Yesterday we mentioned a new paper on PSA screening by Tsodikov et al. in the Annals of Internal Medicine. We have now had the chance to read the full text of that paper, as well as the associated editorial by Vickers in the same issue of the journal.
Here is what this paper tells us.
By conducting a highly sophisticated re-analysis of all data from the PLCO and the ERSPC trials at 11 years of follow-up and by including in that re-analysis a technique that allowed the research team to estimate the “mean lead times” for the patients in the differing groups in these two major prostate cancer screening trials, the authors have been able to show that:
- “… the data from the ERSPC and the PLCO do not provide evidence that screening efficacy (relative to no screening) differed between the trials after we accounted for differences in implementation and setting.”
- “Our estimation of the common effect of screening suggests that it can significantly reduce the risk for prostate cancer death. However. …”
- “… the benefit of screening must be weighed against its potential harms for informed clinical and shared decision making.”
Getting into the details of exactly how the international research team came to these conclusions, and how the use of “mean lead times” or MLTs affected the impact of screening on prostate cancer-specific mortality in specific sets of patients in the trials would be extremely complex and probably not particularly constructive for most readers. So we are going to boil this down to the following:
- The use of mean lead times allowed to authors to take into account how often the men enrolled in the trials got PSA tests and the time from randomization of patients into the trials until prostate cancer was actually found.
- Based on that revision to the analysis, the authors found that the PLCO trial and the ERSPC trial suggested that PSA screening among men of 55 to 69 years at randomization (in the ERSCP trial) or 55 to 74 years at randomization (in the PLCO trial) led to a 7 to 9 percent reduction in risk of death from prostate cancer per year of MLT.
- These reductions in death per year of MLT equate to a roughly 25 to 32 percent reduction in risk of prostate cancer-specific death over 11 years of follow-up in the ERSPC and the PCLO trials.
But let us be very clear:
- There is still no suggestion that these reductions in prostate cancer-specific mortality are associated with a reduction in overall mortality over the same time frame (or over a longer time frame).
What these data do do, however, is
- Correlate the lowering of the annual numbers of prostate cancer-specific deaths since the introduction of PSA testing to the effects of PSA testing on prostate cancer-specific mortality
- Clearly demonstrate the value of PSA testing among those men who, for any reason, believe they may be at higher than average risk for prostate cancer (for familial, genetic, racial, or other reasons)
- Place increasing pressure on the medical community (and on the prostate cancer advocacy community too) to ensure that there is an increasingly sophisticated linkage between the utility of PSA testing to assess possible risk and the value of other methodologies to then make decisions about the need for additional risk assessment, about the need for prostate biopsies, and about the need for interventional treatments in the patients who could reasonably be expected to benefit from treatment (as opposed to suffering the side effects of such treatment without any reasonable expectation of clinical benefit).
This brings us to Vickers’ editorial and to the practical implications of these data.
First, these data provide a high level of support for the current, revised draft guideline on prostate cancer screening issued by the US Preventive Services Task Force (USPSTF) earlier this year. We hope that this revised guideline will be finalized in the very near future.
Second, these data confirm the need to improve, in a highly consistent way, the guidance that is offered to men about the value of individual PSA testing and how frequently it should be carried out in specific and definable sets of men who may be at greater or lesser degrees of risk for prostate cancer.
Third, these data should put to bed the question of whether or not PSA screening reduces risk for prostate cancer-specific mortality. Clearly it does. However, … to quote Vickers:
Unfortunately, the way screening has been implemented in the United States leaves much to be desired. The controversy about PSA-based screening should no longer be whether it can do good but whether we can change our behavior so that it does more good than harm.
In association with this paper and the editorial by Vickers, the Annals of Internal Medicine has released its own summary of this paper for the patient community, which some readers may find helpful.
For yet another, rather more sophisticated analysis of the implications of this important paper, see this video commentary by F. Perry Wilson, MD on the MedPage Today web site.
Editorial note: The “New”Prostate Cancer InfoLink thanks the staff of the American College of Physicians (the publishers of the Annals of Internal Medicine) and another anonymous correspondent for promptly providing us with full text copies of this original article, the editorial by Vickers, and other related information.
Filed under: Diagnosis, Risk | Tagged: mortality, PSA, risk, screening, testing |
THE "NEW" PROSTATE CANCER INFOLINK 
Likely Reason Why This Analysis Found No Difference In Screening Efficacy Between PLCO and ERSPC
The article by Sitemaster stated the following:
“… the authors have been able to show that:
“– the data from the ERSPC and the PLCO do not provide evidence that screening efficacy (relative to no screening) differed between the trials after we accounted for differences in implementation and setting.”
Basically this is good news, because it shows that the PLCO trial also showed a substantial benefit for screening in sharp contrast to the absence of benefit that has been unfortunately widely reported, along with the ERSPC trial where this has been evident for some time. Here are the details.
As is well known, the analysis performed in the original ERSPC and PLCO reports, and echoed prominently by the US Preventive Services Task Force, showed a significant benefit of screening in the ERSPC trial, which has become increasingly evident with follow-ups, and no significant benefit in the PLCO trial.
However, those analyses were based on “Intent to Treat” (ITT) or more clearly “Intent to Screen” (ITS) randomization rather than actual screening. In other words, if you were assigned randomly to the screening group but did not get screened, you would still be counted as having been screened; if you were assigned to the non-screening group but did get screened, you would have been counted as not screened.
Indeed, the group considered as non-screening in the US PLCO trial was actually a “usual care” group; with the high penetration of screening in US primary care practices, there was actually a high proportion of screening in that group. In fact at least one analysis showed more screening in that group than in the group where men were screened per the trial protocol.
Sticking to the ITT/ITS groups, despite “non-compliance” and “contamination,” has the advantage of allowing inferences to be made using statistical tools, a valuable benefit, but it runs aground on the shoals of reality when there is a high degree of non-compliance with randomization, as dramatically illustrated, text-book fashion, in the PLCO trial. It is worth noting and remembering that many in the medical and media communities who were involved in conveying findings, most prominently the US Preventive Services Task Force back then, were woefully unappreciative of this issue. In contrast, many of us saw the problem immediately but were like voices in the desert, including our critical comments on the USPSTF draft recommendation that was basically anti-screening, including smart screening that many of us advocated. The final recommendation ignored our points, an egregious failing!
It appears that the analysis in the subject study actually looked at who had been screened and not screened, regardless of whether men were assigned randomly to the screening or non-screening/usual care groups. This type of analysis has also been the subject of other studies, particularly for the ERSPC, and those studies have shown an even more dramatic impact of screening. This more “realistic” analysis is likely the meaning of the phrase “After differences in implementation and settings are accounted for …” in the current study.
I have not read the full study or the editorial, but I suspect this is the reason that the efficacy of screening did not appear to differ between the trials.
Huge Risk Reduction for Those Being Screened for Prostate Cancer!
This study helps communicate the huge risk reductions attributable to screening in the ERSPC and PLCO trials, despite the fact that both trials are still quite premature in a context where further maturity is likely to amplify the benefits of screening for our long-fused disease. (For instance, the “13 year” update of the ERSPC — meaning 13 years from randomization of supposedly healthy men — reported that follow-up from diagnosis was only 6.4 years in the screening group and 4.3 years in the non-screening group (see just prior to footnote 21). As Sitemaster’s article summarizes: “These reductions in death per year of MLT equate to a roughly 25 to 32 percent reduction in risk of prostate cancer-specific death over 11 years of follow-up in the ERSPC and the PCLO trials.” Hopefully that will catch the attention of the media and medical communities, especially the primary care community and the US Preventive Services Task Force.
Sitemaster also alerts us to a prominent feature of the study that may be used in rebuttal for those opposed to screening, as follows: “But let us be very clear: There is still no suggestion that these reductions in prostate cancer-specific mortality are associated with a reduction in overall mortality over the same time frame (or over a longer time frame).” The key point for us to remember here is that it would require a very large study with very long follow-up to demonstrate such a reduction because prostate cancer death is such a small percentage of overall deaths in a general population, as contrasted with a population of prostate cancer patients! Many studies of prostate cancer patients demonstrate an overall survival benefit as well as the disease-specific benefit found in this study, but it is too hard to see the overall benefit when a large pool of patients who do not have prostate cancer is part of the group being analyzed. I do hope the current USPSTF will understand that; they seem much better tuned in than the prior group that basically discouraged screening.
Let’s Also Promote Screening Value for Those Not at Higher than Average Risk
Sitemaster points out a clear value of screening for one group of men: “What these data do do, however, is … Clearly demonstrate the value of PSA testing among those men who, for any reason, believe they may be at higher than average risk for prostate cancer (for familial, genetic, racial, or other reasons)”
What I hope does not get lost is that a high proportion of challenging cases of prostate cancer is composed of men who are not at higher than average risk. For men whose doctors are able to deliver smart screening, which includes wise follow-up choices such as multiparametric MRI, understanding PSA density and other causes of PSA elevation, etc., and wise management choices such as active surveillance for appropriate risk patients, I’m convinced that widespread smart screening, with an appropriate, brief discussion, is beneficial and deserves our advocacy.
Regarding your quote:
“There is still no suggestion that these reductions in prostate cancer-specific mortality are associated with a reduction in overall mortality over the same time frame (or over a longer time frame).”
This quote is true. But given the statistical noise in trying to detect the effects of PSA screening in a group with high average mortality, there will never be a study that can succeed in statistically detecting such an overall mortality effect unless you push sample size in the experiment (or quasi experiment) into the millions. Given this problem, the sentence you write is not only true now, but is likely to continue to be true for the foreseeable future.
This is so even if the prostate cancer mortality effects are relatively large, and get translated 1 for 1 into overall mortality effects.
So I think it is more accurate to say:
“Given various statistical difficulties, we are unable to determine with sufficient statistical precision whether reductions in prostate cancer specific mortality are translated one for one into reduced overall mortality, or whether or not the overall mortality effects are nil.”
Absence of evidence is not evidence of absence.
Dear Tim:
One can “interpret” the truth any way one wants to. I simply made a simple statement of the truth, because it is important. And if there is no evidence that screening improves overall survival, then there is certainly an argument (as Jim Waldenfals points out) that the value of screening (as compared to selective testing of those at high risk and those who actually want to be tested) is rather limited on a population basis. Whether that argument is valid probably depends on one’s point of view.