• Center on Health Equity & Access
  • Clinical
  • Health Care Cost
  • Health Care Delivery
  • Insurance
  • Policy
  • Technology
  • Value-Based Care

Leveraging AI to Optimize Precision Immunotherapy at Vanderbilt

Commentary
Article

The theme of our recent Nashville Institute for Value-Based Medicine® event was ”Bringing the Future to the Present in Cancer Care.”

At the October 3 Nashville Institute for Value-Based Medicine® event held in conjunction with Vanderbilt University Medical Center (VUMC), presentations focused on incorporating advancements in technology and artificial intelligence (AI) to take cancer care to the next level.

Principal investigator for the Digital Precision Oncology Project, Travis Osterman, DO, MS, FAMIA, FASCO, presented “Leveraging AI to Deliver Precision Immunotherapy,” in which he delivered data on this 5-year-strong initiative that is seeking to determine both the patients who are most likely to have toxicities from immunotherapy and those most likely to benefit from immunotherapy. The project is a collaboration between VUMC and GE Healthcare, “with the goal of ultimately helping oncologists and their patients make better decisions and more informed decisions about their treatment options,” Osterman told The American Journal of Managed Care®.

This transcript has been edited for clarity.

Transcript

What are key takeaways from your presentation, “Leveraging AI to Deliver Precision Immunotherapy”?

This is not a question that's unique to Vanderbilt. I would argue this is one of the top questions in our field, and so there have been many approaches to this. I highlighted about a dozen different approaches, from radiomics to genomics to transcriptomics [to the] gut microbiome. Just about every approach to predicting this question of toxicities and efficacy to immunotherapy has been attempted.

I outlined how our project was a little bit different. Number 1, ours was a pan-tumor sample. We did not limit our study to just lung cancer or just melanoma, which is pretty common. We took any patient that has been treated with immunotherapy and included those in our cohort. That allowed us to include 2200 patients, which doesn't sound like a lot, but I think is one of the larger sample sizes you'll see for this type of project, at least in the US. So, that was a differentiator.

Probably the biggest differentiator is the idea of focusing up front on what it would take to implement this. So, if you required a complex transcriptomic array or a complex analysis of gut microbiome, we don't currently use those in clinical practice today. And so implementing those solutions, even if they showed promise, will be a real challenge. Instead, we, from the beginning, took the route of, “Let's try to leverage data that we're already collecting on our patients, so that if we're able to show a signal, we'll be able to disseminate this in a very widespread way.”

The next point I would say is we focused on something called clinical utility, and I think most physicians will relate to this. I don't need a computer algorithm to tell me a patient's at high risk to have a side effect of a disease. What I need the algorithm to tell me is, how is this different than the average patient? How does this balance against the risk of them, or the benefit that they might get from the treatment? We call that clinical utility. On the inpatient side, we use these things called sepsis scores. The algorithm just gives us a number, and if that number changes, I actually don't know what to do with it.

Our goal, again, from the beginning was providing useful information back to our clinicians and their patients. I think that's probably the last point that differentiates our project.

From an outcome standpoint, we were able to show, and this was published in the Journal of Clinical Oncology: Clinical Cancer Informatics, that with the AUCs [areas under the curve] and the sensitivities and the positive predictive values that we saw with our algorithms, that you could use this to inform, say, clinical trial screenings so that you could take a hypothetical group of 200 patients and you could select the 100 patients that would be most likely to benefit from that clinical trial based on the algorithmic output. And if you did that, you could improve overall survival in the patients that went onto the clinical trial by about 20%. That's important because clinical trials are expensive to run, and the bigger difference you can show in overall survival, the higher survival you can show, it means you can enroll fewer patients and you can run the trial under a shorter timeline. That allows drugs to get through the process and ultimately into the hands of patients more quickly. That's really key.

The second key diagram I showed was, similarly for toxicity, if you took a hypothetical group of 200 patients, applied our algorithm, and said you wanted to find the 100 patients least likely to have a side effect of toxicity—in this case, pneumonitis, which is inflammation of the lung, which is a common side effect of these treatments, about 1 in 8 patients will have this as a toxicity—we were able to reduce that from 9% to 4%. Again, on the pharma side of things, if you have an investigational agent that looks like it has tremendous promise, but it happens to have more toxicities than usual, you have a really tough decision to make to whether to bring that asset forward in future clinical trials, knowing that there are alternatives that maybe don't work as well, but that don't cause harm to patients.

What you would like to do is find the right patients for that drug. Our goal is that algorithms like the ones that we're developing can act as synthetic biomarkers to help companies and patients find the treatments that are most likely to benefit them and least likely to harm them.

How did you see the evening’s theme, “Bringing the Future to the Present in Cancer Care,” reflected in the presentations?

There's really incredible work going on at Vanderbilt across multiple domains. I think Dr Obstein [Keith Obstein, MD, MPH, FASGE, who presented “Through the Looking Glass: Predicting the Future of Colonoscopy”] from gastroenterology really wowed the audience with his work on what's coming in the future of screening colonoscopies. He was able to show not just traditional endoscopic colonoscopies, but new methods of inserting small robots into the colon to be able to do better visualization. I think Dr Idrees [Kamran Idrees, MD, MSCI, MMHC, FACS, who presented “Histotripsy and Its Role in Treating Liver Tumors’”], similarly with histotripsy, was able to highlight a brand-new modality of treatments for our patients with, right now, metastatic hepatocellular carcinoma.

We hope to see this technology used in other cancers that metastasize to the liver, like diseases that I treat, like lung cancer. Certainly, we also heard about the importance of community engagement as we look forward, which was a great way to really wrap the session. Across the board, I think all attendees walked away with, there's just incredible work that's going on at Vanderbilt University Medical Center.

Related Videos
Ruben Mesa, MD
Amit Garg, MD, Northwell Health
dr carol regueiro
Wanmei Ou, PhD, vice president of product, data analytics, and AI at Ontada
Surbhi Sidana, MD, MBBS
Mabel Mardones, MD.
Mei Wei, MD.
Screenshot of Susan Wescott, RPh, MBA
Glenn Balasky, executive director of the Rocky Mountain Cancer Center.
Screenshot of an interview with Stuart Staggs
Related Content
© 2024 MJH Life Sciences
AJMC®
All rights reserved.