Although we are lucky enough to have a marker in the treatment of prostate cancer to predict prognosis before therapy and outcome afterwards, it has taken over 15 years to determine how to best use it, and recent long-term outcome studies have raised several issues along with options for improvement. While little controversy surrounds the prognostic use of this marker in choosing therapy, we are still exploring treatment failure definitions, their application across unlike therapeutic modalities, and the utility of early documentation of recurrence and associated salvage therapy. History When prostate-specific antigen (PSA) first came into clinical use in the mid- to late 1980s, there was great enthusiasm for its use in screening and in the follow-up of prostate cancer after therapy. As studies were conducted, PSA was also found to be a surrogate for tumor burden and proved its value in pretreatment prognosis and therapeutic decision-making. A normal postsurgery PSA level was soon agreed upon, given that with nearly all prostate tissue removed, it had to be very close to zero. Such was not the case with radiation, however, because PSA-secreting prostate was left intact. Thus, the dilemma and debate began. Some of the first publications using PSA as a measure of outcome defined successful radiation as a posttreatment PSA level ≤ 4 ng/mL[1]; it was soon discovered, however, that this level might be normal before therapy but too high for an irradiated prostate with a markedly changed secretory capacity.[2-4] Much discussion ensued as to what the appropriate level should be.[5] We saw a rapid transition to using this marker as an indicator of treatment efficacy as opposed to waiting long periods for local or distant disease to become clinically evident. Outcome studies were reported with a variety of PSA end points, however. Of note were the differences between the clinical outcome measures previously used- local, regional, and distant failure- and the PSA results that appeared to document treatment failure earlier and in higher proportion.[1,6] Consensus Guidelines
In an effort to standardize study reporting using this objective PSA parameter, the American Society for Therapeutic Radiology and Oncology (ASTRO) assembled an expert panel in 1996 and charged the members with reaching a consensus on "the significance of the depth of the PSA nadir, the definition of a rising PSA, the optimal PSA surrogate end point for total eradication of tumor or for relapse after irradiation, and guidelines for using PSA end points for reporting (publishing) success or failure after irradiation."[7] Data on patient outcomes with various PSA characteristics and trends after external-beam radiation along with illustration of various failure definitions were supplied by investigators from seven institutions, who had made significant contributions to the literature in the PSA era. Following presentations by these individuals and additional information obtained from recursive partitioning techniques, the ASTRO panel agreed on four guidelines:

• Biochemical failure is not justification per se to initiate additional treatment. It is not equivalent to clinical failure. It is, however, an appropriate early end point for clinical trials.


• Three consecutive rises in PSA is a reasonable definition of biochemical failure after radiation therapy. For clinical trials, the date of failure should be the midpoint between the postirradiation nadir PSA and the first of the three consecutive rises.


• No definition of PSA failure has, as yet, been shown to be a surrogate for clinical progression or survival; and


• Nadir PSA is a strong prognostic factor, but no absolute level is a valid cutoff point for separating successful and unsuccessful treatments. Nadir PSA is similar in prognostic value to pretreatment prognostic variables.

In addition, the following guidelines were suggested for studies submitted for publication: (1) series should have a minimum observation period of 24 months; (2) PSA determinations should be obtained at 3- or 4-month intervals during the first 2 years after the completion of radiation therapy, and every 6 months thereafter; and (3) patients reported in a series who have had one or two consecutive rises in PSA but not the three consecutive rises necessary for failure should be reported separately in the results.[7] In the 6 years following the consensus conference, the ASTRO definition has been the most frequently used PSA definition of failure after radiotherapy, providing a consistent standard by which studies can be evaluated and compared. Shortly after the conference, six of the participating investigators pooled data on stage T1/2 prostate cancer patients who had a pretreatment PSA and received external- beam radiation at least 2 years prior to analysis. They reported outcome for 1,765 patients with a median follow-up of 4.1 years using the ASTRO definition of failure.[8] The efficacy and durability of irradiation were established and prognostic factors confirmed. Subsequently, singleinstitution studies added to the body of outcome data using a consistent end point for reporting. Deficiencies in the ASTRO Definition
As is not uncommon in any situation where attention is focused on one primary recommendation, the less salient, seemingly minor points that arose from the consensus conference, such as requiring a minimum of 24 months follow-up before reporting a study and in some way indicating patients who had two but not three rises in PSA, were largely overlooked. Additionally, as frequently occurs in piloting a new procedure, more experience with the proposed definition and mature follow-up in the PSA era led to the emergence of deficiencies and interpretative issues. Major criticisms of the ASTRO definition included the lack of consideration of laboratory variation and standard error, the extensive period of time (which was follow-up interval- dependent) required to document three PSA rises, the bias associated with backdating the failure date, the potential for greater sensitivity and specificity of other definitions in predicting clinical failure, and the substantial difference between this and surgical definitions of failure. Several methods have been proposed to deal with the areas of concern. To accommodate the variation in laboratory testing of PSA levels, to reduce the problems associated with PSA values near the lower limit of detection appearing to change by large percentages when increases occurred, and to discount minor fluctuations in PSA production in normal prostate tissue, a definition that quantified the minimum amount of each rise in PSA was proposed-for example, three PSA rises of at least 0.5 ng/mL each.[9] Another method recommended as a way to compensate for small but consecutive increases in PSA level, which also reportedly enhanced the predictive power of the ASTRO definition, was to stipulate a required minimum total PSA level (eg, 1.5 ng/mL) in addition to the requirement of three consecutive rises in PSA.[10] Proposals to decrease the amount of time necessary to document three PSA rises, especially with follow-up intervals of 6 months or more, included defining failure as fewer rises, such as two but of a certain value each, or any elevation above an absolute nadir value, such as 0.2 ng/mL.[9,11] Some authors suggested that the bias introduced by backdating could be remedied by allowing for adequate length of follow-up, perhaps an additional 3 years beyond the point chosen for analysis.[12] The bias stems from reporting failure at the backdated date, when it actually takes considerably longer (ie, time for three PSA rises) to declare a treatment failure. If there is not enough follow-up time available to allow for three rises in PSA, the failure rate may be significantly underestimated at the backdated date. This bias could also be dealt with by moving the reported failure date to the date when the failure was actually determined-in the case of the ASTRO definition, to the date of the third PSA rise.[13] This, of course, would remove the backdating aspect of the definition. Finally, the backdating bias could be handled by a more complicated option that would involve backdating the censor date for patients with one or two PSA rises for whom no additional information was available.[14] To address the lack of uniformity between the definitions of failure used for differing treatment modalities, a single, surgically oriented definition using a solitary cutoff point was proposed.[ 11,15] As might be expected, without appropriate sensitivity and specificity testing, enthusiasm was decidedly lacking. Comparing Failure Definitions External-Beam Irradiation
To expand the work of Shipley et al and the first multi-institutional outcome study,[8] nine institutions recently contributed 4,839 T1/2 prostate cancer patients to a single, combined database.[16] All patients were treated in the PSA era and, therefore, had both pretreatment and a series of posttreatment PSA measurements. These men were treated with definitive external- beam irradiation alone no more recently than 1995, to provide potential follow-up of at least 5 years. Median follow-up was calculated at 6.3 years with 2,049 patients still available for analysis at 5 years, 616 at 8 years, and 179 at 10 years posttreatment. Not only did this database provide the most robust outcome report on external-beam radiation to date, but it also provided a valuable resource with which to test and compare definitions of failure. Although multiple failure definitions have been suggested for various reasons, this large body of data provided a sound basis for the testing and objective comparison of definitions.

• Sensitivity and Specificity-After examining the work on this subject, Thames et al tested 101 definitions of PSA failure with regard to their sensitivity and specificity in predicting clinical failure using the 4,839 patient multi-institution database.[17] A selection of these findings is listed in Table 1. The investigators defined current nadir as the lowest PSA measurement prior to the current measurement. Thus, current nadir will change during follow-up. Absolute nadir is the lowest PSA value during the entire follow-up period as assessed retrospectively. (The absolute nadir is the minimum of all the current nadirs measured during follow-up.) Failure definitions based on two, three, or four PSA rises are dated either at the time of the last rise ("call date"), or are "backdated" to the date halfway between the first PSA rise and the previous PSA. Clinical failure was defined as local disease recurrence, distant metastasis, PSA > 25 ng/mL, or the administration of hormone therapy. Although the last two of these factors are subject to debate, this definition took into account reasonable criterion and opinion to date. The last issue-the administration of hormone therapy before clinical failure-is especially problematic. While the reality of the situation found in any dataset is that hormone therapy is at times administered prior to clinical failure, altering the natural progression of disease, it is not certain that these patients would indeed have developed clinical failure if treatment had not been administered. One must assume, however, that they would be the very ones with the most suspicion of failure such that this circumstance cannot be ignored and the patients simply censored. Therefore, each of these four criteria was included in the definition of clinical failure. To calculate the sensitivity and specificity of each definition of failure, each patient was scored as either true-negative (no biochemical failure and no clinical failure), true-positive (biochemical failure followed by clinical failure), false-negative (clinical failure without a preceding biochemical failure or biochemical failure occurring after clinical failure), or falsepositive (biochemical failure not followed by clinical failure). Then: Sensitivity = true-positive/ true-positive + false-negative Specificity = true-negative/ true-negative + false-positive Thus, the sensitivity of a definition is the proportion of patients with a clinical recurrence who have had a prior biochemical failure, and the specificity is the proportion of patients without clinical failure who have not had a biochemical failure. As seen in Table 1, only four definitions of biochemical failure were both more sensitive and specific than the ASTRO definition. In retrospect, the ASTRO definition appears to have been a remarkably good choice.