Personalized Antibiotic Prescribing Will Help in the Fight Against Superbugs

Israeli researchers have developed machine learning algorithms that allow for the selection of the antibiotic optimal for a specific patient and their condition. They hope that personalized and accurate prescribing of antimicrobial agents will help combat the growing threat of antibiotic resistance.

Last year, the World Health Organization declared bacterial resistance to drugs, particularly antibiotics, one of the main problems in modern medicine, comparing its scale to wars and climate change. Each year, the number of microorganisms that have developed resistance to existing antimicrobial agents is increasing worldwide. This occurs due to misuse, non-adherence to treatment regimens, or incorrect dosing of antibiotics, especially broad-spectrum ones that can be prescribed for a wide range of conditions.

This creates conditions where the weakest bacteria die, while the strongest survive. They actively reproduce and mutate, passing on genes resistant to antibacterial drugs to subsequent generations of microorganisms. As a result, antibiotics cease to work, and bacterial infections that were once considered mild and easily treatable—such as pneumonia, tuberculosis, and salmonellosis—become life-threatening.

The rapidly growing number of so-called superbugs increases the risk of mass epidemics. It also undermines medical capabilities in the fight against cancer, organ transplantation, and prosthetic implantation. Each year, approximately 800,000 people die worldwide from infections caused by drug-resistant bacteria. British specialists estimate that in thirty years, they will kill more people than cancer.

Scientists from the Technion in Israel, together with specialists from the Maccabi Healthcare Services Research Institute, have proposed their solution to the problem, which they see in personalized antibiotic prescribing, that is, the ability to select the optimal drug for each patient considering a multitude of individual factors. In their study, the results of which were presented in a recent issue of the scientific journal Nature Medicine, they focused on studying the characteristics of treating urinary tract infections caused by a range of bacteria, including Klebsiella pneumoniae, E. coli, and Proteus mirabilis.

The team analyzed antibiotic resistance in over 700,000 urine cultures—cultures of microorganisms isolated from urine during bacteriological testing—and developed an algorithm based on more than five million antibiotic purchase cases over the last ten years. This algorithm successfully predicts the degree of infection resistance to antibiotics and provides accurate treatment recommendations.

By analyzing specific cases, the researchers focused on three types of data: demographic data, including information about age, gender, pregnancy status, living in a retirement home, etc.; levels of bacterial resistance based on previous bacteriological urine culture results for the specific patient; and their history of medication purchases.

"In our study, we showed that each of the three types of data plays its role in the effectiveness of antibiotic treatment, and much more importantly than we had assumed," say the scientists. "Even basic demographic data, as it turns out, affects how a specific person's body will respond to antibiotic treatment. People had not even thought about this before."

The authors found that using their algorithm, built on the analysis of readily available data, can reduce the likelihood of prescribing incorrect antibacterial agents by almost 40%.

Their work indicates further changes in the approach to developing therapeutic plans—more often it is based on algorithmic analysis of large data sets, but the researchers themselves assure that their technology, created to help doctors learn more about individual patient needs, is still an example of a patient-centered model of doctor-patient interaction.

At the same time, the scientists emphasize that one should not contradict the other: "This algorithm is as personalized as possible. And at times it may seem that it actually knows a lot about you."