INSIGHT. A new generation of predictive health is emerging at the intersection of artificial intelligence and life science. Researchers at the European Molecular Biology Laboratory (EMBL), the German Cancer Research Centre (DKFZ) and the University of Copenhagen have developed Delphi-2M, a generative AI model that predicts the risk of developing over 1,000 diseases up to 20 years in advance.
The model is based on electronic health records, medical history and lifestyle factors such as smoking and BMI. It was trained on data from 400,000 participants in the UK Biobank and validated on nearly two million patient records from the Danish National Patient Registry. According to the study, the model achieves high accuracy for diseases with clear progression patterns, such as certain types of cancer and heart attacks, but shows lower precision for psychiatric conditions and rare diagnoses.
Delphi-2M is based on a so-called transformer architecture – the same underlying principle used in language models like ChatGPT – and can analyse and provide insights into a very broad range of diseases simultaneously. The model illustrates how AI can contribute to a more holistic understanding of health and risk, rather than replacing established diagnostic methods.
More efficient drug development: AI models could eventually be used to identify risk groups earlier and simulate disease progression digitally. This enables more efficient design of clinical trials – for example through synthetic control groups and digital twins. However, such applications remain in an early research stage.
Emerging market segments: Companies developing tools for data analysis, synthetic data and model validation are becoming increasingly important. This creates opportunities for new players, from technology companies to infrastructure providers, supporting the digital transformation of the healthcare sector.
Partnerships and collaborations: Collaborations between AI companies and pharmaceutical firms are becoming increasingly common, with the aim of improving patient selection, optimising clinical trials and modelling disease progression in virtual environments.
From a regulatory perspective, the field remains complex. The European AI Act and the medical device regulations MDR and IVDR impose high requirements on transparency, data quality and traceability. For developers of predictive models, this means the path from research findings to clinical use may be long. Beyond legal aspects, ethical questions also arise. How do we handle the risk of bias and discrimination when models are trained on historical data? And how should the right not to know one’s future disease risk be interpreted? These questions need to be resolved as the technology matures, so that trust in AI in healthcare can be built on a stable foundation.
AI models for predicting disease risk point towards a future where healthcare becomes more proactive and personalised. For investors, this presents opportunities at the intersection of technology, biology and infrastructure, where early movers may be well positioned as predictive medicine approaches clinical application.
