Predicting your medical future and Artificial Intelligence
Blog vol 6.28. Predicting your medical future and Artificial Intelligence
In September of 2025, Nature published an article about using Large Language Models (LLMs) to help predict disease in individuals. This model is called Delphi-2M, likely named after the Oracle of Delphi, who in ancient Greece channelled prophecies from Apollo to provide guidance for crucial decisions. After centuries, the oracle at Delphi, on the slopes of Mt Parnassus, made her last prophecy around 323 AD when the practise was banned by the reigning Roman emperor (Ecyclopedia Britannica online). In 2025, we now turn to LLMs, like GPT-5 to provide guidance.
The Delphi model was tasked with analyzing the text in a massive databank, the UK Biobank (400,000 strong), then predicting the next word in sequence. The programmers then added in the passage of time as a factor which resulted in more accurate health predictions. These models must have a huge database to work with and be able to adapt as new data is released. The Delphi-2M developers then ran the model through the Danish health system with over 1.9 million patients.
In the British trials, they found that the model had a prediction rate of 0.76, with 0.5 set as completely random and 1.0 as full proof. One cannot go to the bank with these numbers, so to speak, but it is a good start. As more data, like medical imaging and genome sequencing is added, the accuracy will improve.
One asks the question, why go through all the trouble? As diagnosticians, we have lists of common diagnoses at the ready all the time. We need to see LLMs as another tool in our medical bags. This information can someday help with predicting trends in disease and even help health services and teaching institutions to plan. (Read more here).
Back in 1990, I decided to take a one-year residency in Low Vision Rehabilitation at Waterloo out of interest, and in anticipation of an upcoming surge in new cases of age-related macular degeneration (ARMD). However, with the decrease in smoking, improved diet, increased use of UV protection, the use of lutein in AREDS formulas, and new treatment modalities, that surge did not happen despite an aging population. This is something that an LLM could likely have predicted.
Will LLMs like Delphi-2M be able to predict our personal health future? No more than knowing that your mother has glaucoma tells you that you are at higher risk for glaucoma or if you smoke, you are at higher risk for many health issues. An LLM can simply analyze more factors, correlate them, and watch current trends. It is the same sort of thing except on a much larger scale. LLMs cannot account for unpredictables like quitting smoking or taking up running or getting in an accident, etc., all of which can have a huge impact on health outcomes.
An LLM like Delphi-2M is a tool for medical professionals to better help their patients, and health care systems better prepare for the future.
Til next week,
The good doctor
