Risk Stratification for Radiation Therapy

Risk stratification involves assigning patients to categories based on diagnostic risk factors. The goal is to identify those patients who are more or less likely to respond to specific therapies (or active surveillance).  It is an aid to judgment for the patient and doctor, and helps assess prognosis and define the standards of care. It also provides for consistency between research studies so that they are more comparable. Because we depend on those studies for treatment guidelines, we don’t want to change the risk categories frequently.

In 1998, Anthony D’Amico introduced the most widely accepted risk stratification system. It has since been tweaked somewhat by consensus of the National Comprehensive Cancer Network (NCCN). It mainly relies on 3 risk factors – PSA (in 3 groups), Stage (in 3 groups), and Gleason score (in 4 groups) to create 3 risk categories (low, intermediate and high) with 2 sub-categories in each of the three. The “very low risk” sub-category also includes number of positive cores, highest % cancer in those cores, and PSA density. The “very high risk” sub-category also includes number of cores with Gleason score 8-10.

There are competing risk stratification systems. UCSF, for example, uses a system called the CAPRA score that includes age and % positive biopsy cores. Each risk factor is assigned points, and the points are summed to determine the risk category. It is also possible to use nomograms based on historical statistics to help with prognosis. While nomograms will always produce a risk probability as a%, those probabilities may, in some cases, be projected off of a very small dataset and their accuracy is questionable.

A risk stratification system is created through a multistep process. The risk factors are assessed to find the ones that independently predict recurrence after treatment.  For example, stage, Gleason score and PSA, although they are somewhat correlated, independently predict recurrence. Those risk factors are then grouped such that the risk is about the same within the group, but is different between the groups. For example in the NCCN system, Gleason scores of 8, 9 and 10 are all roughly the same at predicting recurrence, but carry much greater risk than lower Gleason scores. Then the risk factors are combined (either by selection or by adding points) such that the risk is about the same within the risk category, but significantly different between risk categories.

D’Amico developed his risk stratification system based on data from patients treated from1989 to 1997. His dataset comprised 888 surgery patients treated at the University of Pennsylvania, as well as 766 patients treated with external beam radiation, 66 patients treated with LDR brachytherapy monotherapy and 152 patients treated with LDR brachytherapy plus ADT at the Joint Center for Radiation Therapy in Boston. He only looked at biochemical progression, which was defined as PSA≥0.2 ng/ml for surgery patients and 3 consecutive rises for radiation patients. External beam radiation was only 67 Gy – far below what is now considered curative. Biochemical recurrence after radiation has since been redefined because the previous definition over-predicted clinical recurrence. Radiation therapies did not include combined modalities, HDR brachytherapy, SBRT or proton therapy.

In 2007, Johns Hopkins validated D’Amico’s risk groups among 6,652 prostatectomy patients. In 2008, the Mayo Clinic validated D’Amico’s data among 7.591 patients treated with radical prostatectomy only. They also broadened outcome data to include clinical recurrence, evidence of systemic progression, overall and cancer-specific survival. I am not aware of any validation studies for external beam radiation or brachytherapy.

Because treatments and outcomes have changed so much for radiation therapies, it may be time to take another look at the risk stratification used for radiation therapy. An Italian group looked at data on 2,493 patients treated at 10 centers between 1997 and 2012. Patients were treated with a median dose of 76 Gy of EBRT and 62% also received ADT (almost half were high risk as defined by NCCN.) They call their risk stratification system the Candiolo Classifier. Like the CAPRA score system, it assigns points to each risk factor. Classification is based on the sum of those points.

They found that age and% positive cores at biopsy significantly added to their model’s ability to stratify the risk of patients. The following table shows the breaks that discriminated best, and the number of Candiolo points assigned to each risk factor.

Risk Factor	Candiolo  (points)	NCCN
Age	<70 (0) ≥70 (22)	NA
% Positive Cores	1-20% (0) 21-50% (29) 51-80% (50) 81-100% (81)	<3 positive cores, ≤50% cancer in a core, and PSA density <0.15 ng/ml/g used in “very low risk” definition. <50% positive cores used in “favorable intermediate risk” definition. >4 cores with GS8-10 used in “very high risk” definition.
PSA (ng/ml)	<7 (0) 7-15 (42) >15 (96)	<10 10-20 >20
Gleason scores	3+3 (0) 3+4 (35) 4+3 (48) 8 (76) 9-10 (106)	3+3 3+4 4+3 8-10 5+(5,4,or 3)
Stage	T1 (0) T2 (17) T3-T4 (58)	T1-T2a T2b-T2c T3a T3b-T4

They defined 5 risk classes that discriminated well with risk of biochemical recurrence. The following table shows the biochemical progression-free survival (bPFS) for each risk class at 5 and 10 years. The relationship is similar for clinical progression-free survival, systemic (metastatic) progression-free survival, and prostate cancer specific survival.

Risk Class	Points	5-yr bPFS	10-yr bPFS
Very Low	0-56	94%	90%
Low	57-116	85%	74%
Intermediate	117-193	80%	60%
High	194-262	67%	43%
Very High	263-363	43%	14%

The Candiolo system beat the 3-tiered (low, intermediate and high risk) NCCN system in predicting all measures of progression after external beam radiation. For bPFS, its concordance index (a measure of how accurate its predictions are) was 72% vs. 63% for the NCCN system. It predicted metastases and prostate cancer survival with an accuracy of 80% vs. 69% for the NCCN system.

The Candiolo Classifier certainly seems to be an improvement, but should be validated by another group of researchers before it gains wider acceptance. Ideally, we would also have data on risk categories suitable for other kinds of radiation therapy, boost therapies, use of adjuvant ADT, and whole-pelvic radiation.

This “new, improved” system raises some interesting questions:

• The D’Amico/NCCN risk stratification system is based on antiquated data and a small dataset for radiation. Is it time for a make-over?

• Do we have to have a single risk stratification system against which all therapies should be assessed? It certainly facilitates comparisons between therapies if we have a single system. However, different risk factors (e.g., age and % positive cores) may be important in determining the risk associated with one therapy but not another.

• At what point has our ability to measure risk factors changed enough that the entire stratification system should be altered? The ability of multiparametric MRIs and advanced PET scans to more accurately assess stage and to target biopsy cores to more suspicious areas may increase the detected risk beyond what it was when the system was first set. Also, the Gleason scoring system and the AJCC staging system has changed over the years.

• How do we maintain comparability with older clinical trials and with our databases if we change our risk stratification? Many trials were established a decade or more ago with pre-set risk groups. When the data mature, will they be hard to analyze? A similar effect occurred when biochemical recurrence after radiation was redefined by the Phoenix consensus in 2005. In many studies, both definitions were presented for a while.

• Can a stratification system from Europe gain acceptance in the US and particularly by the NCCN? How do we get widespread agreement on which system is the “gold standard.” As far as I know, the CAPRA Score is only used by UCSF, even though it is an NCCN member.

• What is the role of other biochemical measures? PHI, 4KScore, PCA3, Oncotype Dx and Prolaris all measure risk. Should any of them be used in a risk stratification system? Should first-degree relatives who have had prostate cancer be included as a risk factor? What about African-Americans? And how should PSA be counted when the patient is taking 5ARis (Proscar or Avodart) for BPH?