J&J Prostate Cancer Trial Shouldn’t Have Been Stopped Early

6/3/12Follow @3nt

One of the more anticipated clinical trials at this year’s American Society of Clinical Oncology (ASCO) meeting was for Johnson & Johnson’s abiraterone (Zytiga) in prostate cancer patients. This study, known as COU-AA-302, was stopped early because independent monitors said the drug was clearly effective. But that decision to stop the trial early is questionable, because the drug hasn’t shown a statistically significant survival advantage. Now because the study was stopped early, it will be much more difficult to say whether this drug actually extends lives.

This trial enrolled 1,088 men around the world who had prostate cancer that had spread, but who hadn’t yet turned to chemotherapy. The patients, at 151 clinical sites in 12 countries, had no visible symptoms or only mild symptoms of their disease. They were randomly assigned to get abiraterone or a placebo.

At ASCO last year, data was presented for the COU-AA-301 trial that showed an overall survival (OS) for abiraterone over placebo of 4.6 months in an even more seriously ill population of prostate cancer patients, who had previously gotten chemotherapy. Based on this evidence of its efficacy in this tough-to-treat population, it was widely expected that the drug would also show efficacy in the pre-chemotherapy setting.

Unfortunately, Johnson & Johnson cannot claim that abiraterone offers a survival advantage, i.e. prostate cancer patients will live longer by taking it prior to chemotherapy. As Dr. Charles Ryan of UC San Francisco stated in his ASCO presentation, “there is a strong trend in overall survival” but, although close, it did not reach statistical significance. That means it’s possible that abiraterone helps these patients live longer, but it could be a result of random chance.

Why did the Data Monitoring Committee (DMC) recommend unblinding the study based on an interim analysis?

That was the question that Susan Halabi, associate professor of biostatistics and bioinformatics at Duke University, sought to answer in her excellent discussion of Ryan’s presentation. Given the complexity of the statistical issues in question, and the difficulties of questioning the decision to end the study early, Halabi gave a very fair review of the issues.

The COU-AA-302 trial had several planned interim analyses, with clearly defined criteria that would show, based on how much study data was available, whether the drug was effective or not.

Halabi presented the COU-AA-302 overall survival data in the context of the O’Brien-Fleming boundary, that graphically represents a curve, one side of which the drug is judged to be “ineffective,” the other side “effective.” Based on this analysis, abiraterone was “ineffective.” But as Halabi went on to say, it was extremely close to the boundary. In that case, why would the independent data monitoring committee (DMC) choose to unblind the study at the time it did the analysis, rather than wait a little longer until statistical significance might be reached?

The judgment of the DMC has to be questioned. The decision to unblind the study means that the “interim” data presented by Ryan is effectively the final data for the randomized trial. Unblinding a study introduces biases, which essentially turns a well-controlled randomized study into an observational one. As Halabi noted in her presentation, the “follow-up data will be biased due to confounding,” —which means other factors besides abiraterone could be influencing whether patients live longer in one group or another. By halting the study early, patients in the placebo group are now being switched over to abiraterone, which makes the comparison between the drug group and the placebo group less clear. Statistically, the COU-AA-302 data only shows a “trend to overall survival,” said Halabi.

The DMC’s decision to terminate the trial early has led to the trial’s failure to show a statistically significant survival advantage. Would it have required long to wait to show overall survival? Probably not. The interim analysis that Ryan presented was based on 43 percent of the expected overall survival events (333 deaths of trial subjects). This took place in the fourth quarter of 2011. The next interim analysis was planned for the second quarter of 2012 based on the expectation that 425 deaths (55 percent of the expected death events) would have occurred. Halabi told me after her presentation that it was highly likely this interim analysis would have crossed the O’Brien-Fleming boundary, and shown a statistically significant overall survival advantage for abiraterone in chemotherapy-naïve patients.

It must be said that the other primary goal in the COU-AA-302 study of radiographic progression free survival (rPFS) was met, as were the secondary endpoints. As Halabi noted, whether rPFS reflects a tangible clinical benefit is unknown. The COU-AA-302 is the first randomized Phase III trial in this prostate cancer population to use it as a primary endpoint, and it is unknown whether the FDA will accept it for regulatory approval purposes.

Perhaps the closeness of the overall survival data to statistical significance, coupled with the significant other end-points influenced the unanimous decision of the DMC to stop the trial at that point. However, this decision is questionable and it was clear that Halabi in her presentation was struggling to answer why didn’t they wait a bit longer before unblinding the data.

Some clinicians may argue that it would be unethical to continue giving men placebo based treatment when faced with evidence that a drug works, but my view is this ignores the reason that we do Phase III clinical trials, which is for drug registration. Will the FDA accept the COU-AA-302 data when abiraterone seeks to expand their indication to the pre-chemo setting? There is a risk that they may not as overall survival remains the gold standard that prostate cancer drugs are currently judged by. Wasn’t it worth waiting for 92 more study deaths, which would have been expected in a matter of months, to get a clear answer to the question that matters most?

There is also the wider issue of evidence-based medicine and how you change the standard of care. Statistical boundaries are set in order to guide decision making, and in the totality of the circumstances, there may be enough data to suggest that abiraterone is effective in the pre-chemo setting, but why not wait a few more months for more definitive evidence?

Rumor at ASCO 2012 was that the short patent-life for abiraterone played a part in this quick judgment. Data monitoring committees are supposed to be independent of a trial sponsor (in this case J&J), and they are supposed to review ongoing clinical trial results in secret. No one at ASCO could speak publicly about the data monitoring committee’s deliberations, and although minutes of DMC meetings are never published, they are available to reviewers at the FDA.

The early interim analysis of the COU-AA-302 trial is an opportunity missed to show an unequivocal survival advantage for abiraterone in the pre-chemo setting. The study failed to show overall survival and no matter what the positive PR spin is put on the data, it is unlikely that the study will ever show statistically significant overall survival for abiraterone. On some level this must be regarded as a failure of drug development. Ending studies in this way does a disservice both to medicine and the men who participated in the trial.

Pieter Droppert is a management consultant, lawyer, science writer and editor of the Biotech Strategy Blog (http://biotechstrategyblog.com) on science, innovation and new product development in the pharma & biotech industries. Follow @3nt

By posting a comment, you agree to our terms and conditions.

  • MyDisqusID2

    An excellent review of what transpired and the results presented. My congratulations on a well-written piece.

  • Herman Grid

    Nice summary, Peter.  While I see your points, I think it would have been ethically questionable for J&J to go against the DSMB recommendation for the sake of statistics as the data had clearly shown a clinically-meaningful benefit across many endpoints. Yes, it’s a “p value-centric” regulatory environment but the data must have been compelling enough to the DSMB to make the call.  Further, my understanding is that the DSMB analysis was on data collected through January 2012, with the unblinding and cross-over initiated in late March 2012, so those additional blinded two months of treament will likely tip the balance in the next planned interim analysis at 55% of events. 

  • Catherina Lucy

    Great post! Got a knowledgeable information.Thanks for the share.