Omission of *absolute* risk reduction from NEJM Editorial on PSA testing

data analysis
, ,
1

An NEJM Editorial last week, discussing long-term follow-up from the European Randomized Trial of Screening for Prostate Cancer (ERSPC), addressed itself to a lack of consensus on PSA screening which [it is claimed] contrasts markedly with general consensus around screening for other cancers. Remarkably, the article cited only a relative risk reduction in this context.

The NNT (or here, NNS) that one would read off the abstract of the accompanying report [2] is pretty large. I know NN_’s are ‘canceled’ on Datamethods :wink: , but it seems to me that even an NNS would have made a reasonable addition to this piece.

image
image2233×1554 1.28 MB

  1. Vickers A. Early Detection of Prostate Cancer — Time to Fish or Cut Bait. N Engl J Med. 2025;393(17):1742-1743. doi:10.1056/NEJMe2509793
  2. Roobol MJ, De Vos II, Månsson M, et al. European Study of Prostate Cancer Screening — 23-Year Follow-up. N Engl J Med. 2025;393(17):1669-1680. doi:10.1056/NEJMoa2503223
2 Likes
2

It’s an absolute reduction in risk of prostate cancer death of 0.002. That’s not far from the reduction of mortality from wearing seat belts or not wearing a large metal belt buckle in an electrical storm. Worth it? We’d need patient utilities to figure that out.

7 Likes
3

Agree. Although mammo and PSA similar RR, can’t just compare based on RR, need to consider adverse effects (and what they mean to individual patients*). Mammogram → core biopsy, which is relatively well tolerated (#) → even mastectomy is relatively well tolerated. PSA → biopsy itself can cause incontinence, impotence → prostatectomy even higher risk.

I cringe as I write this (and I value the editorial), because PSA useful information to have. The lesion is in its interpretation. There is CAPRA, but don’t see much else.

*This is not to say screening decision should be abdicated to patient. Some more discussion on this here.

(#) Hope not making generalizations here with respect to tolerability.

2 Likes
4

I liked your essay on ‘Kindness of Decision Analysis’, Sam! But it brought to mind the “indignity of proof” of this passage from A. N. Whitehead’s Modes of Thought

There is very little large-scale understanding, even among mathematicians. There are snippets of understanding, and there are snippets of connections between those snippets. These details of connection are also understood. But these fragments of intelligence succeed each other. They do not stand together as one large self-evident coordination. At the best, there is a vague memory of details that have recently been attended to.

This succession of details is termed proof. But the large self-evidence of mathematical science is denied to humans.

To give an example, the snippet of knowledge that the addition of 1 and 4 produces the same multiplicity as the addition of 2 and 3, seems to me self-evident. It is a humble bit of knowledge; but, unless I deceive myself, it stands before me with a clarity of insight. I hesitate to claim any such self-evidence when larger numbers are involved. I have recourse to the indignity of proof. Other people have wider powers.

For example, consider Ramanujan, the great Indian mathematician, whose early death was a loss to science analogous to that of Galois. It was said of him that each of the first hundred integers was a personal friend. In other words, his insights of self-evidence, and his delight in such insights, were of the same character as most of us feel for the integers up to the number 5.

Compare Whitehead’s sentiment here with the “relinquishing of control” acknowledged in your post.

3 Likes
5

Thank you—I enjoyed this quote. While unpacking it, I wanted to also expand.

Note the last line of the editorial, “we have to do better than abdicate the responsibility for PSA decisions to individual patients and instead formulate policies that maximize benefits and minimize harms.”

Especially for difficult decisions like PSA screening, there is a disturbing trend where—sometimes under the guise of “shared decision making”—decisions are almost completely offloaded to the patient. I think that the editorial reacts to this trend, but it may go too far in the other direction. In particular, the last line of the editorial seems to sort of separate “individual patients” from (“instead”) “benefits” and “harms.”

The individual patient plays a key role in decision making, always—they define the benefits and harms. As I argued in my post, the better we characterize the patient’s role, and distinguish it from the other roles in a decision (which involve probabilities and optimization), the better things will be. This leads to true “shared decision making,” where the patient, provider (referencing randomized trials), and optimization all play a role. This just a framework, but it serves as a guide.

For example, it helps to define the words “benefits,” “harms,” etc. It shows us how to systematically compare tests—e.g., PSA and mammogram—by taking probability (relative risk) into account, but also forcing adverse effects and individual patient preferences into the conversation.

It is difficult sometimes (for me) to see the details of all the possible scenarios and how decision analysis might apply (especially when we get into complex cases, eg the higher integers), but an overarching framework (a sort of indignity to which we have recourse) in this case is helpful.

4 Likes
6

Beautiful point. It has long been true that very knowledgeable patients choose not to engage in PSA screening because the supporting data have long been quite weak and the consequences of excessive invasive testing and treatment are substantial. It’s not just that tiny benefit from screening but also all the associated biopsies, hormonal treatment, surgery, and anxiety.

We do not “abdicate” decisions that we have no right, or even definitive evidence, to make.

2 Likes
7

Thank you - just an attempt to collect everything into one place:

If interpreted correctly, which is a challenge, PSA can be useful. For this reason, the message in the editorial is valuable.

However, PSA and tests like mammography cannot be compared based on relative risk alone. This is only part of the picture. Mammography often leads to breast biopsy and mastectomy, which are generally well tolerated. PSA can lead to prostate biopsy and prostatectomy, which can cause incontinence and impotence. Overall, any comparison of tests must involve adverse effects and their corresponding patient-specific utilities.

The editorial ends with “we have to do better than abdicate the responsibility for PSA decisions to individual patients and instead formulate policies that maximize benefits and minimize harms.” Although health literate patients with time and resources may be able to make informed PSA decisions, it is a valid point that we must avoid offloading medical decisions entirely to the patient. In the process, however, we must not entirely remove the patient from the equation. The individual patient must define, via utilities, the meaning of concepts like “benefits” and “harms.” If a decision theoretic framework is used, such utilities can be combined with risk estimates from the provider and medical research community, creating a collaboration between individual patients and providers practicing evidence based medicine. Determining how to best do this would require further studies.

Solving the PSA screening problem is nontrivial, and the editorial, and the work it describes, brings us closer to doing so. Decision analysis, as a framework for accounting for risks and benefits in a systematic, patient-centered way, can offer further recourse in the face of such a difficult problem.

8

This strikes me as a plea to stop using binary outcomes in clinical trials. Utilities can be approximated by counting all adverse events according to a ranking consensus. If the new trial had followed everyone for later adverse events such as incontinence post surgery we’d have a better chance of learning which policy is better for patients.

4 Likes
9

Yes - I think especially clinical trials need to stop binarizing outcomes into “primary” (often, mortality, for which the study is powered) and “secondary” (often, adverse effects, as an afterthought in terms of power). Utility is a function of all outcomes, primary and secondary. To make a decision based on utility, therefore, clinical trials need to tell us the joint distribution of all outcomes.

Of course, a joint distribution is harder to estimate than a treatment effect. However, if we can’t estimate an intervention’s joint distribution of benefits and adverse effects, is the intervention safe for clinical use? Similarly, utility is harder to measure than mortality. However, if we can’t estimate an intervention’s utility, is the intervention safe for clinical use?

I like the idea of some approximation to utility, whose distribution we might be able to more easily estimate. Although we still have to measure ranks, we could at least avoid having to estimate a full joint distribution over outcomes and then having to specify the functional form of utility.

Overall, I like that the editorial states that we need to consider both benefits and risks, not treating one as primary and one as secondary—this is what is needed to move toward utility.

2 Likes
10

As Andrew rightly notes in the editorial:

The problem with using the ERSPC to address the question of net benefit is one that is inevitable to the study of early detection in cancer: by the time long-term results are available, diagnostic and treatment methods have evolved, leaving us with an effect estimate for an outdated approach.

The changing landscape of diagnosis and treatment options, specifically for prostate cancer, have evolved notably, even in just the past few years. Those evolving options mean that we are dealing with something of a moving target in trying to establish risk/benefit decision making in this setting, and arguably, here in the US, we are lagging behind other western countries in terms of current and relevant practice guidelines.

Perhaps most prominently, the use of 3T mpMRI as an initial non-invasive pre-biopsy screening option, the use of MRI/US guided fusion biopsies for improved PCa detection, and from a technique perspective, the use of trans-perineal (TP) versus standard trans-rectal (TR) biopsy to reduce procedural bleeding and infection risks, as well as providing better access to the anterior part of the prostate, have all seen increasing utilization and benefit.

I would also envision that, in time, we will likely have better biomarkers for PCa and there are numerous efforts underway in that domain.

From a treatment perspective, for low-risk PCa, the use of active surveillance, has also markedly increased, and provides, at least for a period of time depending upon age and other factors, the opportunity to avoid the side effects of more aggressive interventional treatments.

I can speak with some personal experience, having been on a semi-annual follow up regimen since 2020 due to a spike in my PSA level to 7.6, and with my dad, now 92, as a first degree relative who underwent a robotic RP at age 75 due to a 4+3 Gleason biopsy at the time, along with the attendant side effects.

In mid-2020 I had a 3T mpMRI, which was negative save for an enlarged prostate, and after a second bump in PSA months later with some informed discussion, I had a TP approach biopsy, which was also negative.

Thus, since then and with that information, just dealing with BPH and a fluctuating PSA level between 5.9 and 7.3 for the past three years, albeit the trend is slowly increasing over time. Should there be another marked spike, my urologist and I have agreed to another 3T mpMRI at that time to then facilitate an informed discussion as to next steps.

This approach to my specific care was, at the time back in 2020, less common in the community setting, and after being referred by my primary to a local urologist who turned out to be ambivalent about the use of mpMRI, I ended up self-referring to an academic center some distance away. That initial local urologist now, several years later, openly advertises the use of mpMRI and fusion biopsies on their practice web site. “Crossing the Chasm” anyone?

Given what I do for a living, I would presumably be considered more informed than most prospective patients, not only in having an ability to aggressively review the relevant literature to educate myself along with knowledge of my own risk factors, but also in seeking out physicians who demonstrate a forward thinking perspective. In order to have an informed discussion, both parties need to have a common frame of reference vis-a-vis risks and benefits, with neither one abdicating their role in the discussion. The patient will come to the physician as the presumptive expert in the domain, with some reasonable expectation of being educated as apropos to their choices, so that they can hopefully be an active participant in decision making. While at the end of the day, the ultimate decision maker is the patient, a low-information patient is not helping themselves.

6 Likes
How to conduct and analyze utility-focused trials
11

That is not necessary when using a hierarchical series of outcomes. We just need consensus on which outcome is worse then which other one on a given day. Then use a multistate model, simplest when using discrete time ordinal longitudinal models.

2 Likes
12

Hi All,

In follow up to the discussion, I just became aware of the TRANSFORM Prostate Cancer Screening Trial in the UK:

TRANSFORM Trial

It appears to have been announced in May of 2024, and has just begun enrollment related communications:

BBC: Biggest prostate cancer screening trial in decades begins in UK

Hopefully it will add to our knowledge of improvements in PCa screening processes.

2 Likes
13

I appreciate sharing of your experience. In my opinion, nothing is more important than the patient’s experience. When I was about 5, my grandfather died abruptly of prostate cancer after having chosen to be conservative in treatment. As I understand it, this was partly due to informal advice from a relative and urologist, who was wary of PSA and the aggressive treatment of prostate cancer, and who would often point out in the years that followed (with some guilt I think) that PSA and its downstream procedures appeared a few years after lithotripsy obviated lithotomy (in 1986, the Watson et al. paper was published). My grandfather with prostate cancer ultimately made his own choice, though. Possibly, things worked out in the way he would have preferred; his family never saw him suffer. However, better screening guidelines and better interpretation of information would have been helpful. As mentioned, we need more studies and better models for interpreting information like PSA to help patients better achieve their individual goals of care.

Overall, it sounds like, based on your experience, there are new technologies that guard against the potential downstream adverse effects of falsely elevated PSA levels. I have also noted other potential fixes—e.g., per UpToDate, one can recheck a moderately elevated PSA after a few months (serial testing).

+1 for the point about not abdicating. Probably my main concern with the editorial had centered around this, stemming from the last line, and maybe from an earlier discussion on thresholds for decision making. The issue is when any threshold becomes one-size-fits-all (as we see with statins, which are recommended if ASCVD>=10%), which often nudges the individual patient out of the conversation. My fear was that the last line in the editorial, which seemed to separate individual patients from risks and benefits, could be read as a call for a global threshold like this. I think it’s unlikely this was the intention, though. The decision curve analysis paper appears to empower individual patient preferences; ie, the Pauker threshold reflects individual patient utility.

In the end, a medical decision can be conceptualized as a collaboration between the provider, who gives the probabilities of events, and the patient, who defines utilities on those events (although in practice the provider sometimes helps with this). Then we need to do some kind of math to optimize. Actually, I think we need to do a better job at abdicating optimization to the math.

Although studying something like prostate cancer can be a moving target, I think we need more systematic, utility-based approaches for studying interventions in general.

Re the ordinal models: thank you Dr. Harrell - still thinking more about this.

5 Likes
Optimising models for calibration
14

Sorry to hear about your grandfather. I have also experienced deaths due to prostate cancer, albeit not an immediate family member. In one case, a cardiac surgeon that I used to work with and knew from the time he was a CV surgery fellow back in the late 70s.

In essence that is correct with respect to the general direction the diagnostic process is taking, albeit again, slower here in the U.S. I would not use the term “falsely” elevated, since the elevated measurement is not strictly false, as in an error, it may just not be indicative of PCa.

Curiously, this morning, as I read this new article from the BBC about former UK PM David Cameron being diagnosed with prostate cancer:

Lord Cameron reveals he had prostate cancer

It states:

The Tory peer had a prostate-specific antigen (PSA) test earlier this year, followed by an MRI scan and a biopsy. He was then treated with focal therapy, which targets the area where the tumour is present using methods such as ultrasound waves to destroy cancer cells.

followed by other content referring to current challenges and the aforementioned TRANSFORM trial. That pathway, elevated PSA, followed by a 3T mpMRI, possibly followed by an invasive biopsy, at least today, is how it should flow.

One of the potential benefits of studies like TRANSFORM, is that it may demonstrate the ability to use ‘fast’ 3T MRI, which is essentially a biparametric MRI (bpMRI) rather than the mpMRI. bpMRI eliminates the IV contrast based component of the mpMRI, allowing for faster and cheaper non-invasive imaging, while still gaining the benefits of image quality with 3T MRIs as opposed to 1.5T MRI units. The use of bpMRI may allow for improved access to more men at more facilities, as a pre-biopsy screening tool in time.

As in my case, an elevated PSA alone should not, by default, drive one directly to an invasive biopsy procedure and given that my initial 3T mpMRI was negative, I made the decision, in consultation with the urologist, to not proceed to biopsy. It was the second bump in PSA months later, that gave rise to the subsequent discussion on proceeding to a TP biopsy, based upon two considerations, in addition to my dad’s history, which is a key risk factor for me:

  1. The incidence of a false negative 3T mpMRI is around 15% (notably lower than for a standard 12-core TRUS biopsy) based upon some of the studies that have been done. Albeit, when it misses, it tends to be low risk PCa, in contrast to a TRUS biopsy, which can miss clinically significant PCa.
  2. A second consideration and in effect, a general guideline, is the use of PSA density, which is the PSA level divided by the prostate volume. A bright line of sorts is a PSA density of >= 0.2, and I was right on that line. At or above that value, there is motivation to proceed with a biopsy, even with a negative mpMRI. Lower values (e.g. <= 0.1) would likely be indicative of non-PCa enlargement of the prostate, such as BPH, resulting in the elevated PSA level.

Thus, in my case, it was not one factor that motivated the eventual biopsy, but multiple pieces of information. So, that takes the broad, guideline based approach, and combines it with patient specific information, as it should be, to help improve decision making.

In time, our base of information will improve with more and better studies, to allow us to make better decisions, including what, if any, treatments are apropos, in the face of a positive PCa diagnosis.

5 Likes
15

Me, at age - 73, two years ago: rate of PSA increase prompted a multi-parametric MRI, which informed where to biopsy (pc diagnosed - intermediate unfavorable) followed by a nuclear bone scan (still localized), followed by molecular testing of biopsy sample (Decipher) … guiding the choice of therapy: brachytherapy with 5x SBRT boost, no hormone depravation therapy needed based on Decipher. Considered cured based on PSA of 0.26

image
image653×533 108 KB

5 Likes