The Progression-Free Survival to Overall Survival Connection. How to Interpret – and Why

If a therapy in castrate-resistant prostate cancer (CRPC) shows an improvement in progression-free survival, but not overall survival, should that change practice? The topic is top of mind for us this month based on the results of the ACIS study presented at the ASCO GU 2021 Genitourinary Cancers Symposium by Dana Rathkopf of Memorial Sloan Kettering – and I’ll explore it here. To briefly recap, ACIS compared abiraterone to abiraterone plus apalutamide in men with chemotherapy naïve CRPC.


The study was the natural successor to the COU 302 study of a decade ago, and the mirror image of A031201, a study by the Alliance for Clinical Trials in Oncology that compared enzalutamide to enzalutamide plus abiraterone. A031201 was reported a couple of years ago by Dr. Michael Morris, also of Memorial Sloan Kettering, and was the natural successor to the PREVAIL study with enzalutamide.

The radiographic progression-free survival (rPFS) of the abiraterone arm in ACIS (16.6 months) was nearly identical to the rPFS arm of abiraterone in 302. This is remarkable for a host of reasons and speaks to the consistency of the disease across time. Further, it suggests that, despite a decade of effort, any additional progress that has been made in supportive care did not meaningfully impact the outcome of abiraterone (Abi) treatment in a clinical trial setting. To be clear, we don’t know the rPFS of Abi in a “real world” setting because we don’t know how well clinicians follow the idealized rPFS collection methodology that was performed in both studies. Also, of interest is that the overall survival data from this trial also suggested that we have not made significant strides in the past few years. The overall survival (OS) of 36.2 months for the combination and 33.7  in the monotherapy arm are not that numerically different than what was seen in COU302, the benchmark for ACIS, or the Alliance A031201 study.

Second, following the COU302 study Dr. Morris, myself, and the statistical team from Janssen did a rigorous analysis of the relationship between rPFS and OS and demonstrated that the two correlate statistically in a positive manner with a correlation coefficient of 0.7. That means, directionally, rPFS will correlate with OS. The objective of that analysis was to show statistical correlation potentially as a guide for future study designs (it guided the design of A031201 for sure) but also as a piece of the larger whole of whether rPFS could emerge as a valid stand-alone endpoint, acting potentially as a surrogate for OS. If that was successful, we could design trials with the singular endpoint of rPFS and understand that a study positive for rPFS, by definition, would improve OS. In short, rPFS as a sole endpoint would shorten study timelines, reduce costs, and accelerate progress. That was the hope.

However,  in ACIS, as in A031201, while an improvement in rPFS was observed, an improvement in OS was not. Does this mean they don’t correlate? A deeper dive into this observation, its implications, and its practical clinical utility follow:

To address the potential biological and practical reasons why rPFS could be prolonged while OS may not be, I quote a tweet from Dr. Morris himself: “Some possibilities: 1. The degree of delay of rPFS is not prolonged enough 2. Post-treatment biology is heterogeneous and washes out the benefit of on-treatment effect 3. Clinical events not captured by scans are influencing OS.

 Some detail on each of these points:

  • The degree of delay of rPFS is not prolonged enough

It’s a matter of degree.  There has to be a point when the rPFS advantage is simply not long enough, and the effect gets ‘washed out’ after treatment is complete for unclear or even random reasons. Remember that the way we structure clinical trial designs is about considering the median OS but with confidence intervals. We often forget about the confidence intervals when we look at outcome curves. All it would take would be enough of a variation in the long-term outcome (based on random events including deaths, not from prostate cancer) to widen out the confidence intervals so that they overlap and reduce significance. Depending on the size of the study there may not be enough power to tease out this relationship. Look at the actual survival data- 36 months in the double arm vs 33 in the monotherapy arm. This is a three-month difference on OS, but it's not significant. That’s the power of…study power.

  • Post-treatment biology is heterogeneous and washes out the benefit of on-treatment effect

Post-treatment biology. The emergence of an ultra-aggressive form of CRPC has long been a sort of bogeyman in the treatment of the disease, to the point that, not so long ago, academics and clinicians would argue against treatment with early ADT because it would only potentiate a more aggressive phenotype. LATITUDE, ARCHES, ENZAMET, and CHAARTED put those arguments to bed, I think, clearly demonstrating that earlier more aggressive treatment improves the outcome overall. It does not create some sort of straight edge phenomenon’ where patients are fine until resistance occurs, only to result in rapid death when it does. We just haven’t seen that on a consistent basis (although most of us have seen dramatic examples!)

Indeed, in the last decade a variety of biological explanations for, and descriptions of, such a hyper resistant clinical state have emerged. They go by the terms of anaplastic, neuroendocrine, or triple aberrant prostate cancer (alterations in Rb1/PTEN/TP53).  But, this hyperaggressive resistant CRPC does not result in all cases, and as many as 50% of patients who experience disease progression on abiraterone or enzalutamide or other therapies will obtain substantial clinical benefit from a subsequent therapy.  

Thus, while it may contribute to some degree to the disparity in outcome between rPFS and OS, post-treatment disease biology is unlikely to explain it entirely. Post-study therapy is heterogeneous. This is particularly true in global studies where the ability to deliver subsequent therapy is more highly variable and regulated. In the US for example, a patient on the ACIS study was probably more likely to get docetaxel, cabazitaxel, radium-223 - all in sequence. Whether multiple subsequent treatments contribute meaningfully to clinical outcome is not known, but if it does in some cases it may have been enough to push the OS a bit better in the control arm, for example, widening the confidence intervals enough to result in non-significance.

  • Clinical events not captured by scans are influencing OS. Clinical events not captured by the rPFS endpoint. An underappreciated aspect of rPFS is that is the ‘r’ - suggesting that the only events that matter in the progression of the cancer are those that we can see in black and white – on scans. In my 20 years of focusing on the management of prostate cancer patients and their outcome I have witnessed many distinct types of disease progression – new lesions on scans and growth of nodes and visceral tumors is commonly the course that is run, but patients can develop more disseminated “liquid” disease ( presumably in the bone marrow), growth and progression within one bone met, unexplained declines in status, and of course death from other causes. All of these could create dents in the rPFS endpoint, and reduce the precision with which it mimics the whole disease. We recently published such findings, called unequivocal clinical progression, from the COU-302 study and found that, although such events occur in 20+% of cases ( e.g. clinical events occur without radiographic progression) they did not substantively impact the outcome of the study overall.

Studying endpoints isn’t flashy work, and novel endpoint development rarely gets the statistician to the podium at international meetings. But it is of critical importance for regulatory approvals, study reproducibility, and above all, quantifying how we can promise our patients that a therapy will bring them “clinical benefit”.

Written by: Charles Ryan, MD, B.J. Kennedy Chair in Clinical Medical Oncology, Professor of Medicine and Director of the Division of Hematology, Oncology and Transplantation, University of Minnesota, Minneapolis, Minnesota


Related Content:

ASCO GU 2021: Final Results from ACIS, a Randomized, Placebo-Controlled Double-Blind Phase 3 Study of Apalutamide and Abiraterone Acetate plus Prednisone (AAP) Versus AAP in Patients with Chemo-Naive Metastatic Castration-Resistant Prostate Cancer