Sometimes the practice of medicine seems more of an art than a science. Doctors must weigh the knowledge that they gained during medical school, their understanding of their patients’ medical histories, the available treatment options, and, last but not least, their own experience to recommend a therapy. To enable doctors to provide more personalized, and more effective, care to their patients, IBM has recently developed AI software to make it happen.
Historically, doctors have relied on controlled trials to provide them with information. However, these randomized trials generally have fairly small and, to control other variables, homogeneous groups of subjects. As a result, it can be difficult to effectively extrapolate the results to the real world where there is significantly more variation. IBM’s AI software will help doctors gather and process information about patients in the real world and their treatment outcomes.
To build the software, the team used AI to gather data from Atrius Health’s database with approximately 20 years worth of information on almost 2.5 million patients. To start, IBM’s team focused on only three conditions: type 2 diabetes, hypertension, and high cholesterol. For patients that had the ailments, they pulled patient data from Atrius’ database, and located the times at which the disease was not under control and required a treatment recommendation, which they classified as “decision points.” At these decision points they collected all relevant information about the patient leading up to that point in time. The team was then able to use machine learning to identify additional “decision points” with a combination of the patient data and treatment guidelines. The patients found during this process were then grouped into “precision cohorts,” or patients who closely resembled each other based on their ailments and defining characteristics.
Using these cohorts, the team was able to compare the outcomes of different treatments under similar conditions and ultimately identify the treatments that resulted in the best outcomes.
Armed with information on how different treatments respond in real-world conditions, doctors are able to identify the cohort most similar to their own patients and recommend that treatment that is most likely to be effective in controlling the condition. Although the team focused on only three diseases, this software could be applied to a number of ailments to help provide more personalized and effective care.
AI In Healthcare:
AI seems to have an endless potential to improve healthcare and the treatment of disease, both acute and chronic. Just recently, DeepMind’s AlphaFold had remarkable success in identifying the structure of a protein. For another example of how AI is transforming healthcare, check out Dr. Veysel Kocaman’s session, Natural Language Processing Case-studies for Healthcare Models as part of our Ai+ Training Platform.