Read the Full Video Transcript

Ashish Kamat: Hello, and welcome to UroToday's Bladder Cancer Center of Excellence. I'm Ashish Kamat, Professor of Urologic Oncology and Cancer Research at MD Anderson Cancer Center in Houston. And I'm welcoming today, Dr. Boris Gershman, who's an Assistant Professor of Urology at Beth Israel Deaconess Medical Center in Boston.

Boris, you recently led this publication just last month, which was published, talking about radical cystectomy versus trimodal therapy for muscle-invasive urothelial carcinoma of the bladder. And as I'm sure you're aware, this led to a good amount of discussion in various journal clubs around the country, and especially our current social media journal club, which is Twitter. So we thought we'd invite you to present your work, and then we will have a little discussion after your presentation. So, the stage is yours.

Boris Gershman: I appreciate that. Thanks so much for the introduction, and for the opportunity to speak today on our study. Hopefully, it does generate some provocative discussion.

The paper was called, Radical Cystectomy versus Trimodality Therapy for Muscle-Invasive Urothelial Carcinoma of the Bladder. And here is the author list. The lead author was Dr. Kenneth Softness, one of our excellent residents at Beth Israel Deaconess Medical Center.

And so by way of background, radical cystectomy is the most commonly utilized curative therapy for localized muscle-invasive urothelial carcinoma of the bladder. But it is associated with substantial morbidity of treatments. In this regard, trimodality therapy, which combines chemotherapy, radiation, offers a bladder sparing alternative to radical cystectomy, and notably, has lower morbidity. And although numerous single-arm Phase II and Phase III clinical trials of trimodality therapy have demonstrated outcomes comparable to radical cystectomy in properly selected patients, this still represents a seminal knowledge gap within urologic oncology.

And specifically, the comparative effectiveness of both treatments remains unanswered. There are no randomized clinical trials that have been completed to date, nor is one ongoing. There was a Phase III trial attempted in the UK called the SPARE trial. In this trial, the authors randomized patients with clinical T2 to T3 N0 non-metastatic muscle-invasive bladder cancer, to either radical cystectomy, or trimodality therapy. But unfortunately, the trial closed after completion of the feasibility components because of low recruitment rates and high crossover rates. The authors concluded that randomizing patients to radical cystectomy or trimodality therapy was infeasible, especially due to strong patient preferences, as well as surgeon preferences.

In this regard, emulating a target clinical trial using observational data is an attractive methodological approach when randomized evidence is limited, such as in this clinical setting. The framework for emulating a real or hypothetical target trial has been described by Hernan and Robins. And this framework requires the explicit specification of a target trial to be emulated, which is considered a pragmatic target trial. And the subsequent observational analyses are viewed as an emulation of this pragmatic target trial.

So in this study, we emulated the SPARE trial, because this is really the best example of what intention to be, an evaluation of the compared effectiveness of surgery and chemoradiation for muscle-invasive bladder cancer. You will see the protocol components for the emulation framework on the left here. In designing the observational analyses, we specify the eligibility criteria of the target trial to be emulated. We specify the treatment strategies, the assignment procedures, follow-up period, outcomes, causal contrast, and analysis plan. And one of the advantages of this emulation framework is that, before you start a study, you lend some thinking to how the study should be defined, similar to the work that goes into planning a randomized trial.

And so you have the protocol components for the SPARE trial on the left, and you have our emulated trial on the right. And so you will notice some notable differences and some similarities. The SPARE trial enrolled adults 18 or older, with T2 to T3 N0, M0 urothelial carcinoma of the bladder, without hydronephrosis, who had received three cycles of gem-Cis or similar chemotherapy, and you were fit for radical cystectomy or radiation therapy.

In our hypothetical target trial, because of some of the limitations of the National Cancer Database, we restricted the cohort to ages 40 to 79, with a similar pathologic stage criteria, Charlson, zero to one, to minimize intractable confounding. And we required all patients to receive chemotherapy, as in the specification of the SPARE trial, to use as a surrogate for fitness, for either surgery or radiation therapy.

The treatment arms were defined to mimic those in the SPARE trial. Namely, one treatment arm was multi-agent neoadjuvant chemotherapy, followed by radical cystectomy with lymphadenectomy, defined in the NCDB as five or more lymph nodes removed, versus multi-agent chemotherapy with radiotherapy to the bladder by 3-D conformal techniques, using the dosing scheme, as is outlined below.

Similar to the SPARE trial, the follow-up period started at diagnosis, and ended on the date of last available data, or occurrence of events. And the primary outcome was overall survival. Similar to any randomized clinical trial, we provided both absolute and relative measures of treatment effects, using the Kaplan–Meier method, as well as Cox regression with universe probability treatment re-weighting, to account for the non-randomized nature and confounding within the observational dataset.

So here's what we did. We constructed a cohort in the National Cancer Database meeting the eligibility criteria that we just outlined on the previous slide. We estimated a propensity score for receipt of radical cystectomy, using a flexible logistic regression model. We assessed balance using standardized differences, and only after balance was achieved in the baseline characteristics, we then estimated the associations of each treatment arm with overall survival using the Kaplan–Meier method and Cox regression with stabilized inverse probability of treatment re-weighting. We also conducted several analyses to evaluate for potential heterogeneity of treatment effects, and several pre-specified sensitivity analyses that I will discuss at the end.

In our cohort, we included approximately 2,000 patients, of whom, 1,812 were treated with multi-agent neoadjuvant chemotherapy, followed by cystectomy with lymphadenectomy, and 236 were treated with trimodality therapy, chemoradiation. The median follow-up was 29.0 months, and during that period of follow-up, a total of 838 patients died of any cause.

Here, you can see some selected baseline characteristics. There were differences between the treatment groups with regards to several characteristics, notably, older age and higher Charlson score in patients undergoing chemoradiation, as could be expected from clinical experience. But after re-weighting by stabilized IPWs, baseline characteristics were well balanced.

Here, you can see the primary outcome of overall survival, stratified by treatment group, and the Kaplan–Meier plot, re-weighted by stabilized inverse probability treatment weights. As you can see, the blue curve represents radical cystectomy, and the yellow curve represents trimodality therapy.  There was no statistically significant difference in overall survival across the follow-up periods, with a hazard ratio of 0.87 in the IPW re-weighted Cox regression models.

When we examine the heterogeneity of treatment effects across pre-specified, based on characteristics, you see the main treatment effect here. And when you look at clinical T stage, age, and Charlson index, as specified on the left here, the only difference was among patients with clinical T3 tumors. It appeared that there was an improved overall survival with radical cystectomy, compared to trimodality therapy, with a hazard ratio of 0.42.

We conducted three pre-specified sensitivity analyses. In the first sensitivity analysis, we dropped the requirement for neoadjuvant chemotherapy prior to radical cystectomy. As you can see here, the curves still have a fair amount of overlap, and there is no statistically significant difference in overall survival, with a hazard ratio of 1.01, reinforcing the primary analysis results.

In the second sensitivity analysis, we allowed patients to receive either single or multi-agent chemotherapy, as part of the treatment specification for chemoradiation, or radical cystectomy treatment arms. Here again, you see that there is no statistically significant difference in overall survival after IPW re-weighting, with a hazard ratio of 0.95.

And in the third and final sensitivity analysis, we further restricted the cohort to age 40 to 70, and Charlson zero, to further try to minimize intractable confounding, not measured by the baseline characteristics captured in the National Cancer Database.  Here, you can again see, there was no statistically significant difference in overall survival between the two treatment arms, with a hazard ratio of 1.23, and a wide confidence interval spanning one.

So in conclusion, in observational analyses designed to emulate the SPARE trial, there was no statistically significant difference in overall survival between radical cystectomy and trimodality therapy. We did observe improved survival with radical cystectomy for patients with clinical T3 disease in the analyses that examined the heterogeneity of treatment effects. And we can take these data and further say that, in the absence of randomized data, careful observation analyses, such as these, but in complementary data sets, will be required to inform this critical evidence gap in urologic oncology.

I'd like to acknowledge several individuals in particular, on the author list; Dr. Kenneth Softness is one of our residents who led the authors for this study and did a tremendous job. As well as our biostatistics team, Sumedh Kaul and Aaron Fleishman, without whom this work would not be possible.

So thank you again for your time.

Ashish Kamat: Thank you so much, Dr. Gershman. Obviously, as you alluded to in the background, it's very hard to perform these sorts of randomized studies. Back in the 80s, there were two studies that, maybe even late 70s and 80s, there were two studies that actually got completed. They were small and really did not have any meaningful results. And since then, it's been what now, almost 60 years. There hasn't been a single randomized study that's been completed in this arena. With that in mind, data such as yours, while clearly, you can't really say that this is a randomized study, it is an approximation of a randomized trial. I'm sure, however, you've had a lot of criticism as to the methodology used. So for our audience, could you sort of go through a little bit more detail than you did in your presentation? Some of the pitfalls of this sort of a trial design, or a pseudo trial design, and how you accounted for that?

Boris Gershman: Sure. That's an excellent question. It's a question that sort of comes up for discussion and debate, I think, not infrequently. Let me just start by saying that the emulation framework that's been described in the epidemiological literature is not a surrogate for conducting a randomized trial, nor do we intend for it to sort of be presented as such. It's meant to be a way to conduct analyses using observational data that are more rigorous with more accurate causal inference, than the way we traditionally conduct observational analyses, where we don't specify the target trial to be emulated for causal inference questions.

And so I just want to be very clear, that it's not meant to replace a randomized trial, and if one was feasible, that would be much preferred. But in settings where trials can take many, many years to conduct, for instance, in kidney cancer, it took 20 years to conduct the lymph node dissection trial in the EORTC, in settings such as this, where randomization is likely going to be infeasible.

The only remaining source of evidence we have at the present time, are carefully conducted observational analyses. And so the emulation framework derives a number of benefits from the specification, the explicit specification, of the target trial to be emulated. And more than that, it actually encourages the investigators to acknowledge and identify, before conducting the analyses, limitations in the data set, that can then be either acknowledged as such in qualifying the results, or that can potentially be remedied in sensitivity analyses or other modifications to study design. And so in this regard, obviously, there is tremendous selection bias between surgery and radiation in many settings. And observational analyses can be hindered by residual confounding, and in particular, unmeasured confounding, because no observational data set is going to capture everything.

But the ultimate point that I made in the conclusion there, that complementary analyses and other data sets that have different characteristics that they collect in different covariants, can complement studies such as this, to really add to our evidence base and inform this comparativeness question.

Sometimes it's also asked whether the propensity score approach is sort of the key part, in whether the multi-variable adjustment is any different. And the truth is, it's not. It's not meant to fix the potential for confounding and selection bias. But as you saw in the mini protocol there, by really specifying explicitly before you do the analyses of the components of the trial, things we often take for granted, like how we specify treatment, how we define a lymph node dissection, how we try to minimize intractable confounding in the eligibility criteria, can really be made better by using this emulation framework in the design of these observational analyses.

Ashish Kamat: Right. Thanks. And I'm glad you emphasized that point for our listeners. That the emulation framework does not remove the biases of a database, or what happens at the clinical end. Because obviously, as clinicians, we are taking care of patients, and there is a lot of judgment involved, and that's all reflected in what actually happens to the patient and gets translated into the dataset. You can't control that, but you can make it the best analysis of the dataset that you can. And I don't mean to put words in your mouth, but that's essential, right. Now, did anything about your findings surprise you?

Boris Gershman: Great question. And you can put those words in my mouth. I think that is a very good summary of the benefits of this approach. And also, it allows you to acknowledge the limitations of your data set candidly. Which I think we probably don't do enough when we present such observational studies.

We came in here with a hypothesis, and so we didn't really have a favorite to win, so to speak. There really was equipoise in my mind. I trained at an institution that had a lot of patients undergoing chemoradiation protocols. And so there was equipoise from a clinical standpoint, as well as the research standpoint. So I think the overall findings were not surprising. They reinforced what the single-arm trials of trimodality therapy and the surgical series would suggest, in terms of carefully selecting patients having similar outcomes.

The one finding of significance, wherein, clinical T3 patients seem to have improved survival with radical cystectomy, has to be taken with a grain of salt, and some caution in the inference there. And we make a comment about this in the paper itself. And specifically, it's not clear whether this may be because some of these patients had hydronephrosis, which is traditionally a contraindication to trimodality therapy protocols. But these should be considered exploratory results, and require additional studies to further evaluate this.

Ashish Kamat: Yeah. And again, obviously, the center that you trained at, as you mentioned, has developed and championed trimodal therapy for bladder cancer for many, many years. It's a little unfortunate that it hasn't caught on in the rest of the country, as much as it could have. Because clearly, our patients need the option for bladder preservation, when appropriate.

Now you mentioned hydronephrosis. There are other size criteria, presence of CIS, et cetera, et cetera, that tend to be more North American selection criteria. If you talk to folks across the pond, over in the UK, they don't do any of that. In fact, there are, trials right now underway, where you just look in the bladder, see a tumor, do a biopsy, confirm that it's urothelial, and take the patient straight to radiation therapy. Right? So as these experiential databases evolve, we will have a lot more data. And of course, there are prospective studies ongoing. But let me flip it around to you and at your center, do you follow the classic, I guess, Shipley selection criteria for patients when you refer them for radiotherapy or has your clinical practice informed your judgment and selection criteria for these patients, and have you evolved to have a different selection criterion?

Boris Gershman: That's a great question. The short answer is, I still use the same sort of Shipley criteria where multifocal carcinoma in situ, hydronephrosis, for instance, I think would be associated with a poorer outcome. And it depends on the patient and the case. So I think, it's important to counsel the patient, that if you have one or more of those risk factors, you may not have optimal outcomes with chemoradiation. It's not that therapeutic modality shouldn't be pursued in appropriately selected patients, for instance, patients who are not ideal surgical candidates. But I think the patients should be properly counseled that they may not have outcomes as optimal as a patient with, let's say, a solitary T2 tumor or T3 tumor without hydronephrosis or CIS.

And as you said, as we gain more and more experience globally, and can leverage that data, we will hopefully have better evidence for how to further tailor the selection of patients for trimodality therapy. And in the end, that's really what we are trying to do. It's not about surgery versus radiation. It's about finding the best therapeutic options for the patients that we see, tailored to their needs.

Ashish Kamat: Yeah, absolutely. And it's unfortunate that in the US today, a large proportion of patients don't get any treatment. Right? Let alone, surgery versus radiation. They just do not have access. They can't get any treatment at all. And I'm sure you've seen that in the various database studies that you've done.

In closing, Boris, do you have any thoughts for our audience you want to leave them with?

Boris Gershman: You know what? I think ultimately, I hope this is a generating conversation about trimodality therapy as an alternative to radical cystectomy for muscle-invasive bladder cancer. And in the research community, I hope it generates conversations about how we can better conduct observational studies to answer important causal inference questions, such as this and others, within our specialty.

Ashish Kamat: Great. Thank you again for taking the time, and spending it with us, and stay safe.

Boris Gershman: Thanks so much for having me.