Innovative AI Tool Aims to Forecast Disease Risks Years Ahead

Thu 2nd Oct, 2025

Advancements in artificial intelligence (AI) have opened new frontiers in health prediction, potentially transforming how healthcare systems approach disease management. Researchers have developed an AI tool called Delphi-2M that aims to forecast individual health risks over the next two decades. This innovative model is designed to predict the likelihood of developing various diseases, including cancer, diabetes, and heart conditions.

The development of Delphi-2M involved a comprehensive analysis of data collected from nearly 403,000 individuals participating in the UK Biobank study. By integrating factors such as a person's sex at birth, body mass index, smoking and drinking habits, and previous health events, the AI tool can estimate not only which disease a person may contract but also the timeline for its onset.

In testing, Delphi-2M achieved an accuracy rate of approximately 70% across multiple disease categories, as indicated by the model's area under the curve (AUC) score of 0.7. Although these findings are promising, it is essential to note that the model's accuracy has yet to be validated against real-world health outcomes.

Further validation was conducted using data from the Danish Biobank, where the model demonstrated similar predictive capabilities, indicating its robustness across different datasets.

While Delphi-2M showcases significant potential, the researchers caution that the tool is not yet ready for clinical application. The primary goal of the study was to highlight the capabilities of this AI architecture rather than to suggest immediate implementation in healthcare settings. Delphi-2M employs a transformer network, similar to that used in popular AI models like ChatGPT, but is uniquely adapted to analyze complex interactions among various diseases.

Previous health prediction models have predominantly focused on single diseases and utilized smaller datasets. In contrast, Delphi-2M's architecture allows it to assess risks for multiple diseases simultaneously, enhancing its predictive accuracy. When compared to other models, such as Milton, which relies on traditional machine learning methods, Delphi-2M showed superior performance without requiring extensive data inputs.

Additionally, the open-source nature of Delphi-2M is a noteworthy advancement, allowing researchers to access and adapt the model while ensuring patient privacy. By generating synthetic data that mirrors the original UK Biobank dataset, the developers maintained predictive efficacy while removing identifiable information, facilitating further research in an ethical manner.

Despite the model's promise, the researchers highlighted several challenges. One key concern is the representativeness of the training data. The UK Biobank dataset has been criticized for its lack of diversity, which may impact the model's applicability across different racial and ethnic groups. While preliminary analyses suggested that incorporating demographic factors did not significantly alter predictions, the researchers acknowledged the need for more comprehensive data.

The integration of personal health records, medical imaging, wearable health technologies, and location data could enhance the model's accuracy and utility in the future. However, the potential for privacy breaches and misuse of sensitive information poses significant ethical considerations.

As it stands, Delphi-2M is not intended for direct use by patients or healthcare providers. The model generates generalized predictions, and it is premature to rely on its forecasts for individualized health strategies. Nevertheless, ongoing research and development of models like Delphi-2M could pave the way for more personalized healthcare solutions in the future.


More Quick Read Articles »