AI can now predict diseases you'll develop in 20 years

AI's Crystal Ball: Predicting Your Future Ailments Decades in Advance

Imagine a future where your doctor doesn't just treat your current ailments, but foresees illnesses you might face in two decades. This isn't science fiction anymore, thanks to a groundbreaking AI tool developed by German scientists. Researchers at the Heidelberg Center for Cancer Research have unveiled Delphi-2M, an advanced artificial intelligence system capable of predicting your predisposition to over a thousand diseases. What's truly remarkable is its ability to peer into your health future, sometimes for as long as 20 years.

Unraveling the Code of Future Health

Delphi-2M operates by meticulously analyzing vast amounts of data, including your electronic health records and crucial lifestyle factors. This comprehensive approach allows it to calculate the probability of developing serious conditions such as cancers, various skin disorders, and complex immune system dysfunctions, long before any symptoms manifest. While the initial training for this powerful model was conducted using a single dataset from the UK, its capacity to simulate the risk of such a broad spectrum of diseases offers a revolutionary paradigm shift. It empowers physicians to proactively identify individuals at elevated risk, enabling timely and crucial preventive interventions.

Beyond Single-Disease Prediction: A Generative Leap

As Stephan Feuerriegel, a computer scientist from Ludwig Maximilian University of Munich and one of Delphi-2M's creators, points out, the model's simultaneous multi-disease forecasting capability is nothing short of astonishing. "The ability of this model to predict several diseases at once is truly impressive," he notes. This stands in stark contrast to most existing AI diagnostic and predictive tools, which are typically confined to assessing the risk of a single illness. Moritz Gerstung and his team, driven by the ambition to create a more versatile predictive instrument, ingeniously enhanced a Large Language Model (LLM). This sophisticated LLM, a type of generative pre-trained transformer (GPT) akin to the technology powering popular chatbots like ChatGPT, forms the backbone of Delphi-2M.

How GPT-Inspired AI Predicts Your Health Trajectory

The underlying principle of GPT-like models is to generate outputs that are statistically most probable, based on their extensive training on enormous datasets. The research team ingeniously adapted this principle. Their modified model can now predict the risk of developing an astounding 1,258 distinct diseases by examining an individual's medical history. Delphi-2M takes into account essential personal details such as age, gender, body mass index, and the presence of detrimental habits like smoking or excessive drinking. The AI was trained on the health data of approximately 400,000 UK citizens, generously provided by the UK Biobank, a treasure trove of genetic and health information.

Validation and the Promise of Global Health Impact

The predictive prowess of Delphi-2M has been rigorously tested. In most instances, its predictions either matched or surpassed the accuracy of other state-of-the-art models designed to assess the risk of specific diseases. It even outperformed a standard machine learning algorithm that relies on biomarkers – measurable indicators of biological states – to predict the risk of multiple conditions. Notably, Delphi-2M demonstrated exceptional accuracy in predicting the likelihood of various types of cancer, projecting the possibility of developing each disease within a two-decade window, all based on the information gleaned from a patient's medical records. To further validate its robustness, Moritz Gerstung and his colleagues tested Delphi-2M on the health data of 1.9 million individuals from the Danish National Patient Register, a national database meticulously tracking hospitalizations for nearly half a century. The results were highly encouraging: the model's predictions for patients in the Danish registry were only marginally less precise than those derived from the UK Biobank participants.

This suggests that Delphi-2M possesses the potential to generate reliable health predictions even when applied to national healthcare systems different from the one it was initially trained on.

This crucial finding indicates that Delphi-2M is not confined to the specific data it was trained on and holds the promise of translating its predictive power across diverse national healthcare systems. The research team is now embarking on an ambitious plan to assess Delphi-2M's accuracy using datasets from multiple countries, with the ultimate goal of broadening its global reach and impact. The full findings of this groundbreaking research have been published in the esteemed scientific journal, Nature.