How long is long enough? Length of follow-up on clinical trials for primary treatments

Many of us are faced with the difficulty of choosing a primary therapy based on data from clinical trials with follow-up shorter than our life expectancy. How can we know what to expect in 20 or 30 years? This is quite apart from the fact that most published studies only tell us how the treatment worked for a chosen group of patients treated by some of the top doctors at some of the top institutions – they never predict for the individual case that we really want to know about; i.e., “me.” The issue of length of follow-up is particularly problematic for radiation therapies, although it may be too short for surgery and active surveillance studies as well. How can we make a reasonable decision given the uncertainty of future predictions?

I may have missed some studies, but the longest follow-up studies I have seen for each primary therapy treatment type are as follows:

• HDR brachy monotherapy - 10 years (CET/Demanes)
• HDR brachy+EBRT - 15 years (Kiel, Germany)
• IMRT - 10 years (MSKCC)
• LDR brachy monotherapy - 12 years (UWSeattle & Mt. Sinai)*
• LDR brachy+EBRT - 25 years (RCOG)
• Protons- 10 years (Loma Linda)
• SBRT - 9 years (Katz)
• Robotic RP - 10 years (Henry Ford Hospital, Detroit)
• Laparoscopic RP - 10 years (Heilbronn, Germany)
• Open RP - 25 years (Johns Hopkins)
• Active Surveillance - 20 years (Toronto)

*Mt. Sinai published a study with longer follow-up (15 years); however, all patients were treated from 1988 to 1992, before modern methods were used, and such results are irrelevant (see below) for decision-making today.

On a personal note, I was treated at the age of 57 and had an average life expectancy of 24 years, possibly more because I have a healthy lifestyle and no comorbidities. So there were no data that could help me predict my likelihood of cause-specific survival and quality of life out to the end of my reasonably expected days. What's more, the therapies with the longest follow-up (open RP, brachy boost) also have the highest rates of serious side effects. With my low-risk cancer, there seemed little need to take that risk with my quality of life.

While we may be tempted to wait for longer follow-up, (1) we don't always have that luxury, and (2) there very likely will not be any longer follow-up. Not only is follow-up expensive, there are also the problems of non-response, drop-outs, and death from other causes. The median age of patients in radiation trials is typically around 70, so many will leave the study. The 10-yr Demanes study, for example, started with 448 patients, but there were only 75 patients with 10 years of follow-up. The “10-year” study of IMRT at MSKCC started with 170 patients, but only 8 patients were included for the full ten years! After the sample size gets this small, we question the validity of the probability estimates, and there is no statistical validity in tracking further changes. (It is worth noting that IMRT became the standard of care without longer term or comparative evidence.)

An even bigger problem is what I call irrelevance. Technological and medical science advances continue at so brisk a pace that the treatment techniques ten years from now are not likely to resemble anything currently available (another argument for active surveillance, if that's an option). Dose escalation, hypofractionation, IGRT technology, intra-operative planning, VMAT, variable multi-leaf collimators, on-board cone-beam CT, and high precision linacs - all innovations that have mostly become available in the last 15 years - have dramatically changed the outcomes of every kind of radiation therapy, and made them totally incomparable to the earlier versions. Imagine shopping for a new MacBook based on the performance data of the 2000 clamshell iBook. By the time we get the long-term results, they are irrelevant to the decision now at hand.

What we want to learn from long-term clinical trials are the answer to two questions: (1) Will this treatment allow me to live out my full life? and (2) what are the side effects likely to be? To answer the first question, researchers look at prostate cancer-specific survival. It’s not an easy thing to measure accurately – cause of death may or may not be directly related to the prostate cancer. We usually look at overall survival as well. For a newly diagnosed intermediate risk man, prostate cancer survival is often more than 20 years, so we can’t wait until we have those results to make a decision. Taking one step back, we look at metastasis-free survival, but that is often over 15 years. Sometimes there is clinical evidence of a recurrence before a metastasis is detected (e.g., from a biopsy or imaging). More often, the only timely clue of recurrence is biochemical – a rise in PSA over some arbitrary point. That point is set by consensus. Researchers arrived at the consensus after weighing a number of factors, especially its correlation with clinically-detected progression. Biochemical recurrence-tree survival (bRFS), or its inverse, biochemical failure (BF), is the most commonly used surrogate endpoint.

We might be comfortable if outcomes seem to have reached a plateau. For some of the above studies, we are able to look at some of the earlier reported biochemical failure rates compared to those measures reported at the end of the study (ideally broken out by risk group).

• ·      In the Demanes Study, the 10-year results are virtually unchanged from the 8-year results.
• ·      In the Kiel study of HDR brachy boost, the 5-, 10- and 15-year BF was 22%, 31%, and 36%.
• ·      In the RCOG study of LDR brachy boost, the 10-, 15-, 20- and 25-year BF was 25%, 27%, 27%, and 27%
• ·      In the Mt. Sinai study of LDR brachy, the 8- and 12-yr BF was 12% and 10% for low risk; 19% and 16% for intermediate risk; 35% and 36% for high-risk patients.
• ·      In the MSKCC study of IMRT, the 3-, 8- and 10-yr BF was 8%, 11%, and 19% for low risk; 14%, 22% and 22% for intermediate risk; 19%, 33% and 38% for high risk patients.
• ·      In the Katz SBRT study, the 5- and 7-year BF was 2% and 4% for low-risk, 9% and 11% for intermediate-risk, and 26% and 32% for high-risk patients.
• ·      For comparison, the 5- 10- 15- and 25- year recurrence rates for prostatectomy at Johns Hopkins were 16%, 26%, 34% and 32%.

For most of the therapies, HDR & LDR brachy monotherapy, LDR brachy boost therapy, and SBRT, the failure rates remained remarkably consistent over the years. However, for surgery and IMRT, failure rates increased markedly in later years. Most of us can’t wait 25 or more years to see if a therapeutic option remains consistent or not, and for radiation, the results would almost assuredly be irrelevant anyway.

Ralph Waldo Emerson is misquoted as saying, “Build a better mousetrap, and the world will beat a path to your door.” An important criterion for decision-making when there is only limited data is our answer to the question: Is this a better mousetrap? Arguably, robotic surgery was only an improvement over open surgery, and not an entirely new therapy requiring separate evaluation. It has never been tested in a randomized comparison, and I doubt we will ever know for sure. Arguably, IMRT was simply a “better mousetrap” version of the 3DCRT technique it largely superseded and didn’t need a randomized comparison to prove its worth. Was HDRBT monotherapy just an improvement over HDRBT+EBRT? Was SBRT just an improvement over IMRT, or should we view it as a variation on HDRBT, which it radiologically resembles by design? There are no easy answers to any of these questions. However, as a cautionary note, I should mention that proton therapy was touted as more precise because of the “Bragg peak effect,” yet in practice seems to be no better in cancer control or toxicity than IMRT.

There is also the problem of separating the effect of the therapy from the effect of the learning curve of the treating physician. Outcomes are always better for patients with more practiced physicians. The learning curve has been documented for open and robotic surgery, but less well documented for radiation therapies. Patients treated early (and perhaps less skillfully) in a trial are over-represented in the latest follow-up, and there may be very little follow-up time on the most recently (and perhaps more skillfully) treated patients.

So when do we have enough data to make a decision? That comfort level will vary among individuals. I was comfortable with 3-year data based on choosing a theoretically “better mousetrap”, and many brave souls (thank God for them!) are comfortable with clinical trials of innovative therapies. In the end, everyone must assess for himself how long is long enough. For doctors offering competing therapies and for some insurance companies, there never seems to be long enough follow-up. I suggest that patients who are frustrated by those doctors and insurance companies challenge them to come up with concrete answers to the following questions:

• ·      What length of follow-up do you want to see, and why that length?
• ·      What length of follow-up was used to determine the standard of care?
• ·      Do you need to see prostate-cancer specific survival, or are you comfortable with an earlier surrogate endpoint?
• ·      What is the likelihood of seeing longer term results, and will there be any statistical validity to them if we get them?
• ·      Have outcomes reached a plateau already?
• ·      What evidence is there that toxicity outcomes change markedly after 2 years?
• ·      Will the results still be relevant if we wait for longer follow-up?
• ·      Is the therapy just a “better mousetrap” version of a standard of care?
• ·      Are my results likely to be better now that there are experienced practitioners?