Predictive medicine estimates, before any symptoms appear, the probability that a person will develop a disease within a given time frame, for example 10 years, based on their clinical data, their biomarkers and sometimes their genome. It produces probabilities, not certainties: a high risk is not a disease. Its value depends on the decision it makes possible.
What is predictive medicine?
Predictive medicine is the branch of medicine that seeks to anticipate the onset of a disease in a person who shows no sign of it, using statistical models built from large populations followed over time. It differs from diagnosis, which establishes that a disease is present, and from screening, which looks for a disease that is at an early stage but already there.
The term was born in the field of genetics and later broadened to include clinical risk scores, blood biomarkers and imaging. Today, it is one of the four “Ps” of so-called 4P medicine (predictive, preventive, personalized and participatory), a framework proposed by the American biologist Leroy Hood and discussed in France in the journal médecine/sciences (Billaud and Guchet, 2015). These authors point out that there is no consensus definition, and that the competing terms (precision, stratified, genomic medicine) reflect different visions depending on who uses them.
Here we use “predictive medicine” in the broad sense: any approach that turns individual data into a risk estimate. It is one of the pillars of personalized health.
Predictive medicine, predictive genetic tests and risk scores: what are the differences?
A predictive genetic test looks for a specific inherited variant; a risk score combines a few clinical variables; a biomarker profile measures dozens or hundreds of molecules. All three belong to predictive medicine, but they do not answer the same question, are not intended for the same people and do not have the same level of validation.
| Approach | What it measures | What it indicates | Example | Main limitation |
|---|---|---|---|---|
| Predictive genetic test (monogenic disease) | A high-penetrance inherited variant | A very high lifetime risk, within a given family | Inherited predispositions to certain cancers, familial hypercholesterolemia | Concerns a minority of people; in France, requires a medical prescription, a dedicated consultation and written consent |
| Polygenic score | Thousands of common variants, each with a small effect | A risk relative to the population | Polygenic scores for cardiovascular disease or breast cancer, under evaluation | Calibrated mostly on populations of European ancestry; does not include lifestyle or clinical measurements |
| Clinical risk score | A few variables: age, sex, smoking, blood pressure, cholesterol | An absolute 10-year risk, as a percentage | SCORE2 and SCORE2-OP (European Society of Cardiology, 2021) | Few variables; age carries most of the weight; misses atypical profiles |
| Biomarker profile (metabolomics) | Hundreds of blood measurements from a single sample | Additional predictive information for several diseases | Multi-disease metabolomic profiles (UK Biobank, 2022) | Validated at the level of research populations; not yet included in guidelines for individual decision-making |
All of them provide a probability; they differ in how much of nature and nurture they capture: the genome does not change, metabolic biomarkers shift with lifestyle, and clinical scores mix the two.
What does “predict” mean in medicine?
In medicine, to predict means placing a person in a group whose disease frequency is known, not foretelling their future. Three concepts are needed to read a predictive result correctly: absolute risk, model quality and predictive value.
Absolute risk and relative risk
A “doubled” risk does not mean the same thing if it goes from 1% to 2% or from 15% to 30%. Relative risk compares two groups; absolute risk tells you what applies to you. Scores such as SCORE2 give an absolute 10-year risk, and that is the figure that matters for decisions. We explain how to read it in the article on 10-year cardiovascular risk.
Discrimination and calibration
A model is judged on two qualities: discrimination, its ability to rank people (do those who will become ill have a higher score?), and calibration, its accuracy (when it predicts 5%, do we actually observe 5 events per 100 people?). A model can discriminate well and yet be poorly calibrated in a population different from the one in which it was built. This is the case with volunteer cohorts: in UK Biobank, participants’ mortality was markedly lower than that of the general population (Fry et al., 2017); the associations found there remain generalizable (Batty et al., BMJ, 2020), but absolute risks must be recalibrated before being applied elsewhere, as we explain in the article on UK Biobank.
Predictive value: the arithmetic of false positives
A deliberately simplified example: a disease affects 1 person in 100. A test detects it in 90% of those who have it and is negative in 90% of those who do not. Out of 10,000 people tested, 100 have the disease: the test identifies 90 of them. But among the 9,900 people without the disease, 990 will test positive by mistake. Out of 1,080 positive results, only 90 are correct, or about 8%. The rarer a disease, the more likely a “positive” result is to be wrong: a predictive result is never read in isolation and should not trigger a cascade of tests without careful thought.
What are concrete examples of predictive medicine today?
Three examples illustrate predictive medicine as practiced in 2026: the European cardiovascular scores, multi-disease metabolomic profiles derived from UK Biobank, and biological age indicators.
SCORE2 and SCORE2-OP
Published in 2021 by the European Society of Cardiology, SCORE2 (ages 40–69) and SCORE2-OP (ages 70–89) estimate the 10-year risk of a cardiovascular event, fatal or not, from five variables: age, sex, smoking, systolic blood pressure and non-HDL cholesterol. Calibrated by European region (France is in the low-risk region), they are the most thoroughly validated form of prediction and the best integrated into guidelines.
Multi-disease metabolomic profiles
In UK Biobank, Buergel and colleagues (Nature Medicine, 2022) showed, in 117,981 participants, that a profile of 168 metabolic markers added 10-year predictive information to clinical variables for eight of the 24 diseases studied, including type 2 diabetes, dementia and heart failure; no association was found for breast cancer. These results, validated in four independent cohorts, show both the potential of the approach and its limits. The technique is described in NMR metabolomics.
Biological age and metabolic age
Epigenetic clocks and metabolomic age estimate a “biological age” and derive from it a gap with chronological age. These are trajectory indicators, with no official standard, which we examine in the article on metabolic age.
What are the limits and pitfalls of predictive medicine?
Predictive medicine carries five pitfalls: confusing the population with the individual, overdiagnosing, offering false reassurance, deepening inequalities and promising more than has been proven.
- From population to individual. A 10-year risk of 8% means that 8 out of 100 people with a similar profile will experience the event, not that you will be one of them; and the model ignores whatever it has not measured.
- Overdiagnosis and cascades. Multiplying measurements in people without symptoms inevitably produces out-of-range values and incidental findings, followed by follow-up tests that carry their own risks.
- False reassurance. A low risk is not zero risk, and it does not exempt anyone from organized screening programs or from seeing a doctor if symptoms appear.
- Inequalities. Since early 2026, the French press has been questioning private check-ups that cost several hundred or several thousand euros and are not reimbursed. Prediction reserved for those who can pay for it widens health inequalities. We compare these offerings with the covered pathway in the comparison of private check-ups and the covered check-up.
- The utopian dimension. Billaud and Guchet (2015) note that the promises of personalized medicine have justified massive investments for successes that remain limited, whereas the major health gains have come from universal access to care: this criticism sets the bar for evidence.
Added to this is the question of health data, which is sensitive and whose hosting, use and transmission to the regular doctor must be regulated and consented to.
What ethical framework applies: quaternary prevention?
Quaternary prevention is the ethical framework of predictive medicine: it aims to protect people from medical interventions whose risk outweighs the expected benefit. The Belgian general practitioner Marc Jamoulle defines it as “the prevention of medicine itself,” the understanding of which “could help control the economic and human costs of care” (International Journal of Health Policy and Management, 2015; 4: 61–64). The concept was first formulated in the 1980s. The four types of prevention are defined in our prevention glossary.
Applied to prediction, quaternary prevention translates into a few simple rules:
- A measurement is justified only if its result can change a decision. Estimating a risk with no way to act on it brings nothing but worry.
- Consent is informed, and the right not to know is respected. This is especially true for genetic tests, whose results also affect the family.
- False positives are explained before the test, not after.
- The result is delivered with its uncertainty and with what it does not cover.
- The care pathway remains the standard. Prediction is added to recommended screenings; it does not replace them.
What is the doctor’s role in predictive medicine?
The doctor is the one who chooses what is worth measuring, interprets the result in the context of the person, decides together with them and protects them from excess. Without a doctor, a predictive result is just a number; with one, it is the beginning of a plan.
This role comes into play at four points. Before: establishing the indication, that is, checking that a test answers a question that is useful for this person. During: placing the number in the context of what the model ignores (family history, kidney disease, a complicated pregnancy, financial hardship). After: prioritizing one or two goals in a plan, which is what the Personalized Prevention Plan of Mon bilan prévention, France’s national prevention check-up program, does, fully covered (100%) by French National Health Insurance (Assurance Maladie) at four key ages. Over time: following the trajectory rather than a single data point.
This is the fourth “P,” participatory: the person understands their numbers, knows what can be changed and chooses their priorities. In our practice, we find that this understanding has more influence on actual change than the precision of the model.
Key takeaways
- Predictive medicine estimates, before any symptoms appear, the probability of a disease within a given time frame; it produces probabilities, never certainties.
- Genetic tests, polygenic scores, clinical scores (SCORE2) and metabolomic profiles differ in what they measure and in their level of validation.
- A result is read with three keys: absolute risk, calibration of the model in your population, and the arithmetic of false positives.
- Its pitfalls are overdiagnosis, false reassurance and two-tier prevention; quaternary prevention (Jamoulle) is the safeguard.
- The doctor establishes the indication, interprets, decides with you and follows the trajectory; the covered prevention pathway remains the standard.
What Sokrate lets you do
The Sokrate service applies these principles to the prevention check-up: an adaptive online questionnaire identifies your risks, a doctor writes and signs your Personalized Prevention Plan, and your regular doctor (médecin traitant) receives a copy unless you object. The check-up carried out by an authorized professional is fully covered (100%) by French National Health Insurance, with no upfront payment. As an option, a metabolomic analysis of 249 biomarkers can supplement this pathway when a doctor judges that it may inform a decision. Join the waitlist.