Factors associated with concordant estrogen receptor expression at diagnosis and centralized re-assay in a Danish population-based breast cancer study

Abstract

Background. Estrogen receptor (ER) expression predicts tamoxifen response, which halves the risk of breast cancer recurrence. We examined clinical factors associated with concordance between ER expression at diagnosis and centralized re-assay, and the association of concordance with breast cancer recurrence. Material and methods. We used immunohistochemistry to assess ER expression on archived fixed, paraffin-embedded breast carcinoma tissue excised from women aged 35–69 years, diagnosed 1985–2001 in Jutland, Denmark. We calculated the percentage agreement, positive predictive value (PPV) and negative predictive value (NPV) of ER status at diagnosis and re-assay. We used logistic regression to investigate factors associated with concordance, and its association with recurrence (odds ratios (OR) and associated 95% confidence intervals (95%CI)). Results. ER was re-assayed in 91% of patients (n = 1530). Concordance was better in ER + than ER– tumors (PPV = 94% vs. NPV = 75%). Factors associated with concordance included menopausal status, tumor size, surgical procedure, diagnostic period, lymph node status and time to recurrence. ER + women at diagnosis who re-assayed ER + were less likely to have recurrent disease (OR = 0.49, 95% CI = 0.28, 0.86) than those who re-assayed ER–. In originally ER– women, concordance was not associated with recurrence (OR = 0.97, 95% CI = 0.66, 1.42). Conclusions. Several clinical factors were associated with ER assay concordance. Some women were ineffectively treated with tamoxifen, or required but did not receive tamoxifen. We observed almost exactly the protective effect of endocrine therapy among tamoxifen-treated ER + women whose tumors expressed the ER on re-assay, compared with those ER– on re-assay. Diagnostic pathology results for ER + tumors appear a valid and useful resource for research studies. However, those for ER– tumors have lower validity. Study-specific considerations regarding the aims, diagnostic period, and consequences of including ER– patients with truly ER + disease ought to be examined when using diagnostic pathology results for ER– tumors in research studies.

Endocrine therapy reduces the rate of breast cancer recurrence by approximately one half [1]. Estrogen receptor (ER) expression is the primary predictor of response to endocrine therapy in breast cancer patients [2]. ER status has been routinely assessed during breast cancer prognostic evaluation for more than 30 years [1]. During this period, an evolving series of assays have been used in routine pathology practice to determine tumor ER expression. These ranged from biochemical extraction assays through immunohistochemical (IHC) assays, initially used on fresh, frozen tissue and, more recently, on fixed paraffin-embedded tumor tissue [3]. The advancing techniques have been facilitated by the development of increasingly sensitive and specific commercial antibodies to the ER [3].

The biochemical extraction assay quantitatively scores the ER ligand-binding ability of all cells in a pathology specimen, including not only carcinoma cells, but also pre-invasive and adjacent non-neoplastic cells [3]. This cellular heterogeneity increases the proportion of false positive results, compared with IHC [3–6]. IHC pinpoints the ER location within a specimen, distinguishing ER expression in tumor cells from that in adjacent tissue. However, IHC can produce variable results depending on several factors, including pre-analytical conditions (e.g. duration of fixation), and analytical and interpretative variation (intra- and inter-observer variation) [6]. IHC assays have been used in preference to the more expensive, labor intensive, and less accurate biochemical assays for about 20 years [7].

Earlier studies that examined the concordance between primary (local) testing of ER expression at the time and place of initial diagnosis and centralized retesting of ER expression, found a high level of agreement – approximately 80% [8], 87% [9] and 90% [10]. In the Breast International Group (BIG 1-98) trial, central review carried out retrospectively changed the assessment of receptor status in a substantial proportion of patients [8]. To the best of our knowledge, no studies have examined the association of ER concordance with clinical factors, and few focused on the association of concordance with breast cancer recurrence [5,8,11].

In our Danish population-based case-control study of genetic polymorphisms and prescription drugs that may modify the risk of breast cancer recurrence in patients treated with tamoxifen [12,13], ER expression at diagnosis was the basis for inclusion in the study. Given the continuous improvement of ER assay methods over time, and in order to reduce the potential effect of variability across diagnosing hospitals, we centrally re-assayed ER expression in the tumors of all participants. In the current study, we examine the concordance between the original ER assay and the centralized ER expression re-assay. We identify factors associated with ER concordance and examine the impact of concordance on the rate of breast cancer recurrence.

Material and methods

This study was approved by the Regional Ethics Committee for Biomedical Research (Region Midt & Aarhus County, Denmark), the Danish Data Protection Agency, and the Danish Breast Cancer Cooperative Group.

Study population

The study design has been previously described [12]. Briefly, the source population consisted of female residents of Jutland, Denmark, 35 to 69 years old between 1985 and 2001, who were diagnosed with stage I, II or III breast cancer as defined by the Union for International Cancer Control (UICC) [14], and who were registered with the Danish Breast Cancer Cooperative Group (DBCG) [15].

We divided the source population into three groups: (a) women whose tumor expressed the ER at diagnosis and who were treated with tamoxifen for at least one year (ER + /TAM +), (b) women whose tumor did not express the ER at diagnosis, were not treated with tamoxifen, and who survived for at least one year (ER–/TAM-), and (c) all other breast cancer patients, who were excluded. Follow-up time began one year after the date of breast cancer diagnosis and continued until the date of the first breast cancer recurrence, death from any cause, loss to follow-up (e.g. emigration), 10 years of follow-up or 1 September 2006.

Cases were women with local or distant breast cancer recurrence or contralateral breast cancer occurrence during their follow-up time. We selected ER–/TAM– cases at random, after frequency matching as close as possible to the distribution of stage and calendar period of diagnosis among ER + /TAM + cases.

For each case, we selected without replacement (i.e. once a patient was selected as a control, they were removed from the pool of individuals eligible to be selected as controls; however if they later developed recurrent disease, they also became a case) [16] one control from members of the source population, who were alive and had no recurrence or contralateral breast cancer after the same amount of follow-up time. We matched controls to cases by group membership (ER + /TAM + or ER–/TAM–), (b) menopausal status at diagnosis (premenopausal or postmenopausal), (c) date of breast cancer surgery (caliper matched + /– 12 months), (d) county of residence at the time of diagnosis, and (e) cancer stage at diagnosis (UICC stage I, II, or III). As noted above, controls could become cases later in the follow-up time. When this happened, the tumor was re-assayed only once and the re-assay result was reported with both cases and controls.

Data collection from Danish registries

We used the Danish Civil Personal Registration (CPR) number to link data sets. The CPR number is a unique identifier assigned to all Danish residents who were alive on 1 April 1968, born thereafter, or upon immigration [17].

We collected data from the DBCG registry on patient demographics (year, age, and menopausal status at diagnosis), tumor characteristics (UICC stage, tumor size, node status, and ER expression), therapy characteristics (primary surgery type and adjuvant tamoxifen protocol), and recurrence status.

Data collection from archived tissue samples

Laboratory personnel were blinded to all clinical information, including case or control status, diagnostic ER status, and receipt or non-receipt of tamoxifen therapy.

Tissue processing. We collected formalin-fixed paraffin-embedded tissue blocks from the pathology department archives of treating hospitals [18]. We reviewed hematoxylin- and eosin-stained glass slides and the original pathology reports to identify the blocks to be processed. All tissue blocks were processed using established routine protocols designed to avoid case contamination.

Estrogen receptor re-assay

We re-assayed ER expression using whole sections with a thickness of 2 μm from the original diagnostic paraffin embedded tissues and primary antibody against ERα (clone 6F11; Novocastra, Newcastle-Upon-Tyne, UK). Heat-induced epitope retrieval for ER was achieved by incubation in a Tris/EGTA buffer, pH 9 (VWR-Bie & Bertsen, Denmark) using a microwave oven. Sections were stained on a Lab Vision Autostainer (Thermo Fisher Scientific, Fremont, CA, USA) using the EnVision™ + detection system (Dako). Sections were enhanced using copper sulphate and visualized with horseradish peroxidase and diaminobenzadine. Slides were scored positive for ER when there was distinct nuclear staining of neoplastic cells. A cutoff point of ≥ 10% of positive tumor nuclei was chosen for ER positivity in accordance with previous DBCG recommendations for the diagnostic period of patients included in the current study [14].

Analytic variables

We used the DBCG definition of breast cancer recurrence, i.e. any type of breast cancer or distant metastasis diagnosed subsequent to the initial course of therapy [19]. Given the follow-up time in the source population, all recurrences occurred between one and 10 years after primary breast cancer diagnosis.

We considered the following variables to be potentially associated with concordance between the diagnostic ER status and ER status at re-assay: menopausal status, tumor size, surgical procedure, surgery type, recurrence status (i.e. case or control), node status, time to recurrence (in years) and the DBCG treatment protocol. The DBCG treatment protocol variable incorporates treatment type and time period of diagnosis, and is a proxy for ER assay type at diagnosis (i.e. biochemical vs. IHC) (see www.dbcg.dk).

Statistical analysis

We computed the frequency and proportion of cases and controls, stratified by group (ER + /TAM + or ER–/TAM–) within categories of the independent variables listed above. We calculated the percentage agreement between ER status at breast cancer diagnosis and that obtained from the centralized re-assay of ER status, and positive and negative predictive values (PPV and NPV, respectively), stratified by case and control status. We constructed contingency tables according to changes in ER status on re-assay and calculated associated p-values using χ² tests. We computed models to examine baseline clinical factors associated with concordant ER status, either from positive at diagnosis to positive on re-evaluation, or negative at diagnosis to negative on re-assay. Using these models, we computed odds ratios (OR), adjusted ORs and associated 95% confidence intervals (95% CI), mutually adjusting for all other variables in the model. All statistical tests were two-sided (α = 0.05), and analyses were completed using Stata 11.

Results

The original study identified 541 ER + /TAM + cases of recurrence and their matched controls, as well as 300 ER–/TAM– cases of recurrence and their matched controls [12]. Of these, we completed centralized re-assay of ER expression for 493 ER + /TAM + cases (91%), 488 ER + /TAM + controls (90%), 275 ER–/TAM– cases (92%), and 276 ER–/TAM– controls (92%). The completion proportion did not depend on any demographic, tumor, or treatment characteristics. Failure to complete centralized re-assay resulted mostly from lack of available archived tissue specimens or inadequate specimens available within the block.

Table I shows characteristics of the study population with successful centralized re-assay of ER expression including the frequency and proportion of cases and controls, within group strata, in the categories of the independent variables. More than 90% of the women in the ER + /TAM + group were postmenopausal and had stage II or III disease, due to the DBCG criteria for assignment to standard tamoxifen protocols during the diagnostic era of the study population [15]. Most patients (80%) underwent mastectomy.

Table I. Frequency and proportion of cases of breast cancer recurrence and matched controls with successful centralized re-assay of estrogen receptor expression within group strata (a) expressing the ER and receiving at least one year of tamoxifen therapy (ER + /TAM +), or (b) not expressing the ER, never receiving tamoxifen therapy, and surviving at least one year after diagnosis (ER–/TAM-). Women, 35 to 69 years old, Jutland Peninsula, Denmark, 1985–2001.

	*ER + /TAM + : n, (%)		*ER–/TAM-: n, (%)
	cases	controls	cases	controls
Diagnosis year^† 1985–1993 1994–1996 1997–2001	211 (43) 98 (20) 179 (37)	210 (43) 103 (21) 180 (37)	97 (35) 74 (27) 105 (38)	89 (32) 78 (28) 108 (39)
Age at diagnosis (years) 35–44 45–54 55–64 65–69	15 (3.1) 104 (21) 256 (52) 113 (23)	13 (2.6) 99 (20) 256 (52) 125 (25)	62 (22) 111 (40) 75 (27) 28 (10)	53 (19) 102 (37) 81 (29) 39 (14)
Menopausal status at diagnosis^† Premenopausal Postmenopausal	32 (6.6) 4567 (93)	31 (6.3) 462 (94)	112 (41) 164 (59)	110 (40) 165 (60)
UICC tumor stage at diagnosis^† Stage I Stage II Stage III	9 (1.8) 223 (46) 256 (52)	6 (1.2) 228 (46) 259 (53)	22 (8.0) 142 (51) 112 (41)	25 (9.0) 141 (51) 109 (40)
Tumor size < 2 cm 2- to ≤ 5 cm > 5 cm	159 (33) 288 (60) 36 (7.5)	194 (40) 252 (52) 41 (8.4)	86 (31) 170 (62) 20 (7.3)	78 (29) 175 (64) 19 (7.0)
Lymph node status Node negative Node positive	29 (5.9) 459 (94)	30 (6.1) 463 (94)	74 (27) 202 (73)	104 (38) 171 (62)
Surgery type Breast-conserving surgery Mastectomy	50 (10) 438 (90)	68 (14) 425 (86)	41 (15) 234 (85)	54 (20) 221 (80)
DBCG protocol^§ DBCG82 DBCG89 DBCG99	88 (18) 310 (64) 90 (18)	84 (17) 318 (65) 91 (18)	18 (6.5) 318 (78) 42 (15)	19 (6.9) 219 (80) 37 (13)

*Original ER assay expression tested at the time of breast cancer diagnosis.

^†Variable included in risk-set sampling to match controls to cases.

^§DBCG protocol incorporates type of treatment according to DBCG guidelines and period of diagnosis.

The PPV was slightly higher in controls than cases (96% vs. 92%). The NPV in both groups was similar (75% and 74%, respectively) (Table II). Among those ER + at diagnosis, the unadjusted OR associating confirmation of ER expression on re-assay with recurrence equalled 0.49 (95% CI = 0.28, 0.86). Among those ER– at diagnosis, the unadjusted OR associating failure to confirm the absence of ER expression with recurrence was null (OR = 0.97; 95% CI 0.66, 1.4) (data not tabulated).

Table II. Cross-tabulation of ER status at diagnosis and at centralized re-assay (the assumed gold standard) stratified by case and control status. (NPV, negative predictive value; PPV, positive predictive value).

	ER status at centralized re-assay
Controls
	ER +	ER −	Total
ER status at diagnosis
ER +	474	19	493
ER −	70	205	275
PPV controls: 474/493 = 0.96
NPV controls: 205/275 = 0.75
Cases
	ER +	ER −	Total
ER status at diagnosis
ER +	451	37	488
ER −	72	204	276
PPV cases: 451/488 = 0.92
NPV cases: 204/276 = 0.74
Cases & controls combined:
PPV = 0.94
NPV = 0.75

Crude and mutually adjusted ORs associating the variables with ER concordance among controls are presented in Table III. ER concordance on re-assay was more likely in postmenopausal women, those with larger tumors, tumors excised by breast-conserving surgery, or those diagnosed later in the study period (DBCG89 and DBCG99). A similar profile of factors was associated with ER assay concordance among those who were initially ER + or ER–. Only 4% of ER + tumors at diagnosis failed to express the ER on re-assay, so these estimates were imprecise. Among controls with ER– tumors at baseline, 25% tested ER + on re-assay, so these estimates were more precise. In ER + women only, those with node-positive disease were less likely to re-test ER + on re-assay (adjusted OR = 0.63, 95% CI = 0.32, 1.2).

Table III. Clinical factors and ER concordance among controls on centralized retesting of ER expression by immunohistochemistry. Source population: women with breast cancer aged 35 to 69 years, diagnosed between 1985 and 2001, Jutland Peninsula, Denmark.

Characteristic	ER + who stayed ER + N = 474 (% of total ER + controls)	Odds Ratio (95%CI)	Adjusted Odds Ratio (95%CI)	ER– who stayed ER– N = 205 (% of total ER–)	Odds Ratio (95%CI)	Adjusted Odds Ratio (95%CI)
Menopausal status		p = 0.44	p = 0.25		p < 0.001	p = 0.001
Premenopausal	29 (94)	1.0	1.0	69 (63)	1.0	1.0
Postmenopausal	445 (96)	1.8 (0.40, 8.2)	4.3 (0.36, 50.8)	136 (82)	2.8 (1.6, 4.9)	2.8 (1.5, 5.1)
Tumor size		p = 0.42	p = 0.48		p = 0.04	p = 0.03
≤ 2 cm	184 (95)	1.0	1.0	50 (64)	1.0	1.0
2- ≤ 5 cm	245 (97)	1.9 (0.71, 5.1)	1.9 (0.67, 5.1)	139 (79)	2.2 (1.2, 3.9)	2.3 (1.2, 4.2)
> 5 cm	39 (95)	1.1 (0.22, 5.0)	1.1 (0.22, 5.7)	14 (74)	1.6 (0.51, 4.8)	2.5 (0.75, 8.0)
Nodal status		-	-		p = 0.02	p = 0.17
Node negative	30 (100)	-	-	86 (83)	1.0	1.0
Node positive	444 (96)	-	-	119 (70)	0.48 (0.26, 0.88)	0.63 (0.32, 1.2)
Surgery type		p = 0.68	p = 0.58		p = 0.004	p = 0.005
Mastectomy	408 (96)	1.0	1.0	156 (71)	1.0	1.0
Breast-conserving surgery	66 (97)	1.4 (0.31, 6.1)	1.6 (0.33, 7.4)	49 (91)	4.1 (1.6, 10.7)	4.2 (1.5, 11.5)
DBCG Protocol		p = 0.08	p = 0.07		p = 0.72	p = 0.95
DBCG82	77 (92)	1.0	1.0	13 (68)	1.0	1.0
DBCG89	309 (97)	3.1 (1.1, 8.6)	3.1 (1.11, 8.7)	163 (74)	1.3 (0.49, 3.7)	0.93 (0.30, 2.9)
DBCG99	88 (97)	2.7 (0.67, 10.7)	4.1 (0.48, 34.8)	29 (78)	1.7 (0.48, 5.8)	1.1 (0.27, 4.3)

p-values correspond to test of homogeneity over the entire category of the variable in the logistic regression model.

Crude and mutually adjusted OR associating baseline factors with ER concordance among cases are presented in Table IV. These results were similar to those observed in controls (Table III). Among women with ER + tumors at diagnosis, the median time to recurrence was 3.4 years in those whose tumors re-assayed as ER + (Table IV) and 2.9 years in those whose tumors re-assayed as ER–. Among women with ER– tumors at diagnosis, the opposite pattern pertained; median time to recurrence was 2.0 years in concordant tumors (i.e. ER– on re-assay) (Table IV) and 2.9 years in those ER + on re-assay.

Table IV. Clinical factors and ER concordance among cases of recurrence on centralized retesting of ER expression by immunohistochemistry. Source population: women with breast cancer aged 35 to 69 years, diagnosed between 1985 and 2001 Jutland Peninsula, Denmark.

Characteristic	ER + who stayed ER + N = 451 (% of total ER + cases)	Odds Ratio (95%CI)	Adjusted Odds Ratio (95%CI)	ER– who stayed ER–N = 204 (% of total ER– cases)	Odds Ratio (95%CI)	Adjusted Odds Ratio (95%CI)
Menopausal status		p = 0.02	p = 0.03		p = 0.06	p = 0.15
Premenopausal	26 (81)	1.0	1.0	76 (68)	1.0	1.0
Postmenopausal	425 (93)	3.2 (1.2, 8.3)	4.6 (1.1, 18.1)	128 (78)	1.7 (0.98, 2.9)	1.5 (0.86, 2.7)
Tumor size		p = 0.36	p = 0.32		p = 0.90	p = 0.96
≤ 2 cm	144 (91)	1.0	1.0	65 (76)	1.0	1.0
2- ≤ 5 cm	268 (93)	1.4 (0.69, 2.8)	1.6 (0.76, 3.3)	124 (73)	0.87 (0.48, 1.6)	0.97 (0.51, 1.9)
> 5 cm	35 (97)	3.6 (0.47, 28.5)	3.3 (0.41, 26.9)	15 (75)	0.97 (0.31, 2.9)	1.2 (0.35, 3.8)
Nodal status		p = 0.20	p = 0.37		p = 0.01	p = 0.02
Node negative	25 (86)	1.0	1.0	63 (85)	1.0	1.0
Node positive	426 (93)	2.1 (0.68, 6.3)	1.8 (0.51, 6.2)	141 (70)	0.40 (0.20, 0.82)	0.43 (0.20, 0.89)
Surgery type		p = 0.66	p = 0.63		p = 0.03	p = 0.06
Mastectomy	404 (92)	1.0	1.0	167 (71)	1.0	1.0
Breast conserving surgery	47 (94)	1.3 (0.39, 4.5)	1.4 (0.38, 5.0)	36 (88)	2.9 (1.1, 7.7)	2.8 (0.98, 8.0)
DBCG Protocol		p = 0.01	p = 0.02		p = 0.33	p = 0.72
DBCG82	75 (85)	1.0	1.0	13 (72)	1.0	1.0
DBCG89	294 (95)	3.2 (1.5, 6.9)	2.8 (1.3, 6.4)	156 (72)	1.0 (0.34, 2.9)	1.0 (0.32, 3.2)
DBCG99	82 (91)	1.8 (0.70, 4.5)	3.8 (1.1, 14.0)	35 (83)	1.9 (0.52, 7.1)	1.5 (0.36, 5.9)
Time to recurrence		p = 0.14	p = 0.25		p = 0.008	p = 0.007
(median*)	(3.42)	1.1 (0.96, 1.4)	1.1 (0.93, 1.3)	(2.03)	0.83 (0.52, 0.95)	0.82 (0.71, 0.95)

p-values correspond to test of homogeneity over the entire category of the variable in the logistic regression model. *Median number of years to recurrence.

Discussion

In this large population of Danish breast cancer patients, we observed a high level of concordance between ER expression at diagnosis and centralized re-assay among those originally ER +. Factors associated with concordance included menopausal status, tumor size, surgery type, diagnostic period and time to recurrence. About one quarter of women categorized as ER– at diagnosis re-assayed as ER + — these women had recurrences later on average than those who re-assayed as ER–. Although these women were not assigned tamoxifen treatment, their longer time to recurrence is likely due to the better prognosis in ER + compared with ER– disease, especially in the early years of follow-up. We observed almost exactly the protective effect of endocrine therapy among tamoxifen-treated ER + women whose tumors expressed the ER on re-assay, compared with those whose tumors re-assayed as ER–.

Several issues should be considered when interpreting our findings. The strengths of our study include the use of the DBCG as a source of patient data. The DBCG manages a clinical breast cancer database considered to be one of the most comprehensive in the world, with data quality akin to that of a clinical trial [15]. This facilitated examination of potential associations between patient, tumor, treatment and follow-up characteristics and ER concordance, which were not investigated in the two previous studies [9,10]. Although we had no information on the type of ER assay used in each diagnostic period, we used the DBCG protocols as proxies for ER assay type. The DBCG directs breast cancer diagnosis and treatment in Denmark, so these protocols are likely to be valid proxies for actual practice. During the diagnostic period of patients included in our study, the DBCG recommended an evolving series of ER assays [20] (www.dbcg.dk). These encompassed biochemical assays on fresh tumor tissue in “DBCG82” (1982–1989); biochemical assays in “DBCG89” (1990–1998), unless tumor tissue was scarce (< 0.1 g), in which case IHC was recommended on fresh-frozen tissue, or paraffin-embedded tissue if tumor tissue was very scarce; and IHC in “DBCG99” (1999–2002), virtually all performed on formalin fixed, paraffin-embedded tissue. Throughout the study period, tumors were considered ER + if at least 10 fmol/mg cytosol protein were present in the biochemical assay [6], or if at least 10% of cells stained positive in the IHC assay (www.dbcg.dk).

Other limitations included lack of information on the percentage of cells showing ER expression at diagnosis. Tumors that were ER– at diagnosis but ER + on re-assay, and tumors that were ER + at diagnosis but ER– on re-assay, may have shown weak ER expression at initial diagnosis. One would expect that tumors with weak expression may be most likely to have been misclassified. We could not verify this in our data, as we did not have a quantitative measure of expression at diagnosis with which to correlate the percentage positivity found on centralized re-assay. We set a threshold of 10% in our centralized re-assay, consistent with DBCG guidelines during the period of diagnosis of patients included in our study, to assure a consistent classification criterion. We had no information on progesterone receptor and Her-2 receptor levels in tumor samples and so were unable to correlate these with ER concordance and breast cancer recurrence. Finally, there is evidence that both under- and over-fixation of tumor tissue can give false IHC results [21]. We had no detailed information on variations in the formalin-fixation procedures (including length of fixation) over the period of block collection, which may have impacted ER expression.

Our findings regarding the association between menopausal status and assay concordance agree with published findings related to menopausal status [7] and age [22]. Breast tumors in pre-menopausal women are generally small and aggressive, with involved lymph nodes, and are more frequently ER– than tumors in post-menopausal women. It has been speculated that the biochemical assay underestimates ER positivity in pre-menopausal women, due to their inherently lower ER levels, and, in whom the ER may be occupied by endogenous hormones [22].

It seems counterintuitive that tumors excised by breast-conserving surgery had better concordance than those excised by mastectomy. This also seems to contradict our findings related to tumor size—for which concordance was generally better for larger tumors—since breast-conserving surgery is associated with less advanced disease at diagnosis than mastectomy [23]. Given the small amount of tissue removed during breast-conserving surgery compared with mastectomy, it is likely that such tumors would have been more often initially assessed for ER expression via IHC, especially given the recommendations in the second protocol period. In addition, our previous research shows that breast-conserving surgery in Denmark increased from 1% of all initial breast cancer operations in 1982 to 25% in 2002 [23]. We expect, therefore, that the proportion of breast conserving surgeries correlated with the proportion of tumors assayed for ER status by IHC.

The improvement in ER concordance over time among ER + women may be attributable to several factors. First, the longer storage of the older samples may have adversely affected antigen retrieval. However, FFPE is considered stable and studies using even older archival tissue have shown results similar to ours [9]. Second, the storage conditions at individual hospitals may have affected concordance. However, each hospital contributed tumor blocks from cases diagnosed over the entire range of the study period, so storage conditions were likely unrelated to diagnostic time period. As biochemical assays have a higher rate of error [3], the increasing use of IHC over time is likely to have contributed to the better concordance in the second protocol period, and the best concordance in the most recent protocol period. The switch to IHC reflected both improvement in IHC techniques and the availability of better antibodies for ER expression analysis in fixed, paraffin-embedded tissues. These factors are also likely to have contributed to improved concordance over the study period.

Our results may have clinical implications, suggesting that as many as 25% of the women with ER – tumors actually could have benefited from adjuvant tamoxifen therapy, while only 4% of the ER + women, who would not have benefited from the therapy, actually received it. Breast cancer clinical guidelines have recently advocated endocrine therapy for all women whose tumors express the ER to any degree [24,25]. Our findings of one in four ER– tumors re-assaying as ER +, at a time when 10% or more of a tumor specimen had to express the ER to be considered ER +, supports these recent guidelines.

Current breast cancer clinical guidelines only advocate histopathological characterization of primary breast tumors. Therapeutic decisions for metastatic disease therefore are based on the primary tumor's histopathological evaluation. Despite this, several studies have shown differences in the expression of ER, progesterone receptor, and HER-2 between primary tumor tissue and paired metastatic tissue [26–29]. Taken together with our findings of changes in ER status on retesting, this may indicate the importance of characterizing both primary and later recurrent or metastatic tumors and tailoring patient therapy accordingly.

In conclusion, we found a high level of ER concordance among originally ER + patients but less so among originally ER– patients. Concordance correlated with tumor and treatment characteristics, including protocol era, surgery type, and tumor size. Our concordance proportions are largely consistent with earlier research [9,10], and highlight the reliability of ER data in diagnostic pathology records. There are often significant cost-efficiencies and logistic convenience to be gained by using existing pathology assay results in research studies. As our results show, these are well-justified, particularly among patients categorized as ER + at diagnosis. However, among those categorized as ER–, study-specific considerations regarding the hypotheses, diagnostic era, and consequences of including ER– patients with truly ER + disease ought to be weighed.

Acknowledgements

This work was supported by grants from the US National Cancer Institute at the National Institutes of Health (grant number R01 CA118708); the Danish Cancer Society (grant number DP06117); and the Karen Elise Jensen Foundation. The authors declare no conflict of interest.

Declaration of interest: The authors declare no conflict of interest. However, the Department of Clinical Epidemiology is involved in studies with funding from various companies as research grants to, and administered by, Aarhus University. These include grants from the Lundbeck Foundation and H. Lundbeck A/S (a manufacturer of citalopram and escitalopram). These grants have no direct relation to the present study and supported none of the work reported herein.