Precision medicine: Oncologist focusing on using AI to inform treatment decision-making

As the number of targeted cancer therapies proliferates, so, too, do the decisions that doctors must make. An artificial intelligence (AI) initiative underway at  MUSC Hollings Cancer Center seeks to provide oncologists with insights into customized treatment options for individual patients. Thai Ho, M.D., Ph.D., director of Precision Medicine at Hollings, is leading the effort. He likened the process to using a GPS navigation app to choose a route. “I think one of the good uses of AI is it helps us integrate large volumes of information,” he said. “But ultimately, it's the person running the project that makes the final decision. So if you plot a route from your house to Publix, for instance, and then you look at the various suggested choices, you get to overlay your own knowledge; like you know that Lockwood Drive gets flooded, so you'll pick a route that doesn't require driving on Lockwood.” Similarly, this AI initiative presents doctors with choices based on the results of genomic testing of a patient’s tumor. Tumors, like people, have unique genomic profiles, or a mix of genes. Different mutations can result in different proteins or receptors appearing on the tumor, and researchers are increasingly developing drugs that home in on one of these specific targets.

Beyond those targets, however, there can be other molecular differences among tumors that make a drug more or less likely to work. Cancer makes it even more complicated when multiple gene mutations occur simultaneously, making it challenging to determine which one is the key driver of the tumor growth. Ho said the initiative is currently focusing on identifying tumors with defective DNA repair mechanisms. “We're using AI-based mechanisms to try to identify these tumors, and it helps give us a sense of, ‘Well, do we think this tumor might be more sensitive to drugs that stress the DNA repair mechanisms?’” They’ll then feed their own outcomes from patients treated at Hollings back into the AI program to build on its knowledge base and recommendations further. Looking at those patients whose tumors responded to the drug either far better or far worse than expected can help to refine the treatment suggestions. “We're looking only at the outliers to see, between patients, who really responded well and patients who had no response. What are the key differences? And then we would change the iteration on the AI to further improve its accuracy,” Ho said. Ho emphasized that although the AI algorithm is a work in progress, the drugs being recommended for patients are not. Only drugs that are already an option within the standard of care are offered as recommendations. “The idea is that we're only using it to help make better decisions where we have options within the standard of care, so we're not bypassing drugs that are known to work,” he said. Instead, doctors can incorporate the AI information into their overall decision-making, which must also account for the patient’s other medical conditions and personal preferences when it comes to potential side effects or the amount of time a treatment requires. “We discuss it with the patient, but sometimes the drug isn't the right drug for the patient,” he said. “I think it's sometimes like the difference between a Honda or Toyota, right? You'll still get from A to B, but some people may have certain preferences, and we try to make it work with their lifestyles and their own preferences. The ones who make the decision at the end of the day are the patient and the physician, not the AI.” Ho’s team gets the genomic information by testing samples of the tumor removed during a biopsy or surgery. The test is in addition to the usual tumor testing, and it’s done on an opt-in basis for those doctors who are interested in the additional information. Performing the analysis in-house at Hollings, Ho said, can give patients an extra layer of confidence in the privacy of the information. Importantly, it also means that the testing can be done on a low-cost basis. “Probably less than 5% of our patients pay more than $100 out of pocket, which I think is important because patients need to know how much they are in it for,” he said. “Our patients already face financial toxicity from travel costs, medical expenses and time away from their loved ones.” So far, the new initiative has tested tumors from about 30 patients, but they are already seeing results. “We had a patient who was already getting chemotherapy, and we identified that his tumor might be sensitive to one of these PARP-inhibitor drugs used in ovarian cancer and prostate cancer,” Ho said. “We switched him to a pill, and as a result, he's had a much better quality of life. Yet it's also kept his tumor in check as well.” Looking forward, Ho envisions using this AI to match more patients to clinical trials. “These AI algorithms could potentially identify patients in the MUSC system but beyond the Charleston area. And that's something where we could reach out to the physician and say, ‘Hey, we don't know where you are in the treatment paradigm, but we do have these clinical trials at Hollings that your patient would qualify for,’” Ho said. “This is often genomically driven, so we're looking for particularly rare mutations or mutations that physicians don't necessarily think are druggable – but we have a trial for it.” Eventually, the team wants to make its tumor board – an assembly of medical oncologists, radiation oncologists, pathologists, nurse navigators and others who review a case together to come to a consensus opinion – available to any doctor in South Carolina whose patient has interacted with MUSC at some point. “In the future, we hope to encourage outside physicians to submit their patient cases to our tumor boards if their patients are already within the MUSC system to better serve patients outside the Charleston area,” he said.