C

RpneCwmtt

TOTm

Int. J. Radiation Oncology Biol. Phys., Vol. 65, No. 5, pp. 1487–1493, 2006Copyright © 2006 Elsevier Inc.

Printed in the USA. All rights reserved0360-3016/06/$–see front matter

doi:10.1016/j.ijrobp.2006.03.027

LINICAL INVESTIGATION Prostate

COMPARISON OF BIOCHEMICAL FAILURE DEFINITIONS FORPERMANENT PROSTATE BRACHYTHERAPY

DEBORAH A. KUBAN, M.D.,* LARRY B. LEVY, M.S.,* LOUIS POTTERS, M.D.,†

DAVID C. BEYER, M.D.,‡ JOHN C. BLASKO, M.D.,§ BRIAN J. MORAN, M.D.,� JAY P. CIEZKI, M.D.,¶

ANTHONY L. ZIETMAN, M.D.,** MICHAEL J. ZELEFSKY, M.D.,†† THOMAS M. PISANSKY, M.D.,‡‡

MOHAMED ELSHAIKH, M.D.,§§AND ERIC M. HORWITZ, M.D.��

*Department of Radiation Oncology, The University of Texas M. D. Anderson Cancer Center, Houston, TX; †New York ProstateInstitute, Oceanside, NY; ‡Arizona Oncology Services, Scottsdale, AZ; §Seattle Prostate Institute, Seattle, WA; �Chicago Prostate

Institute, Chicago, IL; ¶Department of Radiation Oncology, Cleveland Clinic, Cleveland, OH; **Department of RadiationOncology, Massachusetts General Hospital, Boston, MA; ††Department of Radiation Oncology, Memorial Sloan-Kettering CancerCenter, New York, NY; ‡‡Department of Radiation Oncology, Mayo Clinic, Rochester, MN; §§Department of Radiation Oncology,

University of Michigan, Ann Arbor, MI; ��Department of Radiation Oncology, Fox Chase Cancer Center, Philadelphia, PA

Purpose: To assess prostate-specific antigen (PSA) failure definitions for patients with Stage T1–T2 prostatecancer treated by permanent prostate brachytherapy.Methods and Materials: A total of 2,693 patients treated with radioisotopic implant as solitary treatment forT1–T2 prostatic adenocarcinoma were studied. All patients had a pretreatment PSA, were treated at least 5 yearsbefore analysis, 1988 to 1998, and did not receive hormonal therapy before recurrence. Multiple PSA failuredefinitions were tested for their ability to predict clinical failure.Results: Definitions which determined failure by a certain increment of PSA rise above the lowest PSA level todate (nadir � x ng/mL) were more sensitive and specific than failure definitions based on PSA doubling time ora certain number of PSA rises. The sensitivity and specificity for the nadir � 2 definition were 72% and 83%,vs. 51% and 81% for 3 PSA rises. The surgical type definitions (PSA exceeding an absolute value) could matchthis sensitivity and specificity but only when failure was defined as exceeding a PSA level in the 1–3 ng/mL rangeand only when patients were allowed adequate time to nadir. When failure definitions were compared by timevarying covariate regression analysis, nadir � 2 ng/mL retained the best fit.Conclusions: For patients treated by permanent radioisotopic implant for prostate cancer, the definition nadir� 2 ng/mL provides the best surrogate for failure throughout the entire follow-up period, similar to patientstreated by external beam radiotherapy. Therefore, the same PSA failure definition could be used for bothmodalities. For brachytherapy patients with long-term follow-up, at least 6 years, defining failure as exceedingan absolute PSA level in the 0.5 ng/mL range may be reasonable. © 2006 Elsevier Inc.

Brachytherapy, Prostate, Implant, PSA, Failure.

adD2peaemeP

fo

c

INTRODUCTION

ecently, there has been considerable interest in revisitingrostate-specific antigen (PSA) failure definitions for exter-al beam radiation therapy. Since the 1996 American Soci-ty for Therapeutic Radiology and Oncology (ASTRO)onsensus Conference which established the three PSA riseith backdating definition (1), multiple criticisms of thisethodology have been expressed (2–4), largely because of

he availability of data with longer clinical follow-up andhe opportunity to apply more sophisticated statistical

Reprint requests to: Deborah A. Kuban, M.D., University ofexas M. D. Anderson Cancer Center, Department of Radiationncology, 1515 Holcombe Blvd, Box 97, Houston, TX 77030.el: (713) 563-2329; Fax: (713) 563-2331; E-mail: dakuban@danderson.org

Presented at the 47th Annual Meeting of the American Society A

1487

nalysis. Therefore, to reevaluate the ASTRO Consensusefinition and explore other definitions, a second Failureefinition Symposium was held in Phoenix in January005. In addition to data on external beam radiotherapyatients, included were failure definition analyses forxternal radiotherapy combined with hormonal therapynd permanent radioisotopic implant. The goal was tostablish a uniform failure definition for all radiationodalities, if possible. The information in this study

xpands on the brachytherapy analysis contributed to theSA Failure Definition Symposium.

or Therapeutic Radiology and Oncology (ASTRO), Denver, Col-rado, October 18, 2005.D.A. Kuban was supported by an Educational Grant from On-

ura.Received Jan 31, 2006, and in revised form Mar 16, 2006.

ccepted for publication Mar 16, 2006.

ptttIsMCpbmPlomPtdo

(dwbitncnddtdl

Pw

pifmp

fvfbf

wscnfpad

afIvuovK(

0PipwdrwSnypl

ar1mpscb

fna(ssiiopapstgaba

t

1488 I. J. Radiation Oncology ● Biology ● Physics Volume 65, Number 5, 2006

METHODS AND MATERIALS

Eleven institutions combined data on 2693 patients treated withermanent radioisotopic implant, I-125 or Pd-103, as solitaryreatment for T1–T2 prostatic adenocarcinoma. Contributing insti-utions included: M. D. Anderson Cancer Center, New York Pros-ate Institute, Arizona Oncology Services, Seattle Prostatenstitute, Chicago Prostate Institute, Cleveland Clinic, Massachu-etts General Hospital, Memorial Sloan-Kettering Cancer Center,ayo Clinic-Rochester, University of Michigan, and Fox Chaseancer Center. All patients had a pretreatment PSA, at least threeosttreatment PSA measurements, were treated at least 5 yearsefore data submission (1988 to 1998), and did not receive hor-onal therapy before recurrence. A total of 20,945 posttreatmentSA measurements were available for the analysis. Median fol-

ow-up was 63 months, and 85% of patients were alive at the timef analysis. 1,831 patients (68%) were treated with I-125 with aedian prescription dose of 160 Gy pre-American Association ofhysics and Medicine Task Group 43 (TG-43) (5), and 862 pa-

ients (32%) were treated with Pd-103 with a median prescriptionose of 120 Gy pre-National Institute of Standards and Technol-gy 1999 Guidelines (6).Based on previous analyses with external radiotherapy patients

2), failure definitions more sensitive and specific than the ASTROefinition were tested as candidate definitions for patients treatedith permanent prostate brachytherapy. Because backdating haseen criticized (2–4) and the consensus of the 2005 Failure Def-nition Symposium was that a PSA failure definition not containhis methodology, only nonbackdated definitions were included:adir plus 2 ng/mL and nadir � 3 ng/mL. To expand upon thisategory of definitions for brachytherapy, nadir � definitions fromadir � 0.5 ng/mL to nadir � 4.0 ng/mL were tested. Nadir wasefined as the lowest PSA level to date. Definitions based on PSAoubling time were also tested (7) as were absolute value defini-ions similar to surgical series (8–10). All proposed PSA failureefinitions were tested for their ability to predict clinical failure:ocal, regional, or distant.

Similar to failure definition testing for external radiotherapy, aSA of 25 ng/mL or more and institution of hormonal therapyere also considered a clinical failure (2).To compare brachytherapy failure definitions with those being

roposed for external beam therapy, a cohort of 4839 externallyrradiated patients with Stage T1–T2 prostate cancer and long-termollow-up was used (2, 11). Specific patient characteristics, treat-ent parameters, and detailed failure definition analysis for this

atient group has been published previously (2, 11).PSA nadir kinetics were analyzed because this can impact the

ailure definition, especially those relying on an absolute PSAalue. PSA bounce, defined as a PSA rise of at least 0.2 ng/mLollowed by return to pre-bounce level, was also investigatedecause this parameter can have a profound effect on false-positiveailure rates.

With regard to the statistical analysis, sensitivity and specificityere calculated as follows: sensitivity � TP/(TN � FN) and

pecificity � TN/(TN � FP) where a true positive (TP) is bio-hemical failure (BF) followed by clinical failure (CF), a trueegative (TN) is no BF and no CF, a false positive (FP) is BF notollowed by CF, and a false negative (FN) is CF without areceding BF or BF occurring after CF. Sensitivity–specificitynalysis, generating receiver operator curves, was done for allefinitions.

Regression analysis, using time-dependent covariates, was done e

s a second method to compare failure definitions (12). The PSAailure variable was assigned a value of 0 at the end of treatment.t retained this value until a PSA failure occurred when theariable was then assigned a value of 1. The value remainednchanged until the end of follow-up or until a clinical failureccurred. For a patient who never experienced a PSA failure, theariable retained a value of 0 until the last follow-up date or death.aplan-Meier analysis was used to compare PSA disease-free

PSADF) outcome for various failure definitions.

RESULTS

The median PSA nadir for all brachytherapy patients was.25 ng/mL at a median follow-up of 2.36 years. However,SA nadir was related to the amount of follow-up time after

mplant such that the median nadir was 0.1 ng/mL foratients with at least 3 years follow-up. In 239 patients whoere PSADF at 8 years posttreatment by the nadir � 2efinition, the median nadir PSA was 0.1 ng/mL with aange of 0–2.1 ng/mL. The 75th, 90th, and 95th percentilesere 0.3 ng/mL, 0.5 ng/mL, and 0.8 ng/mL, respectively.eventy-two percent of patients had a PSA nadir �0.2g/mL, and 91% were at �0.5 ng/mL. Therefore, severalears are typically required to reach a PSA nadir, and allatients cannot be expected to reach a particular absoluteevel despite a nonrising profile thereafter.

Seventeen percent of patients experienced a PSA bouncet a median time of 17 months after implant. The medianise was 0.9 ng/mL with 75th, 90th, and 95th percentiles of.7 ng/mL, 3.3 ng/mL, and 5.5 ng/mL, respectively. Theedian duration of PSA bounce was 14 months. The bounce

henomenon has the potential to significantly affect thepecificity of a PSA failure definition as it significantlyontributes to the false-positive rate. Ninety-five percent ofounces occurred within 5 years of treatment.Sensitivity, specificity, and regression analyses were done

or four categories of failure definitions: consecutive rises,adir plus a designated value, PSA doubling time, and anbsolute, designated PSA value. A receiver operator curveROC) showing the sensitivity (y axis) plotted against thepecificity (1 � specificity on x axis) for these definitions ishown in Fig. 1a, noting the most accurate and those ofnterest. The most accurate definitions (product of sensitiv-ty and specificity) are located in the upper left-hand cornerf the graph. A companion table for the noted definitions isrovided in Table 1. As can be seen, nadir � 1, nadir � 2,nd nadir � 3 are good candidate definitions, and thearticular choice depends on a preference of sensitivity vs.pecificity. One of the main functions of the integer addedo the nadir is to overcome the bounce phenomenon. Thereater the integer, the more false-positive failures second-ry to a bounce can be avoided. However, the trade-offecomes the decreasing sensitivity in detecting a true failures seen in Table 1.

With regard to an absolute value definition, it is apparenthat a PSA failure level in the low range, 0.2–1.0 ng/mL, has

xtremely poor specificity, 34–69% predictive of clinical

frPtptcPmlft�rfsdct

atrPtbttnancrvpidt

patient

1489Failure definitions for brachytherapy ● D. A. KUBAN et al.

ailure, because of the PSA bounce phenomenon and theelatively long period of time necessary to achieve such lowSA levels. Of note is the methodology which must be used

o apply this definition to irradiated patients. Unlike surgicalatients, PSA levels decrease over a prolonged period ofime. Therefore, PSA assessment using an absolute levelannot be used immediately after treatment, but instead, theSA level must be monitored for many years before an assess-ent can be made. If the PSA never reaches the designated

evel or if it reaches that level but then rises above it, aailure is declared. Therefore, although this failure defini-ion category appears to be similar in accuracy to the nadir

definitions, it is only at PSA levels in the 2–3 ng/mLange and only with extensive follow-up. Comparative ROCailure definition analysis for external beam radiotherapy ishown in Fig. 1b and Table 1. As can be seen, the nadir �efinitions, in the nadir � 1 to nadir � 3 range, are goodandidate failure definitions for externally irradiated pa-ients as well.

Fig. 1. (a) Receiver operator curve showing sensitivity andoperator curve showing sensitivity and 1 � specificity for

Table 1. Sensitivity and specificity of faibeam

Definition

Brachythera

% Sensitivity %

2 PSA rises 623 PSA rises 40Nadir � 1 ng/mL 77Nadir � 2 ng/mL 70Nadir � 3 ng/mL 58PSA 0.2 ng/mL 97PSA 0.5 ng/mL 96PSA 1.0 ng/mL 92PSA 2.0 ng/mL 86PSA 3.0 ng/mL 81PSA DT 18 mo 73PSA DT 24 mo 79

Abbreviations: PSA � prostate-specific antigen; D

Failure definitions were also compared by regressionnalysis (Tables 2 and 3). The hazard ratio estimates the riskhat a patient with a PSA failure will develop a clinicalecurrence relative to a patient who has not experienced aSA failure. Comparing the likelihood ratio chi-square sta-

istics, nadir � 2 fits the data best for patients treated byrachytherapy as well as those treated by external beamherapy. When using this parameter, the higher the number,he better the definition fits the data. The absolute value 2.0g/mL definition was accurate in predicting clinical failures well, especially for brachytherapy patients; however, asoted previously, follow-up time and the slow PSA declineharacteristic of irradiated patients are problematic. PSAise definitions and doubling time definitions did not pro-ide nearly as good a fit. PSADF outcome for brachytherapyatients was indeed different depending on the failure def-nition applied. For example, the ASTRO and nadir � 2efinitions were quite similar during the first 5 years afterreatment but then showed an absolute difference of approx-

ecificity for patients treated by brachytherapy. (b) Receivers treated by external beam radiotherapy (EBRT).

finitions for brachytherapy and externalerapy

External beam

ficity % Sensitivity % Specificity

72 6151 8182 7172 8364 8899 997 2694 4886 7178 8177 6883 59

1 � sp

lure deradioth

py

Speci

839482899334526982877267

T � doubling time; mo � months.

idav�dFnvnlbd

awe

easidpd(waltaetbftq

igen; D

1490 I. J. Radiation Oncology ● Biology ● Physics Volume 65, Number 5, 2006

mately 8% at 8 years after therapy (Fig. 2). A much largerifference in PSADF survival, 27% at both 5 and 8 yearsfter implant, was seen between the nadir � 2 and absolutealue 0.5 definitions. In comparing outcome with the nadir

1, nadir � 2, and nadir � 3 definitions, the results ofifferences in sensitivity and specificity are illustrated inig. 3. Nadir � 3 has 12% less sensitivity as compared withadir � 2, which results in a slightly higher PSADF sur-ival. Nadir � 1, on the other hand, is 7% less specific thanadir � 2, leading to more false-positive failures and aower PSADF survival rate. Therefore, outcome as assessedy a PSA surrogate will differ depending on the failureefinition applied.

DISCUSSION

With the ongoing changes in radiotherapeutic methodsnd techniques for prostate cancer treatment in conjunctionith higher doses and varied fractionation schedules, it is

xtremely important that a standard methodology to assess

Table 2. Time-dependent covariat

Definition

-2 Log li

Null model With co

2 PSA rises 4248.78 39003 PSA rises 4248.78 3941Nadir � 1 ng/mL 4248.78 3715Nadir � 2 ng/mL 4248.78 3657Nadir � 3 ng/mL 4248.78 3699PSA 0.2 ng/mL 4248.78 3826PSA 0.5 ng/mL 4248.78 3704PSA 1.0 ng/mL 4248.78 3631PSA 2.0 ng/mL 4248.78 3540PSA 3.0 ng/mL 4248.78 3492PSA DT 18 mo 4248.78 3940PSA DT 24 mo 4248.78 3939

Abbreviations: PSA � prostate-specific ant

Table 3. Time-dependent covariat

Definition

-2 Log li

Null model With co

2 PSA rises 19249.9 18083 PSA rises 19249.9 1805Nadir � 1 ng/mL 19249.9 1703Nadir � 2 ng/mL 19249.9 1671Nadir � 3 ng/mL 19429.9 1674PSA 0.2 ng/mL 19249.9 1816PSA 0.5 ng/mL 19249.9 1793PSA 1.0 ng/mL 19249.9 1746PSA 2.0 ng/mL 19249.9 1695PSA 3.0 ng/mL 19249.9 1670PSA DT 18 mo 19249.9 1791PSA DT 24 mo 19249.9 1798

Abbreviations: PSA � prostate-specific antigen; D

fficacy is employed. To compare treatments fairly, a testednd agreed upon PSA failure definition is necessary. Whiletandardizing failure definitions across all treatment modal-ties for prostate cancer would be ideal, the work reported toate and the analysis presented here indicates that this isossible for at least two commonly applied radiation mo-alities, external beam and permanent radioisotopic implant2, 11). Based upon sensitivity and specificity profiles asell as regression analysis, the nadir � 2 failure definition

ppears to be the best fit for patient groups with varyingengths of follow-up. Even in data sets intended for long-erm outcome analysis, patients may die of other causes orre lost to follow-up after various intervals, and certainlyxtensive follow-up is not available in early analysis of newreatment methods. The nadir � 2 failure definition has alsoeen shown to be a reasonable assessment of PSA failureor patients treated by external beam therapy in conjunc-ion with hormonal therapy, as is the case ever more fre-uently (13).For several years now, criticisms of the ASTRO failure

ssion analysis for brachytherapy

d

p value Hazard ratioChi-square

348.38 �.0001 11.62307.42 �.0001 12.70533.53 �.0001 19.43591.41 �.0001 22.13548.94 �.0001 20.99422.28 �.0001 47.20544.20 �.0001 59.19617.33 �.0001 39.39708.68 �.0001 33.70755.84 �.0001 33.73308.30 �.0001 8.93309.13 �.0001 9.72

T � doubling time.

ssion analysis for external beam

d

p value Hazard ratioChi-square

1167.4 �.0001 10.951194.9 �.0001 10.942213.9 �.0001 25.822532.4 �.0001 27.122509.4 �.0001 27.041087.8 �.0001 141.41311.2 �.0001 48.51788.3 �.0001 33.92295.8 �.0001 25.82542.6 �.0001 25.01333.2 �.0001 10.541262.6 �.0001 12.02

e regre

kelihoo

variate

.40

.36

.24

.37

.84

.50

.58

.45

.09

.94

.48

.65

e regre

kelihoo

variate

2.55.06.17.50.52.28.71.64.27.36.77.3

T � doubling time.

datsckb(chpaAfbCp

dirtmtwrrEuiaapfbnoT

rulSs

tIsoircrcnc2faair

oaPipsvcAodP

m

F

Fb

1491Failure definitions for brachytherapy ● D. A. KUBAN et al.

efinition have been made that were supported by detailednalyses (2–4, 14). This was made possible by the matura-ion of large databases and the application of sophisticatedtatistical analyses. In 1996, when the first failure definitiononsensus conference was held, the study of PSA failureinetics was in its infancy, and the ASTRO definition wasased mainly on clinical data and observational experience1). It was, nonetheless, a unifying effort to report outcomeonsistently using a PSA surrogate, and in that purpose itas served us well. However, it has become evident thatroblems with the ASTRO definition exist, and solutionsre now possible. Of additional note is the fact that theSTRO definition, at the time it was designed, was meant

or application to external radiotherapy only. Althoughrachytherapy patients were not the subject of the 1996onsensus, the ASTRO definition has been applied to im-lant patients as well.One of the most significant problems with the ASTRO

efinition is related to backdating. Because prostate cancers unique in that a serum marker to monitor tumor recur-ence is readily available, it seemed like a good idea at theime of the 1996 Consensus Conference to use this infor-ation and backdate the time of failure to halfway between

he nadir PSA and the first rise (1). Afterward, however, itas shown that this causes significant aberration and mis-

epresentation in Kaplan-Meier outcome curves, biasingesults in favor of data sets with shorter follow-up (2, 3).ssentially, the methodological phenomenon is this: a fail-re is declared at the third PSA rise at time (y). The failures then backdated to time (x) and affects the outcome curvet that point in time. However, patients who do not have (y)mount of follow-up time are also represented among theatients at time (x) such that they may not have beenollowed long enough to document three PSA rises. It haseen shown graphically that when using the ASTRO defi-ition, if follow-up time is adjusted for the same data set,utcome will change just for that reason alone (Fig. 4) (2).

ig. 2. Kaplan-Meier comparison of failure definitions forrachytherapy.

his has the potential to significantly affect comparative b

esults when a newer treatment modality, with less follow-p, is compared with one used previously with longer fol-ow-up. Therefore, at the second PSA Failure Definitionymposium, it was agreed that the chosen failure definitionhould not include backdating.

Another issue associated with a PSA failure definition ishe frequency and consistency of the follow-up evaluations.t has been shown that variability in this time period canignificantly affect failure reporting as well (15). It wouldnly stand to reason that if patients have PSA testing morenstead of less frequently, a consecutive number of PSAises would be detected earlier and failure would be de-lared sooner. Requiring PSA testing at more frequent andegular intervals is probably not a practical solution, espe-ially in retrospective studies. The total amount of timeecessary to document three PSA rises and declare a bio-hemical failure has also been criticized. With the nadir �definition, the median number of PSA rises leading to

ailure is 2, shortening the time interval necessary to detectfailure. While the sensitivity of the nadir � 2 definition is

t least as good as for a 2 PSA rise definition, the specificitys significantly higher for both external beam therapy andadioisotopic implant.

Finally, the ASTRO definition and any definition basedn a specified number of PSA rises are prone to subjectivepplication (16). For instance, consider the posttreatmentSA pattern shown in Fig. 5. Failure would occur at point A

f two similar PSA levels were counted as one rise but atoint B if the definition of three consecutive PSA rises wastrictly applied. One can see that with long follow-up inter-als or multiple instances of fluctuating PSA levels, out-ome curves could be significantly affected. In addition, theSTRO definition does not allow for any degree of physi-logic or interassay variation in laboratory testing, and itoes not account for the effects of androgen recovery onSA expression after withdrawal of hormonal therapy.As shown by the data reported here, there are definitionsore sensitive and specific than the ASTRO definition and

ig. 3. Kaplan-Meier comparison of failure definitions for

rachytherapy.

ar�amfitcsb

stdcsn

ivTnanaclttmfsaftcilp

d 1995

Fmt(Tip

1492 I. J. Radiation Oncology ● Biology ● Physics Volume 65, Number 5, 2006

better fit by regression analysis for both external beamadiotherapy and brachytherapy. Based on these analyses,the nadir � definitions appear to be a good choice to unify

nd standardize PSA failure reporting for both radiationodalities. For both modalities the sensitivities and speci-cities fall within the 60% to 90% range, and because these

wo parameters are inversely related, it is a matter of de-iding which will take precedence. As seen in Table 1, theensitivity for the nadir � 2 definition is 72% for externaleam radiotherapy and 70% for brachytherapy whereas the

0

0.2

0.4

0.6

0.8

1

0 2 4

Fre

e of

Bio

chem

ical

Fai

lure

Followu

Fig. 4. The cohort of 4,839 patients with T1–T2 prostatebeam radiotherapy. Shortened median follow-up times (PSA measurements (longest times first). The American Sbackdating was applied to each subset. Reprinted, withComparison of alternative biochemical failure definitiontreated by external beam radiotherapy between 1986 an

ig. 5. An example of successive prostate-specific antigen (PSA)easurements after radiation and the ambiguities in applying

he American Society for Therapeutic Radiology and OncologyASTRO) definition. Each numbered integer represents a PSA rise.hree consecutive PSA rises is the ASTRO definition of failure in

ts strict form. The bold line represents an ideal, unambiguous

dosttreatment PSA pattern.

pecificities for each modality are 83% and 89%, respec-ively. Sensitivity can be increased by going to a nadir � 1efinition, but specificity is decreased. Sensitivity is de-reased with nadir � 3, but specificity is increased. Regres-ion analysis shows that nadir � 2 fits the data best of theadir � definitions.Sensitivities, specificities, and regression parameters sim-

lar to the nadir � definitions can be seen with absolutealue PSA failure definitions at a PSA level of 2 or 3 ng/mL.his would make sense, of course, because in the case of theadir � definitions, nadirs are usually in the 0.5 ng/mL rangend adding 2, for example, would give a total of 2.5 ng/mL. Ofote, however, is that the nadir � definitions can be appliedt any point in time after treatment, even very early in theourse of follow-up, because the nadir is defined as theowest level to date, and failure is defined as any increase ofhe designated amount above this. On the other hand, withhe absolute value definitions, the entire follow-up periodust be used to determine whether the patient’s PSA ever

alls below the failure level, and if it does, whether itubsequently rises above it. Alternatively, an appropriatemount of time to allow for nadir could be imposed, and theailure definition would only be applied after that point inime, or decreasing PSA values could be censored. In eitherase, the disadvantage of the absolute value PSA definitions that it is more useful in retrospective analyses withong-term follow-up than for real-time evaluation or com-arisons of new treatment methods.Although recently there has been great interest in PSA

6 8 10

rs)

6.3 y5 y4 y

3 y

r and median follow-up of 6.3 years treated by externalnd 5 years) were approximated by uniformly droppingfor Therapeutic Radiology and Oncology definition withission, from Thames HD, Kuban DA, Levy LB, et al.d on clinical outcome of 4,839 prostate cancer patients. Int J Radiat Oncol Biol Phys 2003;57:929–943.

p (Yea

cance3, 4, aociety

perms base

oubling time after PSA failure to predict distant metastasis

admaIet

drtab903vatpwpaaid

0lsv9rybyant

btntviaafo

1493Failure definitions for brachytherapy ● D. A. KUBAN et al.

nd death, this did not appear to be the most accurateefinition of PSA failure according to our analysis. Thisay be due to the effect of relatively common PSA fluctu-

tions, laboratory variations, and the bounce phenomenon.n regression analysis, the best doubling time definition forxternal beam radiotherapy was 18 months and for brachy-herapy 24 months (Tables 2 and 3).

It has been proposed that an absolute value PSA failureefinition in the 0.2 ng/mL to 0.5 ng/mL range be applied toadiation modalities, similar to failure definitions for pros-atectomy patients (8–10). However, we have shown thatlthough the sensitivity of this definition for both externaleam radiotherapy and brachytherapy is quite good (97–9%), the specificity is only 9% and 26% for 0.2 ng/mL and.5 ng/mL, respectively for external beam radiotherapy and4% and 52%, respectively, for brachytherapy. So while theast majority of true failures are detected, unfortunately, sore benign PSA fluctuations, such as PSA bounce. In fact,he specificity of these absolute definitions even for surgicalatients is not high in view of the fact that 51% of patientsho fail by a PSA definition of 0.2 ng/mL and 28% ofatients with a postoperative PSA of 0.4 ng/mL do notppear to have a subsequent rise in PSA (9). In fact, ROCnd regression analyses comparing biochemical failure def-nitions for prostatectomy patients have not been reported to

ate. It might be reasonable to use an absolute definition of w

REFEREN

prostate cancer: Undetectable prostate specific antigen compared

1

1

1

1

1

1

1

.5 ng/mL in brachytherapy patients who have been fol-owed for more than 5 years, because 85% of patients in ourtudy who remained biochemically disease-free had PSAalues at least this low at 5 years. Eighty-nine percent and0% had achieved this level by 6 and 7 years posttreatment,espectively. Additionally, 95% of bounces are seen by 5ears postbrachytherapy. However, only 59% of PSADFrachytherapy patients had a PSA level of 0.2 ng/mL at 5ears after treatment and only 72% at 8 years. Therefore, itppears that the degree of prostatic functional ablation mayot be significant enough to consistently reduce PSA secre-ion to this level.

CONCLUSION

Thorough analysis of biochemical failure definitions foroth external beam radiotherapy, with or without hormonalherapy, as well as for brachytherapy for prostatic carci-oma has now been done. The data presented here show thathe nadir � 2 definition is the best option for patients witharious amounts of follow-up treated by brachytherapy,ncluding for studies with less than 5 years maturity. Anbsolute definition in the 0.5 ng/mL range could only bepplied to brachytherapy patients who have been followedor more than 5 years. The nadir � 2 definition is a goodption for external beam radiotherapy as well such that it

ould unify outcome analysis for both of these modalities.

CES

1. ASTRO Consensus Statement. Guidelines for PSA followingradiation therapy. Int J Radiat Oncol Biol Phys 1997;37:1035–1041.

2. Thames HD, Kuban DA, Levy LB, et al. Comparison ofalternative biochemical failure definitions based on clinicaloutcome in 4839 prostate cancer patients treated by externalbeam radiotherapy between 1986 and 1995. Int J Radiat OncolBiol Phys 2003;57:929–943.

3. Vicini FA, Kestin LL, Martinez AA. The importance of ade-quate follow-up in defining treatment success after externalbeam irradiation for prostate cancer. Int J Radiat Oncol BiolPhys 1999;45:553–561.

4. Horwitz EM, Uzzo RG, Hanlon AL, et al. Modifying theAmerican Society for Therapeutic Radiology and OncologyDefinition of biochemical failure to minimize the influence ofbackdating in patients with prostate cancer treated with 3-di-mensional conformal radiation therapy alone. J Urol 2003;169:2153–2159.

5. Nath R, Anderson LL, Luxton G, et al. Dosimetry of interstitialbrachytherapy sources: Recommendations of the AAPM Radia-tion Therapy Committee Task Group No. 43. American Associ-ation of Physicists in Medicine. Med Phys 1995;22:209–234.

6. Williamson JF, Coursey BM, DeWard LA, et al. Recommen-dations of the American Association of Physicists in Medicineon Pd-103 interstitial source calibration and dosimetry: Impli-cations on dose specification and prescription. Med Phys2000;27:634–642.

7. D’Amico AV, Moul JW, Carroll PR, et al. Surrogate end pointfor prostate cancer-specific mortality after radical prostatectomyor radiation therapy. J Natl Cancer Inst 2003;95:1376–1383.

8. Critz FA. A standard definition of disease freedom is needed for

with the American Society of Therapeutic Radiology andOncology Consensus Definition. J Urol 2002;167:1310–1313.

9. Amling CL, Bergstralh EJ, Blute ML, et al. Defining prostatespecific antigen progression after radical prostatectomy: What isthe most appropriate cut point? J Urol 2001;165:1146–1151.

0. Gretzer MB, Trock BJ, Han M, et al. A critical analysis of theinterpretation of biochemical failure in surgically treated pa-tients using the American Society for Therapeutic Radiationand Oncology Criteria. J Urol 2002;168:1419–1422.

1. Kuban DA, Thames HD, Levy LB, et al. Long-term multi-institutional analysis of stage T1–T2 prostate cancer treatedwith radiotherapy in the PSA era. Int J Radiat Oncol Biol Phys2003;57:915–928.

2. Therneau TM, Grambsch PM. Modeling survival data: Ex-tending the Cox Model. New York: Springer-Verlag; 2000.

3. Pickles T, Kin-Sing C, Morris WJ, et al. Evaluation of theHouston biochemical relapse definition in men treated withprolonged neoadjuvant and adjuvant androgen ablation andassessment of follow-up lead-time bias. Int J Radiat OncolBiol Phys 2003;57:11–18.

4. Kestin LL, Vicini FA, Martinez AA. Practical application ofbiochemical failure definitions: What to do and when to do it.Int J Radiat Oncol Biol Phys 2002;53:304–315.

5. Williams SG. Characterization of the behavior of three defi-nitions of prostate-specific antigen-based biochemical failurein relation to detection and follow-up biases: Comparison withthe American Society for Therapeutic Radiology and Oncol-ogy consensus definition. Int J Radiat Oncol Biol Phys 2006;64:849–855.

6. Williams SG. Ambiguities within the ASTRO consensus def-inition of biochemical failure: Never assume all is equal. Int J

Radiat Oncol Biol Phys 2004;58:1083–1092.