal was to use the AI checks on the ECGs to identify people at higher risk. “ECG is a very common and very cheap test, but that could then be used to guide more detailed testing that could then change how we manage patients and potentially reduce the risk of anything bad happening.
The working principle of the “Life2vec” model is based on a large language model (similar to the language model behind ChatGPT), which analyzes the sequence of events in human life, summarizes patterns and patterns. A person is assigned a code for all parts of their life, such as S52 for forearm fracture, 072 for postpartum hemorrhage, POS3513 for someone who is a computer system technician, etc.The data used to train this model includes health data and labor market dependence of approximately 6 million Danes (according to statistics, as of February 2021, the Danish population was 5.935 million). This is almost the data for the entire Danish population. That is to say, they can predict the time of death for almost all Danes, or some major health events that may occur throughout their lives. The first author of the article, Professor Lehmann from the German Technical University, said, “We attempted to use this model to solve a fundamental question: how likely are we to predict future events based on past conditions? Scientifically, what excites us is not the prediction itself, but the data that enables the model to provide such precise answers.
The “Life2vec” model can predict a variety of outcomes, including subtle differences in personality from early mortality rates. The researchers stated that their framework allows them to identify potential mechanisms that affect lifespan outcomes and the potential for personalized interventions.The original design of Life2vec was not only to predict mortality rates, but also to infer your personality based on your life trajectory.This research article reveals that they selected some questions from authoritative models in personality testing and randomly selected “someone” from a database to conduct personality tests.
For example, people who score high on the question “I prefer working with others rather than working alone” tend to prefer socializing. Then, comparing the results obtained by humans with the model predictions, it was found that the accuracy of the predictions was much higher than that of neural network algorithms.
Life2vec is truly standing from a personal perspective for the first time, providing a possibility to glimpse the future of life from today’s choicesOf course, the “Life2vec” model still has many shortcomings in predicting the time of death, such as its inability to predict how a person will die. For example, algorithms cannot predict whether a person will die in a car accident or poisoning.
But Lehmann believes that within no more than five years, with the training of a large amount of data from countries around the world, perhaps we will see a more accurate model that can predict your more precise time of death, and even the way you die.
What is the purpose of such a large death prediction model? Researcher Lehmann believes that “this model may one day help identify a person’s disease risk, enabling them to take timely measures to maintain their health. It can also be applied to a wide range of health and social issues, such as predicting and early intervention in health problems, or helping governments narrow the wealth gap. However, such applications also bring a lot of privacy, ethical, and data security issues that need to be addressed before this model can help anyone. “One key distinction is that the goal here was to do something that was superhuman – so not replace or speed up something that a doctor could do, but to do something that a doctor cannot do from looking at heart tracing.”
