This article is more than 5 years old. Information may no longer be current.
Treating the primary in metastatic prostate cancer
Click Here to Manage Email Alerts
Rusthoven and colleagues recently reported a post-hoc analysis of 6,382 men with newly diagnosed metastatic prostate cancer treated with androgen deprivation therapy, of whom 538 (8.4%) received radiotherapy to the prostate.
These researchers, who used the National Cancer Data Base (NCDB), claimed that propensity matching — taking into consideration age, year, race, comorbidity score, PSA level, Gleason score, T stage, N stage, chemotherapy, treatment facility and insurance status — ensured that the two numerically unmatched groups were comparable.
As a result, they were comfortable in their conclusion that radiotherapy produced superior median OS (55 months vs. 37 months) and 5-year OS (49% vs 33%). They also noted landmark analyses at 1, 3 and 5 years confirmed their observation.
The researchers were even more comfortable because they cited several other series with similar conclusions, although they were quick to point out that fishing in the sea of the NCDB gave them the biggest reported study to date.
In fairness, they wrote the paper carefully and well, and put appropriate disclaimers and emphasized the need for completion of extant randomized trials.
Limitations of propensity matching
It sounds like important progress, and perhaps one might conclude that patients with metastatic prostate cancer should all receive radiotherapy to the primary tumor. If that’s true, the scenario is certainly becoming complicated — and expensive — as we now also believe that patients with high-volume new presentations of metastatic prostate cancer should have docetaxel-based chemotherapy added to their androgen deprivation.
Although the latter thought is based on clean data from large, randomized trials, I am not certain about this latest offering. I am never happy when investigators extract an 8.4% sample from a large, unselected database, and do unplanned comparisons with confidence. Most know my views on propensity matching — it is a useful tool, but it does not create truth.
The investigators certainly looked at many appropriate variables, including reported “comorbidity status” — whatever that means when recorded by someone who hasn’t seen the patient, reporting the work of a physician who may be focused on other things!
However, they really were not in a position to capture the other, very real determinants of treatment selection bias — namely, patient preference, doctor preference, time availability away from work, and the art of medicine that most physicians practice daily, which leads them to use (or not use) specific treatments in an intuitive and nonrandom fashion.
In my long career, I have absolutely no idea how many times I have chosen to use mitoxantrone in favor of docetaxel — or alternatively no chemotherapy at all — when I have seen a patient who just “didn’t look right.”
Maybe I was correct, and maybe not, but my clinical decisions were often made in a fashion that propensity matching just would not capture. I really do not think that my style of educated, experienced, logic-based and intuitive practice is all that unusual, but I do think that it unintentionally defeats the intent of big-data capture with post-hoc analysis.
Occult biases
However — as the guy who advertises a new type of fruit-cutting device on TV says, “But ... there’s more!” — let’s think about the extraction of only 8.4% of the population of this large database for radiation treatment. That has the potential to represent industrial-strength case-selection bias!
How many were offered it and declined to have it ... and why? How many were referred, where the radiation oncologist felt it was not appropriate? Why might that be? Did the radiation oncologist think the pace of disease was too rapid? Perhaps the ones who received radiation actually had indolent or slowly progressive disease, or had responded to systemic therapy better than those who were not chosen for radiation.
It is quite clear that there were major baseline differences in insurance status between the two initial populations — with private insurance favoring those who received radiotherapy — as well as similar differences in PSA level and N0 status. The numbers are such that it is hard to know the biological or clinical implications of the selections implicit in extracting and matching of very different groups from the initial datasets.
One more key issue bothers me about this study — timing of radiotherapy.
NCDB has precise rules about noting treatment given ab initio; however, it is conceivable that the radiotherapy antedated the presentation of metastases, and it is quite possible that there was, thus, another occult bias — ie, the bias of adding months of survival after the radiotherapy to the OS score (although each of those cases actually represented treatment failure of radiotherapy, with salvage by systemic treatment).
Even that population might be heterogeneous — in other words, those who had low-dose radiation might have represented inadequate treatment of the primary, with subsequent failure; the failures of high-dose radiation might have reflected the emergence of occult metastases years later ... it happens!
The problem with my highly speculative thoughts above is that none of them are specifically addressed in the paper. Thus, this post-hoc analysis of a large dataset (perhaps the size captured the reviewers’ interest), which actually is predicated on a small subset who received added treatment — radiation in this case, either before or after metastatic presentation — might actually be wrong.
I don’t actually fault Rusthoven and colleagues for submitting the paper, and their framing is thoughtful and hypothesis generating, but I am disappointed that the reviewers allowed it to be published in such an important journal without some essential modifications, and with so much important qualifying information buried in an inaccessible Appendix.
The real problem is that the abstract readers may take home the message that this is a new standard of care —although the Journal of Clinical Oncology editorial comment explicitly did not support that view — and that is just not right.
Reference:
Rusthoven CG, et al. J Clin Oncol. 2016;doi:10.1200/JCO.2016.67.4788.
For more information:
Derek Raghavan, MD, PhD, FACP, FRACP, FASCO, is HemOnc Today’s Chief Medical Editor for Oncology. He also is president of Levine Cancer Institute at Carolinas HealthCare System. He can be reached at derek.raghavan@carolinashealthcare.org.
Disclosure: Raghavan reports no relevant financial disclosures.