Dr. Melinda Rushing recently appeared on the Zora Talks podcast to discuss her team’s work using artificial intelligence to improve care for sickle cell disease patients. During the episode, Dr. Rushing explained the basics of sickle cell disease, its disproportionate impact on people of color, and how her team is developing a method called Clinically Guided AI.
She described their approach: “What we’re doing is saying is we’ll take the notes, start pulling out what we know is important and then say, okay, ChatGPT based off of this information, what’s going on with this patient? How severe is this patient’s disease? What does the severity mean? How can we intervene?”
Dr. Rushing emphasized that instead of letting AI search for patterns independently, her team provides ChatGPT with targeted summaries from clinical notes. “Instead of saying having the AI look for these different patterns on its own, we are giving ChatGPT a summary of the information that we want ChatGPT to look at so that ChatGPT can do a better job at telling us what we want to know,” she said.
She outlined three main parameters used in their model: disease severity, summarization of key lab trends, and five-year risk forecasts. Dr. Rushing stated: “With the clinically guided AI, what we did is we have three parameters that we want to pull from the clinical notes. The first one is disease severity. The second one would be a summarization of key lab trends. Then the third one will be these five year risk forecasts.”
Explaining why these were chosen, she added: “We decided on these three parameters because when it comes to the disease severity, clinicians are wanting to understand at this point in time, how severe is my patient or how good or severe is my patient’s disease and where is their health, so that they’re able to then determine what interventions or what treatments they need to continue or discontinue or to seek out next with the lab trends, because you’re looking at months or years worth of clinical notes, but providers can miss little changes that are there that our model can pick up.”
She gave an example regarding kidney function monitoring: “So providing these summarizations, clinicians will be able to see, oh, you know, six months ago they say their creatinine was this level, but even though it’s still the normal range is starting to tick upward. So maybe we need to look into that and see if there’s something going on with their kidneys.”
Regarding long-term forecasting for patients’ outcomes she said: “And then the last one with the five year forecast, the goal with that was to give a give provider or clinicians an idea of what could happen, what negative outcomes could happen in the next five years so that they can intervene earlier and hopefully mitigate or prevent these outcomes from taking place, because they know this ahead of time.”
Dr. Rushing concluded by highlighting their aim for preventative care: “And so all this was the goal is to support this kind of preventative approach and things, let’s not wait till it gets severe, let’s start to identify this early on.”
