Foresight: AI integrated into digital twins to predict your future health

Foresight: AI integrated into digital twins to predict your future health
Foresight: AI integrated into digital twins to predict your future health

Share the post “Foresight: AI integrated into digital twins to predict your future health”

Minority Report, but applied to health. This artificial intelligence tool is called Foresight and can predict your future health. Capable of creating digital twins of patients, it uses more or less the same principle as the large language models used by ChatGPT from OpenAI, Claude from Anthropic or Llama 3 from Meta. Technically, Foresight is a “generative transformer in temporal modeling of patient data.” According to a study, published in April 2024 in the Lancet Digital Healthcarried out by British researchers from King’s College London, the reliability of its projections seems high enough to consider use in the future.

To make this tool work, these researchers first trained it on the basis of numerous anonymized medical records. They then fed Foresight with new health data to create digital twins – virtual duplicates of patients. Once these profiles have been created, it becomes possible to project oneself into various hypotheses and to analyze, thanks to the significant computing power and data analysis of artificial intelligence, the probabilities of developing this or that disease, the performance of treatments and the need, or not, to act in prevention.

Digital twins with an already very interesting reliability rate

For scientists, Foresight – and AI more broadly – ​​could sharply and refine their diagnostics to make more effective treatment decisions. During their experiments, data from a total of 811,000 patients was screened. The researchers asked Foresight to project the health status of American patients based on medical records from Beth Israel Deaconess Medical Center and the MIMIC-III database. The goal was to predict the “next 10 possible health conditions” that could appear in the patient. The AI ​​was correct in predicting their future fitness with an accuracy of 88%.

In contrast, the same reliability tests applied to data from cohorts of British individuals (from King’s College Hospital NHS FT and Maudsley NHS Foundation Trust). In this scenario, the tool accurately predicted subsequent conditions with an accuracy of 68% and 76%, respectively.

Much more varied data taken into account

Foresight “is trained on large amounts of real NHS data and uses free text contained in notes, reports, letters, etc. doctors. Information that is generally ignored but contains rich and detailed information that represents more than 80% of a patient’s record. The tool therefore captures deep and subtle phenotypic information”, said Richard Dobson, professor of medical informatics at King’s College London. James Teo, another co-author of the study and director of data science and AI at King’s College Hospital, believes the predictions represent “possible multiverses” for understanding diseases.

“Our generative AI is able to produce predictions from the health record for any disorder, test, drug, treatment or complication in the future for all types of diseasessaid James Teo on X (Twitter). This digital twin of the patient can provide insights and ‘what if’ scenarios. »

https://twitter.com/jthteo/status/1770714167047025118?ref_src=twsrc%5Etfw

And to add: “In the short term, this technology could support virtual trials to find potential new treatment avenues, predict possible complications and evaluate outcomes. It would also be possible to detect unexpected complications or adverse events not detected by traditional means. Pedagogically, it is also a useful tool for teaching realistic outcomes of imaginary patient scenarios rather than rarefied but improbable textbook scenarios during medical examinations. »

Progress still necessary for these digital health twins

Although the first tests are very encouraging, the time has not yet come for use in real conditions. Before the technology is deployed in the wild, further adjustments and testing are necessary. British researchers have already announced that they are working on a “more precise” tool called Foresight 2.

The question of ethics is also crucial. What about the protection of privacy and personal data? What impact on the patient-doctor relationship if an AI takes part in the diagnosis? We can wonder about the degree of adherence of individuals to a therapeutic protocol that would have been defined with the help of an automatic decision-making system. And to what extent should they be warned of the involvement of an AI in this process?

“For the moment we are not ready to delegate decisions to these systems, especially when a patient’s life depends on it, declared Laurence Devilliers, professor of AI at Sorbonne University on the Inserm website. At the same time, there is an urgent need to build a law, standards and ethical rules to govern the use of predictive systems, in order to minimize the risks of manipulation and dependence. This involves verifying their robustness and compliance with ethical criteria such as the freedom and autonomy of decision-making of humans when using these tools.

SUPPORT WE TOMORROW, SUPPORT INDEPENDENT EDITORIAL
Subscribe to our weekly newsletter
And subscribe to our magazine.

-

-

PREV Treating overall health through singing
NEXT a substance present in these 3 fruits can improve memory