ARTICLE IN PRESS

The Breast (2005) 14, 94–102

THE

BREAST

KEYWORDBreast canSurvival mCancer regQuality ofElderly patAxillary di

0960-9776/$ - sdoi:10.1016/j.b

�CorrespondiE-mail addr

www.elsevier.com/locate/breast

ORIGINAL ARTICLE

A population survival model for breast cancer

F. Straccia,�, F. La Rosaa, E. Falsettinia, E. Riccib, C. Aristeic, G. Bellezzad,G.B. Bolise, D. Fenocchiof, S. Gorig, A. Rullib, V. Mastrandreaa

aDepartment of Hygiene and Public Health, University of Perugia, Umbria Cancer Registry, Via delGiochetto, 06122 Perugia, ItalybDepartment of Surgery, Division of Oncological Surgery, University of Perugia, ItalycInstitute of Radiotherapy Oncology, University of Perugia, Monteluce Hospital, ItalydInstitute of Pathology, University of Perugia, Monteluce Hospital, ItalyeInstitute of Pathology, University of Perugia, S. Maria Hospital, Terni, ItalyfInstitute of Pathology, University of Perugia, Silvestrini Hospital, ItalygDivision of Medical Oncology, Monteluce Hospital, Perugia, Italy

Received 25 May 2004; received in revised form 1 July 2004; accepted 18 August 2004

Scer;odel;istries;care;ients;ssection

ee front matter & 2004reast.2004.08.011

ng author. Tel.: +39 075ess: fabs@unipg.it (F. S

Summary Breast cancer is a major health problem, and disease control dependson an effective healthcare system. A registry-based tool to monitor the quality ofbreast cancer care could be useful. The aim of this study was to develop a populationsurvival model for breast cancer based on the Nottingham Prognostic Model (NPM).To this end, 1452 cases of breast cancer diagnosed in the Umbria Region, Italy, duringthe period 1994–1996 were studied. An extensive search for routinely availablevariants in prognosis and treatment was performed. In about 80% of cases completeinformation on factors included in the NPM was available. The Cox model was used toassess the prognostic value of study factors. Nodal stage was the most importantprognostic factor. In women who did not undergo axillary dissection (17%) the risk ofdeath was twice that in women with no affected nodes, but they receivedchemotherapy with the same frequency. Radiotherapy was also less frequently usedin this group. Grading was a significant prognostic factor only when women over 80were excluded. Population survival models based on data from cancer registries mayprovide a tool that can be used to evaluate healthcare systems and the effectivenessof interventions. The inclusion of older women in our models decreased thesignificance of many established prognostic factors because of the frequency ofincomplete evaluation and less aggressive treatment in these patients. Notundergoing surgical axillary dissection was associated with a worse prognosis andwith less aggressive treatment.& 2004 Elsevier Ltd. All rights reserved.

Elsevier Ltd. All rights reserved.

585 7366; fax: +39 075 585 7317.tracci).

ARTICLE IN PRESS

A population survival model for breast cancer 95

Introduction

Randomized clinical trials are the best availablemethods of evaluating the efficacy of treatmentsor, more generally, interventions undertaken incancer patients.1 However, the results from clinicaltrials are not automatically transferred to practice,and their validity may vary depending on localconditions.2 Thus, evaluation of the effectivenessof specific interventions in actual practice and ofthe performance of an oncological healthcaresystem is important if the quality of care andoutcome for cancer patients are to be improved.

The population survival rate for cancer patientsis an indicator produced by cancer registries of theoverall quality of oncological care.3 This indicatorhas the advantage of being calculated with stan-dardized methods, taking account of all casesdiagnosed in a population (i.e., not being subjectto selection bias), and being available for manypopulations and periods. The main limitation of theindicator is that few explanatory variables areroutinely accounted for. Thus, further study isnecessary to explain the causes of survival differ-ences when they are observed (e.g., amongdifferent geographic areas).4

In order to disentangle the effects of the manyfactors influencing survival and to identify modifi-able factors, population survival models are beingdeveloped.5 A population survival model requirescollection of information on various prognosticfactors. The choice of prognostic factors dependson the cancer site studied and the availability ofinformation; that is to say that only prognosticfactors routinely determined for most patients andrecorded by standard methods in an accessiblearchive can be used.6 Since population survivalmodels are designed to monitor the quality of careand to select effective interventions and healthorganization models, they are mostly useful forfrequent and ‘healthcare-dependent’ cancers.

Breast cancer is the most frequent cancer amonggirls and women in western countries, and theoutcome for patients depends critically on thequality of treatment given and on a timelydiagnosis.7,8 Thus, it seems that a populationsurvival model for breast cancer would be useful,given the social burden in terms of morbidity andmortality and the central role of the health servicein disease control. A number of prognostic factorshave been proposed for breast cancer, and some(e.g., disease stage, grade, and hormone receptor(HR) status) should be routinely measured.6,9 Oneprognostic model for breast cancer, the NottinghamPrognostic Model (NPM), was developed in a clinicalsetting to produce an individual risk score, the

Nottingham Prognostic Index (NPI), and it has beenvalidated repeatedly.9–12In Italy, the healthcaresystem is broken down into Regional HealthSystems, which have growing autonomy. TheUmbria Region, in Central Italy, was the first Regionto establish a cancer registry covering its entirepopulation (820,000 inhabitants). In Umbria, stan-dardized (world) incidence rates increased from 56in 1978–1982 to 65 per 100,000 in the study period;mortality in 1994–1996 was the same as in theperiod 1978–1982 (about 17 per 100,000). Improvedtreatment is the likely explanation for the observedincidence and mortality trends in the absence of apopulation screening program (which has beenstarted since 1998) and with very limited earlydetection activities. The overall breast cancersurvival rate in Umbria is one of the highest inItaly (relative survival was 0.85 at 5 years for caseswith presentation in the period 1994–1996).13 Theavailable data suggest that the quality of oncolo-gical care is rather good in Umbria, as in otherCentral Italian areas (e.g., Emilia Romagna andTuscany).13–15 Registration of detailed prognosticinformation on all breast cancer cases first seen inUmbria is ongoing (started for cases first seen in1994), with the aim of developing a cancer registry-based tool for the evaluation of quality of care andof the efficacy of interventions (e.g., breast cancerscreening for women aged 50–69 years).

The aim of the present analysis was to build up apopulation survival model for breast cancer based onthe validated NPM. The prognostic significance andimpact of incomplete prognostic evaluation and theimportance of factors that are available but are notincluded in the NPM were also explored. Moreparticularly, the prognostic role of not undergoingaxillary dissection (incomplete prognostic evaluation)and the validity of established prognostic factorsamong women 480 years old were investigated.

Patients and methods

We studied the historical cohort consisting of all(N ¼ 1462) cases of infiltrating female breastcancer (ICD-IX 17416) first diagnosed in the UmbriaRegion of Italy over the period 1994–1996. All caseswere identified from the regional cancer registry,the Registro Tumori Umbro di Popolazione.

After reabstracting the cancer registry files, weperformed an extensive search for prognosticfactors in all the regional pathology archives,hospital case records and, when available, personalarchives maintained by surgeons and oncologists. Inaddition, after the aforementioned investigation,general practitioners were contacted for informa-

ARTICLE IN PRESS

Size not available No axillary dissection

(8 %)

(17 %)

(9 %) Grade not assigned

1153 (79 %)

19 1 %

126 9 %

20 1 %

71 5 %

26 2 %

3 0 %

34 2 %

Figure 1 Venn diagram showing frequency and collinear-ity among missing prognostic values.

F. Stracci et al.96

tion on any patients treated outside the Region andany cases for which data were missing.

In the study we took account of age, histologicaltype, and all factors included in the NPI calculation(tumor size (mm), number of affected axillarynodes, and tumor grade). Data on stage, treatment(type of surgery, radiotherapy, chemotherapy,hormonal therapy), and HR status were alsocollected, as were other variables.

Ten (0.7%) death certificate only (DCO) cases(i.e., cases in which the death certificate was theonly source of information) were excluded becausethe date of diagnosis was unknown. The mean ageof death for DCO cases was 80 years. Mean age atdiagnosis for the 1452 study cases was 62.3 years(median 63). Microscopic verification (MV) waslacking for 40 cases (2.8%). Histological type wascoded as ‘carcinoma not otherwise specified(NOS),’ ‘ductal carcinoma NOS,’ ‘lobular carcino-ma’ and ‘other specified types,’ which includedtubular, medullary, mucoid, cribriform, and othertypes. A diagnosis of ‘carcinoma’ or ‘solid carcino-ma’ was available for 49 (3.4%) cases. Grading wasunavailable for 6.8% of cases with MV, and only 4.6%of cases of ductal carcinoma NOS. The reportedgrade for medullary cancers was not used in theanalysis.9 The largest cancer diameter (mm) wasreported for 1259 cases (86.7%) and was estimatedfor 80 further cases (5.5%) as the median diameterof cancers with the same T stage for which thediameters were known. Information on nodal statuswas lacking in 168 cases (11.6%) and was based onclinical/instrumental information in 67 furthercases (4.6%). Classification as ‘no affected nodes(A),’ ‘1–3 affected nodes (B),’ or ‘4 or moreaffected nodes (C)’ was possible for 82.9% of allstudy cases. The mean/median number of nodesexamined nodes was 16 (S.D. 6.4).

Distribution and collinearity of missing values forthe factors included in the NPI are summarized inFig. 1.

Estrogen and progesterone receptor status wasunknown for 25.9% and 32.7% of cases, respectively.

Death certificates and the Registry Officesresponsible for patients’ place of residence werethe sources of information on patients’ vital statusup to 31 December 2001. Only two cases were lostto follow-up. Eleven deaths (0.8%) occurred within1 month after the recorded date of first diagnosis.Twenty women (1.4%) received their main treat-ment in hospitals outside the region. Although thedistribution of prognostic factors was remarkablysimilar over the 3 years of the study (e.g., caseswithout MV Chi-square P ¼ 0:32; Nottingham classi-fication of nodes Chi-square P ¼ 0:060; cases withdistant metastases Chi-square P ¼ 0:43), a max-

imum follow-up time of 5 years was considered inthe analysis to ensure equal potential follow-uptimes for cases diagnosed in different years. Over-all, 340 events were registered (240, or 71%, ofthese were due to breast cancer).

Relations between study factors were analyzedby analysis of variance or a Chi-square test. Astandard proportional hazards model was fitted toour data.17 The proportionality assumption waschecked by graphical methods and by testing.18 TheHarrell’s c-statistic was calculated as an index ofmodel discrimination ability,19 ‘c’ being based on acomparison of the ordering of the model predictedand observed survival between pairs of observa-tions; values near 0.5 indicate that the model hasno discriminating ability. The number of events pervariable (EPV) included was also reported for eachmodel;20 as a rule of thumb EPV equal to or greaterthan 10 is recommended to avoid overfitting.Categorization of continuous factors was used toanalyze the effects of measured prognostic factorstogether with unmeasured factors. Categoriestaken from NPI-based studies were used for cancerdiameter, axillary nodes, and grade.9,10 An alter-native coding21 of cancer type/grade was adoptedto avoid assigning grades to medullary9 andlobular22 cancers. The likelihood ratio (LR) wascalculated to test the overall significance of studyfactors. Data analysis was performed using theStata 8 statistical software.23

Results

‘Lobular cancers’ accounted for 8% and ‘otherspecified cancer types’ accounted for 9% of caseswith microscopic diagnosis. The majority of ductal

ARTICLE IN PRESS

A population survival model for breast cancer 97

carcinomas were assigned a moderately differen-tiated grading (51%), whereas 29% were reported tobe poorly differentiated (G3). Most carcinomas ofspecified types were reported to be well differ-entiated (67%); grades were lacking for 10% ofspecified types, and grade 3 (8%) was reported formedullary cancers only. The mean (S.D.) andmedian cancer diameters were 21mm (14.8) and19mm, respectively. Mean diameter decreasedfrom 23 to 21mm over the study period (F-testP ¼ 0:04). Cancer size varied significantly withmorphology/grade types (Po0:000): ductal carci-nomas G1 and other specified types were thesmallest (median 16mm) and undifferentiatedductal carcinomas, the largest (median 22mm).Axillary nodes were not involved in 55% of caseswith some information on nodal status and in 57% ofcases treated by surgical dissection of the axilla.The frequency of nodal involvement varied sig-nificantly with histological type and grading(Po0:000): more than half of the scarcely differ-entiated ductal carcinomas (29% stage B and 28%stage C) and about 45% of the ductal cancers G2and lobular cancers had axillary nodal metastasesat the time of diagnosis. On the other hand, only21% of other specified cancer types showed nodalinvolvement, and under 7% each had more thanthree affected nodes. Among cases of cancer withmicroscopic diagnosis, the frequency of nonperfor-mance of axillary dissection was higher for otherspecified cancer types (12%) than for ductal cancers(about 10%), though the difference did not reachstatistical significance (P ¼ 0:80). Distant metas-tases were present at the time of diagnosis in 126cases (8.7%); bone was the most frequent site ofmetastasis. G1 ductal cancers (2%) and otherspecified types (1%) had low percentages of distantspread at diagnosis, and G3 cancers had the highestpercentage (7%).

At least one type of HR was positive in 74% ofcases in which HR status was determined. When HRstatus was considered with reference to cancertype, receptor-positive status was found to be mostfrequent among ductal carcinomas in stage G1, andleast frequent among undifferentiated ductal car-cinomas (58%) (Po0:000). When the pattern ofundetermined HR was also taken into account, G3ductal cancers and other specified types showedthe lowest (14%) and the highest (29%) percen-tages, respectively, of undetermined HR status.

Our first proportional hazards model (Table 1,model A) included the same variables as the NPM.However, cases with distant spread were includedand the presence of metastases was accounted forin the model. Moreover, indicator variables forunmeasured factors were included in it.

Distant metastases at diagnosis and having morethan four metastatic nodes were associated withthe highest risks of death. Incomplete prognosticinformation was also associated with hazard ratiosexceeding unity. The morphology and gradingfactor was not significant (LR test P ¼ 0:08). Asecond model restricted to women under 80 years isalso shown in Table 1 (model B). In the age-restricted model, cancer type and grading carriedsignificant prognostic information (Fig. 2). Axillarynodal status, presence of metastases, and unspe-cified cancer types all had larger coefficients afterthe exclusion of very old women, whereas size ofcancer was unaffected.

The group in which cancer diameter was lackingshowed nonproportional hazards: the risk of deathin this group increased rapidly during the first yearof follow-up and then decreased. Accounting forchanging risk level by a time-dependent covariateleft the other coefficients in the model unchanged.

The third proportional hazards model (Table 2)included routinely available prognostic factors inaddition to those included in the NPM, and alsointeractions between the oldest age classes andscarcely differentiated ductal cancers. The non-significant interaction between age of 70–79 yearsand G3 ductal carcinoma was retained in the modelto show the progressive loss of prognostic value ofgrade with age.

Overall, HR status did not prove to have asignificant independent prognostic role. The hazardratios for both lack of determination and negativeHR status were more than unity. The risk of deathdid not decrease significantly with increasingnumber of axillary nodes removed.

Not undergoing surgery explained the excessdeath risk associated with unspecified cancertype/morphology. The excess risk associated withunmeasured diameter was apparently not affectedby adjustment for surgical treatment; this isbecause the prognosis was worse in cases in whichno surgery was performed and the size of thecancer was also not recorded. Inclusion of aninteraction term with no surgery caused the excessrisk associated with unmeasured size to disappear.

Not undergoing axillary dissection was stillassociated with a hazard ratio significantly higherthan unity after adjustment for not undergoingsurgical treatment. A preliminary look at treatmentvariables showed that women in whom no axillarydissection was performed had a probability ofreceiving nonhormonal chemotherapy that wasclose to that in women with no affected nodes(stage A) (Table 3, Fig. 3). This was true even whenthe analysis was restricted to women under 80without distant metastases who were treated

ARTICLE IN PRESS

Table 1 Proportional hazards models including the same variables as in the Nottingham model and presence ofmetastases.

Factor Model A Model B

No. of cases HR 95% CI No. of cases HR 95% CI

Type/grade P ns* P=0.003*

No MV 40 2.5 1.3–4.9 25 4.3 1.7–10.6Carcinoma NS 49 1.8 0.9–3.3 42 3.1 1.3–7.2Ductal NS 51 1.3 0.7–2.5 40 2.2 0.9–5.5Ductal G1 175 Ref. 157 Ref.Ductal G2 585 1.3 0.8–2.1 526 2.1 1.04–4.1Ductal G3 335 1.7 1.02–2.7 312 3.1 1.6–6.3Lobular 115 1.6 0.9–2.9 103 2.6 1.2–5.9Other specified 102 1.0 0.5–2.1 92 1.5 0.6–4.0

Tumor size Po0.0000* P=0.0002*

Not available 113 2.1 1.3–3.4 87 2.1 1.2–3.81–15 516 Ref. 491 Ref.16–30 643 1.9 1.3–2.6 565 1.8 1.2–2.6430 180 2.8 1.9–4.1 154 2.5 1.6–4.0

Axillary nodes Po0.0000* Po0.0000*

No dissection 249 4.7 3.2–6.9 163 4.2 2.6–6.8A 684 Ref. 642 Ref.B 294 2.2 1.5–3.3 278 2.7 1.7–4.0C 225 4.0 2.8–5.7 214 4.6 3.1–7.0

Metastases Po0.0000* Po0.0000*

No 1326 Ref. 1185 Ref.Yes 126 3.1 2.3–4.1 112 4.1 3.0–5.7

Model A: Harrell’s c statistic 79%, EPV 24; Model B: Harrell’s c statistic 81%, EPV 18. Model B was fitted with women over 80excluded.

*LR test.

F. Stracci et al.98

within our Region: 24% of women in whom noaxillary dissection was performed (82 cases) and31% of women with no spread to axillary nodesreceived chemotherapy (Chi-square, P ¼ 0:009).Among patients who did undergo axillary dissec-tion, the use of chemotherapy increased with nodalinvolvement category (trend Chi-square Po0:000).The other main determinant of chemotherapy,besides age at diagnosis, was having a scarcelydifferentiated cancer (56% treated vs. 35% amongall other cases without metastatic disease; Chi-square, Po0:000).

Women with stage M0 disease who were surgicallytreated but did not undergo axillary dissection(n ¼ 129) also seemed to receive radiotherapy lessfrequently than the others (21% vs. 44%). Inparticular, if we exclude women 480 years old,since radiotherapy was hardly used in this agegroup (overall 7%), even the few women (n ¼ 38)who were treated by breast-conserving surgery

(BCS) but did not undergo axillary dissectionreceived radiotherapy less frequently than theother women treated by BCS (40% vs. 75%).

Discussion

The development of a system for follow-up of allcancer cases occurring in a given population andthe publication of population survival figures inlarge comparative studies represented a majorchange since a cancer registry also became a toolthat could be used to evaluate healthcare systemsand effectiveness of interventions. Indeed, the useof cancer registry data to evaluate cancer treat-ment and equity is increasing. Cancer registries, bydefinition, identify all cases of cancer occurring inthe population they cover and may thus contributeto research on the quality and efficacy of treat-

ARTICLE IN PRESS

hazard ratio0 2 4 6 8

0 2 4 6 8

10 12

10 12

canc

er ty

pe

No MV

Carcinoma NS

Ductal NS

Ductal G2

Ductal G3

Lobular

Other specified

hazard ratio

canc

er ty

pe

No MV

Carcinoma NS

Ductal NS

Ductal G2

Ductal G3

Lobular

Other specified

Ductal G1 (ref.)(a)

(b) Ductal G1 (ref.)

Figure 2 Hazard ratios for cancer type and grading for(a) the unrestricted Cox model (Table 1, model A), and(b) the model restricted to patients under 80 years of ageat diagnosis (Table 1, model B). NS=type or grade notspecified; MV=microscopic verification.

Table 2 More comprehensive proportional ha-zards model* including categorized covariates andsignificant interactions among undifferentiatedductal tumors and oldest age classes.

Factor No. ofcases

HR 95% CI

Type/grade Py=0.0005

No MV 40 0.7 0.3–1.5Carcinoma NS 49 1.3 0.6–2.6Ductal NS 51 0.6 0.3–1.4Ductal G1 175 Ref.Ductal G2 585 1.6 0.96–2.5Ductal G3 335 2.6 1.5–4.5Lobular 115 1.7 0.9–3.1Other specified 102 1.2 0.6–2.5G3 Age 70–79 years 69 0.6 0.4–1.1G3 Age 480 years 23 0.3 0.1–0.6

Hormonal receptors P=0.09

All negative 271 1.3 0.9–1.7Positive 808 Ref.Not determined 373 1.4 1.01–1.8

Tumor size Py=0.0001

Not available 113 2.2 1.3–3.61–15 516 Ref.16–30 643 1.5 1.1–2.1430 180 2.31.6–3.4Pyo0.0000Axillary nodes,metastatic nodes

Pyo0.0000

No dissection 249 1.8 1.1–2.9A 684 Ref.B 294 2.3 1.6–3.4C 225 4.9 3.4–7.0Examined nodes (P=0.15)No. of nodesexamined

0.98

0.96–1.0Metastases Pyo0.0000

No 1326 Ref.Yes 126 3.5 2.6–4.7

Surgery Pyo0.0000

Yes 1350 Ref.No 102 3.5 1.9–6.4

Harrell’s c statistic 84%; EPV 13.*Also adjusted for age and marital status.yLR test.

A population survival model for breast cancer 99

ments, but they should be based on clinicallydeveloped models and incorporate more clinicaldetails than hitherto.24 The NPM is regarded as animportant starting point, because it has beenshown to be valid in many different studies.9–12 Afew differences from the model as used in clinicalstudies have been adopted, to avoid selection ofcases: the inclusion of metastatic cases and caseswith missing covariates, and the tentative inclusionof elderly women.

Cancer size and, even more, axillary nodal statuswere confirmed as important prognostic factors inall our population survival models, as they havebeen in the clinical setting.9 The inclusion of veryold women had an effect on many model coeffi-cients. Cancer type and grading became nonsigni-ficant with the inclusion of women over 80. The

influence of elderly patients is not surprising if weconsider that the age class included ‘only’ 11% ofstudy cases but nearly one-fourth of deaths were

ARTICLE IN PRESS

Table 3 Distribution of selected patient, disease, and treatment variables by nodal status among cases withoutmetastases and surgically treated.

Factor No dissection Axillary node dissection: no affected

0 1o=3 43

No. 129 678 281 202Age, mean (S.D.)* 71.7 (15.8) 60.4 (12.8) 61.1 (12.9) 59.8 (12.9)Diameter mean (S.D.)* 24.9 (23.9) 17.3 (10.6) 21.6 (14.4) 27.7 (15.3)T3–4 stage (%)* 18 3 9 17Undifferentiated (%)* 19 20 31 37BCS (%)* 46 51 43 30No radiotherapy %)* 79 57 55 54No chemotherapy (%)* 81 71 42 29No hormonal therapy (%) 46 51 43 30

*Statistically significant: Po0:05:

20

0

40

60

80

100

0

20

40

60

80

100

No Yes No Yes

no axillary dissection negative nodes

1–3 affected nodes 4 or more affected nodes

Use of chemotherapy

Per

cent

Figure 3 Use of chemotherapy by nodal status among cases without distant spread of disease at the time of diagnosis(Chi-square Po0:000).

F. Stracci et al.100

observed in this age class during the 5-year follow-up. Patients over 80 years of age had a cancer-specific mortality hazard ratio of 6 adjusted forother factors relative to the 40–49-year age class.Both incomplete prognostic evaluation and under-treatment of very old women were likely causes ofthe decreasing significance of many prognosticfactors.25–27 It is likely that in elderly womenpresenting with advanced disease prognostic eva-luation was incomplete, so that the selective lackof determination of grading among more aggressivecancers resulted in a loss of prognostic value in theolder age classes. A different biology of breastcancer in the elderly is another, albeit less likely,explanation.28 Most studies based on the NPM

excluded the oldest age classes, and claims ofvalidity were therefore not warranted in thosecases.9–12 Missing data may be more than merely asource of bias in population studies. From a publichealth perspective, incomplete prognostic evalua-tion would be of interest if it could identify caseswith worse prognosis in which suboptimal treat-ment was given. In the present study the missingcategories were associated with hazard ratioshigher than unity, as in other studies,29 and alsowith unstable coefficients, reflecting collinearityamong missing variables and the low number ofcases with incomplete information. Most of theexcess risk associated with incomplete prognosticevaluation was explained by the fact that the

ARTICLE IN PRESS

A population survival model for breast cancer 101

patients concerned did not undergo surgery, prob-ably because of the presence of very advanceddisease or severe comorbidities.

No dissection of the axilla, however, was asso-ciated with a higher risk of death even afteradjustment for surgical treatment and age. Thisfinding has also been reported by otherauthors.30,31 In addition, not undergoing axillarydissection was associated with less frequent che-motherapy among elderly women, but also amongwomen under 80 who had no distant disease andwere surgically treated. Radiotherapy was also lessfrequently used among women who did not undergoaxillary dissection. Thus, the prognostic effect ofno axillary surgery was likely to be a consequenceboth of (1) not removing cancerous tissue and of (2)decreasing the probability of adjuvant chemother-apy and radiotherapy. Moreover, treatment vari-ables for cases not undergoing axillary dissectionwere similar to those in cases without nodalmetastases; this finding also holds true for caseswith moderately differentiated to undifferentiatedcancers (n ¼ 83), despite the higher risk of nodalinvolvement associated with less well differen-tiated cancers. Locally advanced disease was morefrequent among cases without axillary dissection,but chemotherapy and radiotherapy were also lessfrequently given in T3–T4 cases in which no axillarydissection was performed. The presence of seriouscomorbidities is another factor that may havecontributed to both less aggressive treatment andhigher mortality. Indeed, most women who did notundergo axillary dissection (about 65%) were over70 years old. More aggressive treatment of olderwomen and the assessment of nodal status in thesepatients could probably lead to survival improve-ments.32,33 The overall proportion of cases in whichaxillary dissection was performed (84%) was similarto that reported from another population-basedEuropean study (87%)34 and was at the lower end ofthe range reported from US studies.27

Two additional prognostic factors were tested:HR status and the number of axillary nodesremoved. Negativity for both ERs and PRs wasassociated with a nonsignificant excess risk, as wasthe lack of size determination. HR status is useful indecision-making about hormonal therapy and wasfound to be an independent prognostic factor insome but not all prognostic studies.35,36 Number ofnodes has recently been proposed as a prognosticfactor.37 An increasing number of removed nodeswas associated with a nonsignificant risk reductionin our study.

In conclusion, we have succeeded in constructinga population survival model for breast cancer basedon the NPM. Potential targets of intervention to

improve survival rates in our region were alsoidentified. The prognostic importance of nodalstatus was confirmed. The diffusion of surgicaltechniques to evaluate nodal status while reducingside effects (e.g., sentinel lymph node) willprobably reduce the number of cases in whichinformation on this factor is missing and, hopefully,improve survival. The opportunity for chemother-apy and radiotherapy should be carefully consid-ered in cases not receiving axillary dissection(especially in cases with an unfavorable covariatespattern). The inclusion of very old women in thesurvival models has caused problems because of themore frequent incomplete prognostic evaluationand the different treatment. More aggressivetreatment of breast cancer in elderly women isnecessary to improve prognosis.25–26,33 Cancerregistries should be closer to clinical practice toallow a more thorough understanding of clinicalprocesses and evaluation of quality of care. The useof a validated model (e.g., the NPM) would allowmore informative comparisons both among popula-tion-based studies and with clinical studies.38

Acknowledgments

Financial support for this study was received fromthe University of Perugia, Italy, and from theEpidemiology Section of the Umbria RegionalHealth Department.

The authors thank the Umbrian general practi-tioners and the Regional pathologists, surgeons,radiotherapists and oncologists who collaborated inthe study. Special thanks go to Professor Jan WillemCoebergh for helpful suggestions.

References

1. MacMahon S, Collins R. Reliable assessment of the effects oftreatment on mortality and major morbidity. I. Clinicaltrials. Lancet 2001;357:373–80.

2. Black N. Why we need observational studies to evaluate theeffectiveness of health care. BMJ 1996;312:1215–8.

3. Micheli A, Gatta G, Verdecchia A. Studying survival of cancerpatients in different populations: its potential and role.Tumori 1997;83:3–8.

4. Quinn MJ, Martinez-Garcia C, Berrino F. Variations in survivalfrom breast cancer in Europe by age and country,1978–1989. Eur J Cancer 1998;34:2204–11.

5. Sant M, Allemani C, Capocaccia R, et al. Stage at diagnosis isa key explanation of differences in breast cancer survivalacross Europe. Int J Cancer 2003;106:416–22.

6. Page DL, Jensen RA, Simpson JF. Routinely availableindicators of prognosis in breast cancer. Breast Cancer ResTreat 1998;51:195–208.

ARTICLE IN PRESS

F. Stracci et al.102

7. Jatoi I, Miller AB. Why is breast-cancer mortality declining?Lancet Oncol 2003;4:251–4.

8. Botha JL, Bray F, Sankila R, Parkin DM. Breast cancerincidence and mortality trends in 16 European countries. EurJ Cancer 2003;39:1718–29.

9. Elston CW, Ellis IO, Pinder SE. Pathological prognosticfactors in breast cancer. Crit Rev Oncol Hematol 1999;31:209–23.

10. Haybittle JL, Blamey RW, Elston CW, et al. A prognosticindex in primary breast cancer. Br J Cancer 1982;45:361–6.

11. Balslev I, Axelsson CK, Zedeler K, Rasmussen BB, CarstensenB, Mouridsen HT. The Nottingham Prognostic Index appliedto 9,149 patients from the studies of the Danish BreastCancer Cooperative Group (DBCG). Breast Cancer Res Treat1994;32:281–90.

12. Kollias J, Murphy CA, Elston CW, Ellis IO, Robertson JF,Blamey RW. The prognosis of small primary breast cancers.Eur J Cancer 1999;35:908–12.

13. Rosso S, Casella C, Crocetti E, Ferretti S, Guzzinati S.Survival of cancer patients in Italy in the nineties; figuresfrom the Cancer Registries. Epidemiol Prev 2001;25(suppl.3):1–375.

14. La Rosa F, Patavino VM, Epifani AC, Petrinelli AM, Minelli L,Mastrandrea V. Ten-year survival and age at diagnosis ofwomen with breast cancer from a population-based study inUmbria, Italy. Tumori 1996;82:441–3.

15. Scorpiglione N, Nicolucci A, Grilli R, et al. Appropriatenessand variation of surgical treatment of breast cancer in Italy:when excellence in clinical research does not match withgeneralized good quality care. Progetto Oncologia Femmi-nile. J Clin Epidemiol 1995;48:345–52.

16. World Health Organisation. Manual of the internationalstatistical classification of diseases, injuries and causes ofdeath (9th Rev., vol. I). Geneva: WHO; 1977.

17. Cox DR. Regression models and life tables (with discussion).J R Stat Soc, Ser B 1972;34:187–220.

18. Hosmer DW, Lemeshow S. Applied survival analysis. NewYork: Wiley; 1999.

19. Harrell Jr FE, Lee KL, Mark DB. Multivariable prognosticmodels: issues in developing models, evaluating assumptionsand adequacy, and measuring and reducing errors. Stat Med1996;15:361–87.

20. Peduzzi P, Concato J, Feinstein AR, Holford TR. Importanceof events per independent variable in proportional hazardsregression analysis. II. Accuracy and precision of regressionestimates. J Clin Epidemiol 1995;48:1503–10.

21. Fentiman IS, Hyland D, Chaudary MA, Gregory WM. Prognosisof patients with breast cancers up to 1 cm in diameter. Eur JCancer [A] 1996;32:417–20.

22. Sinha PS, Bendall S, Bates T. Does routine grading of invasivelobular cancer of the breast have the same prognosticsignificance as for ductal cancers? Eur J Surg Oncol2000;26:733–7.

23. Stata Corp. Stata Statistical Software (Release 8.0). CollegeStation (US): Stata Corporation; 2003.

24. Malin JL, Kahn KL, Adams J, Kwan L, Laouri M, Ganz PA.Validity of cancer registry data for measuring the quality ofbreast cancer care. J Natl Cancer Inst 2002;94:835–44.

25. Vercelli M, Capocaccia R, Quaglia A, Casella C, Puppo A,Coebergh JW. Relative survival in elderly European cancerpatients: evidence for health care inequalities. The EURO-CARE Working Group. Crit Rev Oncol Hematol 2000;35:161–79.

26. Bouchardy C, Rapiti E, Fioretta G, et al. Undertreatmentstrongly decreases prognosis of breast cancer in elderlywomen. J Clin Oncol 2003;21:3580–7.

27. Malin JL, Schuster MA, Kahn KA, Brook RH. Quality of breastcancer care: what do we know? J Clin Oncol 2002;20:4381–93.

28. Alberg AJ, Singh S. Epidemiology of breast cancer in olderwomen: implications for future healthcare. Drugs Aging2001;18:761–72.

29. Lash TL, Silliman RA, Guadagnoli E, Mor V. The effect of lessthan definitive care on breast carcinoma recurrence andmortality. Cancer 2000;89:1739–47.

30. Du XL, Freeman JL, Nattinger AB, Goodwin JS. Survival ofwomen after breast conserving surgery for early stage breastcancer. Breast Cancer Res Treat 2002;72:23–31.

31. Orr RK. The impact of prophylactic axillary node dissectionon breast cancer survival—a Bayesian meta-analysis. AnnSurg Oncol 1999;6:109–16.

32. Du XL, Key CR, Osborne C, Mahnken JD, Goodwin JS.Discrepancy between consensus recommendations andactual community use of adjuvant chemotherapy in womenwith breast cancer. Ann Intern Med 2003;138:90–7 Erratum:Ann Intern Med 2003;139:873.

33. Holmes CE, Muss HB. Diagnosis and treatment of breastcancer in the elderly. Cancer J Clin 2003;53:227–44.

34. Voogd AC, Coebergh JW, Repelaer van Driel OJ, et al. Therisk of nodal metastases in breast cancer patients withclinically negative lymph nodes: a population-based analy-sis. Breast Cancer Res Treat 2000;62:63–9.

35. Rampaul RS, Pinder SE, Elston CW, Ellis IO. NottinghamBreast Team. Prognostic and predictive factors in primarybreast cancer and their role in patient management: TheNottingham Breast Team. Eur J Surg Oncol 2001;27:229–38.

36. Mirza AN, Mirza NQ, Vlastos G, Singletary SE. Prognosticfactors in node-negative breast cancer: a review of studieswith sample size more than 200 and follow-up more than 5years. Ann Surg 2002;235:10–26.

37. Weir L, Speers C, D’yachkova Y, Olivotto IA. Prognosticsignificance of the number of axillary lymph nodes removedin patients with node-negative breast cancer. J Clin Oncol2002;20:1793–9.

38. Altman DG, Royston P. What do we mean by validating aprognostic model? Stat Med 2000;19:453–73.

The Breast (2005) 14, 83–84

THE

BREAST

Editorial

Commentary:Axillary dissection in breast cancerupdated

In this issue, Stracci et al. report effects on survivalin 1452 breast cancer patients.1 Arguably, theirmost interesting finding is an increased risk ofbreast cancer death in cases who had no axillarysurgery. As this is an observational rather than anexperimental result, there are a number of possibleinterpretations. These include

(1) Those not receiving axillary surgery are a high-risk group a priori, who are largely ineligible forextensive surgery or adjuvant chemotherapydue to age or comorbidity. This may explainpart but not all of the finding. Of those who hadany surgery at all, only 11% had no axillarysurgery, and this small group may differ system-atically from the overall patient population.However, the relative hazard associated with noaxillary surgery, although considerably attenu-ated when adjusted for whether any surgerytook place, is still significant, suggesting anapproximate doubling of the hazard of deathfrom breast cancer. This suggests that there is asurvival advantage associated with axillarydissection independent of fitness for surgerygenerally.

(2) The fact of having axillary surgery may be asurrogate for more thorough diagnostic workup,staging and therapy in general. While this isprobably the case, it is unlikely to be the fullexplanation, and in any case, axillary dissectionis arguably an integral part of the betterworkup, staging and treatment.

(3) There is a real survival advantage of axillarydissection and appropriate treatment. This isalmost certainly true. The survival advantagemay be smaller than indicated by the results ofStracci et al., due to the two other possible

process described above, but it is almostcertainly present.

Axillary dissection of some kind is recommendedroutinely in Oncology textbooks,2,3 and there is anevidence base for this practice.4 There are,however, untoward sequelae of axillary clearanceand of axillary sampling followed by radiotherapy inthe positive case.3 The most promising resolution ofthese two phenomena is sentinel node biopsy, inwhich not usually more than two sentinel nodes areexcised, and further axillary surgery performedonly in those with at least one positive sentinelnode.5 Early results from trials of sentinel nodebiopsy are encouraging.6,7 Long-term follow-up ofthese studies is necessary to confirm that thesurvival benefit of this procedure is similar to thatobserved for axillary surgery by Stracci andcolleagues.1

References

1. Stracci F, La Rosa F, Falsettini E, Ricci E, Aristei C, et al. Apopulation survival model for breast cancer. The Breast.doi:10.1016/j.breast.2004.08.011

2. Mansel R. Local Treatment and Reconstruction. In: SouhamiRL, Tannock I, Hohenberger P, Horiot JC editors. OxfordTextbook of Oncology. Oxford: Oxford University Press; 2002.p. 1737–44.

3. Sacks NPM, Baum M. Treatment. In: Allen-Mersh TG editor.Bailey and love’s surgical oncology. London: Chapman & Hall;1996. p. 3551–72.

4. Shukla HS, Melhuish J, Mansel RE, Hughes LE. Does localtherapy affect survival rates in breast cancer? Ann Surg Oncol1999;6:455–60.

5. McIntosh SA, Purushotham AD. Lymphatic mapping andsentinel node biopsy in breast cancer. Br J Surg1998;85:1347–56.

ARTICLE IN PRESS

www.elsevier.com/locate/breast

0960-9776/$ - see front matter & 2004 Elsevier Ltd. All rights reserved.doi:10.1016/j.breast.2004.09.007

6. Goyal A, Newcombe RG, Mansel RE, Chetty U, Ell P, et al.Sentinel lymph node biopsy in patients with multifocal breastcancer. Eur J Surg Oncol 2004;30:475–9.

7. Veronesi U, Paganelli G, Viale G, Luini A, Zurrida S, et al. Arandomized comparison of sentinel-node biopsy with routineaxillary dissection in breast cancer. New Engl J Med2003;349:546–53.

Stephen W. DuffyProfessor of Cancer Screening, Cancer Research UKCentre for Epidemiology, Mathematics and Statis-

tics, Wolfson Institute of Preventive Medicine,Charterhouse Square, London EC1M 6BQ, UKE-mail address: Stephen.duffy@cancer.org.uk

ARTICLE IN PRESS

Editorial84