The prognostic value of PSA kinetics in untreated men with prostate cancer

The question of whether PSA kinetics (e.g., PSA doubling time and PSA velocity) can be used to accurately project risk for prostate cancer progression and mortality is still not definitively known (although it is much debated).

In a new paper in the Annals of Oncology, Thomsen et al. have used data from patients in the placebo arm of the Scandinavian Prostate Cancer Group 6 (SPCG-6) trial — i.e., men with effectively untreated, localized prostate cancer who were simply on watchful waiting — to see whether PSA kinetics did, in fact, correlate with their mortality rates over a 13-year time frame.

The SPCG-6 trial included 263 men randomized to the placebo arm of the study who were initially diagnosed as having clinically localized prostate cancer and who remained on placebo for at least 18 months after study entry. Thomsen et al. used the available data for each of these 263 men to assess a spectrum of PSA kinetics data, including their

  • PSA doubling time (the time it takes for a patient’s total serum PSA level to double, e.g., from 10 ng/ml to 20 ng/ml)
  • PSA velocity (the absolute annual increase in total serum PSA, in ng/ml/year)
  • PSA velocity risk count (PSA-VRC) a score, based on serial PSA velocities and the number of times that they exceed a threshold of 0.4 ng/ml/year)

All patients included in the study survived for at least 2 years after diagnosis and had a minimum of three PSA determinations available.

Here are the authors’ core findings:

  • Of the 263 eligible patients in the SPCG-6 trial
    • 116 (44.1 percent) had a PSA of ≤ 10 ng/ml at trial entry (Group A).
    • 76 (28.9 percent) had a PSA of 10.1 to 25 ng/ml at trial entry (Group B).
    • 71 (27.0 percent) had a PSA of > 25 ng/ml at trial entry (Group C).
  • Average (median) follow-up of the 263 patients was 13.6 years.
  • For the men in  Group A there was no significant association between PSA kinetics and prostate cancer-specific mortality (PCSM) at 13 years.
  • For the men in Group B, there was a clearly significant association between PSA kientics and PCSM at 13 years.
    • For men with PSA doubling times ≤ 3 years, PCSM = 62.0 percent.
    • For men with PSA doubling times > 3 years, PCSM = 16.3 percent.
    • For men with a PSA velocity > 2 ng/ml/year, PCSM = 48.0 percent.
    • For men with a PSA velocity ≤ 2 ng/ml/year, PCSM = 11.0 percent.
    • For men with a PSA-VRC of 2, PCSM = 45 percent.
    • For men with a PSA-VRC of 0 or 1, PCSM = 3.8 percent.
  • For the men in Group C there was again no significant association between PSA kinetics and PCSM at 13 years.

The authors conclude that

magnitude changes in 13-year risks of [prostate cancer-specific] mortality that can be indicated by PSA kinetics depend on PSA level in patients with localized [prostate cancer] who were managed observationally. Our results question PSA kinetics as surrogate marker for [prostate cancer-specific] mortality in patients with low and high PSA values.

One is tempted to wonder what the results would have shown if the patients had initially been grouped by Gleason score (e.g., a Gleason score of ≤ 3 + 4 = 7 or a Gleason score of ≥ 4 + 3 = 7) as opposed to by PSA level. The numbers of patients would almost certainly have been too small to group them by both PSA level at study entry and by Gleason score too.

14 Responses

  1. These results won’t make any difference to the attitude of many urologists who will simply decide to order a biopsy once the threshold of 4 ng/ml or even lower is reached.

    The problem is the attitude and competence of many urologists, not the availability of appropriate strategies. They are basically a law unto themselves.

  2. Dear Michelle:

    Just being angry about whatever happened to a member of your family isn’t going to help anyone. The path to change is through the better education of both physicians and the public in general. We all tend to be resistant to change. It is a part of normal human nature, and far from being something unique to the urology community.

  3. Puzzling Results for PSADT and PSA velocity, risk factors and 13 year mortality due to prostate cancer (“PCSM”)

    The result for Group A, which showed no significant difference in mortality for these factors, is understandable and expected. Using the Memorial Sloan Kettering Cancer Center doubling time calculator and a little trial and error, at the 13-year point PSAs would still be far below the range associated with mortality. For Group A (PSA less than or equal to 10) the average initial PSA was probably around 6; at 13 years, with a doubling time of exactly 3 years, the projected average PSA would be 121. Even at the maximum PSA of 10 for Group A, the projected average PSA at 13 years with a 3-year PSA doubling time would be 207, still well below the lethal range. (Granted, some men have disease that does not produce much or any PSA, but these cases are likely too rare to have affected results for the 116 men in this group.)

    Results for Group B, where an effect was found, also make sense.

    What doesn’t make sense to me is the absence of an effect with men who start with a PSA of greater than 25 — Group C. There should have been an effect, but there was none. Perhaps it just happened by chance for the 71 men in this group that, though they started with relatively elevated PSAs, their PSA doubling times were disproportionately low, perhaps with a disproportionately high number of benign growth and/or infection, as well as the diagnosed cancer, accounting for the substantially elevated PSAs. That seems possible but unlikely to me. Another explanation is that men in this group might have tended to get effective treatment sooner than the men in Group B, but the abstract gives no indication one way or the other.

    I have a hunch there is an explanation for the apparent lack of effect in Group C in the context where there was an effect in Group B, but I’m puzzled.

  4. Dear Jim:

    I am not sure that I necessarily agree with your statement that there “should have been” a difference in effect on mortality rates among the men in Group C at all.

    If one was to make the not unreasonable presumption that all the men in Group C had at least micrometastatic disease (TxNxM0) by the time that they had a PSA of at least 50 ng/ml (i.e., at about 4 years of follow-up or less), you are making the presumption that there is necessarily some type of association between PSA doubling time/velocity in untreated men with disseminated disease and time to their death. However, I know of no evidence to support such a presumption. And we are never going to be able to determine this today because it would be unethical not to treat such men (assuming that they had any symptoms of disseminated diseases).

  5. A Possible Explanation for the Puzzling Difference In PC Mortality Between Groups B and C in This SPCG-6 Study

    In overview, I suspect that, with both Groups B and C being rather small from a statistical analysis standpoint (76 in Group B and 71 in Group C), a circumstance that makes it difficult to achieve significance and where chance factors and “statistical noise” can have a large impact, Group B happened by chance to include a disproportionately large number of men with PSA doubling times in the high risk or dangerous range, while Group C happened by chance to include a disproportionately small number of men in that high risk/dangerous range, and that this chance factor was primarily behind the differences in mortality.

    I still have not seen the full study, but I have a hunch that the numbers of men in both Groups B (PSA baseline in the 10.1 to 25 range) and C (PSA baseline > 25) who also had short PSA doubling times as defined in the study (less than or equal to 3 years, or in other words up through 3 years) were small enough that chance rather than a real effect determined the study results. Key context here is that the chosen PSA doubling time research cut point of less than or equal to 3 years combines two essentially different groups: men with dangerously short PSA doubling times and men with fairly safe PSA doubling times, as further considered below; my strong impression, based on reading a number of studies, is that this point is well established.

    Regarding the numbers in the study in each group with "short" (less than or equal to 3-year) PSA doubling times, while the abstract is silent on this, it is possible to make some rough but reasonable approximations based on other studies. I took a look at one of the earlier Toronto active surveillance (AS) studies by Dr. Laurence Klotz's team because the complete paper is available online, thinking of him because of his well-known interest in PSADT as a tool for managing AS. In this paper we can read the team's 2003 analysis of 231 men qualifying for PSA doubling time analysis in their AS cohort at that time. (Later analyses have been published but are not available as complete papers online.) Of the 231, 6 (2.6%) had a PSA doubling time of between 6 months and 1 year, 20 (8.7%) from 1 to 2 years, and 26 (11.3%) from 2 to 3 years. It is reasonable that the distributions of PSA doubling times for these Toronto study patients and the patients in the subject study are fairly similar, and similar to like groups in other studies.

    A second and complementary key point is that there has been research on prostate cancer mortality and doubling time that indicates substantially different outcomes for men in different PSA doubling time ranges up to a limit of a PSA doubling time of 15 months rather than 3 years. One instance of this research is the Johns Hopkins pool of radical prostatectomy recurrence patients, initially studied by Freedland and colleagues. They found that PSA doubling time was the most important of three factors (later collapsed to two) in forecasting whether a recurrence was serious or not, and the range of PSA doubling time outcomes they looked at was 15 months or more versus shorter ranges of less than 3 months, 3 to 8.9 months, and 9.0 to 14.9 months. They observed that survival at 15 years was very high at 87% even for men with Gleason 8 and higher cancer if the PSADT was 15 months or longer. In sharp contrast:

    — For men with a PSADT of 9 to 14.9 months, 15-year survival was 72% with a Gleason of 8 or higher and 86% if lower;
    — For men with a PSADT of 3.0 to 8.9 months, 15-year survival was 30% with a Gleason of 8 or higher and 59% if lower; and
    — For men with a PSA of 3.0 to 8.9, 15-year survival was 2% with a Gleason of 8 or higher and 19% if lower.

    While the Freedland group is a post-surgery group rather than an initial watchful waiting group, I strongly suspect the differences in survival at 15 years in the Freedland group are mostly due to PSA doubling time and not to how patients were initially managed. In short, while this is admittedly rough, it gives us a benchmark for looking at PSADT in the subject study. The Freedland results suggest that the subject study used too broad a range for assessing the mortality associated with PSADT. I would like to see a redone analysis using a cut point of 18 months. I suspect that the mortality observed in the subject study would be rather strongly associated with PSA doubling times of 18 months. (Patients in the subject study group had to have at least 18 months in the observation arm of the study to qualify for analysis.) Instead, by almost certainly lumping in a majority of patients with fairly non-risky PSADTs in the less than or equal to 3 year groups (Group B and Group C), true results regarding the significance of PSA doubling times were likely obscured. (A quick note: Results from the Freedland group patients would almost certainly be much better if those patients were to have been diagnosed and treated more recently, especially if starting out with today's technology.)

    Another important problem related to all this is that the numbers of patients in Groups B and C with PSA doubling times up to 3 years had to be quite small. In the Klotz study above, 52 of 231 men, or 23% (52/231) had a PSA doubling times of less thaN OR 3 years. Applying 23% to Groups B and C, we get:

    – Group B: 76 men x 23% = 17; and
    – Group C: 71 men x 23% = 16.

    Those groups are quite small for any type of meaningful analysis. Now if we recognize that more than half of each group likely had PSAs in the fairly safe range of 15 months or greater, we can estimate that perhaps only about 8 men in each group had a fairly risky PSA. It's no wonder that no significant result was observed in Group C. It's a little surprising that results considered meaningful were observed in Group B.

  6. Dear Jim:

    You are comparing apples and oranges again. … The patients in the Klotz papers and the patients in the Freedland study are both so utterly different to those in the SPCG-6 study that this type of presumptive analysis of data from an abstract is meaningless.

  7. Dear Jim:

    You are comparing apples and oranges again. … The patients in the Klotz papers and the patients in the Freedland study are both so utterly different to those in the SPCG-6 study that this type of presumptive analysis of data from an abstract is meaningless.

  8. Dear Sitemaster,

    I do not (and did not) see results from the Klotz and Freedland papers as offering proof, partly because the study populations are quite different, the point you make. However, while the populations are different, all the patients have been diagnosed with prostate cancer and the results pertain to the significance of doubling time. To me the comparisons I’m making indicate the plausibility of the interpretation of the subject study that the results are due to chance and not to a real relationship or non-relationship between PSA doubling time and prostate cancer-specific mortality. I’m looking for the best data on the distribution of PSA doubling time for an untreated population and have used the Klotz paper as a tentative proxy. I’m looking for the best data relating PSA doubling time to mortality and have used the Freedland study as a tentative proxy.

    As the subject paper addresses the important issue of the relationship of PSADT and mortality, I’m trying to get a copy of the complete paper to see if it contains information that can support or refute, fully or partially, the points I have been offering.

  9. Dear Jim:

    (1) The Klotz and Freedland patient cohorts are very poor “proxies” for the SPCG-6 patient cohort. Indeed, in my view they are so bad as to be near to meaningless.

    (2) Any statistical assessment of a data set based on something like 250 patients divided into three or more subgroups comes with significant limits as to its evaluation for statistical differences between the groups. These are small numbers of patients at best, which is why, in my view, you are over-analyzing the whole issue. One would need more like 1,000 patients to be able to produce a reasonably reliable assessment of the validity of PSA doubling time as a marker for risk in patients like these, and even then the patients’ baseline PSA levels, clinical stages, and Gleason scores would be critical factors that might affect the significance of PSA doubling time.

    You are trying to make a silk purse out of a sow’s ear based on data from a study that was never originally designed to assess the relevance of PSA doubling time in this population. And it is worth bearing in mind that Klotz and his colleagues have generally stopped worrying about PSA velocity as a marker for risk in men on active surveillance.

  10. Now Viewing Complete Thomsen Paper; Question About a Flaw in Table 2

    I have now studied and analyzed the complete paper (though without some supplementary information available only by subscription).

    I intend to post an analysis seriously questioning the authors’ conclusion that PSADT is not that useful with baseline PSAs of 25 (and that their paper supports usefulness for PSADTs from 10.1 to 25, which I believe, but which is probably not well supported by the paper), but I would like to amplify the analysis with numbers from Table 2, “Percent cumulative and 13-years mortality (95% confidence interval) according to PSA kinetic estimates following two years of observation stratified by baseline PSA.

    The problem is that Table 2 appears to have inserted numbers in the wrong blocks for at least the “n” column, and that flaw may have carried over to the other columns in the table. (I suspect the flaw did carry over as corresponding numbers and graph results for Figure 1 do not seem to line up with the Table 2 data, even with my guess at the correct positions for “n” as follows.) Here are the details for anyone who wants to try to visualize this, but it will be tough without the actual table to look at and other information from the complete paper. The table has three major sections, the first for men with baseline PSAs 25, which Sitemaster has labeled Groups A, B and C respectively. There are six rows of data in each section: two for PSADT (below and above 3); the next two for PSA velocity (above and below a value); and the final two for PSA velocity risk count (again, above and below a value). Unfortunately, Group A data for n (the number of men in the study for Group A, which was 116 men) is entered as the first pair of rows in all three sections; Group B data (76 men) is entered as the second pair of rows in all three; and Group C data (71 men) is entered as the third pair of rows in all three sections. I suspect the correct arrangement is, for section 1, keeping the first two rows as they probably represent PSADT, then picking up the first two rows under section 2 and moving them to section 1 for PSA velocity, then picking up the first two rows for section 3 and moving them to section 1 for PSA velocity risk count, with corresponding adjustments for sections 2 and 3.

    Will someone volunteer to contact the corresponding author and check this? If no one else will commit to this on this blog, I will check myself in a few days. The corresponding author is Dr. Frederik Birkebaek Thomsen, at, Telephone +45 35457125.

    Just for the record, having put together research reports with complicated figures and tables myself, I appreciate the difficulty and especially the opportunities for error where you know what you intended and just mixed it up. The easiest time to spot mistakes is immediately after publication.

  11. Dear Jim:

    You seem to be the person who is concerned with all of this so if you want to contact the authors please go ahead. Since I don’t have the full text of the paper I am not able to check what you think you are seeing. However, since I have long acquaintance with the senior author of the paper (Dr. Iversen), who is a very highly regarded prostate cancer European researcher, I think it is highly unlikely that the paper is flawed in the way you are suggesting (although these things are known to happen).

  12. Okay, I’ll go ahead. Frankly, I wish someone could confirm the flaw, though I have checked about four times. I always find these errors hard to acknowledge as errors, but I’ve seen enough of them (and made enough myself) to realize it is possible.

  13. Hi Mike.

    I overlooked this post when it first appeared.

    It is important to clearly distinguish between use of PSA test results as a screening tool [not supported in the literature], with the use of changes in PSA kinetics over time to provide information about how a particular patient responds to treatment. The latter is on much firmer ground. Response to radiation treatment, for example, is partially determined by tracking the length of time it takes for a patient to achieve a “nadir”. A biochemical relapse can also be monitored by tracking changes in PSA levels, from the time they become detectable until they reach some critical value — say 0.2 ng/mL according to some study protocols.

  14. Hi Jim.

    I made my comment to Mike’s original post on this topic before I read the correspondence you two have had regarding possible flaws in the study.

    You are right, transcription errors do happen in preparing a document for publication and sometimes the statistics performed on a data set are pretty arcane and don’t lend themselves to a transparent analysis.

    There are many dimensions to the “apples and oranges” comparisons Mike alluded to in his comments.

    One of the most salient occurs when PSA values are compared across individuals and those values may range from 50 (i.e., several orders of magnitude). Several studies have suggested that it is necessary to convert raw PSA scores into log values when comparing results across such a span of test results, which are not normally distributed, since the level of “detectability” [the lowest value that the test result can report] does not provide an absolute zero level [because some PSA levels may have been present in the blood, but were too low to create a “detectable” level].

    This limitation is similar to that seen in Centigrade and Fahrenheit temperature scales, contrasted with a Kelvin scale, which does have a true zero value. Accordingly, it is meaningless to make statements like “the two-fold increase in temperature from 20 to 40 degrees is the same as from 80 to 160 degrees on a Centigrade scale.

    Sorry for the intrusion, hope some of this was helpful.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: