Rethinking risk stratification for radiation therapy

In 2016, we looked at the Candiolo risk stratification system for radiation therapy. To our knowledge, it has not been prospectively validated or widely adopted. In the intervening 5 years, a number of things have changed:

  • Active surveillance (AS) has become the treatment of choice for many patients with low-risk prostate cancer, and for some with favorable intermediate-risk disease.
  • We have data from the first large randomized trial (ProtecT) of external beam radiation vs. surgery vs “active monitoring” — demonstrating 10-year oncological equivalence for favorable-risk patients.
  • Multiparametric MRI is increasingly used to find higher grade cancer. (We won’t discuss whether this has been a net benefit, as Vickers et al. doubt.)
  • Multiparametric MRI has also been used for staging by some doctors.
  • Multiparametric MRI has been used to detect local recurrence.
  • Decipher and other genomic tests of biopsy tissue have been used to independently assess risk.
  • PSMA PET scans have recently been FDA approved by the FDA for unfavorable risk patients to rule out distant metastases.
  • PSMA PET and Axumin PET scans have been approved by the FDA to determine radiographic recurrence.
  • NCCN has added the distinction between favorable and unfavorable intermediate-risk disease, as described by Zumsteg et al.
  • The use of brachytherapy has declined.
  • Several new hormone therapies (abiraterone, enzalutamide, apalutamide, and darolutamide) have been approved for metastatic patients.

Prognostic vs Predictive Risk Stratification

There is a new staging system called “STAR CAP.” It shows a patient’s prognosis of dying in 5 years or 10 years from prostate cancer (prostate cancer-specific mortality or PCSM) after availing himself of whatever standard therapies he may have chosen. This was an enormous undertaking. The researchers looked at the records of 19,684 men with non-metastatic prostate cancer (those with positive pelvic lymph nodes were included) who were treated at 55 sites in the US, Canada, and Europe between January 1992 and December 2013. Treatment may have consisted of radiation of any kind (7,263 patients) or prostatectomy (12,421 patients). Any one patient may have also have had ADT and salvage therapy. He may have also had docetaxel (2004) and Provenge (2010) therapy; Xofigo was approved in May 2013, so some few may even have had this form of therapy too. Follow-up ended in December 2017. The patients were split equally into “training” and “validation” cohorts. Secondarily, they validated it using 125,575 men in the SEER database. It has also been independently validated in Europe for prostatectomy patients.

The research team used five risk factors (except for pelvic lymph nodes [N stage]) to assign points (similar to CAPRA and Candiolo, in the following groupings:

  • Age: ≤ 50, 51-70, 71+
  • T stage: T1, T2a,b, T2c/T3a, T3b/T4 (based on physical examination, not imaging)
  • N stage: N0. N1 (based on CT)- note: only 22 patients were N1 in the training cohort
  • Gleason score: 6, 3 + 4, 4 + 3, 4 + 4/3 + 5, 4 + 5, 5 + 3/5 + 4/5 + 5
  • Percent positive cores: ≤ 50 percent, 51-75 percent, 76-100 percent
  • PSA: ≤ 6, > 6-10, >10-20, > 20-50, > 50-200 ng/ml

It divides patients into nine risk groups (three low (IA-C), three intermediate (IIA-C), and three high (IIIA-C)) based on how likely they are to die of their prostate cancer after their therapies. Interested patients can use this handy nomogram.

Their system outperforms the AJCC prognostic stage groups (8th edition) or the NCCN system if they were used to predict prostate cancer mortality.

Their system is necessarily limited by the risk factors available in the large databases they used to train and validate their model. That means that there may be risk factors that are not accounted for, including:

  • Genomic risk
  • Percentage of Gleason pattern 4 in Gleason 3 + 4 = 7 (this may be important in determining prostatectomy risk and risk of staying on AS; it is often not reported on biopsies)
  • Multiparametric MRI for staging and tumor volume
  • PSA density and perineural invasion
  • Use of 5-ARIs (Proscar or Avodart)
  • Use of PSMA PET scans to better select patients for local therapy

The STAR CAP system is also limited by how prostate cancer mortality is ascertained. For example, if a man dies of a blood clot in his lungs, heart, or brain, was that because the cancer increases blood clots, or was that a competing cause of death?


For most patients with localized prostate cancer, their cancer is not likely to be lethal after well-done therapies, at least not for a long time. Patients who are correctly diagnosed with localized PCa and treated for it will usually die of something else — their prognosis is excellent. What patients want to know is which therapy gives them the best chance of a cure and what side effects they can reasonably expect — their predicted outcomes are more important than their prognosis.

The wise advocate often counsels patients to try to stay in the present moment, and not be concerned with what may or may not happen down the line. The patient is rightly concerned with making the best treatment decision he can make given what he currently knows about his cancer. If his cancer progresses, there are potentially curative salvage therapies for both surgery and radiation. If his cancer progresses after salvage therapy, his cancer can often be managed with a variety of systemic therapies for many years. The list of systemic therapies is growing rapidly. It doesn’t help the patient to know the percentage of patients who died in the past, given the therapies that were available then. (The STAR CAP cohort goes back to 1992!) The patient wants to know his odds of a given therapy working for him now — a predictive model.

A good example of such a predictive model is the Memorial Sloan-Kettering (MSK) nomogram for predicting prostatectomy outcomes. It is based on the outcomes of over 10,000 men and is continually updated. Like STAR CAP, CAPRA, and Candiolo, it includes patient age and percentage of positive cores, as risk factors. While it also provides 10-year and 15-year prostate cancer survival estimates (also, see this MSK nomogram that uses comorbidities and actuarial survival tables to calculate 10- and 15-year survival probabilities), it tells the patient what his probability for progression-free survival (PFS) is if he is like the average man with his risk characteristics who chooses prostatectomy as his treatment. They define “progression-free survival (PFS)” as a PSA of less than 0.05 ng/ml and no evidence of clinical recurrence. It also shows the probability of adverse pathology after prostatectomy.

I know of no such comparable nomogram for radiation therapies. What is needed is a large predictive model for each of the major types of radiation therapies: external beam radiation, brachytherapy monotherapy, and the combination of external beam radiation and brachytherapy. It also needs to include whether whole pelvic treatment and androgen deprivation therapy (and its duration) are used with the radiation to the prostate itself.

Building such a database is an enormous undertaking. No one institution has enough primary radiotherapy patients to create a reliable sample for all risk strata and for modern best practice. Unlike surgery, which has changed little in its effectiveness over time (even nerve-sparing surgery didn’t change that), the effectiveness of radiation therapy changed a lot with dose escalation. Perhaps ASTRO or a multi-institutional consortium can create a registry to hold the data.

While patients making a treatment decision want to compare predictive outcomes across the treatments available to them, there are many reasons why such comparisons are difficult. The only valid way of comparing treatments is via a prospective randomized trial, like ProtecT. As we saw in the MSK nomogram, PFS or biochemical recurrence-free survival (bRFS) depends on the definition of PSA recurrence. MSK uses a PSA of 0.05 ng/ml as their definition of PSA progression after prostatectomy. Radiation therapies define biochemical recurrence as “nadir + 2.0 ng/ml.” It is impossible to say if these are comparable benchmarks. Perhaps future definitions of local recurrence after radiotherapy will include detection by mpMRI or one of the PSMA radioindicators that are not urinarily excreted that are in trials now.

The patient also needs to understand his likelihood of incurring the side effects associated with each treatment. ProtecT again provides the only direct comparison, but that is limited to prostatectomy, external beam radiation, and active monitoring. We know that side effects may increase with brachy boost therapy, use of ADT, and whole pelvic treatment.

Case Examples

(1) A 65-year-old man in good health, recently diagnosed with

  • Gleason 4+3, 7 cores out of 12 were positive
  • Clinical stage T1c (nothing felt by DRE)
  • Bone scan/CT negative
  • PSA of 7.5 ng/ml

Here’s how the various staging systems categorize him:

  • STAR CAP: Stage IIB  (IIA-C is intermediate risk)
    • 5-year PCS =1.1 percent; 10-year PCSM = 4.4 percent
  • CAPRA Score: 6 — high risk (6-10 is high risk)
  • AJCC Prognostic Stage Group: IIC (IIA-C is intermediate risk)
  • NCCN: Unfavorable intermediate risk
    • Recommended treatment options: RP + PLND, EBRT + ADT (4-6 months), brachy boost therapy ± ADT (4-6 months)
  • Candiolo score: 162 (intermediate range is 117-193)
    • 5-year bPFS= 80 percent; 10-year bPFS=60%
  • MSK pre-op nomogram:
    • 10-year and 15-year PCSM = 1 percent
    • 5-year PFS = 5 8 percent; 10-year PFS = 42 percent
    • Organ confined = 34 percent; EPE=63 percent; N1=14 percent; SVI=16 percent
  • Multi-institutional SBRT consortium (Kishan et al.) reported 7-year bRFS of 85 percent for unfavorable intermediate-risk (NCCN)
  • 10-year bRFS was reported (Abugharib et al.) to be 92 percent for brachy boost therapy among unfavorable intermediate-risk (NCCN) with relatively high late-term urinary toxicity
  • 5-year bRFS was reported (Kittel et al.) to be 81 percent for low dose rate brachytherapy monotherapy among unfavorable intermediate-risk (NCCN)

So brachy boost therapy is far more successful than surgery for unfavorable intermediate-risk patients. SBRT monotherapy may be better than either EBRT or LDR brachytherapy monotherapy because of the higher biologically effective dose.

(2) A 55-year-old man in good health, diagnosed with

  • Gleason score 3 + 4 = 7 (10 percent pattern 4)
  • 3/12 positive biopsy cores
  • Perineural invasion
  • Clinical stage T1c
  • PSA 4.5 ng/ml

Here’s how the various staging systems categorize him:

  • STAR CAP: Stage IC  (1A-C is low risk)
    • 5-year PCSM = 0.5 percent;   10-year PCSM = 2 percent
  • CAPRA score: 2 (0-2 is low risk)
  • AJCC Prognostic Stage Group: IIB (IIA-C is intermediate risk)
  • NCCN: favorable intermediate risk
    • Recommended management options: AS, EBRT, brachytherapy monotherapy, RP ± PLND
  • Candiolo score: 86 (low risk 57-116)
    • 5-year bPFS = 85 percent; 10-year bPFS =74 percent
  • MSK pre-op nomogram: 10-year and 15-year PCSM = 1 percent
    • 5-year PFS = 90 percent; 10-year PFS = 83 percent
    • Organ confined= 77 percent, EPE=21 percent, N1=2 percent, SVI=2 percent
  • Multi-institutional SBRT consortium (Kishan et al.) reported 7-year bRFS of 91 percent for favorable intermediate-risk (NCCN)
  • 5-year bRFS was reported (Kittel et al.) to be 90 percent for low dose rate brachytherapy monotherapy among favorable intermediate-risk (NCCN)

So, all therapies for favorable intermediate-risk patients have “success” rates in the same range (85-91 percent at ~ 5 years), independent of the chosen therapy. This is consistent with what we saw in the ProtecT trial. However, he isn’t a good candidate for AS because of his biopsy-detected perineural invasion (see this link).

(3) A 72-year-old man with a heart stent but otherwise healthy, diagnosed with

  • Gleason score 4 + 5 = 9
  • 8/12 positive biopsy cores
  • Clinical stage T3a (felt bulge)
  • PSA 15 ng/ml, neg. bone scan/CT

Here’s how the various staging systems categorize this patient:

  • STAR CAP: Stage IIIB (IIIA-C is high risk)
    • 5-year PCSM = 6 percent; 10-year PCSM = 21.2 percent
  • CAPRA score: 8 (6-10 is high risk)
  • AJCC Prognostic Stage Group: IIIC (IIIA-C is high risk)
  • NCCN: high/very-high risk (two high-risk features)
    • Recommended treatment options: EBRT + ADT (1.5-3 yrs), brachy boost therapy + ADT (1-3 yrs), RP + PLND
  • Candiolo score: 256 (high risk 57-116)
    • 5-year bPFS= 67 percent; 10-year bPFS= 43 percent
  • MSK pre-op nomogram: 10-year PCSM = 4 percent; 15-year PCSM = 10 percent
    • 5-year PFS =12 percent; 10-year PFS = 7 percent
    • Organ confined= 1 percent, EPE=99 percent, N1 = 71 percent, SVI = 79 percent
  • Kishan et al. reported that for Gleason 9/10 patients at UCLA and Fox Chase, 10-year bRFS was 70 percent for brachy boost therapy, 60 percent for EBRT, and 16 percent for prostatectomy. While surgery by itself is inferior to radiation therapies for these very high-risk patients, surgery+ salvage RT has success rates that seem to be closer.

In this case, age and the heart stent probably rule out surgery. His expected lifespan argues against watchful waiting. Brachy boost therapy and 18 months of adjuvant ADT (with cardiologist agreement) is a preferred option. Pelvic lymph nodes should be treated because of the high risk of pelvic lymph node invasion. If possible, a PSMA PET scan should be used to rule out distant metastases.

For patient decision-making, prognostic risk groups like STAR CAP, AJCC, and CAPRA are useless. The NCCN risk groups were based on prostatectomy bRFS. Counts of positive cores already used in the NCCN schema help differentiate very low-risk from low-risk, favorable intermediate-risk from unfavorable intermediate-risk, and high-risk from very high-risk patients. It is not clear that age is a risk factor that determines the oncological success of any therapy (although it undoubtedly affects toxicity). As we can see from these prototype cases, we are more needful of a risk stratification system/nomograms for the various radiation therapies similar to the MSK pre-op nomogram.

Editorial note: This commentary was written by Allen Edel for The “New” Prostate Cancer InfoLink.

4 Responses

  1. Testing and treatment was one of the worst mistakes I have ever made in my life. All my doctors lied to me and I now have a long list of side effects. I get no help from any of my doctors. Do not get tested or treated is my advice. Read the USPSTF recommendation vary carefully. A sugar coated turd is still a turd.

    [NB: John provided a link to the following URL which appears to be inaccurate:

  2. Yes, and the MSK pre-op nomogram updates are a great thing … though I guess any increases in effectiveness of surgery would mostly come from newer salvage/adjuvant treatments … otherwise I think you have said surgery hasn’t changed much over a long time period.

    I’ll note that the PREDICT nomogram used data where radiation was the most prevalent treatment … something like 2:1 I believe. The authors state that PREDICT is most useful for men considering treatment vs AS/watchful waiting. For my variables (including GS 4 + 5 = 9, PSA 7.5 ng/ml, 2/14 cores positive for cancer, etc,). PREDICT shows

    — With no treatment: 10% cancer deaths at 10 years, and 15% at 15 years
    — With treatment: 5% cancer deaths at 10 years and 8% at 15 years

    MSK shows

    — 98-99% cancer-specific survival (CSS) at 10 and 15 years.

    Like STAR-CAP, PREDICT was derived from much community hospital data in something like years 2000 to 2010?

    Is MSK that superior in treatment mortality?

  3. Dear John:

    You are by no means the first person to feel the way you do. However, the question of whether you did or did not need treatment would depend on a detailed assessment of information available at the time of your diagnosis. And there are many men who would feel very strongly that getting tested for risk of prostate cancer is a very different decision than a decision about whether to have treatment (of whatever type might seem to be appropriate) after a diagnosis of clinically significant prostate cancer

  4. Dear Dale,

    The PREDICT nomogram is from the NHS and is based on men in the UK only. Diagnosis of localized prostate cancer during that time period typically came later than in the US, so it is not surprising that mortality is higher. It also has too few high-risk men to reliably base statistics on.

    One of my pet peeves is when nomograms don’t show the sample size on which the estimate is based. They should also show the 95% confidence interval.

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: