Radiation is one of the most common treatments for cancer. How patients are dosed depends largely on cancer type, location, and stage, but this decision is mostly based on population studies, even though patients with similar types of cancer often respond differently to the same amount of radiation. Among the multiple efforts to face that challenge, a research team has developed a model—based on the gene expression of a tumor—that predicts the biological effect of a given radiation dose, with the aim of optimizing treatment doses in each patient.
In a study published August 4 in The Lancet Oncology, the team tested whether their model, known as genomic-adjusted radiation dose (GARD), predicted clinical outcome following radiation treatment in cohorts already reported in literature. They found that GARD scores, which aim to forecast tumor responsiveness to treatment, were associated with time to first recurrence and overall survival in patients treated with radiotherapy, while the magnitude of the prescribed dose alone was not associated with these clinical outcomes.
“Radiation oncology is lagging behind medical oncology in having genomic tests that help us understand both who should receive radiation and how much radiation they should get,” says Wendy Woodward, a radiation oncologist at the University of Texas MD Anderson Cancer Center who did not participate in this study but previously collaborated with one of its authors on a review article. There are currently various efforts to translate tumor gene expression into biomarkers that can inform treatment—for instance, to recommend whether a given patient should be treated with radiotherapy at all, and if so, whether it should be alone or in combination with other treatments.
But “what is unique about GARD, if it turns out to be clinically validated, is that it’s the only one of these tests that’s really trying to give us information about what dose we should be giving people,” says Woodward.
GARD is based on an earlier model proposed by Javier Torres-Roca, a radiation oncologist at Moffitt Cancer Center in Tampa, Florida, and his colleagues. In 2009, his team published a molecular assay they developed, dubbed the radiosensitivity index (RSI), to determine how susceptible tumors are to radiation. The RSI was based on their finding that in multiple human cancer cell lines, the expression of ten genes could predict tumor response to a given radiation dose. In 2016, the researchers integrated the RSI and the radiation dose received by a patient into a model—GARD—that aims to predict the biological effect of a given dose.
See “Bursting Cancer’s Bubble”
Aiming to test GARD, in the current study the authors gathered 11 independent published clinical cohorts reporting patients’ tumor gene expression and radiation doses, as well as their outcomes after treatment. The resulting dataset comprised a total of 1,615 patients (1,218 treated with radiation and 397 without it) affected by any of seven different cancers, including breast cancers, pancreatic cancers, and melanomas.
“It is definitely a strength that they are looking across multiple tumor types,” as it would indicate that GARD isn’t specific to a particular type, says Ane Appelt, a clinical radiotherapy researcher at the University of Leeds.
Using the data available, the authors calculated the GARD for each patient treated with radiotherapy. Two patients receiving the same dose might have different GARD values, with the higher score indicating a tumor likely to be more sensitive to that treatment. The team asked how the therapeutic benefit of radiation was forecast by a higher GARD score as compared to that predicted by a higher received dose. In a hazard model where values above 1.0 indicate an association with worse outcome and those below 1.0 refer to an association with a better outcome in treated patients, they found that a higher GARD score was significantly associated with better clinical results, with a relative hazard of 0.98 per unit GARD for time to first recurrence and 0.97 for overall survival over the time period reported by each study. Using the same hazard model, the prescribed dose showed values of 0.99 and 1.0, respectively.
In addition to calculating GARD for those treated with radiation, authors calculated a sham GARD for those who did not receive radiotherapy; the calculation assumed a standard dose given to patients. The sham GARD serves as a control, explains Torres-Roca. If the expression of the genes assessed with the RSI has an effect on outcome that is independent of radiation treatment, then both the GARD and the sham GARD values would yield predictions with similar accuracy. But GARD was only accurate in predicting the outcomes of patients treated with radiation. Thus, “GARD is not just distinguishing good and bad tumors; it’s actually distinguishing the therapeutic impact of radiation,” says Torres-Roca. He and other authors on the paper have licensed RSI and GARD to Cvergenx, a company he cofounded.
Melvin Chua, a clinician scientist at the National Cancer Centre Singapore who was not involved in the study, says he finds the results underwhelming. The difference between how well GARD and the received dose alone predict clinical outcome is small, he says, and it could be driven by a few of the analyzed cohorts. For instance, two of the included clinical studies with the highest association between GARD and overall survival involved only 10 and 55 patients, respectively. He says that whether GARD predicts clinical outcome is still an open question. “Ultimately, like any molecular assay to be used in the clinic, GARD has to withstand the rigors of a prospective clinical trial,” he adds.
According to Appelt, one of the main limitations of this pooled analysis is that it “is all done on retrospective data” and on patient cohorts whose treatment was not randomized. Thus, it is possible that GARD’s apparently good performance in the study could be driven by other unknown factors and not by GARD itself. The authors “have done very robust statistical analyses to try to account for this, which makes the results promising,” she says, but they need to do a similar analysis with data from randomized studies and, ultimately, perform a prospective trial. If GARD’s usefulness is validated in those scenarios, she adds, “this could fundamentally change how we prescribe radiotherapy and could have a significant impact on outcome in cancer patients.”
Torres-Roca agrees that “a randomized trial would be the gold standard to demonstrate the predictive value of RSI.” His team has plans to soon start clinical trials to test the effectiveness of GARD in pinpointing the radiation doses that patients should receive. Those are not planned to be randomized, but he says he foresees randomized studies coming after that.
Woodward says that she “would absolutely support the development of a clinical trial to either demonstrate that [GARD] works or that it doesn’t.” Moreover, she notes, “this isn’t the only potentially predictive radiosensitivity assay out there, and the right thing to do is to design a trial where you can test them all.”
Clarification (August 30): The paragraph on relative hazards calculated in the study has been revised to clarify that the values stated for GARD are per unit GARD.