BACKGROUND - Information on tumor stage and grade are used to assess cancer prognosis and to produce standardized comparisons of end results over time. Changes in the interpretation of classification schemes can alter the apparent distribution of cancer stage or grade in the absence of a true biologic change. Since the introduction of prostate-specific antigen testing, the reported incidence of low-grade prostate cancer has declined. To determine whether this decline is in part a result of Gleason score reclassification during the same time period, we documented the potential impact of reclassification between 1992 and 2002 on clinical outcomes.
METHODS - A population-based cohort of 1858 men who were < or = 75 years of age at diagnosis of prostate cancer in 1990-1992 was assembled retrospectively from the Connecticut Tumor Registry. Histology slides of the diagnostic prostate tissue were retrieved and reread in 2002-2004 by an experienced pathologist blinded to the original Gleason score readings. Prostate cancer mortality rates for the cohort calculated using the original Gleason score readings were compared with those calculated using the contemporary Gleason score readings. Statistical tests were two sided.
RESULTS - The contemporary Gleason score readings were statistically significantly higher than the original readings (mean score increased from 5.95 to 6.8; difference = 0.85, 95% confidence interval = 0.79 to 0.91; P < .001). Consequently, the Gleason score-standardized contemporary prostate cancer mortality rate (1.50 deaths per 100 person-years) appeared to be 28% lower than standardized historical rates (2.08 deaths per 100 person-years), even though the overall outcome was unchanged. This apparent improvement in mortality held for all Gleason score categories.
CONCLUSIONS - In this population, a decline in the reported incidence of low-grade prostate cancers appears to be the result of Gleason score reclassification over the past decade. This reclassification resulted in apparent improvement in clinical outcomes. This finding reflects a statistical artifact known as the Will Rogers phenomenon.