GOG Sends Rebuttal Letter to Oncolytics Biotech
The Clinical Cancer Letter received a rebuttal letter for publication from the GOG Foundation and its president, Philip DiSaia, regarding “Study: Reolysin-Paclitaxel Combination Demonstrates Higher Response Than Paclitaxel Therapy Alone” which appeared in the March 2016 issue.
The rebuttal was addressed to Brad Thompson, president and CEO of Oncolytics Biotech Inc., relating to issued by the company and an interpretation of the results of the study GOG-0186H, a phase II study evaluating Reolysin plus paclitaxel compared to paclitaxel alone in patients with persistent or recurrent, ovarian, fallopian tube or primary peritoneal cancer.
The GOG Foundation’s letter follows:
Dear Doctor Thompson,
We have recently become aware of an independent interpretation of the data from GOG-0186H that differed from the interpretation and results presented by David Cohn, M.D. at the recent Society of Gynecologic Oncology (SGO) Annual Meeting on Saturday, March 19, 2016 on behalf of NRG Oncology/The GOG Foundation, Inc. (NRG/GOG). We disagree with your analysis of the data, and this letter serves as a response to your analysis and a rebuttal to your findings.
The primary endpoint of GOG-0186H was the progression-free survival’s (PFS) hazard ratio (HR) of treatment (paclitaxel and Reolysin) to control (paclitaxel alone), which guided the trial’s sample size and required number of events to detect a clinically significant reduction in the hazard rate by 37.5% with approximately 80% power while restricting the level of significance to 10%. The study’s principle conclusion depends on the results of this endpoint. Additional secondary (subordinate) endpoints were included to gain a fuller spectrum of potential treatment activity. These endpoints were Kaplan-Meier (KM) estimates of PFS and overall survival (OS), hazard ratios of death for OS, tumor response by RECIST, and tumor response by modified CA-125 criteria (among others). For the purposes of characterizing response by treatment, complete (or full responses) were distinguished from partial responses, but when describing overall responses for the purposes of efficacy evaluation, it is common practice to group complete with partial responses together (e.g., see Table 9 of the Final Analysis). It is important to note that the probabilities of false positive results are not necessarily controlled for secondary or exploratory analyses. That is to say, the chances of seeing what is called “a significant result” (usually associated with a p-value < 0.05) can be much higher than what is advertised. Unless an analysis plan is meticulously specified a priori, the true value of the p-value can be very difficult to determine. The significance of a result may be judged when adjusting p-values for multiple testing (e.g., experiment-wise error rates). Without these adjustments, results from secondary analyses are not taken with the same level of confidence as the primary analysis.
The primary endpoint was analyzed in this study with a 3-outcome design. If the observed HR was greater than 0.857, the regimen was declared uninteresting. If the observed HR was less than 0.757, then the regimen was declared to be worthy of further investigation. Lastly, if the HR was between these values, then the regimen was declared to be non-definitively negative. In this instance, NRG/GOG could consider further study of the regimen if evidence in favor of its efficacy could be found (e.g. in other studies). The observed value of the HR was 1.1, so the primary conclusion of the study was that the regimen is uninteresting and not worthy of further investigation. (See the Figure on the top of page 3).
Regarding the secondary (subordinate) endpoints, the KM curves for PFS and OS indicated little or no benefit from the experimental regimen. The HR for OS also indicated no significant advantage, but the study was underpowered for this objective as expected for a phase II study. However, the initial analysis and the long-term follow-up studies were not promising with regard to OS. The analysis of response by RECIST showed no advantage to the combination regimen (9 responses in the paclitaxel regimen versus 8 in the combination arm). An alternative to PFS is event-free survival (EFS) which takes times of patients beginning subsequent therapies into account. This analysis also indicated little or no benefit from Reolysin. Finally, the proportion responding by CA-125 (grouping both partial and full responses) was roughly the same (18.5% control vs. 22.2% ITT; 34.5% control versus 30.8% among patients evaluable for CA-125 response). All of the secondary analyses were consistent with the primary conclusion of the study – the regimen was not supported by any of these measures of efficacy.
An abstract titled, “Study: Reolysin-Paclitaxel Combination Demonstrates Higher Response than Paclitaxel Therapy Alone”, published March 31, 2016 in The Clinical Cancer Letter (and patterned after a press release entitled “Intent-to-Treat Analysis Shows Statistically Significant Reduction in Tumour Burden as Measured by CA-125” dated March 21, 2016 accessed from the Oncolytics Biotech, Inc. website concluded that the combination regimen had more CA-125 responses than the control regimen. This analysis was restricted to those who had Full Responses versus those who had any other response, including those who were not evaluable for response by CA-125. The abstract reports, “An analysis of progression free survival stratified by measurable disease and platinum-free interval (test arm: n=54, 43 progressions, and control arm: n = 54, 48 progressions) was performed and demonstrated a median PFS of 4.4 months for the test arm, and 4.3 months for the control arm.”
The abstract draws its conclusions mostly on a secondary endpoint. We did not restrict our attention to those with full responses only as a matter of ordinary procedure. If one looks at response by RECIST, we see that 3 patients had CRs on the control regimen and 1 patient with a CR on the test arm, which is not as expected from the CA-125 responses. Analysis of response is commonly restricted to those who are evaluable for response. The abstract used an intent to treat analysis. The number of patients who are not evaluable by CA-125 is greater in the control arm than in the test arm, so the intent-to-treat analysis inflates the proportion responding on the test arm relative to the control arm. There appears to be a typographical error with the number of patients who progressed on the arms investigated. There were 48 patients who progressed on the test arm and 43 patients who progressed on control arm. Lastly, there was a long-term follow-up study of OS to see if a trend could be detected. There were 76 patients who died at the time of this analysis. The KM estimates did not reveal an advantage to the addition of Reolysin to the paclitaxel control arm. The estimated HR was 1.00 (90% CI 0.69 ~ 1.47) when the analysis was not stratified, and it was 0.91 (90% CI 0.62 ~ 1.33) when the analysis was stratified. Taken in whole, the NRG Oncology Statistical and Data Management Center feels that the original analysis and conclusions are valid and the analysis performed by Oncolytics Biotech, Inc. misrepresents the GOG-0186H data.
Philip J. DiSaia, M.D.
President, GOG Foundation Inc.