Artificial intelligence and personalized medicine

The expression personalized medicine has been circulating for years, often as an advertising slogan. Behind it, however, lies a concrete change in the way a diagnosis is constructed, and artificial intelligence is accelerating that transition.

At IMAGE REGENERATIVE, in Milano and St. Moritz, the precision assessment collects genetic, metabolic and functional tests in a single report, built following exactly this logic.

Let's see what it really means to personalize a treatment, what role artificial intelligence plays today, which data enters the reasoning and which risks need to be carefully managed.

What is personalized medicine

Personalized medicine starts from an elementary observation, namely that people with the same diagnosis respond differently to the same treatment. The reasons for this variability lie in genetic makeup, metabolism, body composition, environment and habits, and traditional medicine manages them through successive attempts.

The personalized approach reverses the sequence. First the individual profile is measured, then the intervention with the highest probability of working on that specific profile is chosen. It is a methodological change that requires much more input data and an ability to read them together rather than one at a time.

A terminological clarification helps with orientation. Personalized medicine and precision medicine are used as synonyms, but the latter more properly indicates the ability to identify subgroups of patients with common characteristics, while the former refers to adaptation to the individual. In clinical practice the two things converge.

It is also worth clarifying what this approach is not. Personalizing does not mean multiplying tests until filling a folder, nor proposing a different protocol to each person on principle. It means choosing, among the available options, the one most likely to work on that person, and knowing how to give up everything else. Selection is as much part of the method as collection.

The role of artificial intelligence in medicine

The contribution of artificial intelligence concerns a very concrete problem of scale. A human being reasons well on about ten variables at a time, while a complete clinical profile contains hundreds, with reciprocal relationships that no eye can hold together. Machine learning algorithms identify patterns within that mass of data and make them usable.

The most mature applications today are three. The first is support for the interpretation of diagnostic images, where models signal suspicious areas to the radiologist or dermatologist, who maintains the final decision. The second is risk stratification, that is, estimating the probability that a person will develop a certain condition over time. The third is the integrated reading of heterogeneous data, where the algorithm relates test results that originate in different laboratories and at different times.

On this ground there is a precise institutional reference. The World Health Organization guide on ethics and governance of artificial intelligence for health, published in 2021, identifies six shared ethical principles for the use of these tools in healthcare, from protecting personal autonomy to model transparency, from responsibility for decisions to equity in access. The document insists on a point worth reporting, namely that technology should be evaluated for the real benefit it brings to patients and not for its sophistication.

Data-based diagnosis and prevention

The transition from reactive to predictive medicine depends on the quality and variety of data collected. A blood count alone says little, while the same blood count read together with metabolic profile, body composition and sleep quality begins to tell a trajectory.

The data entering this type of assessment belong to different families:

• Genetic data, which describe individual predispositions and help understand which fronts are worth investing in for prevention.
• Metabolic and lipidomic data, which photograph the current functioning of the organism and are the most sensitive to lifestyle changes.
• Markers of oxidative stress and inflammation, which signal ongoing processes before they produce recognizable symptoms.
• Functional parameters, from body composition to strength to recovery capacity, which measure a person's physiological reserve.
• Behavioral data, particularly sleep, circadian rhythms and physical activity, now collectible continuously and no longer only through patient reporting.

Clinical value arises from the intersection. In our experience, the integrated report intercepts early imbalances that individual tests, read separately and perhaps months apart, let pass unnoticed, and this moves forward the moment when intervention is still possible with simple measures.

AI and longevity: predicting to prevent

Longevity medicine has a different objective from traditional medicine, because it deals with people who are well and want to remain so. This changes the type of question, which no longer concerns the diagnosis of a present disease but the estimation of a future trajectory.

In this area, predictive tools find their most natural application. Epigenetic clocks estimate biological age from DNA methylation patterns, cardiometabolic risk models combine dozens of parameters to project a trend over time, and the analysis of sleep and activity data identifies gradual deteriorations that no one notices living them day by day.

However, honesty is needed about the maturity level of these tools. A biological age is a useful path indicator for measuring whether an intervention is working, not a prediction about lifespan, and should be presented as such. In the pathways we follow, these data serve to establish priorities and verify changes over time, while clinical decisions remain with the physician who knows the person.

There is also a practical advantage that is only noticed by using them. An indicator moving in the right direction supports adherence much more than a generic recommendation, because it makes visible a change that would otherwise remain abstract for months. Those who see a measured parameter improve tend to maintain the habits that produced it, and this motivational effect is a concrete part of the clinical outcome.

Opportunities and risks of AI in medicine

The opportunities are concrete and already visible. Early diagnosis gains sensitivity, prevention becomes targeted instead of generic, physician time is freed from repetitive activities, and community medicine accesses specialist skills that previously required a referral center.

The risks deserve equal attention, and must be named precisely. Here are the four that weigh the most.

• Bias in training data, because a model built on poorly representative populations works worse on those who do not resemble them, with an effect that falls on already disadvantaged groups.
• The opacity of complex models, which makes it difficult to understand why a system produced a certain indication and complicates both clinical verification and informed consent.
• The protection of health data, among the most sensitive information that exists, with the need to know where it is stored, who processes it and for how long.
• The question of responsibility, because a wrong recommendation produced by an algorithm still falls on the professional who accepted it, and this requires clear rules before adoption.

There is an additional risk, less discussed and very practical. Easy access to predictive tools generates an excess of tests and anxiety, with healthy people chasing values without a clinical project behind them. The selection of which data to collect, and especially which to ignore, is as much part of medical work as their interpretation. An out-of-range value without clinical relevance produces cascading tests that cost time, money and worry, and the task of those reading the report is also to stop that chain when it leads nowhere.

The future of personalized medicine

The direction is reasonably clear, and concerns less the arrival of new tools and more their integration. Genetic, metabolic, functional and behavioral data will tend to converge into a single profile updated over time, instead of remaining in separate reports that no one ever reads together.

The second change concerns measurement continuity. The current model photographs health status once or twice a year, while wearable devices and repeated analyses allow tracking a trend, which for prevention is much more useful information than a single value. An isolated datum may depend on a bad night or a period of stress, while a series of repeated measurements distinguishes occasional variation from real trend.

The third concerns the role of the people involved. Artificial intelligence is a decision support tool and its value depends on who uses it, the quality of the data it receives and the clinical relationship in which it is embedded. The part of medical work that consists of listening, contextualizing and deciding together with the person remains the decisive one, and it is also the reason why a personalized pathway always begins with a consultation and not with a report.

Frequently asked questions

Will artificial intelligence replace the doctor?

In mature applications available today, the algorithm plays a support role, signaling elements that deserve attention and estimating probabilities, while the clinical decision remains with the professional. International guidelines insist on human supervision as a requirement, precisely because responsibility for therapeutic choice cannot be delegated to an automatic system.

Are personalized medicine and precision medicine the same thing?

They are almost always used as synonyms. Precision medicine indicates in the strict sense the ability to identify subgroups of patients with common characteristics to better target therapies, while personalized medicine refers to adaptation to the individual. In daily clinical practice, the two perspectives overlap.

What happens to the data collected in this type of assessment?

Health data fall under special categories protected by European data protection regulation, and their processing requires specific informed consent. It is correct to ask the facility where they are stored, for how long, who accesses them and whether they are shared with third parties, and to receive verifiable answers before starting.

Is artificial intelligence needed for a longevity pathway?

It is not a requirement. What really matters is the quality of the data collected and their integrated reading by a physician, which can occur even without algorithmic tools. The contribution of technology becomes relevant when the variables to be related are many and must be followed over time.

Content reviewed by the medical-scientific committee of IMAGE REGENERATIVE. The information provided is for informational purposes and does not replace specialist consultation.