AI for choosing pediatric hypertension treatment: MIPT model picks the right drug 98% of the time

Over the past five years, the number of children with arterial hypertension in Russia has grown by 17%. Doctors choose treatment by trial and error: they pick one of five approved drugs and wait 2–3 months to see whether it works. Anastasia Adamson, a master's student at MIPT, has developed a machine learning model to address this. It analyzes 154 clinical features and predicts an effective drug with 98% accuracy, so doctors can choose the right treatment on the first attempt.

Months of wasted waiting

Arterial hypertension is becoming increasingly common in children. On the surface, the solution looks simple: the doctor picks one of five approved drugs and starts therapy. But this runs into a fundamental problem in medicine: every body is different. Lisinopril works for one child, another responds better to Amlodipine, a third does well on Nebivolol, and a fourth needs a combination approach. The doctor has no way of knowing this in advance.

If the choice turns out to be wrong, a month or two is lost. During that time, the child's blood pressure stays elevated, the blood vessels take on extra strain, and the risk of left ventricular hypertrophy and other complications rises. Then a new drug is prescribed, and the waiting starts again. This cycle of failed attempts affects not only the patient's physical health but also the child's mental state: months of uncertainty and ineffective treatment take their toll.

How the model predicts the outcome

Adamson trained the algorithm on data from 272 patients. What matters most, though, is not the volume of data but its depth. The model draws on 154 features that together form a complete clinical picture of the child:

• Demographics: age, weight, body mass index, height
• Cardiology: cardiac ultrasound, left ventricular dimensions, echocardiography
• Vascular system: carotid artery ultrasound, intima-media thickness
• Hemodynamics: systolic and diastolic blood pressure under different conditions
• Hormonal status: renin, aldosterone, and catecholamine levels
• Laboratory diagnostics: kidney and liver function, electrolytes, urinalysis

The machine learning algorithm processes this mosaic of data and finds hidden patterns. It learns which combinations of parameters indicate that a given drug will be effective. The result: with 98% accuracy, the model predicts which of the five drugs will work for a specific child, before the first prescription is written.

Doctors' hunches, now in numbers

The most interesting part is that the model found statistical relationships that doctors had suspected for years but could not rigorously prove. For example, it identified a clear correlation between excess weight in children and the high effectiveness of Lisinopril in that subgroup. Doctors had noticed this in practice, but there was no formal documentation.

"It's not a replacement for the doctor, but a powerful support tool,"

Adamson says.

The final decision still rests with the pediatrician or cardiologist. But now the specialist gets scientifically grounded advice instead of pure guesswork.

What this means for pediatrics

Adamson's research shows how artificial intelligence is making its way into real-world medical practice. Instead of unsystematic trial and error, doctors will have a personalized prediction based on big data analysis. That means cutting therapy selection time from months to days, shortening periods of uncontrolled blood pressure, and lowering the risk of complications for every patient. AI is not a magician here but an assistant that expands a clinician's diagnostic and therapeutic capabilities.