Combating antibiotic resistance, one customized prescription at a time

Researchers are hopeful tailored prescriptions can slow the evolution of antibiotic-resistant bacteria.

Pills medicine medication treatment (photo credit: Srdjan Zivulovic / Reuters)
Pills medicine medication treatment
(photo credit: Srdjan Zivulovic / Reuters)
Israeli researchers are using machine learning algorithms to optimize which patients receive what antibiotics, technology that could provide a new response to the growing threat of antibiotic resistance. 
Overuse of antibiotics, especially broad-range antibiotics that can be prescribed for a variety of conditions, accelerates the evolution of antibiotic-resistant bacteria. This has the potential to make mild infections — like pneumonia, tuberculosis, and salmonellosis — more lethal. 
Now, researchers are hoping to slow this process by tailoring antibiotic prescriptions to patients.
To customize antibiotic prescriptions to patients, Technion researchers Professor Roy Kishony and Dr. Idan Yelin collaborated with a team led by Professor Varda Shalev at Kahn-Sagol-Maccabi Research and Innovation Institute at Maccabi Healthcare Services (KSM).
For this study, published July 4 in Nature Medicine, the team focused on studying antibiotic prescriptions for urinary tract infections, which involve a number of bacteria like Klebsiella pneumoniae, E. coli and Proteus mirabilis.
They analyzed antibiotic resistance in more than 700,000 urine cultures and developed an algorithm based on more than five million cases of antibiotic purchases made over 10 years. This algorithm predicts an infection’s resistance to antibiotics and provides a treatment recommendation accordingly. 
In analyzing these case studies, the researchers looked at three sets of data: demographic data — including a patient’s age, gender, pregnancy status, and retirement home residence — resistance levels measured in previous urine cultures, and the patient’s drug-purchase history.
“We showed that each one of these three correlations was significant, much more than is perceived,” Yelin said. Even basic demographic data was shown to be correlated with particular responses to antibiotics. “This was something people were not aware of,” he said.
The team found that their algorithm — built on readily-accessible data — is capable of reducing the likelihood of prescribing the wrong medication by about 40%. 
Their breakthrough represents a growing shift in medical research toward treatment plans based on algorithmic analysis born out of Big Data, but Yelin states that their technology, designed to help physicians learn more about what an individual needs, is still patient-centric.
“These things don’t have to contradict,” he said. “This algorithm is as personal as anything can be. It sometimes feels like an algorithm is not personal, but it really knows a lot about you.”