Prostate specific antigen levels during radical radiation therapy and the prediction of outcome in localized carcinoma of the prostate

Abstract

There has been substantial interest in the effect of radiation therapy upon serum prostate specific antigen (PSA) levels in patients managed by radiation therapy and their ability to predict the eventual outcome. At our institute, an observational prospective longitudinal study was begun in 1989 to identify prognostic factors for biochemical relapse from among several variables, including PSA levels measured prior to treatment, during treatment, and post-treatment, and to summarize what happens to PSA levels over the course of treatment with radical radiation therapy. A total of 142 patients with adenocarcinoma of the prostate (T 1–4, N 0, M 0) were radically irradiated (6–7 weeks) between February 1989 and January 1991. Serum PSA levels were recorded weekly during radiation therapy in 117 patients. Of these 117, weekly PSA measurements ranged in completeness from 95 to 113 cases. A number of statistical tests were performed on the data with investigative/exploratory intent. There were 60 biochemical relapses documented in the whole group of 142 patients, with a maximum follow-up of 4.6 years and median follow-up of 3.3 years. Of the candidate prognostic variables tested by univariate analysis, the following emerged as statistically significant (i.e. P<0.05): each of the four pretreatment factors (absolute PSA value, dichotomous PSA (normal versus above normal), T category and Gleason score); the treatment variables namely, the end-of-treatment PSA, the slope of PSA, and the absolute change in PSA from pretreatment to the end of treatment; and, among post-treatment variables, the first follow-up PSA, the absolute change in PSA from pretreatment to first follow-up, and the return to normal of an above-normal pretreatment PSA by first follow-up. The majority of these factors were then subjected to multivariate Cox proportional hazards (PH) regression analyses. The end-of-treatment PSA and T category were consistently identified as independently statistically significant factors associated with biochemical relapse. The Gleason score was selected less consistently, and never when defined on a categorical scale. Therefore, using a Cox PH model with the variables of end-of-treatment PSA and T category, both defined on a categorical scale, we developed three prognostic groups with good, intermediate and poor prognoses ( χ 2 TREND = 40.7; P< 0.0001). Their 3-year biochemical relapse-free rates, for example, were: 91%, (standard error (SE) 5%); 64% (SE 9%); and 24% (SE 6%), respectively. However, substitution of the baseline value of PSA, which was also strongly associated with outcome, and using the data from all 142 patients, provided similarly distinct prognostic groups ( χ 2 TREND = 41.6; P< 0.0001), with corresponding 3-year relapse-free rates of: 92% (SE 4%); 79% (SE 7%); and 30% (SE 6%). Mean weekly PSA levels measured during treatment were found to have a negatively sloping or decreasing tendency. A statistically significant decrease in PSA occurred from pretreatment to the end of treatment ( t 116 = 7.5; P< 0.0001); the geometric mean of the ratio of end-of-treatment PSA to PSA at pretreatment was 0.7 (95% CI 0.6–0.8). The end-of-treatment PSA and T category emerged as independently statistically significant prognostic variables predicting biochemical relapse. Using the fitted Cox PH model with these two variables, three distinct prognostic groups were identified. The results using pretreatment PSA instead of end-of-treatment PSA produced similarly distinct prognostic groups. Mean weekly PSA levels measured during treatment exhibited a decreasing tendency, and a statistically significant decrease in PSa from pretreatment to the end of treatment was observed.