Artificial intelligence is revolutionizing the field of preventive and personalized medicine. A recent study published in the journal Nature has presented significant advances in predicting diseases that could emerge within a 20-year timeframe, thanks to an innovative model called Delphi-2M. This technology is not only seen as a key tool for anticipating health problems, but its use also raises questions about potential misuse by insurers and financial entities.
The Importance of Early Prediction in Health
In preventive medicine, the ability to foresee diseases decades in advance represents a crucial advance that could help modify lifestyles or design more effective health policies. As people age, it is common for them to experience variable health episodes, ranging from periods of good health to the development of chronic conditions. Each individual experiences these changes differently, influenced by factors such as genetics, lifestyle, and socioeconomic status.
To evaluate a patient's future health, it is essential to have a complete view of their medical history. It is not merely about analyzing isolated diagnoses, but about studying their evolution over time, understanding the interactions between various diseases, and recommending specific interventions.
Advances of the Delphi-2M Model
Researchers from the European Bioinformatics Institute, DKFZ (German Cancer Research Center), and several Danish institutions have developed Delphi-2M, a model that integrates the same technology powering language models like ChatGPT. This model is capable of studying health patterns from medical histories, lifestyle habits, and pre-existing conditions.
Moritz Gerstung, head of the Artificial Intelligence Division in Oncology at DKFZ and co-author of the study, comments, “The most unexpected finding was that the model can predict over 1,000 diseases. We would have expected it to work for some, but fail for many others. This illustrates just how interconnected many diseases are and underscores the need to investigate the underlying mechanisms that connect them.”
Delphi-2M has been trained on data from approximately 400,000 individuals in the UK and validated with records from nearly two million patients in Denmark. The effectiveness of the model allows for projections of health trajectories at both the population and individual levels over two decades.
A Probabilistic Approach
Like weather predictions, the projections from Delphi-2M do not guarantee certainties but instead calculate probabilities regarding the risk of developing certain diseases over a specific period. For instance, when assessing the risk of a heart attack over the next 10 years, the model achieves a 70% accuracy rate. However, when the forecast is extended to 20 years, the accuracy rate drops to 14%, which still exceeds the 12% achieved by knowing only the individual's age and sex.
For example, the model estimates that men aged 60 to 65 in the UK Biobank cohort have an annual risk of heart attack ranging from 4 in 10,000 to 1 in 100, depending on their medical history and habits. Meanwhile, the average risk for women is lower, although variability remains similar. The model’s results have been validated by comparing its predictions to the actual incidence of cases in different age and sex groups, confirming that the calculated risks accurately reflect population trends.
Comparison with Other Prediction Models
Delphi-2M shows a level of accuracy comparable to models specifically designed for individual diseases, such as dementia or myocardial infarction, and outperforms algorithms that predict mortality. In the case of diabetes, however, the glycosylated hemoglobin A1c (HbA1c) indicator remains more reliable.
The study also identifies diseases that can increase the risk of others, such as mental disorders or certain types of tumors in the female reproductive system.
Ethical and Application Challenges
Despite its potential, predicting future diseases raises questions about the impact on individuals' mental health. Gerstung suggests that more studies need to be conducted to explore how this knowledge can truly benefit patients. He believes that applications of AI in medicine should be subjected to randomized clinical trials, where one group receives medical care alongside AI support and another does not, to measure any significant differences in the benefits obtained.
Equally relevant is the risk of discrimination by entities such as insurers, which could use this data to exclude patients they view as less profitable. Guillermo Lazcoz, a member of the Research Ethics Committee at the Carlos III Health Institute, argues that the use of AI in large healthcare databases introduces new risks, including the identification of individuals from data that should be anonymous.
Protection Mechanisms and Regulations
Despite the risks, Lazcoz emphasizes that there are numerous layers of control in Europe to prevent the misuse of biomedical data. Access to biobanks, for example, requires researchers to justify their scientific criteria and limits the use of samples to authorized purposes. According to Mikel Recuero, a researcher at the University of the Basque Country and a data protection expert, regulations concerning identifiable data require restrictions on their use to prevent applications in the insurance and financial sectors.
The new European regulation on the health data space reinforces this stance by prohibiting commercial decisions based on genetic information. While risks are present, there are ethical, regulatory, and legal mechanisms in place aimed at limiting the misuse of information and ensuring social benefit in projects handling these data.
A New Era of Synthetic Data
One of the most interesting advances of the Delphi-2M model is its ability to generate synthetic health data. Based on partial information, the model creates complete trajectories that retain the statistical properties of real data, without relating to specific individuals. This not only protects patient privacy but also enables the training of other AI models without needing access to sensitive clinical data.
Thanks to this capability, potential impacts on public health could be explored, such as what would happen if obesity in a population increased by 5%.
Conclusions
While algorithms already exist that predict the risk of certain diseases, the approach of Delphi-2M encompasses the complexity of human health, characterized by the interaction among multiple diseases. In a context of an aging population, the ability to foresee the burden of diseases and formulate preventive policies will be crucial in addressing the challenges posed by the future of health.
For more information on advancements in artificial intelligence and its application in health care, readers are encouraged to continue exploring the content of this blog.
