Influence of follow-up bias on PSA failure after external beam radiotherapy for localized prostate cancer: results from a 10-year cohort analysis.

PURPOSE: To estimate the biases inherent in prostate cancer outcome that arise from different failure end points and variations in follow-up time and intensity using a cohort of men with long follow-up. METHODS AND MATERIALS: The study cohort consisted of 205 men with T1-T2N0-Nx prostate cancer treated with conventional radiotherapy between 1991 and 1993. The median follow-up was 103 months. Outcome was assessed using different definitions of biochemical failure, including the American Society for Therapeutic Radiology and Oncology (ASTRO) consensus criteria and the "nadir plus two" criteria (any rise of 2 ng/mL greater than the current nadir). Patient subgroups were created according to where patients had received their last 2 years of follow-up. Patients were also stratified according to whether they were initially present in the departmental database (under regular surveillance) or were uncovered after more vigorous investigation (previously "lost to follow-up").
RESULTS: In this series with maximized follow-up, the 10-year biochemical disease-free survival rate did not change significantly with varying definitions of failure, 49% and 45% for ASTRO and "nadir plus two" criteria, respectively. Patients followed by outside physicians (n = 99) were faring better at 10 years than those followed at the treating institution by either their radiation oncologist (n = 50) or their medical oncologist or urologist (n = 52). This was by all measures of outcome, including overall survival, and metastasis-free survival. Patients previously lost to follow-up (n = 43) who were tracked down also appeared to be doing better than those on our database for whom information had been readily available (n = 161). This, however, may have been an artifact of the ASTRO criteria, which underestimates failure when insufficient prostate-specific antigen values are available.
CONCLUSION: The ASTRO definition of failure underestimates late failure. This bias may be compensated for by the use of cohorts with long follow-up or the use of the "nadir plus two" definition of failure. The use of institutional prostate cancer databases may overestimate failure rates because patients followed outside of the treating institution fared better with respect to both survival and biochemical recurrence. Vigorous attempts to obtain follow-up beyond the hospital walls may correct this bias.