Are genomic tests for prognostic risk at diagnosis really helpful (yet)?

One of the critical and unanswered questions about the value of genomic/genetic testing in relation to the diagnosis and prognosis of prostate cancer is, “What are the currently available tests actually telling us compared to the clinical data?”

A presentation to be given at the upcoming meeting of the American Urological Association has made a start in providing an answer to this question, but there is still a long way to go.

The presentation to be given by Alam et al. in May looks at comparative data from three different genomic tests (the Decipher, Prolaris, and OncotypeDx tests) when they were used to try to determine whether a cohort of patients, all newly diagnosed with prostate cancer, were good candidates for active surveillance. However, this study was small and it was not a prospective clinical trial. The data come from a retrospective chart review which identified patients who had been given at least two of these tests — at the patients’ request — between 2014 and 2017 at one US-based hospital.

Patients who had favorable results based on these tests were considered to meet genomic criteria as good candidates for active surveillance, and favorable results were defined as follows for each of the three tests:

  • OncotypeDx test: > 70 percent probability of disease being organ-confined and Grade Group 1 or 2 disease at surgery
  • Decipher test: ≤ 3 percent probability of prostate cancer-specific mortality at 10 years
  • Prolaris test: ≤ 3 percent probability of prostate cancer-specific mortality at 10 years

All patients were also evaluated on the basis of the standard NCCN risk criteria alone and were considered appropriate candidates for active surveillance if they had very low-, low-, or favorable intermediate-risk disease according to those criteria.

Here are the key study findings as presented in the abstract of the presentation:

  • Just 22 patients had received at least two different genomic tests.
    • 12/22 patients received both the Decipher and Prolaris tests.
    • 8/22 patients received both the Prolaris and the OncotypeDx tests.
    • 2/22 patients received both the Decipher and the OncotypeDx tests.
  • 21/22 patients were considered to be suitable candidates for active surveillance based solely on their NCCN risk criteria.
  • For the 12 patients who received the Decipher and the Prolaris tests,
    • The level of agreement between the two tests was 67 percent.
    • The kappa (κ) statistic was 0.31 (p = 0.276).
  • For the eight patients who received both the Prolaris and the OncotypeDx tests,
    • The level of agreement between the two tests was 75 percent.
    • The κ statistic was 0.39 (p = 0.168).
    • The Prolaris tended to favor active surveillance over surgery.
  • For the two patients who received both the Decipher and the OncotypeDx tests,
    • The level of agreement between the two tests was 50 percent.
    • The κ statistic could not be calculated (because the sample size was too small).
  • When the results of the individual tests were compared to the NCCN risk criteria alone,
    • For the Prolaris test (N = 20), the level of agreement was 75 percent and the κ statistic was 0.21 (p = 0.117).
    • For the Decipher test (N = 15), the level of agreement was 60 percent and the κ statistic was 0.15 (p = 0.268).
    • For the OncotypeDx test (N = 10) the level of agreement was 50 percent and the κ statistic could not be calculated.

What does the κ statistic tell us? It is a way to assess the degree to which agreement between two different ways of measuring something (in this case the level of agreement between any two of the methods of assessing risk) is above or below chance. Thus a κ statistic of 0.50 is equal to chance and a κ statistic of ≥ 0.6 suggests a moderate degree of agreement between the results of the two different ways of measuring something.

So what are these data telling us?

Basically, they suggest the following:

  • There are some very real differences between the data provided by the three different genomic tests.
  • The results of the Prolaris test come closest to conforming with the prognostic outcomes suggested by the NCCN risk criteria alone.
  • In no case did the κ statistic exceed 0.50, which means that all of the results could have been achieved on the basis of pure chance.

Thus, what the study can not tell us is whether any one of these tests is “better” or “worse” at projecting whether an individual patients is

  • A good candidate for active surveillance
  • A better candidate for active surveillance than what we can already tell from the NCCN risk criteria alone

One of the conclusions reached by the authors states succinctly that:

This study highlights the difficulty of interpreting genomic tests for prostate cancer.

The only way we are going to be able to tell whether any one genomic test is better than all of the other available ones is to give multiple genomic tests to several hundred men and then follow those men over time to see whether the predictions of any one test are significantly more accurate than the predictions of the NCCN risk criteria alone and better than  all of the other genomic tests too in projecting the actual outcomes of the patients at at least 5 and 10 years of follow-up.

In the mean time, it is impossible to know whether any one of these tests is actually more helpful than the NCCN risk criteria in projecting risk of disease progression for a man who is otherwise a reasonable candidate for active surveillance. We need long-term, prospective data to allow us to make such decisions.

2 Responses

  1. I think you could make the question more precise. Genomic testing costs money, you have NCCN classification for free. The NCCN classification works. So genomics should improve outcomes of treatment by at least 10 percent to be cost effective. You can take this estimate and calculate the sample size needed for a prospective study to show such a gain. Then you look at the number of patients in the study to see if it has a chance to give an answer.

  2. Another factor is the huge expense of these tests ($3,000 to $4,000). Even if insurance/Medicare agrees to cover it, it will raise premiums for all of us.

    I’ve noticed they seem to be offered more or less routinely to prospective AS patients at some institutions. I think their only value may be for patients who are borderline AS-eligible, say with high volume Gleason 6 or low volume Gleason 3 + 4. Even then, I’m not sure if anything is actually gained over just tracking and watching closely. To know that, we would need a clinical trial where prospective AS patients are randomized to get a genomic test or to follow a rigorous AS protocol and see if there is any difference in metastasis-free survival.

    I was intrigued by another abstract that found that the combined use of phi < 24.4 and PIRADS ≤ 3 in men on AS missed only 4% of grade reclassifications found by biopsy. Since mpMRI should be used in confirmation biopsies anyway, the only additional cost was phi (which I'm told costs only $125). This seems to be a lot less expensive and a lot more useful than a genomic test.

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 )

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: