Artificial intelligence in Romanian hospitals: treatments tailored to patients' genetic profiles

Romania's healthcare system is piloting a project in which artificial intelligence analyses patients' genetic profiles and lifestyles to establish personalised diagnoses and treatment plans. The project was discussed in an interview with Adevărul by Assoc. Prof. Dr Beatrice Mahler from the Marius Nasta Institute in Bucharest, and physician Mihail Pautov.

According to the two specialists, the technology can deliver faster diagnoses, optimised treatment regimens, and the ability to prescribe the right molecule for each patient from day one. The system enables genomic evaluation of both the patient and the micro-organisms involved in infections.

What benefits does AI bring to medical practice

Beatrice Mahler explains that artificial intelligence is becoming "an indispensable partner in optimising clinical workflows," owing to the vast volumes of data it can process and the speed of its analysis — factors that can make a life-or-death difference for patients.

She highlights two new directions opened up by the technology: on one hand, understanding a patient's genetic profile allows for the rapid creation of personalised treatment regimens, in contrast to the current approach based on cohort studies involving thousands of patients and statistical averages. On the other hand, genetic analysis of micro-organisms can offer not only a rapid diagnosis but also new treatment solutions for infections — ones "we probably cannot yet anticipate."

For common chronic conditions — hypertension, diabetes, cardiovascular disease — Mahler argues that genetic analysis could enable the prescription of "the right molecule, at the optimal dose, from the very first day, maximising the chances of recovery and minimising risks for the patient."

Why genetics matters in treatment

Physician Mihail Pautov used an analogy to illustrate the role of genetics: even if two people had different eye colours (green and blue) without noticing this difference themselves, eye colour could influence their response to medication, their need for sunglasses, or the amount of sleep they require. Similarly, although humans are 99% genetically identical, that 1% difference can decisively influence a patient's response to treatment.

According to Pautov, current medical guidelines remain useful, but they are based on statistical averages that do not account for each individual's particularities. He cites hypertension as an example, noting that it is influenced by factors such as alcohol consumption, smoking, a sedentary lifestyle, and a high-fat diet — factors that do not carry the same weight for every patient. Genetics could reveal precisely which factors truly matter for each individual.

In the field of pharmacogenetics, Pautov explains that medication doses are currently set as averages for large populations — for example, 500 milligrams of paracetamol for one million people. However, due to individual differences in metabolism, a fast metaboliser might need 700 milligrams to achieve a therapeutic effect, whilst a slow metaboliser might find those same 500 milligrams toxic to the liver, requiring as little as 200 milligrams.

Nevertheless, Pautov cautions that simply scanning the population's genes will not immediately yield answers, which is why projects such as the one discussed are necessary, alongside the ROGEN project — also developed by a University of Medicine and Pharmacy — which aims to collect over 10,000 genomes in its first phase. Another project mentioned is BioMedLake, an artificial intelligence-based initiative designed for the collection, analysis, and interpretation of large volumes of genetic data too complex to be processed manually by research teams.

Who has the final say: the doctor or the AI

When asked what she would do if an AI recommendation contradicted her clinical experience, Beatrice Mahler replied that technology "cannot replace human reasoning or the physician's responsibility," and that it remains a tool in service of medical decision-making.

Mahler also drew attention to the risks involved: vulnerabilities relating to the security of genetic data, potential algorithmic errors (AI "hallucinations" or biases arising from incorrect training data), and the risk of cyberattacks, in the context of today's global information warfare.

Mihail Pautov likewise emphasises that the physician remains the one who makes the final decision and bears full moral and medico-legal responsibility for any errors. He notes that AI recommendations should be treated "with a healthy degree of scepticism" and filtered critically — not dismissed outright, but integrated into medical practice as supporting tools.