A modeling system for the prognosis of prostate cancer should not only be able to predict risk among groups of patients, but for individual patients as well, Thomas Pisansky, MD, said in presenting a multiple prognostic index at the First Sonoma Conference on Prostate Cancer.
We do conduct studies on groups of patients, Dr. Pisansky said, but we see patients one at a time in the examination room. We should be able to provide a predictive index for individual patients.
In developing a prognostic model, Dr. Pisansky, of the Division of Radiation Oncology at the Mayo Clinic, acknowledged that he, as well as other investigators, are still on an evolutionary course.
The multiple prognostic index he presented relies on an equation that takes into account T stage, Gleason score, and prostate-specific antigen (PSA). While calling the index an enhanced prognostic system, he acknowledged that it is a phase II prognostic factor exploratory analysis that requires confirmation in other studies.
I do not, in contemporary practice, look at these pieces of information as directing therapy. I dont use them to say to a patient: You should get treatment A or B, Dr. Pisansky stated.
It is our hope, he said, that the prognostic groupings provided in the index will lead to phase III prognostic factor testing and therapeutic trials to determine which therapy is preferable in patients, according to more uniformly defined and homogeneous prognostic groupings.
First Foray Into Developing Index
In a previous studyOur first foray into combining multiple factors into a prognostic index we took favorable tumor grade, that is, grade 1 or 2, and a favorable PSA category, that is less than 13, and termed it the low-risk group, Dr. Pisansky said.
The high-risk category included those patients who had two unfavorable factorstumor grade of 3 or 4 and a PSA that exceeded 13. Patients who had one adverse prognostic factor were assigned to the intermediate risk category. He called this early effort a simplistic, though highly useful method to distinguish between prognostic groupings.
In this first investigation, we had only 241 patients, with relatively short median follow-up. It was obvious that we needed a large sample size, and longer follow-up.
In the next study, the sample size was increased to 500. Patients were treated solely with radiation therapy for clinical stage T1 - T4 (as determined by digital rectal examination), N0, M0 prostate cancer.
With a median follow-up of 43 months, 69 patients (14%) had clinical evidence of relapse within 5 years. In addition, 40 patients had biochemical relapse based on postradiation PSA level.
Logistic regression analysis was used to estimate risk of relapse as a function of the individual variablespretherapy clinical stage, Gleason score, and PSA leveland combinations of factors. Relative risk ranged from 1.32 to 5.72 (Table 1).
Enhanced Prognostic System
We then began to formulate what Ive termed an enhanced prognostic system, Dr. Pisansky stated.
He listed some of the principles the enhanced system is based on.
- Make no a priori assumptions about anything.
- Efficiently capture very complex interactions between variables, and test the prognostic significance of putative prognostic factors.
- Provide some information about the relative importance of these factors, and provide accurate predictions of relapse and group relapse rates.
- Because of the relatively short duration of follow-up, and the fact that there were relatively few cause-specific, or even overall mortality events, we could not select survival as the end point.
- And perhaps foremost, we felt that a system thats easy to apply in the clinical setting, and one which would span a variety of clinical practices, from a small community hospital to the most prestigious of academic centers, was appropriate for wide distribution.
Risk Score Equation
The three variables were used to construct the following risk-score equation:
Risk score (R) = (1.07 tumor stage value) + (1.21 Gleason score value) + (1.22 loge PSA), where tumor stage T1-T2 = 0; T3-T4 = 1; Gleason score 2-6 = 0; 7-10 = 1
A receiver operating characteristic analysis was used to identify risk score values to group patients into low, intermediate, and high relapse risk categories. The goal was to characterize a group with a 90% or greater relapse-free rate at 5 years (low-risk), a group with a 50% or greater relapse risk at 5 years (high-risk), and for the intermediate group to be representative of the entire patient group, Dr. Pisansky stated (Table 2).
The relapse-free probabilities at 5 years after radiation therapy were: 92% for the low-risk group and 24% for the high-risk group. And for the intermediate-risk group, the relapse-free rate was 67%, which closely paralleled the 70% relapse-free rate for the overall study population, Dr. Pisansky noted.
Narrowing Down Relapse Risk
Dr. Pisansky presented an example to show the importance of a multiple predictive index in estimating risk. As an example, in our study population of patients with clinical classification T2b disease, overall, 23% of patients had clinical or biochemical failure at 5 years. However, the range of relapse was from about 0 for the low-risk patients up to 90% for the high-risk group.
One could improve the range of risk for that T2b patient if you included the Gleason score, but not by much. If the Gleason score was 6, the range of relapse at 5 years was somewhere between 0 and 70%.
However, when adding the third variable, PSA, one can see that at a PSA value of 10, with a Gleason 6, T2b tumor, theres an estimated 10% risk of relapse at 5 years, with a confidence interval spanning from 8% to 13%. So, we have narrowed the 0 to 90% T2-b relapse risk down to 8% to 13%.