For centuries, humanity has asked: *“If we could know future diseases in advance, how many lives could we save?”* Modern medicine has greatly advanced in diagnosis and treatment, but in terms of preventive medicine, the limitations have remained significant. Most predictive tools focus on a single disease or only assess short-term risks.

For example, cardiovascular prediction tools calculate the likelihood of disease within five years based on blood pressure, cholesterol levels, and smoking habits. However, they do not take into account the possibility that the same person may simultaneously develop diabetes, kidney disease, or a stroke. Human health unfolds within complex interactions and patterns of multimorbidity, but existing systems have not been able to address this comprehensively.

With aging populations and rising chronic diseases burdening healthcare systems, 'long-term multi-disease prediction' has become not just a research issue but a social necessity. It is precisely here that 'Delphi-2M', developed by the European Molecular Biology Laboratory (EMBL) and its research collaborators, emerges as a groundbreaking solution.

The most distinctive feature of Delphi-2M is its 'scale and comprehensiveness'. The model was trained on more than 400,000 participant records from the UK Biobank combined with data from 1.9 million patients in the Danish National Patient Registry. Through this massive dataset, the AI does not merely predict the probability of individual diseases but also detects 'interactions and cascading risks between diseases'.

The core elements are:

* 'Multi-disease prediction': Unlike single-disease models, Delphi-2M predicts more than 1,000 diseases simultaneously.

* 'Long-term forecasting': Instead of short-term risks, it estimates probabilities decades into the future.

* 'Interaction detection': If obesity and hypertension occur together, the model does not just add their risks but also captures how they jointly raise the likelihood of cardiovascular complications, stroke, or kidney disease.

Because of this design, Delphi-2M functions as a '“personal health simulation engine.”' It can forecast how an individual’s risk for numerous diseases will evolve over 10, 20, or 30 years if current lifestyle patterns remain unchanged—providing a foundation for personalized management strategies.

Delphi-2M attracted global attention in 2025 when it was published in 'Nature'. In trials predicting major conditions such as cardiovascular disease, diabetes, and sepsis, the model demonstrated accuracy equal to or better than conventional single-disease prediction tools.

Particularly, in validation using the Danish National Patient Registry, the AI showed its ability to evaluate long-term risks by learning 'patterns of patient medical histories'. For instance, if someone was diagnosed with obesity in their 30s and began hypertension medication in their 40s, the model could calculate not just each disease’s individual risk but also the compounded probability of cardiovascular complications in their 50s.

The research team described this as a '“breakthrough from single-dimensional prediction to multidimensional simulation.”' This achievement was not merely a technical advance but also a redefinition of the very concept of preventive medicine.

What Delphi-2M opens up goes far beyond simple risk forecasting.

1. 'Personalized Health Management'

Medical guidelines so far have been based on the “average patient.” Now, strategies can be tailored to each individual’s genes, medical history, and lifestyle. For example, two patients with hypertension may face entirely different risks—one more prone to diabetes, the other to kidney disease.

2. 'Efficient Allocation of Healthcare Resources'

Identifying high-risk groups early enables early interventions. This means that instead of spending enormous costs in emergency rooms or intensive care units, resources can shift toward cost-effective management at the primary care level.

3. 'Innovation in Pharmaceuticals and Insurance'

The model can be applied to select participants for clinical trials or design risk-based insurance products. With personalized drug development on the rise, predictive AI can significantly improve trial efficiency.

Ultimately, Delphi-2M has the potential to move healthcare from treatment-centered to 'prevention- and prediction-centered systems'.

Although Delphi-2M’s vision is compelling, its practical realization still faces several obstacles.

* 'Data Privacy': Since millions of sensitive medical records are used, strict safeguards for personal data and ethical standards are essential.

* 'Generalization Issues': The model is trained primarily on European populations; ensuring the same accuracy for Asian, African, or other groups will require additional training and validation.

* 'Explainability': For real-world clinical adoption, doctors must be able to explain results to patients. Questions like *“Why is this person’s 20-year diabetes risk 60%?”* require clear, interpretable answers from the model.

* 'Institutional Adoption': To incorporate predictions into national insurance or healthcare systems, broad consensus and regulatory frameworks are needed.

Thus, Delphi-2M remains a technology standing '“on the bridge from research to clinical practice.”'

If Delphi-2M is widely implemented, its societal effects will be profound.

* 'National Health Policy': In aging societies such as Korea, Japan, and Europe, shifting to prevention-based policy could significantly cut medical costs.

* 'Insurance and Finance': Risk-based personal data could create new types of insurance products and financial models tailored to individuals.

* 'Healthcare Inequality': While developed nations and wealthy populations may reap the most benefits, underserved regions risk falling further behind. International cooperation and data-sharing frameworks will be vital.

* 'Industrial Competitiveness': Countries that lead in AI-driven predictive healthcare will gain advantages across healthcare, biotech, and data industries. For Korea in particular, early participation and technological capacity will be decisive.

The vision Delphi-2M presents is clear: a transition to '“medicine that manages health before the onset of disease.”'

Over the next 10–20 years, as models are trained with diverse international datasets and as ethical and institutional frameworks solidify, individuals will be able to anticipate their health futures and adjust their behavior accordingly. For instance, if AI predicts a 70% likelihood of heart attack within 15 years, an individual can change their diet and schedule regular checkups immediately.

Ultimately, this could transform not just healthcare but entire social structures—cutting healthcare costs, maintaining labor productivity, and ensuring sustainability in aging societies. By merging AI with preventive medicine, humanity could achieve a 'paradigm shift from managing diseases after they occur to managing them before they arise.'

Nature, 2025, “AI model Delphi-2M predicts risk of over 1,000 diseases decades in advance”, EMBL & collaborators.