Radiation oncologists Jacob Scott, MD, DPhil, and Javier Torres-Roca, MD, discuss the benefits of using GARD in a pooled pan-cancer analysis and why there is a call to action to integrate GARD-based radiotherapy dosing in oncology.
The biological effect of radiotherapy, as quantified by the genomic-adjusted radiation dose (GARD) model, was a predictor for time to first recurrence and overall survival (OS) in patients with cancer who received treatment with radiation, according to radiation oncologists Jacob Scott, MD, DPhil, and Javier Torres-Roca, MD.
Moreover, because GARD is predictive of radiotherapy benefit, as seen in results of a study published in Lancet Oncology, Scott and Torres-Roca recommend the integration of genomics into radiation dosing decisions.
“The genomic adjusted radiation dose is the first metric of the biological effects of radiotherapy, and what we have found is that giving the same dose of radiotherapy produces very heterogeneous effects on each individual patient,” Torres-Roca said.
In an interview with OncLive®, Scott, an associate professor at Case Western Reserve University School of Medicine and a radiation oncologist with Cleveland Clinic, and Torres-Roca, a senior member in the Department of Radiation Oncology at Moffitt Cancer Center and a professor of oncologic sciences at the University of South Florida Morsani College of Medicine, discussed the benefits of using GARD in a pooled pan-cancer analysis and why there is a call to action to integrate GARD-based radiotherapy dosing in oncology.
Scott: In 2005, Dr Torres-Roca and his team came up with a method to look at tumor genomics in the form of gene expression to try to predict which tumors were radiation-sensitive vs which are radiation-resistant. This is something that every radiation oncologist or oncologist knows from their clinical practice; some patients respond great, and some patients do not. What has been frustrating is that, over the years, that information has not been actionable. Therefore, [assumptions must be made] about how patients will respond to the treatment, and then [we must] give treatments based on what is best for the average patient. The problem is that the best treatment for the average patient is not very good for most patients. If we are treating the average patient, that means half the time we are giving more [treatment] than we should, and half the time we are giving less [treatment] than we should. If we are blinded to those differences, the only real option is to treat the average patient.
Dr Torres-Roca published a result that showed how to take gene expression and dichotomize patients into good or bad groups. Over the course of the years since I joined the team, we have worked to move that from a dichotomous prediction that shows good vs bad responders, to a more continuous prediction, which we have termed ‘GARD,’ or genomic-adjusted radiation dose. It takes the tumor gene expression profile and the radiation dose given and provides a prediction about how that radiation dose will affect the individual patient. It is a sliding scale prediction, which is a continuous variable. Currently, most patients get the same dose and receive that over a somewhat protracted course of radiation of 5 or 6 weeks. [The dose given is] usually 2 Gy per day. The problem is that some patients respond beautifully to that dose, and some patients respond very poorly.
There is a concept of biological effects that is written into our textbooks, but we have had to assume that is the same from patient to patient. Now, with the concept of GARD, it is possible to make a quantitative prediction about how each patient will respond. For example, if there are 2 patients who receive 2 Gy, but 1 responds twice as well as the other, there is now a quantitative measure of how the patient will relatively benefit from that radiation dose. The real motivation behind this study was to take all the available data in the literature and try to see if this predicted benefit of radiation therapy was associated with outcomes of interest clinically.
Torres-Roca: The idea is that when prescribing a dose each day, we are delivering a physical dose of radiation that is measured very accurately. When this dose interacts with the patient, it causes a biological effect that is very heterogeneous. Like everything else in life, everyone responds differently to interventions. What [we are trying to understand] here is what happens to the patient.
Torres-Roca: We identified patients in public data sets that had all the information required to calculate GARD. To calculate GARD, we needed to know the radiosensitivity index [RSI], which is a metric of radiosensitivity that was developed a few years ago based on gene expression, and the gene expression. Furthermore, [we needed to know] the dose of radiotherapy the patient received, because the equation used includes both the radiosensitivity and the radiotherapy dose. This then gives us an actual GARD for each patient.
[Among] all the data sets, [we identified] a total of 1615 unique patients across 7 different cancer types. This includes breast cancer, glioma, endometrial cancer, lung cancer, pancreatic cancer, melanoma, and head and neck cancer. Moreover, [we had data on] clinical outcomes for all patients. For some patients we had OS data, and for other patients there was only [information available on] whether they had recurred or not. Additionally, for some patients there were [data on] whether they had recurred distantly or locally. Following this, all recurrences were pooled, and we used [this as an] event of first recurrence.
We performed a full Cox analysis where we integrated all the data sets, and we determined whether GARD was associated with OS or recurrence for these patients. Critically, GARD was treated as a continuous variable. What was found in the pooled analysis, was the biological effect of radiotherapy, as quantified by GARD, was associated with both recurrence and with OS, but notably, the dose prescribed is not. In other words, the biological effect of radiation, as quantified by GARD, tells us more about what is going to happen to the patient than the actual dose delivered out of the machine. This is critical, [as we are] trying to make the argument that this is a better parameter than the physical dose. The way to move forward in radiation oncology is [obviously] still delivering the physical dose [of radiation], but [also understanding] the biological effects that are being induced [in the patient]. [This provides] much more information on what is happening to [each] patient.
There was also a negative control, which means there were patients treated without radiation, and we tested whether GARD predicted [outcomes] in those [patients]. To do this, we had to calculate what is referred to as a ‘sham-GARD,’ where we assumed the patients received radiation even though they had not, and calculated sham-GARD. Sham-GARD was not associated with outcome in patients treated without radiation.
Finally, we did an analysis of whether there was an interaction between GARD and whether the patient had received radiation therapy or not. This was critical because we wanted to demonstrate whether GARD was predicting the therapeutic benefit of radiotherapy, or if it was predicting that patients would do well, independent of the treatment that they received. In this figure, there is a high association between GARD and outcome in patients treated with radiation, but there is no association between GARD and outcome in patients without radiation. This demonstrates that it is predicting the therapeutic benefits of radiotherapy, which is what we need to be able to personalize an individualized radiation treatment.
Scott: There are different ways to define levels of evidence. For biomarkers and studies based on archival tissue, like this one, we are very close to level 1 evidence. This is a measure that we believe is ready to be used in our clinical practice and that will help us learn to do a better job in radiation therapy. Some of that is going to be hard, because we will have to relearn some of the things we do. Currently, practice is based on empirical trials with dose escalation that have been done over the years, showing what the average patient does best with.
What is being suggested here is that within known safe limits of dose, it is time to be thinking of a clinical decision support tool. Very soon, this will be a test available to any clinician to order, like an oncotype test, which gives you an idea of relative benefit. The key finding in the paper for clinicians to understand is in the final figure, which takes this abstract analysis of relative hazard per unit GARD, where we show that, strikingly, for each individual increase in unit GARD, the outcome improves. We try to reset this abstract concept back into the context of individual diseases for absolute benefits. Looking at this data, there is a nomogram which shows for each individual unit increase of GARD what the percent increase of absolute benefit would be for each disease site that we analyzed.
The critical thing to think about is that because each patient responds differently, each patient's GARD will change differently for dose. For example, if we wanted to have 5 extra units of GARD delivered to a patient and the nomogram predicts a 20% absolute benefit in survival, that sounds great. Twenty percent is a huge increase in survival, and of course every clinician would want to give that to their patient. The problem is that to getting 5 units of GARD increased to a patient can be very different from 1 patient to the next. One patient may require a single extra fraction of 2 Gy radiation, which is almost always deliverable in a safe way. In that patient, that would be an easy discussion to have and an easy decision to make as a clinician. However, for other patients who are more radio-resistant, getting 5 units extra units of GARD would be completely unsafe and not worth it, as the toxicities would strongly outweigh the relative possible benefit changes.
Trying to think about how to move forward with tradeoffs between toxicity and benefit is going to be a hugely important thing that we are all going to have to learn together. We have a strong place to start, and a solid footing based on the analysis completed. However, although 1600 patients are a big number, it is not as big as we are used to in some other diseases. Overall, these patients have the information we need. Going forward, it is necessary to add to our collective understanding as a field and continue to measure the parameters that we need to continue to learn. For the community physician, the answer is that we're ready right now. There is level 1 evidence to let us adjust our dosing within safe limits. To start going beyond that, we're going to need more information and controlled trials, but we are ready to move forward right now with clinical decision support tools.
Torres-Roca: Something that might be useful for radiation oncologists to understand is that what we are doing is adding a dimension to dose. We have only considered the physical dimension, and now, we have added a biological dimension. In many ways, that is essentially what happened when integrating computed tomography into the treatment planning system. Back when I started training, we had standard fields that we would use to radiate patients, and when we developed the systems to integrate anatomy into our treatment planning, we realized that we were treating a 3-dimensional system that was different for each patient. Not everybody had the same size prostate, or the same size bladder, or they did not have their rectum in the same place.
By adjusting our treatments based on that information and adding that new dimension, we had a new tool to optimize how we prescribed radiotherapy. This is essentially the same thing. We are adding a dimension that allows us to understand dose much better, and that is going to help us optimize dose better. Overall, I want to make this available for physicians to use. That is the next step, and of course, we are going to be doing more trials in parallel with that. However, we are at a point where we need to get the era of genomics started in radiation.