Purpose: To assess the merit of the American Society for Therapeutic Radiology and Oncology (ASTRO) definition of biochemical failure after external beam radiotherapy for prostate cancer by testing alternative prostate-specific antigen (PSA) failure definitions against the "gold standard" of clinical failure and to study the effect of backdating the time of failure.
Methods and materials: Nine participating institutions agreed to submit follow-up results for all patients with clinically localized prostatic cancer (Stage T1b, T1c, T2, N0M0) treated between 1986 and 1995 by external beam radiotherapy only, to doses of >or=60 Gy, with no androgen deprivation before treatment. A total of 4839 men met the study criteria, with a median follow-up time of 6.3 years. The prediction of clinical failure by 102 definitions of biochemical failure was assessed using various quantitative measures.
Results: Four definitions were superior as measured by the sensitivity, specificity, positive and negative predictive values, and hazard of clinical failure after biochemical failure: two rises of at least 0.5 ng/mL backdated, PSA level at or greater than the absolute nadir plus 2 ng/mL at the call date, and PSA level at or greater than the current nadir plus 2 or 3 ng/mL at the call date. The absolute nadir was the lowest measured PSA level during all of follow-up, and the current nadir was the lowest PSA measured previous to a particular PSA measurement during follow-up. With the possible exception of patients in the low-risk group, the likelihood of ultimate clinical failure decreased as the time of biochemical failure increased. Failure definitions based on PSA levels >0.2 or 0.5 ng/mL were inferior to other definitions. Backdating the failure time introduced bias into the estimate of freedom from biochemical failure, which was increasingly overestimated at shorter median follow-up times. This bias can be circumvented either by using a failure definition based on the call date or by backdating the censoring times of patients with one or two rises who could potentially have failure at a future (unobserved) time. A short follow-up time as such does not result in bias unless the failures are backdated; in the absence of backdating, it is the precision of failure-free survival that is increasingly compromised as the follow-up time is reduced.
Conclusion: The ASTRO failure definition ended the confusion resulting from different failure definitions that had been in use, and it did so accurately enough that it is probably not necessary to recalculate previously published results. Nevertheless, for the current pooled analysis of outcome in 4839 men with a 6.3-year median follow-up, other definitions of biochemical failure were superior as assessed by various quantitative measures of concordance of biochemical and ultimate clinical failure. An additional disadvantage of the ASTRO definition is the bias introduced by backdating failures, as well as the necessarily retrospective nature of its application. Some "current" definitions, but not those based on the PSA level rising above a fixed threshold, have significantly higher sensitivity and specificity, do not lead to biased estimations of biochemical disease-free survival, and are directly applicable during patient counseling. These are all issues that would play a role in replacing the ASTRO consensus definition.