2 min read
We know about smart algorithms that can diagnose diseases once you input a patient’s symptoms, but how about taking it one step further? What if your smart algorithm could tell you exactly what medicine you should use to treat the disease? According to researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), they have developed an algorithm that can do just that.
The recommendation algorithm developed by the CSAIL researchers predicts the probability of a urinary tract infection (UTI) patient getting treated by first- or second-line antibiotics. Using this information, the machine learning model recommends a specific treatment fairly quickly. This is a massive breakthrough because conventionally, as a doctor, you would have to experiment with a number of treatments, some of which are likely to fail and will result in a lot of precious time being consumed. This algorithm, however, makes it much easier to land upon the perfect treatment for a disease (in this case, UTI).
According to the CSAIL team, their model was trained on data from more than 10,000 patients from Brigham & Women’s Hospital and Massachusetts General Hospital. It is so effective that it will allow clinicians to reduce their usage of “backup” antibiotics by 67%. In fact, when it came to selecting the appropriate first-line drug for a patient, the algorithm was right more than 90% of the time. And whenever clinicians chose the wrong first-line drug, the algorithm corrected them almost half the time.
The team of researchers do admit that they haven’t tested their algorithm on more complex forms of UTI, nor have they experimented with randomized control trials. Moving forward, they will increase the diversity of their sample size to improve recommendations across race, ethnicity, and socioeconomic status as well.
MIT professor and research co-author David Sontag believes that the bets thing about this algorithm is its ability to allow clinicians to better evaluate their decision-making when it comes to selecting appropriate treatments.
“What’s exciting about this research is that it presents a blueprint for the right way to do retrospective evaluation,” he said. “We do this by showing that one can do an apples-to-apples comparison within the existing clinical practice.”
