Artificial Intelligence Predicts Future Diseases Up to 20 Years in Advance

Advancements in artificial intelligence (AI) are paving the way for predictive tools capable of forecasting an individual's health risks decades ahead. This revolutionary approach aims to shift healthcare focus from reactionary treatments to proactive prevention, potentially transforming patient care.

A recent study has introduced Delphi-2M, an AI-driven model designed to predict the likelihood of developing various diseases over the next 20 years. Developed by a European research team, Delphi-2M utilizes data from nearly 403,000 participants in the UK Biobank to analyze and forecast health outcomes such as cancer, diabetes, and heart disease.

The model predicts not only which disease a person might develop but also estimates the timing of disease onset. It considers factors like sex at birth, body mass index, smoking and alcohol consumption habits, and a history of prior illnesses. The AI achieved a prediction accuracy with an area under the curve (AUC) of approximately 0.7, indicating about 70% reliability in theoretical settings.

To validate its robustness, Delphi-2M was tested on Danish Biobank data, where it demonstrated similar predictive capabilities. The core of this system employs a transformer network architecture, similar to that used in language models like ChatGPT. Researchers modified this architecture to utilize disease and time-specific data, enabling the model to analyze complex interactions across multiple health conditions.

Compared to traditional models, Delphi-2M shows a slight advantage in predictive accuracy and offers the benefit of being open-source. Researchers created synthetic data that emulate the UK Biobank while safeguarding patient identities, enabling widespread access without compromising privacy. Additionally, Delphi-2M requires fewer computational resources than comparable transformer models, facilitating easier deployment.

Despite its promise, the study emphasizes that Delphi-2M is still in the development phase and not yet ready for clinical application. Future iterations could incorporate richer datasets, including electronic health records, medical imaging, and wearable technology data, to enhance its predictive power.

Key challenges remain, particularly concerning data diversity and privacy. The current training data lack sufficient representation of various racial and ethnic groups, which could impact the model's accuracy across populations. Additionally, ethical considerations around data security and personal privacy need addressing before these tools can be integrated into routine healthcare.

While early diagnosis and disease prevention through AI are promising, experts caution that such technologies must be approached carefully, ensuring accuracy, privacy, and fairness. As research progresses, these models may eventually allow personal health predictions tailored to individual data, heralding a new era in personalized medicine.

Source: https://medicalxpress.com/news/2025-10-ai-tool-diseases-years-advance.html