Machine Learning Predicts Financial Toxicity in Stage IIA Non-Seminoma Treatments - Peter Sullivan

November 1, 2024

Peter Sullivan discusses a study on financial toxicity in testicular cancer treatment. The study employs large language machine learning models and Markov modeling to predict financial toxicity patterns in stage IIA non-seminoma patients, comparing retroperitoneal lymph node dissection (RPLND) versus chemotherapy treatment approaches. Using the COST-FACIT questionnaire and a 10-year horizon analysis, the research demonstrates that while RPLND has higher initial costs, it results in lower long-term financial toxicity compared to chemotherapy, which shows increasing financial burden over time due to ongoing treatment costs and complications. Dr. Sullivan explores the importance of discussing financial toxicity with patients and the need for better support resources. The discussion concludes with insights into expanding this research methodology to other cancers and the importance of prospective trials to validate these findings across medical specialties.

Biographies:

Peter Sullivan, DO, Urology Resident, UT Houston, Houston, TX

Zachary Klaassen, MD, MSc, Urologic Oncologist, Assistant Professor Surgery/Urology at the Medical College of Georgia at Augusta University, Wellstar MCG, Georgia Cancer Center, Augusta, GA


Read the Full Video Transcript

Zachary Klaassen: Hi, my name is Zach Klaassen. I'm a urologic oncologist at the Georgia Cancer Center in Augusta, Georgia. I'm delighted to be joined on UroToday by Dr. Peter Sullivan, who is a urology resident at UT Houston. Today we're going to be discussing some awesome data presented at the AUA South Central Sectional Meeting discussing financial toxicity in testicular cancer treatment. Peter, thanks very much for joining us on UroToday.

Peter Sullivan: Hi, Dr. Klaassen. Thank you again for having us today on UroToday. I look forward to discussing this topic with you. As Dr. Klaassen mentioned, my name's Peter Sullivan. I'm a urology resident here at UT Houston, and I had the fortune of teaming up with some of the people at the University of Kansas to discuss the financial toxicity in those with testicular cancer, especially those with stage IIA non-seminoma.

So a little outline of what we'll be discussing. And so financial toxicity is really an underreported outcome in patients with testicular cancer and really an unreported outcome in a lot of different disease states. It's been shown in the data that multicycle chemotherapy regimens have been associated with worsening financial toxicity. And Dan Joyce published an article recently that looked at the different treatments for stage IIA seminoma based on the SEMS trial, and it was actually found to favor RPLND in regard to cost-effectiveness.

And now with the advent of different AI software and machine learning platforms, they can be used to help predict the cost that's associated with these various treatment strategies. And then on top of that, we can utilize our patients and have them answer some validated questionnaires that really help us understand their degree of financial toxicity that they experience. And more specifically, we use the COST-FACIT questionnaire, which is a validated questionnaire that looks at the degree of financial toxicity in patients with cancer.

And so the purpose of this study was to develop a large language machine learning model that is based on assumptions from the literature, and then apply that model to a COST-FACIT questionnaire to then help us predict the degree of financial toxicity in patients that have stage IIA non-seminoma. And on the right, what you'll see is this is actually a picture of the COST-FACIT questionnaire. It consists of 12 questions on a Likert scale and then a grading rubric, and it's tallied up, and then the higher the score is associated with the less degree of financial toxicity.

And so what we did is we created a Markov model for those with IIA non-seminoma based on the various different treatments and the outcomes that are described in the different literature. And so then we applied this to a large language machine learning model and used it to apply previous assumptions, looking at the different outcomes and the costs associated with those outcomes and the different treatments. And then these costs were then applied to the COST-FACIT questionnaire. And this was based on the proportionality of those to identify potential financial stress in different toxicities based on the various treatment strategies and the outcomes. And so on the right you'll see some of the assumptions for our model. And so on the first assumption we looked at the financial toxicity and we used a scale from one to four, and one being no financial toxicity at all up to four, which is the most severe.

And then you'll look down at the cost. And so the cost of RPLND in this study based on some of the literature was a little over $58,000 for the procedure. And then as far as the chemotherapy cohort, the different studies suggest that three cycles of BEP is a little over $104,000. And so then the model we used for the long-term toxicities was that some of the ones described that were most prominent were the infertility and the cardiotoxicity. And this was much more prominent in the chemotherapy arm rather than the RPLND cohort.

And then the probability of recurrence. We utilized 22% for the RPLND and 5% for the chemotherapy arm based on previous data. And one of the utility values that we used was that the financial burden will increase with more severe toxicities and that this influences the transition to a higher financial toxicity state. And so we mapped this out on a 10-year horizon, and what we did is we used a baseline of a number one where everyone had no financial toxicity.

And so what you'll see is our results, we looked at it in three different stages, and this was based on years on that 10-year horizon. And so the initial stage was that years one through three, and what it showed was that the majority of patients in the RPLND arm remained in the no toxicity or the mild toxicity states. And then on the curve for the chemotherapy in those one to three years, there was a more noticeable progression of those toward the mild and the moderate toxicity. And that was often driven by the more amounts of treatment cycles and the different acute side effects associated with those treatment cycles.

And then as far as the next stage of it, we looked at the years four through six, and those were the mid-stages. And so for the RPLND cohort, most of those patients, they had progressed to a mild or at least a moderate toxicity state. But the proportion of the severity remained very low, where if you look at the chemotherapy curve, there was a significant proportion of those patients that had now reached the moderate toxicity with an uptick of those going into the severe toxicity.

And this was due to the continued expenses and the treatment for some of those longer-term toxicities and complications that were associated. And then the last part was the years seven through 10, which is the long-term component. So for the RPLND group, the probability of remaining in the moderate toxicity was much higher and only a smaller group of them moved to the severe toxicity. Now then if you look at the chemotherapy, there was a much higher number of those patients that ended up in the severe toxicity state by year 10. And this is likely due to the financial burdens from the long-term side effects and the cost of managing the different complications such as infertility or different cardiovascular diseases from the drug toxicities.

And so in conclusion, we know that we can use large language modeling software and use different Markov models associated with various outcomes of treatments for the disease of stage IIA non-seminoma to help us predict the financial toxicity that patients experience. And so from this model, it showed that the RPLND was generally a little bit more favorable in terms of long-term financial toxicity, although the initial cost with the surgery is high, the subsequent financial burden remained low over time, even in those patients that progressed and had chemotherapy. Then on the other hand, the results for chemotherapy resulted in a higher cumulative financial toxicity due to the ongoing treatment costs that everyone had associated with them and the risk of the side effects and the need for some more intensive management of those toxicities.

Zachary Klaassen: Wonderful. This is fantastic work, Peter. Congratulations. Really taking some really complex methodology and the Markov model applied to that and really having some clear conclusions on financial toxicity for these patients. So I guess my first question is how do we take this information that we're discussing today and counsel our patients even tomorrow in the clinic?

Peter Sullivan: Yeah, you're correct. It is a tough thing to look at and especially in these patients that are very vulnerable, such as the testicular cancer patients. And so I think it behooves us to actually start today and start discussing it with our patients. And now that our understanding of this is really small in regards to the understanding that we have of the disease. We don't know a whole lot of concrete things of how patients truly experience the different types of financial toxicity, but we know that it exists and we know that it varies based on the different treatments that we discuss with them. So at least being upfront with the patients and letting them know that that exists and that could be a burden to them and that we're here for them and try to counsel them as best as we can with the information that we have and trying to create different modeling strategies to help predict things to come.

Zachary Klaassen: Yeah, absolutely. And when we look at our cancer centers, you're at UT Houston, I'm in Georgia. We all have great resources for social work, we have great resources in terms of psycho-oncology. In your opinion, what specific resources do we need to help these patients, particularly this vulnerable population like you mentioned, when it comes to financial toxicity?

Peter Sullivan: Yeah, it's a great question because I think that that answer has such a broad spectrum. And so I think what we can do is we can first start with the small things and assess where the patient is at as far as not only just from a financial standpoint, but what their social situation is too, whether they have family, what kind of job they have and things of that sort. And start small. And so you can assess for... Try to find different ride assistance or different meal vouchers or child care for various appointments or treatments to help—the things that we can do right now. And then as far as we're able to develop stronger models, then we can help further educate our patients in a more concrete fashion of the different financial toxicity that could be associated with the different various treatments.

Zachary Klaassen: Yeah, absolutely. As I look at this model and the work that you guys have done, how are you thinking about expanding this? Could this be expanded into say, advanced prostate cancer? Is there more work to be done in testis? I mean, we know there's financial toxicity with advanced disease for sure, and some of that comes into the chemotherapy data that you mentioned. What's the next steps for this work?

Peter Sullivan: Yeah, exactly like you mentioned it, there's actually been a little bit done in prostate cancer already. Dr. Joyce and Dr. Boorjian at Mayo Clinic came out with a study looking at some financial toxicities in prostate cancer patients. And it is really a good idea to help this subset of patients that have such a long cancer survivorship, like the patients with testis cancer, which is very fortunate. But that can impact them for many more years than say, some of the other cancers that are usually a later onset of age.

And so with this, we need to really prove what we're doing. So there needs to be a lot more prospective studies that help really validate these assumptions that we've made with the different machine learning models and these questionnaires, such as the COST-FACIT. And then as we are able to validate those, then we can help develop more modeling tools to help in the present day of our patient's current state at their expected disease decision point.

Zachary Klaassen: Yeah, absolutely. It's been a great discussion. Maybe a couple of take-home messages for UroToday listeners.

Peter Sullivan: Yeah, thanks again for having me on. I've really enjoyed talking with you.

Zachary Klaassen: Of course.

Peter Sullivan: It's been good to see you again. So I think the biggest take-home is to have increasing awareness that this exists. It's here and you can't run from it. And to try to meet patients where they're at to the best of your busy clinic's ability. And to understand that the degree can change based on the different treatment that we recommend or that patients decide. I think this is a good area that some of these AI and machine learning platforms can really help us predict some of this stuff in the current age while we're waiting on some of those more validated models to confirm what we are doing, that we've had done with prospective trials. And so speaking of prospective trials, I think we need more prospective trials regarding financial toxicity because it's not just in testicular cancer patients, it's not just in cancer patients, it's in all facets of medicine. And so I think that can be a big movement in medicine going forward.

Zachary Klaassen: That's well said. Congratulations again on this great work, Peter, and thanks for your time and expertise, sharing with our UroToday listeners.

Peter Sullivan: Well, thank you all very much.