PROGNOSTIC INDEX SCORE AND CLINICAL PREDICTION

C

PfboL

w

Y12m

YHDA

Int. J. Radiation Oncology Biol. Phys., Vol. 64, No. 5, pp. 1401–1409, 2006Copyright © 2006 Elsevier Inc.

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

doi:10.1016/j.ijrobp.2005.11.015

LINICAL INVESTIGATION Breast

PROGNOSTIC INDEX SCORE AND CLINICAL PREDICTION MODEL OFLOCAL REGIONAL RECURRENCE AFTER MASTECTOMY IN BREAST

CANCER PATIENTS

SKYE HONGIUN CHENG, M.D.,*†# CHENG-FANG HORNG, M.S.,† JENNIFER L. CLARKE, PH.D.,¶††

MEI-HUA TSOU, M.D.,‡ STELLA Y. TSAI, M.D.,*# CHII-MING CHEN, M.D.,� JAMES J. JIAN, M.D.,*#

MEI-CHIN LIU, M.D.,§ MIKE WEST, PH.D.,¶ ANDREW T. HUANG, M.D.,§**AND LEONARD R. PROSNITZ, M.D.#

Departments of *Radiation Oncology, †Research, ‡Pathology, §Medical Oncology, and �Surgical Oncology, Koo Foundation SunYat-Sen Cancer Center, Taipei, Taiwan; ¶Institute of Statistics and Decision Sciences and Departments of #Radiation Oncology,

**Medicine, and ††Biostatistics and Bioinformatics, Duke University, Durham, North Carolina

Purpose: To develop clinical prediction models for local regional recurrence (LRR) of breast carcinoma aftermastectomy that will be superior to the conventional measures of tumor size and nodal status.Methods and Materials: Clinical information from 1,010 invasive breast cancer patients who had primarymodified radical mastectomy formed the database of the training and testing of clinical prognostic and predictionmodels of LRR. Cox proportional hazards analysis and Bayesian tree analysis were the core methodologies fromwhich these models were built. To generate a prognostic index model, 15 clinical variables were examined fortheir impact on LRR. Patients were stratified by lymph node involvement (<4 vs. >4) and local regional status(recurrent vs. control) and then, within strata, randomly split into training and test data sets of equal size. Toestablish prediction tree models, 255 patients were selected by the criteria of having had LRR (53 patients) or noevidence of LRR without postmastectomy radiotherapy (PMRT) (202 patients).Results: With these models, patients can be divided into low-, intermediate-, and high-risk groups on the basisof axillary nodal status, estrogen receptor status, lymphovascular invasion, and age at diagnosis. In the low-riskgroup, there is no influence of PMRT on either LRR or survival. For intermediate-risk patients, PMRT improvesLR control but not metastases-free or overall survival. For the high-risk patients, however, PMRT improves bothLR control and metastasis-free and overall survival.Conclusion: The prognostic score and predictive index are useful methods to estimate the risk of LRR in breastcancer patients after mastectomy and for estimating the potential benefits of PMRT. These models provideadditional information criteria for selection of patients for PMRT, compared with the traditional selectioncriteria of nodal status and tumor size. © 2006 Elsevier Inc.

Breast cancer, Mastectomy, Radiotherapy, Prediction model, Prognostic score.

otfa(b

CPHYDoRYto

INTRODUCTIONostmastectomy radiotherapy (PMRT) clearly reduces therequency of local regional recurrence (LRR) in high-riskreast cancer patients (1). It also seems to impact favorablyn survival (2). The delineation of patients at high risk forRR is controversial, however. Conventionally, the number

Note—An online CME test for this article can be taken atww.astro.org under Education and Meetings.Reprint requests to: Skye H. Cheng, M.D., Koo Foundation Sun

at-Sen Cancer Center, Department of Radiation Oncology, No.25, Lih-Der Road, Pei-Tou District, Taipei, Taiwan. Tel: (�886)-2897–0011, ext. 310; Fax: (�886) 2-2897–2233; E-mail: [email protected] in part by research funds from Koo Foundation Sun

at-Sen Cancer Center and in part by a grant from the Nationalealth Research Institutes of Taiwan (Contract Project 1997, No.D01-86IX-CR601S).
cknowledgments—The authors thank the members of the Breast A
1401

f involved axillary lymph nodes and the size of the primaryumor are considered, and PMRT is generally recommendedor those with 4 or more involved axillary lymph nodesnd/or those with large primary tumors (T3 or greater)3–5). However, three randomized trials in which a survivalenefit for PMRT was demonstrated included primarily

ancer Team at Koo Foundation Sun Yat-Sen Cancer Center: Drs.o-Sheng Yang and Ben-Long Yu (Department of Surgery), Drs..H. Lin and M.Y. Lee (Department of Pathology), Drs. Kwan-ee Chan and Christopher K.J. Lin (Department of Radiology),rs. Tran-Der Tan, Cheng-I Hsieh, and Nei-Min Chu (Departmentf Medical Oncology), and Dr. Yu-Ling Chung (Department ofadiation Oncology) for patient care; and Yen-Chun Lin, Yueh-un Yu, Yi-Wen Chang, and An-Chen Feng in the Clinical Pro-

ocol Office for data collection, data entry, data quality control, andutcome analysis.Received April 2, 2005, and in revised form Nov 15, 2005.

ccepted for publication Nov 15, 2005.

http://www.astro.org

pbLw

tehmh

prdttBap

T

wdYp

Pocwfa

danciiwff

w

A

M

F

H

P

N

N

E

E

L

N

S

A

A

(Continued)

A

1402 I. J. Radiation Oncology ● Biology ● Physics Volume 64, Number 5, 2006

atients with 1–3 positive lymph nodes (6–8). Thus, aetter definition of what constitutes a patient at high risk forRR and who would be expected to benefit from PMRTould be valuable.Several other prognostic factors, such as estrogen recep-

or (ER) status, age, lymphovascular invasion (LVI), andxtracapsular extension of tumor in axillary lymph nodesave previously been identified as predictive for LRR afterastectomy (9–11). The interaction between these factors,

owever, is largely unknown.The aim of this study was to develop more sophisticated

rediction models for LRR after mastectomy, by usingeadily available clinical data in a fashion analogous to theevelopment of the International Prognostic Index for pa-ients with non-Hodgkin’s lymphomas (12). To accomplishhis, both traditional Cox proportional hazards models andayesian classification trees were used to estimate the prob-bility of LRR after mastectomy for individual breast canceratients (13–15).

METHODS AND MATERIALS

reatment policiesBetween April 1999 and December 2001, 1,143 patients under-

ent modified radical mastectomy as initial treatment for newlyiagnosed invasive breast cancer at the Koo Foundation, Sunat-Sen Cancer Center in Taipei, Taiwan. Adjuvant treatmentolicies were as follows.Postmastectomy radiotherapy. Before 1997, patients received

MRT if they had involvement of 4 or more axillary lymph nodesr a primary tumor �5 cm in size or a resection margin positive orlose. After 1997, patients with 1–3 axillary lymph nodes involvedere also candidates for PMRT if combined with another risk

actor, specifically a primary tumor �3 cm, ER-negative status,ge �40 years, or the presence of LVI.

The technique for PMRT included radiation fields specificallyirected to the ipsilateral chest wall and internal mammary chainnd supraclavicular lymph nodes with CT-based treatment plan-ing. The heart was largely excluded from the radiation fields. Theentral lung distance of the tangents fields was limited to a max-mum of 3 cm. Internal mammary chain nodes were either includedn wide tangent fields if the included lung was acceptable or treatedith a separate photon/electron field. The full axilla was excluded

rom the radiation fields. The dose of radiation was 45–50 Gy in 25ractions (16).

Adjuvant systemic therapy. Before 2000, node-positive patients

Table 1. Clinical characteristics of patients in the training andtest data sets (Continued)

CharacteristicTraining Test

P value*(n � 506) (n � 504)

djuvant chemotherapyYes 369 362 0.70No 137 142

* Chi-square test.† Favorable histology: medullary, tubular, or mucinous carcinoma.

Table 1. Clinical characteristics of patients in the training andtest data sets

CharacteristicTraining Test

P value*(n � 506) (n � 504)

ge (y)�35 44 44 0.8836–40 72 7441–50 194 181�50 196 205enstruation statusPremenopausal 312 299 0.45Postmenopausal 194 205

amily historyNo 451 453 0.871st or 2nd degree 52 49Others 3 2

istologyFavorable† 26 24 0.36Infiltrating ductal 430 416Other invasive 50 64

athological tumor size(cm)

�2.0 230 227 0.89�2.0 276 277

o. of axillary nodesdissected

�10 24 15 0.3210–19 226 21120–29 192 209�30 64 69

o. of axillary nodespositive

0 248 250 0.761–3 144 1424–9 70 61�9 44 51

xtracapsular extensionNegative 378 366 0.58Positive 121 133Unknown 7 5

strogen receptor statusNegative 137 154 0.29� 102 111�� 137 110�� 130 129

ymphovascular invasionAbsent 266 268 0.35Focal 85 100Prominent 155 139

uclear grade1 93 77 0.372 158 1683 249 259

urgical marginsNegative 493 483 0.43Close (�2 mm) 7 14Positive 1 2Unknown 5 5

djuvant radiotherapyYes 137 123 0.33No 369 381

djuvant hormonal therapyYes 363 362 0.98No 143 142

ere treated with one of four chemotherapy regimens—AC (doxoru-

b5csrlpplwp

P

actfrmmsas

S

pestwA�pstspruwawt(ra

sacLpc

paegfi

oecwwfiap

ohelto

pewttFm

cacAAd

wr1Tc

t

A

E

L

L

A

1403Prediction of LRR after mastectomy ● S. H. CHENG et al.

icin, cyclophosphamide), CMF (cyclophosphamide, methotrexate,-fluorouracil), CAF (cyclophosphamide, doxorubicin, 5-fluoroura-il), or CEF (cyclophosphamide, epirubicin, 5-fluorouracil)—as wereelected node-negative patients (those who were ER/progesteroneeceptor–negative or premenopausal patients with a �2-cm primaryesion). Patients with 4 or more positive lymph nodes during thiseriod received A followed by CMF (17). After 2000, node-ositive patients with ER negative were treated with AC fol-owed by paclitaxel; node-positive patients with ER positiveere treated with A followed by CMF. Patients with 10 or moreositive nodes received dose-dense ACT.

rognostic factorsFifteen tumor-, patient-, and treatment-related factors were ex-

mined for their impact on LRR (Table 1). We included allharacteristics that have been previously reported in the literatureo influence LRR, regardless of the strength of the evidence. Theseactors might be characterized as patient related (e.g., age), tumorelated (e.g., ER status), or treatment related (e.g., adjuvant che-otherapy). Of the 1,143 patients who underwent modified radicalastectomy during the study, 1,010 patients who have data on all

ix clinical risk factors (age at diagnosis, primary tumor size,xillary lymph node status, nuclear grade, LVI, hormonal receptortatus) constitute the study group.

tatistical analysisTwo statistical approaches were used in this study; these ap-

roaches complement each other by using different types of math-matical models (linear and nonlinear) to characterize the relation-hip between LR control and the prognostic factors under study. Inhe first approach, Cox proportional hazards regression modelsere used to assess the prognostic significance of risk factors (14).ll patients were stratified by lymph node involvement (�4 vs.4) and LR status (recurrence vs. control), and then, within strata,

atients were randomly assigned to one of two subsets of equalize for model training and validation. Patient characteristics in theraining and test data sets were well balanced (Table 1). Thetep-down regression analysis of LR control in the training sam-les evaluated all 15 prognostic variables mentioned above. Du-ation of LR control was calculated from the first day of treatmentntil the day when chest wall or regional lymph node recurrenceas observed. Local regional control estimates were calculated

ccording to the methods of Kaplan and Meier (18). Log–rank testsere used to assess the statistical significance of the difference in

he probabilities of LR control between specific patient subsets19). The model results were used to define levels of risk of LRecurrence (see below); the predictive ability of these levels wasssessed by application to the validation set.

The Kaplan-Meier estimates of LR control in specific patientubsets were compared with results from the second statisticalpproach, based on prediction tree models. These tree modelslassify patients into subgroups with higher or lower probability ofR control, according to selected risk factors (e.g., lymph nodeositive �3 vs. �3, age �40 vs. �40 years). Estimates of LRontrol were calculated for each subgroup.

The details of tree generation and Bayesian analysis have beenublished previously (13, 20, 21). In brief, Bayesian methods ofnalysis are used to fit candidate tree models to the training data;ach model uses different risk factors to assign patients to sub-roups, which vary in their probability of LR control. The best-
tting trees are selected, and from each of these trees an estimate
f LR control for each patient is calculated. This provides severalstimates of LR control for each patient; the final estimate of LRontrol for a patient is a weighted average of these estimates,hereby the estimates from better-fitting trees are given largereights. The risk factors used to subgroup patients in the best-tting models provide insights into which clinical predictor vari-bles and interactions of clinical predictor variables impact therobability of LR control.We selected 255 of the 1,010 patients for training and construction

f the prediction tree models. The selection criteria were patients whoad any evidence of LR recurrence within 5 years (n � 53) and novidence of LR recurrence without PMRT for a minimum fol-ow-up of 5 years (n � 202). The predictive ability of the modelo assess probability of LR control is assessed by cross-validationn the training set.

RESULTS

Table 1 shows the clinical characteristics and treatmentarameters for the 1,010 patients in the study group, dividedqually into training and test subsets. These subsets wereell balanced with regard to these risk factors. In general,

he patients were young, with 23% aged �40 years at theime of diagnosis. Despite this, 71% were ER positive.orty-nine percent were node negative, and 45% had pri-ary tumors 2 cm in size or less.Regarding treatment characteristics, 72% of patients re-

eived adjuvant chemotherapy, 72% adjuvant hormonal ther-py, and 26% adjuvant radiotherapy. Chemotherapy regimensonsisted of CMF in 19.7%, CAF in 38.3%, CEF in 1.6%,C in 12%, A-CMF in 14.5%, ACT in 2.6%, dose-denseCT in 4.7%, and miscellaneous in 6.6%. The radiationose ranged from 45 to 66 Gy (median, 50 Gy).With a median follow-up of 48 months, 869 patients (86%)

ere without evidence of disease, whereas 141 (14%) hadelapsed. The initial sites of relapse were local regional in.6% of patients, distant in 8%, and both combined in 4%.he 5-year probability of any LRR, either isolated or inombination with distant metastases, was 7.2%.

Multivariate analysis of both the training data set and theest set revealed four clinical characteristics significantly

Table 2. Cox proportional hazards analysis of risk of 5-yearlocal regional recurrence in the training data set

Risk factor Hazard ratio 95% CI

ge (y)�40 vs. �40 2.5 1.1–5.6

strogen receptor statusNegative vs. positive 2.6 1.1–5.7

ymph node positive1–3/4–9/�9 vs. 0 3.1/16.1/45.8 1.1–8.7/3.4–76/

11–190ymphovascular invasionProminent vs. absent/focal 3.3 1.4–7.7

djuvant radiotherapyYes vs. no 0.05 0.01–0.15

Abbreviation: CI � confidence interval.
n � 506.

asacan(

Tmrcph

onisas

oicrgLa

pwcd1Acd

tn


ssociated with LRR. These were axillary nodal status, ERtatus, LVI, and age at diagnosis. In addition, the use ofdjuvant radiotherapy was significantly associated with LRontrol. The best grouping of these clinical risk factors wasge (�40 vs. �40 years), ER (negative or positive), axillaryodal involvement (0 vs. 1–3 vs. 4–9 vs. �9), and LVIabsent/focal vs. prominent).

Hazard ratios for these risk factors are shown in Table 2.he four pretreatment characteristics were used to design aodel to predict an individual patient’s risk of LRR. The

elative risks associated with each of the risk factors wereomparable (hazard ratios of 2.5–3.3) with the exception ofatients with 4–9 positive nodes or �9 positive nodes, withazard ratios of 16.1 and 45.8, respectively.A prognostic index score was then developed for the risk

f LRR. The index score was defined as the sum of theumber of risk factors present with each risk factor receiv-ng a value of 1, except for lymph node status, which wascored as 1 for 1–3 nodes positive, 2 for 4–9 nodes positive,nd 3 for �9 nodes positive. Thus, patients could receive acore from 0 to 6.

Fig. 1. Probability curves for (a) local regional (LR) conton the prognostic index score, which divided patients in
interval.
Risk groups were defined by comparing the relative riskf LRR in patients with different index scores and combin-ng categories with similar relative risks. Thus, patientsould be grouped as low risk (score 0 or 1), intermediateisk (score 2 or 3), or high risk (score 4–6). These threeroups showed distinctive differences, not only in the risk ofRR but also in terms of metastasis-free survival and over-ll survival (Figs. 1a–c, Table 3).

The low-risk group (score 0 or 1) comprised 547atients, 522 of whom did not receive PMRT and 25 ofhom did. There were no significant differences in out-

ome whether PMRT was given or not. In the interme-iate-risk group (score 2 or 3) there were 328 patients,26 of whom received PMRT, 202 of whom did not.lthough LRR was significantly reduced in patients re-

eiving PMRT (p � 0.0003), there were no significantifferences in metastasis-free survival or overall survival.In the high-risk group (score 4–6) there were 135 pa-

ients, 109 of whom received PMRT, 26 of whom didot. Postmastectomy radiotherapy significantly improved

metastasis-free survival, and (c) overall survival, based-, intermediate-, and high-risk groups. CI � confidence

rol, (b)to low

L(

B

rfuwppmddm3Tafa

tlst5w

Ffiniasogi

C

T

radiotherapy.


R control, metastasis-free survival, and overall survivalp � 0.0001, 0.001, and 0.0002, respectively.)

ayesian tree analysisTo provide an initial indication of robustness and accu-

acy, we evaluated the predictive probability of LR controlor individual patients of the 255-patient training data set,sing leave-one-out cross-validation; the tree model processas recomputed repeatedly, each time leaving out one sam-le and then predicting it on the basis of the rest. Therediction tree methodology generated six best-fitting treeodels (Fig. 2). These models identified 5 of the 15 candi-

ate risk factors as important (lymph node status, age atiagnosis, ER status, LVI, and primary tumor size). Threeajor tree models (Trees 2, 43, and 13, weighting 42%,

3%, and 16%, respectively) are presented in Figs. 3A–C.he estimates of 5-year LR control from these trees wereveraged to produce overall estimates of 5-year LR controlor each patient (the tree numbers are automatically gener-ted by MATLAB software; MathWorks, Natick, MA).

As shown in Fig. 3, axillary lymph node involvement ishe first risk factor selected to split patients into high- andow-risk subgroups, followed by age at diagnosis and ERtatus. Patients with a prior condition of lymph node nega-ive and age �35 years in Tree 13 are estimated to have a-year LR control probability of 79%. In contrast, patients

Table 3. Five-year local

GroupNpa

ox modelLow-risk patients (score 0–1)

No PMRTPMRTTotal

Intermediate-risk patients (score 2–3)No PMRTPMRTTotal

High-risk patients (score 4–6)No PMRTPMRTTotal

ree modelLow-risk patients (predictive index �0.94)

No PMRTPMRTTotal

Intermediate-risk patients (predictive index 0.71–0.94)No PMRTPMRTTotal

High-risk patients (predictive index �0.71)No PMRTPMRTTotal

Abbreviations: LR � local regional; PMRT � postmastectomy

ith 1–3 positive lymph nodes, age �38 years, and ER f

ig. 2. Summaries of predictor variables and the clinical tree model:ve clinical variables (age at diagnosis, primary tumor size, lymphode status [lnpos], estrogen status [erlevel], and lymphovascularnvasion [lvi]) (columns) that appear in the selected top trees (rows),nd the levels (boxed numbers) of the trees in which they define nodeplits. The probability of each tree and the overall probability ofccurrence of each of the clinical factors across the set of trees are alsoiven. Weighted average of estimates from individual trees are shownn parentheses in the column headings; utility estimates of clinical risk

regional control and survival

o. oftients

5-y LR controlprobability (%)

5-y metastasis-freeprobability (%)

5-y overallsurvival (%)

522 98 92 9525 100 84 84

547 98 92 95

202 85 81 82126 99 86 87328 90 83 84

26 36 33 21109 85 56 60135 77 53 54

518 98 92 9433 100 82 89

551 98 91 94

173 89 83 8780 100 86 90

253 93 86 88

59 32 41 40147 88 65 65206 77 60 59

actors in tree models are shown in parentheses in the row headings.


Fig. 3. Three most common trees: the boxes at nodes of a tree indicate the number of patients, whereas the number next tothe box represents the model-based estimate of 5-year LR control survival probability. (a) Tree 2 is the most important treeof our tree models, weighted 42% (see Fig. 2); the first node is number of positive axillary lymph nodes, with a cutoff valueof 3; the second nodes are age 38 years and estrogen receptor moderately positive; the third level of split is axillary lymphnode positive, followed by estrogen receptor moderately positive, and so on. (b) Tree 43 is the second most important tree,weighted 33%; similar to Tree 2, patients initially are split by number of positive axillary lymph nodes positive, with a cutoffvalue of 3; age 38 years and estrogen receptor strongly positive are the second nodes; lymph node negative and estrogenreceptor moderately positive are the third nodes, and so on. (c) Tree 13 is the third most important decision tree, consistingof 16% of our prediction tree models, which means of lower weight in the prediction of recurrence. Patients with a priorcondition of lymph node negative and age �35 years would have a 5-year probability of LR control of 79%. Abbreviations:
ER status: negative (�); node weakly positive (�); node densely positive (��); node strongly positive (��).

mm

dpumric1u

Cp

p�pnsPwpm

gta

cptst

nmPahtmct

tbhTs

imn

(Conti


oderately positive in Tree 43 have excellent 5-year esti-ates of LR control, at 95%.Patients with good LR control generally had a high pre-

ictive index, and patients with LR recurrence had a lowredictive index. Both prediction patterns had small predictionncertainty. As compared with Cox proportional hazardsodels, the tree models focuses on predicting individual

isk of LR recurrence. The computer program to estimatendividual risk is available at http://www.kfsyscc.org/unit/po/cpo.php. We applied the prediction tree models to all,010 patients to obtain a predictive index for each individ-al patient.

omparing Bayesian classification tree models with Coxroportional hazards modelA grouping of the 1,010 patients based on their individual

redictive index is shown in Table 3. A predictive index of0.71 (high-risk group) is probably the best cutoff value for

atients to undergo PMRT; PMRT in this group improvedot only LR control but also metastasis-free and overallurvival. In contrast, patients did not obtain benefit fromMRT if their predictive indices were �0.94. For patientsith predictive indices between 0.71 and 0.94, PMRT im-roved LR control (from 89.4% to 100%, p � 0.005) butade no difference in metastasis-free or overall survival.Multivariate analysis of treatment effects in the high-risk

roups defined by either prognostic index score or predic-ion tree models revealed that PMRT, adjuvant chemother-

Fig. 3.

py, and hormonal therapy each significantly improved LR n

ontrol. Postmastectomy radiotherapy also significantly im-roved metastasis-free survival and overall survival, whereashe improvement in metastasis-free survival and overallurvival due to either adjuvant chemotherapy or hormonalherapy is evident but not consistently significant (Table 4).

DISCUSSION

The role of PMRT in the management of breast carci-oma has been and is controversial since its introductionore than 5 decades ago. It was soon established thatMRT significantly decreased LRR after surgery. Innumer-ble phase II studies and more than 30 randomized trialsave shown a consistent relative risk reduction in LRR (ofwo thirds to three quarters), as summarized in three recenteta-analyses (22–24). The absolute benefit depends, of

ourse, on the relative risk of an LRR, being greatest inhose patients at highest risk for LRR.

Survival benefits obtained from PMRT are the center ofhe controversy. Many early trials failed to show a survivalenefit. The complexities of these trials and the pitfalls thereinave been extensively reviewed by Recht and Edge (25).hree major recent trials, however, have convincingly shownurvival benefits, as has a recent meta-analysis (6–8, 22).

Patients have generally been selected for PMRT (andnclusion in these trials) on the basis of nodal involve-ent, traditionally grouped as 1–3 nodes involved or �4

odes positive. Increasing number of involved lymph

nued)

odes generally correlates with both the risk of systemic

http://www.kfsyscc.org/unit/cpo/cpo.php.

http://www.kfsyscc.org/unit/cpo/cpo.php.

sainsCfeivwpio

rmmiiv

oAptAt

nPdCaipars(i

iy1Tpo

hew

ftm1astewcfitPtbp

piippp5ddpa


pread as well as that of LRR. The three trials mentionedbove, though comprising mostly patients with 1–3 pos-tive nodes, either showed a greater benefit for the �4ode-positive group or were insufficiently powered forubset analysis (26). An expert American Society oflinical Oncology (ASCO) panel developing guidelines

or PMRT concluded that although there was strongvidence for increased local control in both nodal group-ngs, the evidence for improvement in disease-free sur-ival and overall survival was more consistent for thoseith �4 nodes involved, compared with the 1–3 node-ositive group. The panel concluded, “the evidence wasnsufficient to make recommendations for the routine usef PMRT in patients with 1–3 positive nodes” (5).The panel also addressed the influence of other tumor-

elated characteristics (e.g., grade, ER status, LVI, pri-ary size) as well as patient-related factors (e.g., age,enopausal status) and concluded that evidence was

nsufficient to know how these variables should factornto the decision of whether to use PMRT. Further in-estigations in these areas were strongly recommended.Unfortunately, evidence from phase III trials on some

f these points will not be soon forthcoming. A Northmerican Breast Intergroup study looking specifically atatients with 1–3 positive lymph nodes and randomizinghem to PMRT or not has closed owing to poor accrual.

European study is planned but is years from comple-ion (2).

Given this background, we set out to develop a prog-ostic index to guide decisions regarding the use ofMRT that might be more informative than a simpleivision into 1–3 or �4 lymph nodes involved. Usingox proportional hazards analysis, we identified patientge, ER status, and LVI, in addition to nodal status, asmportant prognostic variables and were able to groupatients into three distinct groups with low, intermediate,nd high risks of LRR. Although PMRT increased localegional control in all groups, statistically significanturvival benefits were only seen in the high-risk groupTable 4). The great majority of patients with �4 nodes

Table 4. Multivariate analysis for treatment eoverall survival

Treatment factors LR

By prognostic index model (n � 135)Adjuvant radiotherapy 0.1Adjuvant chemotherapy 0.1Adjuvant hormonal therapy 0.3

By prediction tree models (n � 206)Adjuvant radiotherapy 0.1Adjuvant chemotherapy 0.0Adjuvant hormonal therapy 0.2

Abbreviation as in Table 2.

nvolved fall into this high-risk group, but so do signif- s

cant numbers of patients with 1–3 positive nodes (e.g., aoung woman with ER-negative disease with LVI and–3 nodes involved would fall into the high-risk group).he model does suggest that PMRT is not necessary foratients with 1–3 lymph nodes involved in the absence ofther risk factors.The predictive tree model gave similar results. Only the

ighest-risk group, with a predictive index of �0.71, wouldxperience survival benefits from PMRT, although all groupsould have decreased LRR.These results are consistent with recent calculations per-

ormed by Olivotto et al. (27), suggesting a ratio betweenhe number of LRRs prevented and survival of approxi-ately 4 to 5 to 1. Thus, an absolute reduction in LRR from

5%, for example, to 3% would be expected to lead to anbsolute survival benefit of 2% to 3%. A demonstration oftatistically significant survival improvement would requirehousands of patients in trials, similar to what has beenntered in trials of adjuvant systemic therapy. In patientsith a much higher risk for LRR, the absolute benefit is

orrespondingly greater, and larger absolute survival bene-ts are expected as well. The ASCO expert panel concluded

hat, similar to systemic therapy, it was highly likely thatMRT achieved proportional reductions in the risk of LRR

hat were the same for all treated groups, the absoluteenefit depending on the relative risk of the event, asreviously stated.In summary, we have developed a prognostic score and

redictive index for LRR based on easily obtainable clinicalnformation that seems to provide more information regard-ng the risk of LRR than a simple division of patients byrimary size and the number of lymph nodes involved. Thisrognostic score and predictive index might lead to im-roved guidelines for selection of patients for PMRT (Table). Confirmation of these results by other groups is highlyesirable. As with adjuvant systemic therapy, the ultimateecision as to whether to undergo PMRT rests with theatient and relates to what level of risk she finds acceptablend how much treatment she is willing to undertake for a

on local regional recurrence/metastasis-free/h-risk patients

Hazard ratio (95% CI)

enceMetastasis-free

survival Overall survival

–0.4) 0.4 (0.2–0.9) 0.36 (0.2–0.8)–0.3) 0.43 (0.2–1.1) 0.57 (0.2–1.3)–0.9) 0.85 (0.5–1.6) 0.59 (0.3–1.1)

–0.2) 0.47 (0.3–0.8) 0.58 (0.3–1.0)–0.2) 0.25 (0.1–0.5) 0.37 (0.2–0.7)0.5) 0.68 (0.4–1.2) 0.49 (0.3–0.8)

ffectsin hig

recurr

6 (0.061 (0.044 (0.14

0 (0.057 (0.034 (0.1–

mall survival benefit.

1

1

�2�


Table 5. Impact of postmastectomy radiotherapy on patient outcomes, according to prognostic score and predictive index

Prognostic scoreor predictive

index

Improvementof LRcontrol

Improvementof

metastasis-freesurvival

Improvementof overallsurvival

2/�0.94 No No No–3/0.71–0.94 Yes No No3/�0.71 Yes Yes Yes

Abbreviation as in Table 3.

REFERENCES

1

1

1

1

1

1

1

1

2

2

2

2

2

2

2

2

1. Griem KL, Henderson IC, Gelman R, et al. The 5-year resultsof a randomized trial of adjuvant radiation therapy after che-motherapy in breast cancer patients treated with mastectomy.J Clin Oncol 1987;5:1546–1555.

2. Whelan T, Levine M. More evidence that locoregional radia-tion therapy improves survival: What should we do? J NatlCancer Inst 2005;97:82–84.

3. Overgaard M, Christensen JJ, Johansen H, et al. Evaluation ofradiotherapy in high-risk breast cancer patients: Report fromthe Danish Breast Cancer Cooperative Group (DBCG 82)Trial. Int J Radiat Oncol Biol Phys 1990;19:1121–1124.

4. Fowble B, Gray R, Gilchrist K, et al. Identification of asubgroup of patients with breast cancer and histologicallypositive axillary nodes receiving adjuvant chemotherapy whomay benefit from postoperative radiotherapy. J Clin Oncol1988;6:1107–1117.

5. Recht A, Edge SB, Solin LJ, et al. Postmastectomy radiother-apy: Clinical practice guidelines of the American Society ofClinical Oncology. J Clin Oncol 2001;19:1539–1569.

6. Overgaard M, Hansen PS, Overgaard J, et al. Postoperativeradiotherapy in high-risk premenopausal women with breastcancer who receive adjuvant chemotherapy. Danish BreastCancer Cooperative Group 82b Trial. N Engl J Med 1997;337:949–955.

7. Overgaard M, Jensen MB, Overgaard J, et al. Postoperativeradiotherapy in high-risk postmenopausal breast-cancer pa-tients given adjuvant tamoxifen: Danish Breast Cancer Coop-erative Group DBCG 82c randomised trial. Lancet 1999;353:1641–1648.

8. Ragaz J, Jackson SM, Le N, et al. Adjuvant radiotherapy andchemotherapy in node-positive premenopausal women withbreast cancer. N Engl J Med 1997;337:956–962.

9. Dunst J, Steil B, Furch S, et al. Prognostic significance of localrecurrence in breast cancer after postmastectomy radiother-apy. Strahlenther Onkol 2001;177:504–510.

0. Katz A, Buchholz TA, Thames H, et al. Recursive partitioninganalysis of locoregional recurrence patterns following mastec-tomy: Implications for adjuvant irradiation. Int J Radiat OncolBiol Phys 2001;50:397–403.

1. Sundquist M, Thorstenson S, Klintenberg C, et al. Indicatorsof loco-regional recurrence in breast cancer. The South EastSwedish Breast Cancer Group. Eur J Surg Oncol 2000;26:
357–362.
2. A predictive model for aggressive non-Hodgkin’s lymphoma.The International Non-Hodgkin’s Lymphoma Prognostic Fac-tors Project. N Engl J Med 1993;329:987–994.

3. Pittman J, Huang E, Nevins J, et al. Bayesian analysis ofbinary prediction tree models for retrospectively sampled out-comes. Biostatistics 2004;5:587–601.

4. Cox D. Regression models and life-tables. J R Stat Soc 1972;34:187–220.

5. Huang E, Cheng SH, Dressman H, et al. Gene expressionpredictors of breast cancer outcomes. Lancet 2003;361:1590–1596.

6. Cheng SH, Jian JJ, Chan KY, et al. The benefit and risk ofpostmastectomy radiation therapy in patients with high-riskbreast cancer. Am J Clin Oncol 1998;21:12–17.

7. Cheng SH, Tsou MH, Liu MC, et al. Unique features of breastcancer in Taiwan. Breast Cancer Res Treat 2000;63:213–223.

8. Kaplan EL, Meier P. Nonparametric estimation from incom-plete observation. J Am Stat Assoc 1958;53:457–481.

9. Mantel N, Haenszel W. Chi-square test with one degree offreedom: Extensions of the Mantel-Haenszel procedure. J AmStat Soc 1963;58:690–700.

0. Pittman J, Huang E, Dressman H, et al. Integrated modeling ofclinical and gene expression information for personalized pre-diction of disease outcomes. Proc Natl Acad Sci U S A 2004;101:8431–8436.

1. Kass R, Raftery A. Bayes factors. J Am Stat Assoc 1995;90:773–795.

2. Whelan TJ, Julian J, Wright J, et al. Does locoregional radi-ation therapy improve survival in breast cancer? A meta-analysis. J Clin Oncol 2000;18:1220–1229.

3. Favourable and unfavourable effects on long-term survival ofradiotherapy for early breast cancer: An overview of therandomised trials. Early Breast Cancer Trialists’ CollaborativeGroup. Lancet 2000;355:1757–1770.

4. Cuzick J, Stewart H, Rutqvist L, et al. Cause-specific mortal-ity in long-term survivors of breast cancer who participated intrials of radiotherapy. J Clin Oncol 1994;12:447–453.

5. Recht A, Edge SB. Evidence-based indications for postmas-tectomy irradiation. Surg Clin North Am 2003;83:995–1013.

6. Ragaz J, Olivotto IA, Spinelli JJ, et al. Locoregional radiationtherapy in patients with high-risk breast cancer receivingadjuvant chemotherapy: 20-year results of the British Colum-bia randomized trial. J Natl Cancer Inst 2005;97:116–126.

7. Olivotto IA, Truong PT, Chua B. Postmastectomy radiation
therapy: Who needs it? J Clin Oncol 2004;22:4237–4239.

Documents

PROGNOSTIC INDEX SCORE AND CLINICAL PREDICTION