← Back to Library

It's time for primary prevention in medicine

Dr. Eric Topol · Ground Truths · ·13 min read
Commentary by Hex Index staff

Dr. Eric Topol makes a provocative claim that could redefine the entire trajectory of modern medicine: the era of merely treating disease is ending, and the dawn of true primary prevention is finally here, powered not by better drugs, but by artificial intelligence that can read the "grammar" of human health. While the medical establishment has long accepted that we are failing to stop cancer, heart disease, and neurodegeneration before they strike, Topol argues that recent breakthroughs in large language models and biosensors have finally provided the missing tools to predict and avert these conditions decades in advance.

The Grammar of Health

Topol begins by dismantling the comforting illusion that current medicine is effective at stopping disease. He notes that despite 75 years of effort since the concept was coined, "we've yet to achieve any substantive primary prevention with the notable exception of vaccinations." The current system, he argues, is reactive, focused on "treating these diseases, which has limited success for many cancers and even less, thus far no disease-modifying impact, for neurodegenerative diseases."

It's time for primary prevention in medicine

The core of his argument rests on a new study published in Nature by Moritz Gerstung and colleagues, which introduced Delphi-2M. This is not a standard diagnostic tool but a transformer-based large language model that treats human health events like a sentence. "What if an individual's arc of health events was like a sentence, a sequence whereby you could predict the next condition like you would with a transformer architecture A.I. autocompletion of words?" Topol asks. By training on the health records of 400,000 people in the UK Biobank and validating against 1.9 million in Denmark, the model learned to predict the timing and occurrence of over 1,000 diseases.

The results are striking. The model achieved an area under the curve (AUC) of 0.76 across all diseases and a near-perfect 0.97 for predicting death. Topol emphasizes that this model can project a person's health "20 years ahead," identifying clusters of risk that human clinicians miss. "This represents a jump from my prior piece on precision medical forecasting," he writes. The implication is profound: we are moving from guessing who gets sick to knowing exactly when and how, allowing for intervention before the first symptom appears.

Critics might note that an AUC of 0.76, while impressive for a broad model, still leaves significant room for error in individual predictions, and the reliance on historical hospital data introduces biases where sicker populations are overrepresented. However, Topol is candid about these limitations, acknowledging that the model did not yet include polygenic risk scores or epigenetic clocks, layers that could push accuracy even higher.

"We've gotten used to large language models that predict the next word in a sentence, but just imagine how powerful a large health model will be when all the other layers of data beyond those utilized in Delphi-2M are integrated."

The Personal Health Agent

Prediction is only half the battle; the other half is action. Topol argues that even with perfect data, humans struggle to change their behavior without guidance. To bridge this gap, he highlights a new development from Google DeepMind and Columbia University: the Personal Health Agent (PHA). This is not a simple chatbot but a multi-agent system where one AI analyzes data, another acts as a medical expert, and a third serves as a health coach, all coordinated by an orchestrator.

The system was designed to handle the complexity of real-world health data, including unstructured notes and continuous sensor feeds. Topol reports that in benchmark tests, clinicians found 65% of the summaries generated by the PHA useful, compared to just 30% for a base model. More importantly, experts ranked the PHA as the most effective system for helping users achieve their goals in 80% of cases. "The teamwork is coordinated by a project manager or orchestrator that harmonizes the work and output of the 3 agents," Topol explains. This addresses a critical failure point in digital health: the inability to synthesize disparate data streams into actionable, personalized advice.

However, the path forward is not without friction. Topol admits the system has not yet undergone a prospective study in the real world to prove it actually prevents disease outcomes. There are also valid concerns regarding data privacy, security, and the risk of users becoming over-reliant on algorithmic advice. "It clearly will require clinician oversight," he cautions, positioning the AI as a complement to, not a replacement for, human judgment.

The Convergence of Data Layers

The most compelling part of Topol's argument is the convergence of these AI models with emerging biological technologies. He points to epigenetic organ clocks that can measure the accelerated aging of specific organs like the heart or brain from a single blood sample, and continuous protein monitoring sensors that track inflammation in real-time. These tools provide the "biological truth" that AI models need to refine their predictions.

Topol envisions a future where these layers are integrated into a comprehensive prevention strategy. "Knowing that a person's organ or immune system is out of step with the rest of their body for accelerated aging may be considered a tipoff to guide preventing a disease," he writes. For instance, if an immune clock shows senescence, the risk of cancer spikes, prompting immediate preventative measures. He specifically highlights the potential to prevent Alzheimer's in high-risk individuals using blood biomarkers like p-tau217, a shift from managing dementia to stopping it before it starts.

This synthesis of data is where the economic argument becomes undeniable. Topol notes the "profound economic benefit of primary prevention for reducing the cost of such treatments, such as tailored oncology drugs or support of people with dementia in long-term care facilities." The current system is financially unsustainable because it waits for catastrophe; the new model aims to intercept the trajectory of disease.

"The new era of primary prevention would not be possible without it. It will take time to get this to be standard of care, that is real prevention, but I believe it will ultimately wind up being seen as the most important contribution of A.I. for promoting human health."

Bottom Line

Topol's argument is a powerful, evidence-backed vision that shifts the medical paradigm from reactive treatment to proactive interception, leveraging AI to finally realize the 75-year-old promise of primary prevention. While the technology is nascent and faces hurdles regarding bias, privacy, and clinical validation, the convergence of large health models, multi-agent AI, and advanced biosensors represents a genuine inflection point. The strongest part of this case is the demonstration that we can now predict the future of health with enough accuracy to act; the biggest vulnerability remains the translation of these algorithms into equitable, real-world clinical practice.

Sources

It's time for primary prevention in medicine

by Dr. Eric Topol · Ground Truths · Read full article

Primary Prevention means a disease or condition is averted. The term was coined and introduced by Leavell and Clark in the late 1940s. Now, about 75 years later, we’ve yet to achieve any substantive primary prevention with the notable exception of vaccinations that prevent infectious diseases. Of the 3 major age-related diseases that I focused on in SUPER AGERS—cardiovascular, cancer, and neurodegenerative—we have not prevented the latter two. Remember screening for cancer (such as mammography, colonoscopy, PSA, or total body MRI as some have advocated without adequate data) is a secondary prevention, with the objective of finding cancer at an early stage. Nothing meaningful has yet been shown to prevent neurodegenerative diseases. While there has been some preemption of cardiovascular disease with the use of lipid panels and cholesterol lowering drugs, heart attacks (and heart disease) and strokes remain the number 1 and 3 medical causes of death in the United States, respectively. And heart disease deaths are ticking up, adjusted for age.

Instead our healthcare now is centered on treating these diseases, which has limited success for many cancers and even less, thus far no disease-modifying impact, for neurodegenerative diseases (Alzheimer’s and Parkinson’s). No less, there’s the profound economic benefit of primary prevention for reducing the cost of such treatments, such as tailored oncology drugs or support of people with dementia in long-term care facilities. I hope this brief review will convince you that primary prevention is a great and largely unfulfilled need, that in light of recent advances it should be given the highest priority. The key problem is that we haven’t had the ability to do it. Until now.

This Ground Truths post is about the new and exciting opportunity for achieving primary prevention. We’ll start with a groundbreaking study just published in Nature by Moritz Gerstung and his colleagues.

A Large Language of Health Model Predicting Individualized Risk for >1,000 Diseases — in Multiples and When.

What if an individual’s arc of health events was like a sentence, a sequence whereby you could predict the next condition like you would with a transformer architecture A.I. autocompletion of words (hopefully a helluva lot better that we have now on our phones!)? That is the essence of generative A.I.’s power for text, using tokens of words (and word fragments) with extraordinary contextualization, giving it the ability to write high quality text, even books. Delphi-2M is a transformer, large ...