Chapter  172:  HER2 Testing to Manage Patients With Breast Cancer or Other Solid Tumors

Key Question 1: Concordance and Discrepancy of HER2 Methods

HER2 assay results are influenced by multiple biologic, technical and performance factors. Since many aspects of HER2 assays were standardized only recently, these disparate influences confound the existing literature that compares results of different methods. Discordances between immunohistochemistry (IHC) and fluorescent in situ hybridization (FISH) results might arise in one of three ways. They may be artifacts of one accurate and one inaccurate test or of two inaccurate tests, as preanalytic, analytic, and postanalytic practices can vary among laboratories within a study, as well as among studies. Interobserver variability can play a role. Alternatively, discordances may reflect a threshold issue, either related to changes in threshold definitions over time, or an inherent problem of using a continuous measure to classify patients dichotomously. Finally, discordant test results might accurately reflect a variation among patients with respect to the biologic mechanisms that can increase membrane levels of the HER2 protein. This clearly affects the interpretation of evidence on the use of “HER2 status” to predict treatment or disease outcomes, which presumes accurate classification by tissue assays.

Notably, there is no recognized gold standard to determine the HER2 status of tumor tissue, which also precludes consensus on one “best” HER2 assay. Recent guidelines acknowledge present uncertainty, permit clinicians and laboratories to choose an initial well-validated and properly performed HER2 assay method, and recommend confirming results with an alternative assay when initial tests are equivocal. The ASCO/CAP expert panel (Wolff, Hammond, Schwartz, et al., 2007a) defines equivocal HER2 assay results as IHC 2+, or HER2 gene copy number from 4.0 to 6.0, or HER2/CEP17 ratio from 1.8 to 2.2, if ISH is the first or only assay.

Key Question 2: HER2-Negative or -Discrepant Breast Cancer

Currently available evidence on outcomes of trastuzumab added to chemotherapy for most HER2-equivocal, -discordant, or -negative patients may generate hypotheses, but is too weak to test hypotheses. Most of this evidence is from post-hoc analyses on subgroups not directly randomized or stratified by HER2 status. Scant but intriguing evidence suggests the hypothesis that some patients currently classified as HER2 negative may benefit from adjuvant trastuzumab. Data reported from a post-hoc subgroup analysis of one adjuvant trial (NSABP B31) showed significantly longer DFS and relapse-free interval (RFI) in FISH-negative IHC ≤2+ patients given trastuzumab than in patients managed without trastuzumab, whether the analysis did or did not include those who were IHC 0. However, analysis of data from another similar adjuvant trial (NCCTG N9831) found no significant differences. Both were interim analyses of trials in which fewer than 25 percent of subjects had reached a failure event. Followup analyses from these trials will be of interest.

CALGB 9840 investigators also analyzed a subgroup of metastatic FISH-negative patients that either had (n=38) or did not have (n=103) polysomy 17; overall response rate (ORR) was significantly higher with versus without trastuzumab for those with polysomy 17, but was identical with or without trastuzumab for those without polysomy 17. However, a study in the adjuvant setting (Reinholz, Jenkins, Hillman, et al., 2007) reports no impact of polysomy 17 on benefit from trastuzumab. Additionally, other studies report conflicting data on association of polysomy 17 with overexpression of HER2 protein.

Key Question 3: Breast Cancer Patients Receiving Chemotherapy (3a) or Hormonal Therapy (3b)

For Question 3a, across all three treatment settings (adjuvant, neoadjuvant, or advanced/metastatic), currently available evidence comparing chemotherapy outcomes in HER2-positive and HER2-negative patient subgroups may generate hypotheses, but is too weak to test hypotheses. In the only study that prespecified multivariate subgroup analysis by HER2 status, interaction of assigned adjuvant treatment (with or without paclitaxel) with HER2 status to predict outcome was not statistically significant (ratio of hazard ratios [HRs]=0.85; p=.41). All other evidence is from post-hoc analyses on subgroups not directly randomized, selected, or stratified by HER2 status, and used data from secondary or correlative analysis on patient subgroups with archived tissue samples. It is uncertain whether these subgroups were well balanced. No studies for Question 3a used trastuzumab for HER2-positive patients.

Available evidence focuses on three types of adjuvant chemotherapy: cyclophosphamide plus methotrexate plus fluorouracil (CMF), regimens with an anthracycline, and paclitaxel after or with doxorubicin (Adriamycin®) plus cyclophosphamide (AC). Evidence from two studies (one randomized, controlled trial and one series) suggests HER2-positive patients may benefit less from CMF (smaller improvements in OS and DFS) than HER2-negative patients. Only one of four randomized, controlled trials reports a statistically significant interaction that suggests HER2-positive patients (but not HER2-negative patients) benefit from including an anthracycline in their treatment regimen. Given the highly statistically significant result favoring anthracycline therapy for the entire population (N=14,000) of breast cancer patients included in the Early Breast Cancer Trialists' Collaborative Group (EBCTCG 2005) patient-level meta-analysis, a rigorous test of this hypothesis is necessary before one can conclude that omitting anthracyclines from adjuvant chemotherapy regimens would not worsen outcome for HER2-negative patients.

Two trials compared different doses or frequencies of anthracycline-based regimens. One reported statistically significant interaction of cyclophosphamide, doxorubicin, and fluorouracil (CAF) dose with HER2 status to predict treatment outcome, but the second showed no relationship. One study found that adding paclitaxel after AC improves OS and DFS for HER2-positive patients, but may not improve these outcomes for HER2-negative patients. In contrast, the only randomized, controlled trial with a prespecified multivariate subgroup analysis found no difference by HER2 status in outcomes of concurrently added paclitaxel. Thus, for each of the adjuvant chemotherapy regimens compared, available evidence is too weak to rule out the possibility that HER2-negative patients may benefit from using the added drug or higher dose.

Evidence on whether HER2 status predicts complete response (pCR) to neoadjuvant chemotherapy is limited to four uncontrolled series (retrospective analysis in three). Data are lacking to directly compare any neoadjuvant regimens. There is also limited evidence on differences by HER2 status for outcomes of chemotherapy for advanced or metastatic disease, with most studies lacking statistical power.

For Question 3b, four studies addressed use of tamoxifen in various breast cancer patient populations, and two compared tamoxifen with aromatase inhibitors. None of these studies included trastuzumab. There were no trials that stratified randomization by HER2 status or randomization to therapy directed by HER2 results or not. Less informative designs were used, including post-hoc multivariate analyses in five randomized trials and one post-hoc multivariate analysis in a single-arm study. Data are too weak to reach new conclusions about differences between subgroups based on HER2 status in effects of specific hormone therapies for patients who are hormone-receptor positive.

Key Question 4: Plasma or Serum HER2 (sHER2) in Patients Treated for Breast Cancer

Of 13 included studies, three were randomized trials and 11 were single-arm designs. The evidence is weak on whether sHER2 predicts outcome after treatment with any regimens in any setting. Evidence primarily focused on first-line or second- and subsequent-line treatment of metastatic disease using variety of regimens. Studies used different thresholds for a positive sHER2 result and varied on whether patient selection required positive tissue HER2 status. One randomized and two single-arm studies performed multivariate analysis, although reporting lacked sufficient detail. Univariate analyses provide very limited information value, suggesting candidate variables for future multivariate analyses. Overall, the evidence is too weak to assess whether sHER2 predicts disease progression, treatment response, or outcomes of any specific treatment regimen.

Key Question 5. Serum or Tissue HER2 Testing in Malignancies of Lung, Ovary, Head and Neck, or Prostate

With respect to use of serum or tissue HER2 testing for malignancies of lung, ovary, head and neck, or prostate, the evidence is quite weak. Studies were heterogeneous regarding treatment regimens and thresholds for positive HER2 test results. Of 22 studies addressed for the four types of malignancies, there were no randomized trials that could have analyzed HER2 by treatment effect interactions. Six multivariate analyses in single-arm designs were performed, all of which were poorly described; it is unclear if they were well conducted. Data from these exploratory analyses did not consistently find that HER2 status predicts treatment results. Univariate analyses provide very limited information value, at best suggesting candidate variables for future multivariate analyses.

Types of Participants

For Key Questions 1–4, populations of interest are patients with breast cancer, with separate analyses for early stage patients receiving adjuvant therapy and those undergoing treatment for metastatic disease.

For Key Question 5, populations of interest are patients with cancers of the lung, ovary, prostate, and head and neck.

Types of Outcomes

In general, outcomes should be standard, valid, reliable, and clinically meaningful. Two types of outcomes are relevant to Key Question 1:

• Diagnostic accuracy (e.g., analytic sensitivity, specificity, reliability, etc.);

• Concordance between assay methods; and

Multiple levels of outcomes will be addressed for Key Questions 2 through 5:

• Lead time for detection of progression, recurrence or metastasis.

• Patient management decisions, which may be altered by test results;

• Primary (health) outcomes, which may be affected through management changes guided by test results, such as:

  • Duration of survival, disease-free survival, progression-free survival, and/or time to failure or progression.

  • Quality of life.

  • Palliation of measurable symptoms.

  • Treatment-related adverse effects.

• Secondary (intermediate) outcomes include:

  • Objective clinical response rates (complete and partial responses; separately and summed).

  • Pathologic complete response rates in patients undergoing neoadjuvant therapy followed by surgery.

  • Response durations.

Health outcomes will be given greatest emphasis. However, it will likely be necessary to construct causal pathways to connect assay results to health outcomes through patient management decisions.

Types of Interventions

The interventions of interest for Key Questions 1, 2, 3, and 5 are tissue assays to evaluate tumor HER2 status by:

• Immunohistochemistry;

• Fluorescence in-situ hybridization;

• Chromogenic in-situ hybridization;

• Polymerase chain reaction; or

• Other methods.

The interventions of interest for Key Question 4, and also of interest for parts of Key Question 5, are assays to measure serum concentration of the HER2 extracellular domain.

Practice Settings

Interventions relevant to Key Questions 1–5 are used in the following settings:

• Pathology and laboratory medicine.

• Hospitals.

• Outpatient surgery facilities.

• Office-based practices.

Types of Studies

Following are study selection criteria specific to each key question.

HER2 assay results are influenced by multiple biologic, technical and performance factors. Since many aspects of HER2 assays were not standardized until very recently, we could not isolate effects of these disparate influences on assay results and patient classification.

This challenged the validity of using systematic review methods to compare available assay technologies. For that reason, we provide a narrative review of the following factors influencing HER2 test results and their use to classify patients: biologic processes, assay methods, and sources of variability.

Key Question 2. For patients who are not unequivocally HER2-positive, what is the evidence on outcomes of treatment targeting the HER2 molecule (trastuzumab, etc.), or on differences in outcomes of a common chemotherapy or hormonal therapy regimen with versus without additional treatment targeting the HER2 molecule, in:

• a)

  Breast cancer patients characterized by discrepant HER2 results from different tissue assay methods performed adequately; and

• b)

  For those with HER2-negative breast cancer?

Search Strategy

Electronic databases. The following databases were searched for citations. The full search strategy is displayed in Appendix A ^*. The search was not limited to English-language references; however, foreign-language references without abstracts were disregarded.

The MEDLINE® search was performed through 2/23/07. The EMBASE® search was performed through 2/23/07. The Cochrane Controlled Clinical Trials Register search was performed through 2/23/07. Search updates limited by the Cochrane clinical trial filter were performed for all 3 databases on 4/25/08.

Additional sources of evidence. The Technical Expert Panel and individuals and organizations providing peer review were asked to inform the project team of any studies relevant to the key questions that were not included in the draft list of selected studies.

We also examined the bibliographies of all retrieved articles for citations to any relevant study that was missed in the database searches. In addition, we sought studies published in conference published in conference proceedings and abstracts from the American Association for Clinical Chemistry (AACC), American Society of Clinical Oncology (ASCO), College of American Pathologists (CAP) and the San Antonio Breast Cancer Symposium (SABCS) over the past two years.

Search Screen

Search results were stored in a ProCite® database. Using the study selection criteria for screening titles and abstracts, a single reviewer marked each citation as either: 1) eligible for review as full-text articles; 2) ineligible for full-text review; or 3) uncertain. Citations marked as uncertain were reviewed by a second reviewer and resolved by consensus opinion, with a third reviewer to be consulted if necessary. Using the final study selection criteria, review of full-text articles was conducted in the same fashion to determine inclusion in the systematic review. Of 6,337 citations, 666 articles were retrieved and 70 selected for inclusion (Figure 1). Records of the reason for exclusion for each paper retrieved in full-text, but excluded from the review, were kept in the ProCite® database (see Appendix B, Excluded Studies).

Data Elements

The data elements below were abstracted, or recorded as not reported, from included studies. Data elements to be abstracted were defined in consultation with the Technical Expert Panel.

Data elements from intervention studies (randomized, controlled trials, prospective single-arm studies, and retrospective consecutive series of identically treated patients) were:

• Critical features of the study design (for example, patient inclusion/exclusion criteria, number of subjects, use of blinding)

• Patient characteristics, including:

  • Age

  • Gender

  • Race/ethnicity

  • Disease and stage

  • Disease duration

  • Performance status

  • Other prognostic characteristics (e.g., estrogen or progesterone receptor status)

• HER2 assay techniques (tissue versus serum, IHC, FISH, PCR, ELISA, scoring methods, cutoffs);

• Treatment protocols (for example, regimen, dose, frequency, duration)

• Patient monitoring procedures (for example, followup duration and frequency, outcome assessment methods) and

• The specified key outcomes and data analysis methods (including techniques for assessing associations between HER2 findings and outcomes and methods for assessing treatment effect interactions)

Evidence Tables

Templates for evidence tables were created in Microsoft Excel® and Microsoft Word®. One reviewer performed primary data abstraction of all data elements into the evidence tables, and a second reviewer reviewed articles and evidence tables for accuracy. Disagreements were resolved by discussion, and if necessary, by consultation with a third reviewer. When small differences occurred in quantitative estimates of data from published figures, the values obtained by the two reviewers were averaged.

Evidence Hierarchy

Table 3 shows the framework for evaluating how informative different designs and analytic strategies would be to predictions of outcomes according to HER2 status. The most informative scenario would be a trial in which randomized assignment to treatment groups would be stratified by HER2 status or patients were randomized to receive treatment guided by HER2 results or not (Conley and Taube, 2004). An adequately powered stratified randomization would allow valid inferences of treatment by HER2 interactions. Randomized trials generally are preferred because they convey the possibility of determining differences in the relative efficacy of two treatments, whereas single-arm studies can only assess the association between HER2 status and outcomes after a single treatment regimen. Subgroup analyses in randomized trials should ideally assess the significance of treatment effect interactions. Prespecified subgroups analyses guard against the problems of data dredging.

Post-hoc subgroup analyses may generate hypotheses, but may not support strong inferences about differential effectiveness. Multivariate subgroup analyses in randomized trials may be useful if the subgroup variable introduces imbalances between different variable by treatment combinations, particularly when only a subset of patients have tumor or serum specimens available. An alternative to multivariate subgroup analysis is cross tabulation of treatment by HER2 level results. The weakness of this approach is failure to control for imbalances in any important prognostic factors, particularly if the patients analyzed are a subset of those randomized. A formal test of interaction is preferred for any trial subgroup analysis. In single-arm (identically treated) studies, multivariate analyses may identify whether a variable is a significant independent predictor of treatment outcome while taking into account the separate influences of other predictors. The least informative situation would be a single-arm study that presents univariate comparisons of HER2 groups.

Assessment of Study Quality

As stated, to assess the quality of predictive studies, we adapted the REMARK statement (McShane, Altman, Sauerbrei, et al., 2005). A checklist based on portions of REMARK and other sources (Gould Rothberg, and Bracken, 2006; Altman and Riley, 2005; Altman, 2001a, 2001b; Altman and Lyman, 1998; Brocklehurst and French, 1998; Altman, Lausen, Sauerbrei, et al., 1994; Simon and Altman, 1994) was developed. Table 4 identifies good quality characteristics that we looked for in predictive studies, including: prospective design; prespecified hypotheses about relation of marker to outcome; large, well-defined, representative study population; marker assay methods well-described; blinded assessment of marker in relation to outcome; homogeneous treatment(s), either randomized or rule-based selection; low rate of missing data (≤15 percent); sufficiently long followup; well-described, well-conducted multivariate analysis of outcome. Decision rules for evaluating each quality item are described in the table.

For assessing the quality of randomized trials, the general approach to grading evidence developed by the U.S. Preventive Services Task Force (Harris, Helfand, Woolf, et al., 2001) was applied.

• a

  The quality of randomized, controlled trials will be assessed on the basis of the following criteria:

  • Initial assembly of comparable groups: adequate randomization, including concealment and whether potential confounders (e.g., other concomitant care) were distributed equally among groups.

  • Maintenance of comparable groups (includes attrition, crossovers, adherence, contamination).

  • Important differential loss to followup or overall high loss to followup.

  • Measurements: equal, reliable, and valid (includes masking of outcome assessment).

  • Clear definition of interventions.

  • All important outcomes considered.

  • Analysis: Adjustment for potential confounders, intention-to-treat analysis.

  Definition of ratings based on above criteria:

  • The rating of intervention studies encompasses the three quality categories described here.

  • Good: Meets all criteria: Comparable groups are assembled initially and maintained throughout the study (followup at least 80 percent); reliable and valid measurement instruments are used and applied equally to the groups; interventions are spelled out clearly; all important outcomes are considered; and appropriate attention is given to confounders in analysis. In addition, for randomized, controlled trials, intention to treat analysis is used.

  • Fair: Studies will be graded “fair” if any or all of the following problems occur, without the fatal flaws noted in the “poor” category below: In general, comparable groups are assembled initially but some question remains whether some (although not major) differences occurred with followup; measurement instruments are acceptable (although not the best) and generally applied equally; some but not all important outcomes are considered; and some but not all potential confounders are accounted for. Intention to treat analysis is done for randomized, controlled trials.

  • Poor: Studies will be graded “poor” if any of the following fatal flaws exists: Groups assembled initially are not close to being comparable or maintained throughout the study; unreliable or invalid measurement instruments are used or not applied at all equally among groups (including not masking outcome assessment); and key confounders are given little or no attention. For randomized, controlled trials, intention to treat analysis is lacking.

• b

  The quality of included prospective single-arm intervention studies and retrospective consecutive series of identically treated patients was assessed based on a set of study characteristics proposed by Carey and Boden (2003), as follows:

  • Clearly defined question.

  • Well-described study population.

  • Well-described intervention.

  • Use of validated outcome measures.

  • Appropriate statistical analyses.

  • Well-described results.

  • Discussion and conclusion supported by data.

  • Funding source acknowledged.

Biologic Processes that Influence Cell Membrane Levels of HER2 Protein

Genes such as those in the epidermal growth factor (EGF) receptor family (HER1 through HER4) affect cellular function through the proteins they encode. The HER2 gene is expressed and HER2 protein is found in membranes of all breast and other epithelial cells, and cut-points between “normal” and “overexpressed” levels of HER2 protein are imprecise. Nevertheless, studies have associated increased amounts of HER2 protein in cell membranes with more aggressive behavior of breast and other epithelial cancers and may predict treatment outcomes (Slamon, Clark, Wong, et al., 1987; Esteva, Pusztai, Symmans, et al., 2000; Rowinsky, 2004; Hynes and Lane, 2005; Ettinger, 2006; Serrano-Olvera, Duenas-Gonzalez, Gallardo-Rincon, et al., 2006).

Expression of HER2 and similar genes is a sequential process that (in a simplified overview) includes the following steps: transcription of DNA to messenger RNA (mRNA); processing mRNA to mature, translatable messages; and translation of mature mRNA to synthesize the protein's amino acid sequence. For many proteins (including HER2), additional steps required to produce functional molecules include: post-translational modification (e.g., glycosylation), three-dimensional folding, assembly of multi-subunit proteins, and movement to the relevant cellular site or organelle (not necessarily in this sequence).

We will discuss each of the following biologic mechanisms that potentially may increase the amount of HER2 protein in cell membranes:

• A

  Increased gene copy number (i.e., more than diploid amounts of HER2 DNA in cell nuclei), by:

  • 1

    HER2 gene amplification, or

  • 2

    Chromosome 17 polysomy;

• B

  Elevated HER2 protein levels in cells with diploid amounts of HER2 DNA, by

  • 1

    Increased rate of HER2 gene expression; or

  • 2

    Decreased degradation (increased stability) of HER2 mature message and/or protein.

Tissue Assays Routinely Used in Clinical Practice to Determine HER2 Status of Breast Tumors

In current clinical practice, assays used to classify breast cancer patients with respect to HER2 status detect either HER2 protein or HER2 DNA (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007). Research laboratories use assays for HER2 mRNA to study molecular mechanisms and biologic regulation. They are technically more difficult than protein and DNA assays, and measure less-stable molecules. Although real-time reverse transcription polymerase chain reaction (RT-PCR) methods recently were adapted to measure HER2 mRNA in fixed, paraffin-embedded tissues and compared with IHC and ISH assays (Capizzi, Gruppioni, Grigioni, et al., 2008), RT-PCR assays for HER2 mRNA are still uncommon in clinical management of patients with breast cancer and thus are not included in this review.

Each method used to determine HER2 status applies results of a quantitative or semiquantitative assay to assign a binary (“yes/no”) classification. Thus, test results with each assay can vary with different scoring systems and thresholds for positivity. As discussed in a following section (“Postanalytic Factors”), scoring and thresholds may depend on choice of reagents to detect, visualize, and quantitate analytes. Scoring systems and thresholds also have changed over time, with standardized approaches recommended quite recently (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007). Data are lacking to determine whether differences in treatment outcome as a function of HER2 status are affected by reclassifying patients with currently recommended scoring systems and thresholds.

Methods to detect/measure amount of HER2 protein. Immunohistochemistry (IHC) is the assay used most widely for classifying HER2 status of breast cancer patients, since it uses techniques and equipment long used by most clinical pathology laboratories for other proteins such as estrogen and progesterone receptors (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007; Laudadio, Quigley, Tubbs, et al., 2007; Ross, Fletcher, Linette, et al., 2003). The assay incubates thin slices of fixed tissue on a microscope slide with an antibody to HER2, washes off unbound antibody, then visualizes bound antibody. Because IHC preserves tissue architecture and cellular structure (morphology), it permits scoring to focus on antibody specifically bound to membranes of invasive breast cancer cells. IHC also permits permanent storage of stained slides if later re-evaluation is needed.

IHC scoring systems consider the proportion of antibody-stained invasive cancer cells and the intensity of staining, a partly subjective judgment. Besides the U.S. Food and Drug Administration (FDA) -approved IHC kits (HercepTest™ and PATHWAY™; see Table 2, Introduction), various antibodies to HER2 protein are commercially available as analyte-specific reagents that can be used for independently developed (so-called “home-brew”) assays (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007; Laudadio, Quigley, Tubbs, et al., 2007; Ross, Fletcher, Linette, et al., 2003; Hicks and Kulkarni, 2008). Some are polyclonal, with a mix of antibody molecules that may recognize different binding site (epitopes) on the HER2 protein. Others are monoclonal, homogeneous molecules that recognize a single epitope. These differences may lead to discrepant results with different antibody reagents (Press, Hung, Godolphin, et al., 1994). Other sources of variability in IHC results are discussed in the following section, “Sources of Variability in Classifying HER2 Status.”

Protein assays on homogenized tissue may use antibody to visualize HER2 after separating proteins in a solid matrix (Western blots), or quantitate HER2 by enzyme-linked immunosorbent assay (ELISA). These assays destroy the analyzed tissue samples. Additionally, tissue extracts may mix proteins from cytosol, membranes, and other organelles; and also from multiple cell types: normal breast, inflammatory cells, in situ tumor, and invasive cancer. HER2 levels of in situ breast tumor cells often are elevated, for uncertain reasons and with inadequately studied clinical implications (Allred, Clark, Tandon, et al., 1992; Hoque, Sneige, Sahin, et al., 2002; Collins and Schnitt, 2005). Guidelines stress avoiding areas of ductal carcinoma in situ when scoring assay results (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007). Nearly all clinical studies on HER2 protein assays to predict treatment outcomes used IHC on tissue slices rather than assays on tissue homogenates, and assigned HER2 status by amount of HER2 protein in membranes of invasive breast cancer cells.

Methods to detect/measure HER2 gene copy number or amount of HER2 DNA. In situ hybridization (ISH) is the most commonly used method to measure HER2 gene copy number in tissue samples from breast cancer patients (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007; Ross, Fletcher, Linette, et al., 2003; Hicks and Kulkarni, 2008). It uses a labeled probe complementary to the DNA sequence of interest (here, a unique segment from the HER2 gene). Double-stranded DNA in cell nuclei of the fixed tissue sample is denatured so the probe can hybridize (bind) to its complementary sequence, then unbound probe is washed away. As with IHC, tissue preparation for ISH preserves tissue and cell morphology, and scoring focuses on invasive breast cancer cells.

The gene-specific probes are visualized in one of three ways: by fluorescence (FISH), a chromogenic reaction (CISH; uses digoxigenin), or silver deposition (SISH; uses dinitrophenol for enzymatic metallography). FISH requires a fluorescence microscope (more expensive and unavailable in some smaller pathology laboratories), while CISH and SISH use routine light (brightfield) microscopy. Three FDA-approved kits are available for HER2 testing by FISH (PathVysion®, Inform™, and HER2 FISH pharmDx™), while kits for CISH (SPoT-Light) and SISH (EnzMet GenePro™) are not yet approved (see Table 2, Introduction). Slides prepared for FISH testing lose fluorescence, thus, cannot be stored for later review. In contrast, slides prepared for either CISH or SISH can be archived and re-evaluated. Additionally, it is sometimes difficult to identify invasive tumor cells with fluorescence microscopy. All three ISH methods require more time per sample than IHC for slide scoring. Because they were developed recently, fewer clinical studies used CISH or SISH than FISH to classify HER2 status of breast cancer patients.

In ISH assays, pathologists count fluorescent (FISH) or dark-colored (CISH, SISH) spots visible above the nucleus to measure HER2 gene copy number: two in diploid cells; more in cells with amplified HER2 or polysomy 17. Typically, one determines gene copy number for multiple invasive cancer cells on the slide, and averages results for the tissue sample. In some ISH assays, slides are hybridized simultaneously with two probes that fluoresce in or show different colors, to permit copy number measurement for the HER2 gene and chromosome 17 centromere (CEP17). With this approach, HER2 gene status is defined by the ratio of HER2 to CEP 17 copy numbers: greater than 2 if amplified, but approximately 2 if unamplified whether chromosome 17 polysomy is absent or present.

Early research studies extracted DNA from tissue homogenates and measured amounts of the HER2 gene by Southern or slot blots, or by quantitative polymerase chain reaction (PCR) assays. Southern blots first separate DNA molecules by their mobility in a matrix, while slot blots use the mixed extract. Each selectively visualizes the DNA sequence of interest by hybridizing to labeled probes as in ISH. PCR assays amplify (selectively replicate) DNA sequences of interest in vitro, detect them by fluorescent or other probes, and quantify the starting amount using standard curves. As with protein assays on tissue homogenates, these techniques dilute DNA from invasive cancer cells with DNA from surrounding normal tissues and inflammatory cells (Laudadio, Quigley, Tubbs, et al., 2007; Ross, Fletcher, Linette, et al., 2003). They also consume the samples they analyze. Southern and slot blots are less sensitive than PCR and require substantially larger amounts of DNA. Southern blot assays also are labor intensive and less widely available in clinical pathology labs. The remainder of this review focuses on IHC and ISH methods, the only HER2 assays with FDA-approved kits available for clinical use.

Sources of Variability in Classifying HER2 Status

Accurately determining HER2 status depends on proper performance of preanalytic, analytic, and postanalytic steps (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hicks and Kulkarni, 2008; Hanna, O'Malley, Barnes, et al., 2007; Laudadio, Quigley, Tubbs, et al., 2007; Ross, Fletcher, Linette, et al., 2003). Preanalytic steps are those involved in obtaining, preserving (fixing), and storing tissue samples prior to staining and analysis. Analytic steps prepare and stain fixed tissue samples with antibody to HER2 for IHC, or prepare and hybridize them to HER2 gene probe for ISH, then visualize tissue-bound antibody or probe. Postanalytic steps score test results, classify patients, and assure test quality, consistency, and reproducibility. Some processes for these steps are the same for IHC or ISH, but many differ.

Preanalytic: tissue processing and storage. HER2 tests can use tissue from core (incisional) biopsy or tumor excised for biopsy, lumpectomy, or mastectomy (Wolff, Hammond, Schwartz, et al., 2007a). Tissue sources can be the primary tumor or a lymph node or distant metastasis (Carlson, Moench, Hammond, et al., 2006). While uncommon, studies have reported discordances in HER2 status between primary tumor and metastases (for references, see Carlson, Moench, Hammond, et al., 2006). Retesting HER2 status if metastases develop after a long disease-free or progression-free interval may be warranted, depending on where and how HER2 status of the primary tumor was determined.

Tissues are prepared and preserved for assays by slicing larger samples, fixing in a denaturing solution, and embedding fixed tissue for long-term storage (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007). Factors that may influence test results include: edge, retraction, or crush artifacts with some core needle biopsies; time from excision to slicing, and to fixation; type of and time in fixative; choice of embedding material; and conditions and duration of storage for fixed and embedded tissues.

Guidelines seeking to standardize methods were not published until recently (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007), although prior consensus conferences (cited in the guidelines) recommended many of the same methods. Importantly, the recommended preanalytic steps are identical for tissues to be tested by IHC or ISH; these are summarized in a following section (see Table 5 in “Current Guideline Recommendations”). Systematic reviews conducted for the American Society of Clinical Oncology/College of American Pathologists (ASCO/CAP) and National Comprehensive Cancer Network (NCCN) guidelines (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006) reported data were lacking to evaluate effects of nonadherence on test results for some aspects of tissue processing. The published guidelines did not include evidence tables summarizing effects of nonadherence on test results for those aspects of tissue processing that have been evaluated comparatively.

Notably, the literature review for this report showed that most studies reporting concordance and discordance rates of different IHC and ISH assays used archived samples, fixed and embedded elsewhere than the laboratory performing the HER2 assays. With exceptions, most publications did not report adequately on adherence to guideline or prior (consensus) recommendations for tissue processing.

Analytic: performing HER2 assays. Analytic steps for processing thin sections of fixed and embedded tissue cut onto glass slides differ for IHC and ISH assays (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007). Each begins by deparaffinizing thin tissue sections, but IHC assays use an antigen retrieval step that optimizes antibody binding to HER2 protein while ISH assays first unwind (denature) cells' double-stranded DNA so that the probe can hybridize to its complementary sequence. The temperature and duration of heating used to bake tissue sections on slides, as well as the conditions used for antigen retrieval, can introduce variability in IHC results. Each assay incubates slides with an analytic reagent (antibody for IHC; probe for ISH), removes unbound reagent in one or more washing steps, and incubates with other reactants to visualize bound analytic reagent. Some steps can be automated, which improves consistency and reproducibility if equipment is well-maintained and regularly calibrated. In addition to reagent choice (which antibody, for IHC; which DNA probe, for ISH), varying the conditions (temperatures, durations, etc.), solutions, and reactants used for each step can affect test results, as can poorly maintained or calibrated automated equipment.

While FDA-approved kits include protocols with optimized methods for each analytic step, guideline publications report that approximately half of surveyed laboratories did not adhere completely to protocol methods (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006). The guidelines stress the need to train and periodically assess the skills of staff conducting these assays, and that each run should include standardized positive and negative controls. They also emphasize that each laboratory offering HER2 testing services should validate its test results against a previously validated test, and that laboratories departing from protocol-specified methods with FDA-approved kits, and those using independently developed assays with analyte-specific reagents, should validate test results against established methods and develop their own standard protocols.

As with preanalytic steps, most published studies did not adequately report information needed to evaluate complete adherence with guideline or prior (consensus) recommendations on all analytic steps. Studies that used FDA-approved kits rarely commented on protocol adherence in the methods sections of their reports, and studies that used independently developed assays rarely described assay validation against approved kits.

Postanalytic factors. IHC scoring systems and positivity thresholds have changed over time, and these changes likely alter the proportion of patients classified as HER2 positive (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hicks and Kulkarni, 2008; Hanna, O'Malley, Barnes, et al., 2007; Laudadio, Quigley, Tubbs, et al., 2007; Ross, Fletcher, Linette, et al., 2003). Some studies on archived tissues classified tumors as HER2 positive if any invasive cells showed strong, complete membrane staining (e.g., Paik, Bryant, Park, et al., 1998; Houston, Plunkett, Barnes, et al., 1999; Paik, Bryant, Tan-Chiu, et al., 2000). Others classified samples as HER2 positive if 1 percent or more of invasive cells were stained (e.g., MacGrogan, Mauriac, Durand, et al., 1996; Elledge, Green, Ciocca, et al., 1998; Di Leo, Larsimont, Gancberg, et al., 2001); yet others, only if 50 percent or more were stained (e.g., Agrup, Stal, Olsen, et al., 2000; Berry, Muss, Thor, et al., 2000; Colozza, Sidoni, Mosconi, et al., 2005). Few studies adopted (or adapted) Allred's system (Harvey, Clark, Osborne, et al., 1999; developed for IHC assays of estrogen receptors), which rates the proportion of stained invasive cells (from 0 to 5) and the intensity of staining (from 0 to 3), then adds for a final score between 0 and 8.

The scale recommended in FDA-approved IHC kits (0 to 3+; developed for HercepTest™ but also used with PATHWAY™) requires membrane staining in 10 percent or more of invasive cells for scores greater than 0. The scale assigns positive scores by staining intensity and totality of membrane staining: 1+ is faint or barely perceptible staining that is incompletely circumferential; 2+ is moderate intensity but complete circumferential staining; and 3+ is strong intensity and complete circumferential staining (www.dakousa.com/prod_downloadpackageinsert.pdf?objectid_105073003). However, some studies that used this scale defined HER2-positive cases as those scored 2+ or 3+, while others classified only those with a score of 3+ as HER2 positive. The ASCO/CAP guideline retains the original definitions for scores of 0 to 2+, but recommends scoring IHC 3+ only if more than 30 percent of invasive breast cancer cells show dark, homogeneous, circumferential membrane staining in a “chicken wire” pattern (Wolff, Hammond, Schwartz, et al., 2007a). Adequate data are lacking to compare accuracy or concordance for this wide variety of scoring systems and thresholds used to classify patients' HER2 status by IHC alone. However, in one recent study (Hameed, Chhieng, and Adams, 2007), three pathologists blinded to FISH results scored IHC-stained slides from 98 breast cancer cases separately using cut-offs of 10 percent, 30 percent, and 50 percent of stained cells to classify samples as HER2+. Specificity of IHC versus FISH was 82 percent, 86 percent, and 87 percent, respectively, for the three increasing cut-offs, while concordance rates of 3+ cases with FISH were 59 percent, 64 percent, and 65 percent.

Scoring and categorizing results of ISH assays also varies (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hicks and Kulkarni, 2008; Hanna, O'Malley, Barnes, et al., 2007; Laudadio, Quigley, Tubbs, et al., 2007; Ross, Fletcher, Linette, et al., 2003). Guidelines stress that precision and accuracy depend on the number of cells counted and averaged, on accurately identifying and only counting invasive cells, and on counting invasive cells from two or more separate areas of each tumor on either the same or sequential slide(s) (Wolff, Hammond, Schwartz, et al., 2007a). With assays estimating gene copy number per cell without normalizing to a CEP17 probe, most published studies using FISH classified tissues averaging more than 4.0 copies per cell as HER2 positive (for references, see Wolff, Hammond, Schwartz, et al., 2007; Carlson, Moench, Hammond, et al., 2006; Laudadio, Quigley, Tubbs, et al., 2007a). Most published studies using CISH scored samples HER2 positive if the average gene copy number per cell was greater than 5, although some followed the manufacturer's recommendation and defined low-level amplification as copy numbers between 6 and 10. In contrast to published studies with FISH or CISH, recent guidelines consider average scores greater than 6.0 as FISH positive, scores less than 4.0 as FISH negative, and scores between 4.0 and 6.0 as equivocal (ASCO/CAP) or borderline (NCCN). Most studies that normalized to CEP17 classified HER2 to CEP17 ratios greater than 2.0 as HER2 positive (for references, see Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Laudadio, Quigley, Tubbs, et al., 2007). The guidelines consider a HER2/CEP17 ratio greater than 2.2 as positive, a ratio less than 1.8 as negative, and ratios between 1.8 and 2.2 as equivocal (ASCO/CAP) or borderline (NCCN). As with IHC scoring and thresholds, data are lacking to evaluate consequences of the newer classification criteria on accuracy or concordance.

Guidelines and reviews caution that assigning HER2 status is partially subjective and potentially inconsistent because IHC and FISH scoring criteria are variably interpreted and applied by different raters (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hicks and Kulkarni, 2008; Hanna, O'Malley, Barnes, et al., 2007; Laudadio, Quigley, Tubbs, et al., 2007; Ross, Fletcher, Linette, et al., 2003). Expert panels and reviewers emphasize that image analysis methods, using digital microscopy and automated cellular imaging systems (e.g., Bloom and Harrington, 2004; McCabe, Dolled-Filhart, Camp, et al., 2005; Tubbs, Pettay, Swain, et al., 2006; Ciampa, Xu, Ayata, et al., 2006; Tawfik, Kimler, Davis, et al., 2006; Moeder, Giltnane, Harigopal, et al., 2007), can decrease inter-rater variability and thus improve scoring consistency, accuracy, and precision, particularly for IHC assays. However, this requires careful validation and periodic recalibration of automated systems against standardized positive, negative, and equivocal control samples. Nevertheless, a study testing agreement between pathologists reported that use of digital microscopy to score IHC improved concordance with FISH and also decreased inter-rater variability (Bloom and Harrington, 2004).

Postanalytic steps also include reporting elements that should be provided to clinicians ordering HER2 testing, as well as quality assurance procedures (laboratory accreditation and proficiency testing; competency assessment for pathologists). However, these issues are outside the scope of this report. Readers are referred to recommendations in current guidelines (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007).

Is There a “Best” Method to Determine HER2 Status from Breast Tumor Tissue?

Although many studies reported concordance and discrepancy rates for collections of breast tumor tissue tested for HER2 status by IHC with different antibodies, or by IHC and ISH assays, or by multiple ISH assays, current evidence does not suggest one HER2 assay is superior to all others (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hicks and Kulkarni, 2008; Laudadio, Quigley, Tubbs, et al., 2007; Ross, Fletcher, Linette, et al., 2003). As described previously, preanalytic, analytic and postanalytic methods varied between studies, and all studies preceded guidelines for standardizing these methods. Additionally, data are lacking to fully evaluate effects of nonadherence with certain guideline recommendations on test results. Thus, it is difficult (perhaps impossible) to isolate effects of individual factors that contribute to discordance. As detailed above, these include differences in:

• Fixing and embedding tissues, preparing and staining them for assays, or scoring and classifying test results;

• Inherent differences in antibody binding, epitope stability, or antigen retrieval when comparing different antibodies used for IHC;

• Different biologic mechanisms that can increase membrane HER2 protein, when comparing IHC assays versus ISH assays; or differences in sensitivity and specificity of diverse DNA probes and visualization techniques when comparing different ISH methods.

Identifying one “best” HER2 test clearly requires better comparative data than presently available, with assays that standardized key aspects of preanalytic, analytic, and postanalytic steps in HER2 assay methods.

The lack of a gold standard to determine breast tumors' HER2 status also prevents agreement on one “best” HER2 assay. Furthermore, seeking a single gold standard may be unrealistic, since HER2 status is used in different ways. The optimal assay (or combination of assays) may differ for HER2 as a prognostic marker, as a marker to predict clinical benefit from trastuzumab, or as a marker to predict benefit from a chemotherapy drug class (e.g., an anthracycline or a taxane). For example, HER2 gene amplification may best predict tumor aggressiveness hence prognosis, while membrane density of HER2 protein may best predict trastuzumab binding to tumor cells and thus clinical response. Furthermore, HER2 may only be a surrogate marker for other molecular alterations that more directly impact tumor cell sensitivity to certain chemotherapy drugs (e.g., anthracyclines).

Outcomes of well-designed and adequately powered comparative clinical trials with sufficient followup duration may be a gold standard to evaluate HER2 assays as predictors of treatment benefit. However, even the large randomized, controlled trials on adjuvant trastuzumab (Romond, Perez, Bryant, et al., 2005; Piccart-Gebhart, Procter, Leyland-Jones, et al., 2005; Slamon, Eiermann, Robert, et al., 2005; Joensuu, Kellokumpu-Lehtinen, Bono, et al., 2006) may not have adequately standardized preanalytic steps at local hospitals, did not test all patients with at least two assays, treated few patients with discordant results by different assays conducted in central laboratories; and presently lack sufficient followup to compare outcomes in subgroups of the main treatment arms (see “Results and Conclusions, Key Question 2”).

Current guidelines acknowledge present uncertainty, permit clinicians and laboratories to choose an initial HER2 assay method, and recommend confirming results with an alternative assay when initial tests are equivocal (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007).

Current Guideline Recommendations

Current guidelines recommend very similar algorithms for using well-validated IHC and ISH assays to classify breast cancer patients with respect to HER2 status (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007). The algorithm shown in Figure 2, describes possible results, decision-making, and confirmatory testing when IHC is the initial test. All three guidelines agree that an IHC score of 3+ is definitively HER2 positive, a score of 0 or 1+ is definitively HER2 negative, and a score of 2+ is equivocal and requires ISH followup testing to determine HER2 status. In contrast to the other guidelines, the NCCN Task Force (Carlson, Moench, Hammond, et al., 2006) did not specify that an IHC 3+ score requires complete membrane staining in more than 30 percent of invasive cells. The ASCO/CAP expert panel recommended this change from FDA labeling (which requires staining in more than 10 percent of invasive cells), primarily to decrease the number of patients with false-positive results who might be given trastuzumab but are unlikely to benefit (Wolff, Hammond, Schwartz, et al., 2007a). This recommendation anticipates that true positives with equivocal IHC results will be correctly classified by followup ISH. However, data are currently lacking to test this hypothesis.

Figure 3 provides a similar algorithm if FISH is the initial test. The guidelines suggest that well-validated alternatives (CISH or SISH, currently available in the U.S. only as independently developed assays) probably can replace FISH. The algorithm considers HER2 gene copy numbers from 4.0 to 6.0 or HER2/CEP17 ratios between 1.8 and 2.2 as equivocal ISH results. It recommends additional cell counting, retesting by a reference laboratory, or followup testing by IHC before classifying equivocal cases. The other guidelines agree with this recommendation (Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007). No studies reviewed for this report followed this recommendation; thus, data are lacking to determine whether confirmatory followup testing on patients with equivocal ISH results improves the accuracy of HER2 status as a predictor for treatment outcomes.

Importantly, the guidelines' treatment recommendations are not identical for all patients whose assay results remain in the equivocal range after additional cells are counted, a different assay method is used, and/or testing is repeated on another tumor section. The recommendation depends on whether the patient would have been included in or excluded from key randomized, controlled trials. For example, patients with HER2/CEP17 ratios 2.0 or greater but less than 2.2 were included and randomized in the adjuvant trastuzumab trials. Therefore, the guidelines view current evidence as too weak to deny such patients adjuvant therapy that includes trastuzumab. In contrast, patients with HER2/CEP17 ratios 1.8 or greater but less than 2.0 were excluded from these trials, and the guidelines view current evidence as too weak to support including trastuzumab in their adjuvant therapy regimens. Figures 2 and 3 include information on trial eligibility of patients whose test results are equivocal by each HER2 assay.

Interestingly, a recent study reported on 17 patients with breast core biopsy specimens showing invasive carcinoma and equivocal FISH results (HER2/CEP17 ratios between 1.8 and 2.2) (Striebel, Bhargava, Horbinski, et al., 2008). These patients were subsequently re-evaluated by IHC and FISH testing on resection specimens. For 10 of the 17 cases, equivocal results obtained with biopsy specimens were definitively resolved by retesting of resection specimens. Four patients were classified HER2 positive and treated with trastuzumab, while six were classified HER2 negative and managed without trastuzumab.

Other recommendations in the ASCO/CAP guideline focus on good laboratory practices for each preanalytic, analytic, and postanalytic step of IHC and ISH assays (Wolff, Hammond, Schwartz, et al., 2007a). They provide a more explicitly detailed set of recommendations than included in the other two guidelines (Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007). Table 5 reprints the summary of recommendations from the ASCO/CAP guideline. The remainder of this narrative review for Key Question 1 summarizes evidence published after these guidelines on the following four topics, and discusses unresolved issues and uncertainties:

• Concordance and discordance of different assay methods

• Discordance between central and local laboratory results

• Validation and proficiency testing

• Reports on polysomy 17

Evidence Reported Post-ASCO/CAP Guidelines on Concordance and Discrepancy of HER2 Assay Results

Concordance/discordance of different assay methods. Evidence reviewed by the ASCO/CAP expert panel (Appendixes C and G in Wolff, Hammond, Schwartz, et al., 2007a) led to consensus definitions for unequivocal IHC and ISH results. As shown in Figures 2 and 3, and in Table 5, the panel defined unequivocal HER2-positive results by IHC (i.e., 3+) as greater than 30 percent of invasive cells strongly stained in a homogeneous, circumferential “chicken-wire” pattern, and by ISH as HER2 gene copy number per cell greater than 6 or HER2/CEP17 ratio greater than 2.2. They defined unequivocal HER2-negative results by IHC as scores of 0 or 1+, and by ISH as HER2 gene copy number per cell less than 4.0 or HER2/CEP17 ratio less than 1.8. Equivocal results (defined as 2+ by IHC, HER2 gene copy number from 4.0 to 6.0, or HER2/CEP17 ratio from 1.8 to 2.2) probably imply low-level HER2 amplification and/or overexpression, and should not be considered discordant, whether results of followup testing are positive or negative. Some but not all of these samples may actually have an amplified HER2 gene, but require additional testing to define the patient's correct HER2 status. The ASCO/CAP expert panel found insufficient evidence to determine whether breast cancer patients with equivocal HER2 results benefit from HER2-targeted therapy, although as discussed above, some patients included in adjuvant trastuzumab trials fit this category (also see “Results and Conclusions, Key Question 2”).

For purposes of this review, discordant results are operationally defined as unequivocally positive results by one assay method and unequivocally negative results by a different assay method on sections from the same tumor, with both assays conducted using good laboratory practices, as recommended in the ASCO/CAP guideline (Wolff, Hammond, Schwartz, et al., 2007a). Presently, evidence is lacking to estimate discordance rates from studies that followed all ASCO/CAP recommendations on tissue preparation, testing practices, scoring systems, and thresholds to classify HER2 status of breast cancer patients. Therefore, in the following sections, we summarize evidence on discordance rates reported after the guideline was published by studies that used scoring systems and thresholds similar to those originally specified in U.S. Food and Drug Administration (FDA) -approved kits for IHC and ISH assays.

Investigators from the National Surgical Adjuvant Breast and Bowel Project's (NSABP) central pathology laboratory and colleagues at NSABP-approved reference laboratories conducted IHC (HercepTest™) and FISH (PathVysion®) assays on formalin fixed, paraffin embedded tumor blocks (Paik, Kim, Jeong, et al., 2007; Paik, Kim, and Wolmark, 2008). They reported results with both assays for 1,787 of 2,043 patients enrolled in the NSABP B31 randomized, controlled trial on adjuvant therapy with versus without trastuzumab (Romond, Perez, Bryant, et al., 2005). Of these, they found FISH-negative, IHC 3+ discordant results in 31 cases (1.7 percent). They also reported FISH-positive, IHC 0, 1+, or 2+ results in another 125 cases (7 percent), but did not separately report the proportion of those who tested FISH positive and IHC 0 or 1+.

Central and reference laboratory results with both IHC (HercepTest™) and FISH (PathVysion®) assays also are available (Perez, Romond, Suman, et al., 2007) for 1,779 of the 2,535 patients registered in a similar randomized, controlled trial conducted by the North Central Cancer Treatment Group (NCCTG N9831; Romond, Perez, Bryant, et al., 2005). Investigators reported discordant IHC 3+, FISH-negative results in 53 cases (3 percent), and FISH-positive, IHC 0, 1+, or 2+ results in 218 cases (12.3 percent). Here again, separate results were not reported for the proportion who tested FISH positive and IHC 0 or 1+. Data presently are unavailable on IHC/ISH discordance rates from three other randomized, controlled trials of adjuvant trastuzumab (Piccart-Gebhart, Procter, Leyland-Jones, et al., 2005; Slamon, Eiermann, Robert, et al., 2005; Joensuu, Kellokumpu-Lehtinen, Bono, et al., 2006).

In a retrospective study, a Canadian central reference laboratory used HercepTest™ and three other HER2 antibody IHC assays to retest tumors from patients diagnosed with metastatic breast cancer between 1999 and 2002, and compared the IHC results with central lab FISH using PathVysion® (O'Malley, Thomson, Julian, et al., 2008). Among 505 patients initially classified HER2 positive by IHC in local labs and treated with trastuzumab for metastatic disease, concordance between central IHC and central FISH ranged from 88.9 percent to 90.9 percent, depending on the HER2 antibody used. Concordance between IHC and FISH was highest (92.2 percent) when all four HER2 antibody assays were used to test each sample, and tumors were only classified IHC positive if positive by 2 or more assays. In a sequential sample of 205 invasive breast tumors locally classified IHC negative, from patients diagnosed with metastasis, concordance of central IHC and central FISH ranged from 93.7 percent to 99 percent for individual antibody assays, and was 98.1 percent if tumors were only classified IHC negative if negative by 2 or more assays. However, this study did not report FISH/IHC discordance rates separately by IHC score.

A study from Greece that separately compared IHC results (using HercepTest™ and two other methods) from central and regional laboratories versus central FISH (PathVysion®) reported on 375 breast tumors tested centrally by IHC and FISH (Papadopoulos, Kouvatseas, Skarlos, et al., 2007). FISH-positive, IHC 0/1+ discordances were seen in six cases (1.6 percent; 11.5 percent of 52 IHC 0/1+ cases), while FISH-negative, IHC 3+ discordances were seen in three cases (0.8 percent; 9.4 percent of 32 IHC 3+ cases). Another study from three Greek hospitals compared IHC results (CB11 antibody) with FISH (PathVysion®) for 194 resected breast cancer patients, and also with CISH (SpoT-Light) for 159 of these patients (Kostopoulou, Vageli, Kaisaridou, et al., 2007). This study reported no FISH-positive cases and only one CISH-positive case among 94 IHC 0/1+ patients. Of 30 patients with IHC 3+ results, one (3.3 percent) was FISH negative and CISH negative.

A study from Germany on patients evaluated for inclusion in a trial of trastuzumab for metastatic breast cancer reported central IHC (HercepTest™) and FISH (PathVysion®) results for 289 patients (Hofmann, Stoss, Gaiser, et al., 2008). Investigators reported no FISH-positive cases among 100 patients scored IHC 0/1+, and nine FISH-negative but IHC 3+ cases (8.4 percent of 107 scored IHC positive; 3.1 percent of all patients evaluated).

A small study (n=55) compared two dual-probe (i.e., for HER2 and CEP17) FISH kits (PathVysion® and HER2 FISH pharmDx), a single-probe FISH kit (Inform; HER2 only) and the SpoT-Light CISH kit versus two IHC assays (HercepTest™ and an independently developed test) (Cayre, Mishellany, Lagarde, et al., 2007). Investigators reported results with each assay (and with different positivity thresholds for Inform and SpoT-Light) separately for each sample. Four of 55 (7.3 percent) cases tested IHC 3+ with HercepTest™ and ISH-negative by all assays (other than a threshold of more than four signals for Inform). Three of the four were scored less than 3+ by independently developed IHC. All cases scored FISH positive by two or more kits also were scored IHC 3+ by HercepTest™.

Another small study (n=54) used the HercepTest™ and PathVysion® kits on all samples (Kuo, Wang, Chang, et al., 2007). Three cases (5.6 percent) that tested FISH negative were scored 3+ by IHC. In contrast, no cases that tested FISH positive were scored IHC 0 or 1+.

A systematic review abstracted data from 17 studies (all published before the ASCO/CAP guideline; pooled N=8,419) on FISH/IHC concordance (Dendukuri, Khetani, McIsaac, et al., 2007). Selection criteria sought studies that included consecutive patient series or a random sample, reported agreement between IHC and FISH using standard thresholds, and used assays licensed in Canada to select patients for trastuzumab therapy. All studies used PathVysion® for FISH; 16 used HercepTest™ and one used PATHWAY™ for IHC. Ten combined results for patients scored IHC 0 or 1+, and separately for those scored IHC 2+ or 3+ (pooled N=4,641); seven reported results separately for each IHC score (pooled N=3,778). Using Bayesian meta-analysis, they estimated proportions of breast cancer patients with each of the four possible IHC scores and proportions with each IHC score with positive results by FISH. Table 6 summarizes estimated IHC/FISH discordance rates based on results of the Dendukuri and coworkers' meta-analysis.

Three small studies (combined N=211) conducted outside North America compared results of different ISH methods. An Australian study on 49 breast cancer samples reported that each case (n=20) scored highly positive (greater than 10 signals/cell) by FISH, and seven of 10 cases scored low-positive (5–10 signals/cell) by FISH, also scored positive by CISH (Bilous, Morey, Armes, et al., 2006). Each sample scored IHC 3+ by HercepTest™ also tested CISH positive. A study from Germany reported agreement in 95 of 99 breast tumor samples tested by FISH (PathVysion®) and SISH, an overall concordance of 96 percent (Dietel, Ellis, Hofler, et al., 2007). Finally, a study from Poland compared FISH, CISH, and SISH on 63 breast tumor specimens selected for 2+ or 3+ staining by IHC (Sinczak-Kuta, Tomaszewska, Rudnicka-Sosin, et al., 2007). Investigators reported and interpreted multiple statistical tests (Pearson chi-square tests with p<0.01; gamma correlation coefficients of 0.89 to 0.96; Spearman rank correlation coefficients of 0.70 to 0.79; and Kappa coefficients of 0.38 to 0.58) for separate two-way comparisons of assay results (i.e., CISH versus FISH, FISH versus SISH, and SISH versus CISH) as evidence for good agreement between the methods, but did not report concordance or discordance rates. Larger studies are needed to estimate more reliably rates of concordance and discordance between FISH or IHC and newer ISH methods (CISH, SISH). Furthermore, FDA-approved kits for CISH or SISH are not yet available.

To summarize, evidence from seven studies and a meta-analysis reported after the ASCO/CAP guideline (Wolff, Hammond, Schwartz, et al., 2007a) suggests variable but perhaps non-negligible rates for FISH-negative, IHC 3+ discordance (albeit by the older definition of strong, complete membrane staining in greater than 10 percent of invasive cells), ranging from 0.5 percent to 7.3 percent of breast cancer cases. The meta-analysis also estimated that 0.6 percent (95 percent CI: 0.1–1.3 percent) of cases might be scored IHC 0 and FISH positive, while 1.8 percent (95 percent CI: 0.8–3.0 percent) of cases might be scored IHC 1+ and FISH positive. However, data are unavailable to estimate discordance rates for either group using the current ASCO/CAP definition of IHC 3+ (greater than 30 percent of invasive cells stained).

Disagreement between central and local laboratory results. Evidence reviewed by the ASCO/CAP expert panel demonstrated disagreement between central and local laboratory HER2 test results in approximately 20 percent of cases (Wolff, Hammond, Schwartz, et al., 2007a). This included data from the first 104 patients registered for NSABP B31, showing disagreement in 18 percent of cases (Paik, Bryant, Tan-Chiu, et al., 2002), which resulted in a protocol amendment limiting HER2 testing to 23 approved laboratories. The evidence also included data from NCCTG N9831 showing agreement in 88.1 percent of 813 cases rated FISH positive, 81.6 percent of 1,063 cases scored IHC 3+ by HercepTest™, and 75.0 percent of 636 cases scored IHC 3+ by non-HercepTest™ assays (Perez, Suman, Davidson, et al., 2006). Finally, it included data from a community-based clinical study on trastuzumab for metastatic breast cancer showing 77 percent agreement on samples scored IHC 3+ by local laboratories, but only 26 percent agreement on samples locally scored IHC 2+ (Reddy, Reimann, Anderson, et al., 2006). Based on the available evidence, the panel recommended specific measures for assay validation, self-assessment, accreditation, and proficiency testing by laboratories conducting HER2 assays. In the following section, we summarize new evidence comparing local versus central laboratory results, published since the ASCO/CAP review. Although published after the ASCO/CAP guideline, these studies preceded the guideline and scored samples as originally recommended by manufacturers and FDA labeling.

Final data from NSABP B31 showed disagreement on HER2 status in 174 of 1,787 cases (9.7 percent) classified HER2 positive by local laboratories but HER2 negative by both FISH (PathVysion®) and IHC assays in central or reference laboratories (Paik, Kim, Jeong, et al., 2007; Paik, Kim, and Wolmark, 2008). Data presently are unavailable on rates of disagreement between local and central laboratories from three other randomized, controlled trials of adjuvant trastuzumab (Piccart-Gebhart, Procter, Leyland-Jones, et al., 2005; Slamon, Eiermann, Robert, et al., 2005; Joensuu, Kellokumpu-Lehtinen, Bono, et al., 2006).

A small study compared central and local laboratory IHC results on breast tumor samples initially scored IHC 2+ locally and found FISH positive after referral for central laboratory confirmation (Barrett, Magee, O'Toole, et al., 2007). Investigators reported that of 153 IHC 2+ cases referred to the central laboratory for FISH confirmation, 29 (19 percent) had amplified HER2 genes. With repeat IHC in 25 of the 29, the central laboratory scored 18 cases (72 percent) as IHC 3+ and agreed with the local laboratory score of IHC 2+ in only 7 cases (28 percent). Since the central laboratory did not repeat IHC testing for the 124 cases with nonamplified HER2 genes by FISH, the overall rate of agreement with local results cannot be determined.

A larger study compared IHC results in local (regional) and central laboratories (Papadopoulos, Kouvatseas, Skarlos, et al., 2007). Of 458 available samples, 369 were tested by IHC both regionally and centrally and scores agreed for 296 (80.2 percent). Disagreement was greatest among samples (n=11) scored IHC 3+ by regional laboratories (63 percent concordance). Concordance was better among those (n=20) scored IHC 0 or 1+ and those scored IHC 2+ (n=338) at regional laboratories (85 percent and 80 percent, respectively).

A central reference laboratory analyzed tumor specimens from 315 of 399 (79 percent) patients randomized to capecitabine with or without lapatinib, using both IHC (antibody not reported) and FISH (PathVysion®), seeking confirmation of local laboratory results that classified these patients HER2 positive thus eligible for this randomized, controlled trial (Cameron, Casey, Press, et al., 2008). Central testing found 241 of 315 (77 percent) HER2 positive, including 211 with IHC 3+ results and 30 with IHC 2+, FISH-positive results.

In the Canadian study cited previously, central laboratory testing of breast tumor tissue samples confirmed the IHC-positive status of 79.3 percent to 89.6 percent of 505 cases found IHC positive by local laboratory results (O'Malley, Thomson, Julian, et al., 2008). Among 205 cases found IHC negative by local labs, central IHC testing confirmed local results in 94.8 percent to 100 percent of cases. The concordance rates varied, depending on which of four IHC assays the central laboratory used.

To summarize, data reported after publication of the ASCO/CAP guideline (Wolff, Hammond, Schwartz, et al., 2007a) confirm the estimate of approximately 20 percent disagreement between local (or regional) and central laboratories with respect to HER2 assay results. Data are presently lacking to evaluate the effects of adherence to guideline recommendations for preanalytic, analytic, and postanalytic steps on rates of local/central disagreement.

Validation and proficiency testing. Since these issues are outside the scope of this evidence report, interested readers are referred to current guidelines for specific recommendations on best practices to validate assays and test laboratory proficiency (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007). Evidence reviewed by the expert panel included a summary of results from 2004 and 2005 surveys of laboratories participating in CAP-sponsored interlaboratory comparisons of IHC results, using tissue microarrays as the test material (Fitzgibbons, Murphy, Dorfman, et al., 2006). The key finding was that 97 of 102 laboratories (95 percent) in 2004 and 129 of 141 laboratories (91 percent) in 2005 correctly scored 90 percent or more of the test cases. In the following section, we briefly summarize evidence published after the ASCO/CAP guideline. Again, these studies scored samples as originally recommended by manufacturers and FDA labeling.

An international study compared five pathology reference centers (from Netherlands, Canada, France, Belgium, and Germany) on assay scoring and HER2 status classification for separate samples tested by IHC (n=20) or by FISH (n=20) (Dowsett, Hanna, Kockx, et al., 2007). Agreement was uniform among centers on HER2 status classifications for all 20 IHC test cases, although some scoring differences were noted, and some equivocal cases (i.e., those scored IHC 2+) required FISH confirmation to determine HER2 status. Agreement was uniform among centers 16 of 20 (80 percent) FISH test cases. Each of the other four cases was scored in the equivocal range (HER2/CEP17 ratio 1.7–2.3).

A similar international study (from Netherlands, Australia, Canada, France, and Germany) compared results from five central laboratories on 211 breast cancer specimens tested by CISH, FISH and IHC (van de Vijver, Bilous, Hanna, et al., 2007). Each central laboratory sent unstained sections from samples they tested to four other (“outside”) central laboratories. Investigators reported uniform agreement by CISH in the “outside” laboratories on 73 of 76 cases (96 percent) scored highly amplified (HER2/CEP17 greater than 4.0) by FISH in the initial laboratory. Similarly, “outside” CISH uniformly agreed with 94 of 100 (94 percent) cases initially scored as not amplified by FISH (HER2/CEP17 less than 2.0). Among 35 cases scored as equivocal by initial FISH testing (HER2/CEP17 2.0–4.0), 20 were scored as CISH positive and 15 were scored as CISH negative. Overall interlaboratory concordance was 95 percent for cases with normal HER2 gene copy number (1–5) and was 92 percent for cases with 6 or more copies of the HER2 gene.

A brief report by investigators from the Italian Network for Quality Assessment of Tumor Biomarkers (INQUAT) and the United Kingdom National External Quality Assessment Service (U.K. NEQAS) highlighted the importance of including both preanalytic and analytic steps in proficiency testing programs (Paradiso, Miller, Marubini, et al., 2007). The U.K. NEQAS program for HER2 testing focuses on preanalytic aspects of the IHC assay, while the INQUAT program focuses on intra- and interlaboratory variability in scoring a set of fixed and stained IHC slides. Twelve Italian laboratories participated in both quality control programs during 2003, and only one achieved high-quality performance in preanalytic processing steps and in intra- and interlaboratory reproducibility. Some laboratories that achieved high-quality performance in preanalytic steps did not score slides reproducibly, or vice versa. Three of the 12 laboratories did not perform adequately on either preanalytic or analytic steps.

A recent study covalently attached fixed and unfixed samples of synthetic HER peptide to glass microscope slides with unstained sections of invasive breast carcinomas (Vani, Sompuram, Fitzgibbons, et al., 2008). The peptide fragments were used as positive analyte controls on slides distributed to 192 laboratories participating in the CAP 2006 HER2-B proficiency testing survey. Stained slides were returned and centrally reviewed (n=109 laboratories), permitting participants to evaluate sources of variability in HER2 staining performance. Investigators reported suboptimal staining in 20 of 109 slides (18.3 percent). Of these, seven cases (35 percent of the 20 failures) were attributable to errors in the antigen retrieval step, four (20 percent) were attributable to problems with the antibody staining protocol, and nine (45 percent) had problems with both.

In summary, two studies published subsequent to the ASCO/CAP review (Wolff, Hammond, Schwartz, et al., 2007a) reported similar results on interlaboratory comparisons. Overall, the available evidence shows 90 percent or greater agreement between high-volume reference laboratories in North America, Europe, and Australia. Scoring differences between laboratories occur most often with cases of low-level amplification or low-level overexpression. Results reported before and after the ASCO/CAP review (and other guidelines) support considering such cases as equivocal results, with confirmatory testing needed to classify HER2 status. Collaborative data from Italy and the United Kingdom suggest that quality control programs must evaluate all steps (preanalytic, analytic, and postanalytic) in HER2 testing. Positive analyte controls confirmed that antigen retrieval and antibody staining are persistent sources of interlaboratory variability in IHC results.

Reports on polysomy 17. The ASCO/CAP expert panel (Wolff, Hammond, Schwartz, et al., 2007a) interpreted evidence from two studies (Downs-Kelly, Yoder, Stoller, et al., 2005; Ma, Lespagnard, Durbecq, et al., 2005) as not supporting an association of polysomy 17 (defined as three or more copies of CEP 17) with HER2 protein or mRNA overexpression. However, one of these (Ma, Lespagnard, Durbecq, et al., 2005) reported increased HER2 protein (IHC 3+) in a subset of patients with polysomy 17 and HER2/CEP 17 ratios less than 2. In the following section, we summarize evidence published subsequent to the ASCO/CAP guideline.

Nine studies have reported data on polysomy 17 and HER2 status of breast cancer patients since the ASCO/CAP review. Of these, seven have been published in full (Dal Lago, Durbecq, Desmedt, et al., 2006; Torrisi, Rotmensz, Bagnardi, et al., 2007; Corzo, Bellosillo, Corominas, et al., 2007; Beser, Tuzlali, Guzey, et al., 2007; Hyun, Lee, Kim, et al., 2008; Kostopoulou, Vageli, Kaisaridou, et al., 2007; Hofmann, Stoss, Gaiser, et al., 2008) and two were reported at meetings with slides or video available on line (Kaufman, Broadwater, Lezon-Geyda, et al., 2007; Reinholz, Jenkins, Hillman, et al., 2007). Three studies reported no association of polysomy 17 with HER2 protein and/or mRNA overexpression (Dal Lago, Durbecq, Desmedt, et al., 2006; Torrisi, Rotmensz, Bagnardi, et al., 2007; Corzo, Bellosillo, Corominas, et al., 2007). In contrast, five other studies reported increased levels of HER2 protein in some cases with polysomy 17 and unamplified HER2 genes (Hyun, Lee, Kim, et al., 2008; Kaufman, Broadwater, Lezon-Geyda, et al., 2007; Reinholz, Jenkins, Hillman, et al., 2007; Kostopoulou, Vageli, Kaisaridou, et al., 2007; Hofmann, Stoss, Gaiser, et al., 2008). The ninth study did not report data on overexpression of HER2 protein or mRNA; this study reported chromosome 17 polysomy in two of 11 patients with HER2 gene amplification and in seven of 39 patients with unamplified HER2 genes (Beser, Tuzlali, Guzey, et al., 2007). In one study (Hofmann, Stoss, Gaiser, et al., 2008), seven of nine discordant IHC 3+/FISH-negative patients had chromosome 17 polysomy, and six of 26 patients with polysomy 17 responded to trastuzumab therapy for metastatic disease. However, all six responders were scored 3+ by IHC.

In contrast to conclusions of the ASCO/CAP review (Wolff, Hammond, Schwartz, et al., 2007a), evidence published subsequently reopens the question of whether chromosome 17 polysomy has implications for classifying patients' HER2 status. Five of eight new studies found polysomy 17 to be associated with protein (and/or mRNA) overexpression in at least some patients with nonamplified HER2 genes, while three of eight found no association.

Implications for Remainder of this Report

Discordances between IHC and FISH results might arise in one of three ways. They may be artifacts of one accurate and one inaccurate test. Alternatively, they may reflect a threshold issue, either related to the changes in threshold definitions over time, or an inherent problem of using a continuous measure to classify patients dichotomously. Finally, discordant test results might accurately reflect a small number of different patients with respect to the biologic mechanism that increases membrane levels of the HER2 protein. Present data could not tease apart the many factors reviewed here (preanalytic, analytic and postanalytic) that might have contributed to discordances in HER2 assay results. This clearly affects the interpretation of evidence on key questions that address use of “HER2 status” to predict treatment outcomes, even in nonbreast malignancies (Key Questions 2, 3, and 5). Furthermore, it also affects interpretation of evidence on the added clinical utility of serum measurements for patients with known tissue status, since this presumes accurate classification by tissue assays. Future studies reporting outcomes as a function of HER2 status should report separately on patients with concordant, equivocal, and discordant assay results.

Study Selection

The search strategy for studies on HER2 testing in breast cancer yielded 3,218 citations. Initial review selected 74 citations potentially relevant to Key Question 2 for retrieval and review as full articles. We used the ASCO/CAP expert panel's definition (Wolff, Hammond, Schwartz, et al., 2007a) of equivocal HER2 assay results: IHC 2+, or HER2 gene copy number from 4.0 to 6.0 or HER2/CEP17 ratio from 1.8 to 2.2 if ISH is the first or only assay. We defined discordant results as unequivocally positive results by one assay method (i.e., IHC 3+, HER2 gene copy number greater than 6.0, or HER2/CEP17 ratio greater than 2.2) and unequivocally negative results by a different assay method on another tissue section from the same tumor. Four trials (eleven reports; see Table 7 and “Available Studies” for citations) met selection criteria for data abstraction and compared outcomes with versus without a drug targeting HER2, for breast cancer patients with equivocal, discordant, or unequivocally negative HER2 assay results. Three trials randomized patients to chemotherapy with versus without trastuzumab; the fourth randomized patients to chemotherapy with versus without lapatinib, a tyrosine kinase inhibitor active against HER1 and HER2. Trials and their results are summarized in Tables 7–9; detailed abstraction data can be found in Appendix Tables II-A–II-I ^*.

Available Studies and Reports

Table 7 includes two trials on adjuvant trastuzumab with data for Key Question 2 (NSABP B31 and NCCTG N9831). Each reported post-hoc analyses on interim results for small subgroups of resected breast cancer patients inadvertently randomized to chemotherapy with or without trastuzumab in trials seeking to randomize only HER2-positive patients. Similarly, a trial on chemotherapy with or without lapatinib for locally advanced or metastatic disease (EGF100151) also intended to randomize only HER2-positive patients (Cameron, Casey, Press, et al., 2008; Geyer, Forster, Lindquist, et al., 2006). In each of these trials, local laboratory HER2 testing initially classified all randomized patients as HER2 positive. However, central or reference laboratory retests subsequently identified small subsets as equivocal, discordant, or HER2 negative. Only one trial (CALGB 9840) intentionally randomized HER2-negative metastatic breast cancer patients (referred to as “HER2 non-overexpressors” by study authors), and directly tested whether adding trastuzumab to chemotherapy improved outcomes.

One trial on trastuzumab in adjuvant therapy (NSABP B31) reported data on post-hoc subgroup analyses in a brief published communication (Paik, Kim, and Wolmark, 2008). Another adjuvant trastuzumab trial (NCCTG N9831) compared local, central, and reference laboratory results of HER2 testing in a published article that did not report outcomes (Perez, Suman, Davidson, et al., 2006). Both trials reported subgroup outcomes in meeting abstracts, with slides available online (B31: Paik, Kim, Jeong, et al., 2007; N9831: Perez, Romond, Suman, et al., 2007, and Reinholz, Jenkins, Hillman, et al., 2007). A single, published report provided baseline characteristics and preliminary outcomes data for patients randomized to treatment arms common to B31 and N9831 (Romond, Perez, Bryant, et al., 2005). Data were reported in this publication on each trial separately and both trials combined.

Two trials on patients with advanced or metastatic disease published full reports with subgroup analyses (Seidman, Berry, Cirrincione, et al., 2008; Cameron, Casey, Press, et al., 2008). The EGF100151 trial on chemotherapy with or without lapatinib (Cameron, Casey, Press et al., 2008) also published an earlier report (Geyer, Forster, Lindquist et al., 2006), but without results of repeat HER2 testing by a central or reference laboratory or analyses relevant to Key Question 2. CALGB 9840, the only preplanned analysis relevant to this key question, is on a HER2-negative (i.e., non-overexpressor) subgroup randomized to chemotherapy with or without trastuzumab within a larger trial studying an unrelated question (Seidman, Berry, Cirrincione, et al., 2004, 2008). CALGB 9840 also is the source of all patients in the subgroup analyzed post-hoc in CALGB 150002 (Kaufman, Broadwater, Lezon-Geyda, et al., 2007).

Treatments and Subgroups Compared

Adjuvant therapy. Two trials (NSABP B31, NCCTG N9831) investigated outcomes of adjuvant doxorubicin plus cyclophosphamide (AC; every three weeks for four cycles), followed by paclitaxel (P; every three weeks for four cycles), with versus without trastuzumab (+/-TRZ; weekly for 12 months, beginning concurrently with paclitaxel) in women with fully resected early breast cancer. Outcomes are as-yet unreported for a third arm of N9831, which began trastuzumab therapy after all eight cycles of chemotherapy (AC→P→TRZ). Both B31 and N9831 limited eligibility to HER2-positive patients, defined as FISH-positive/IHC unknown, IHC3+/FISH-unknown, or IHC2+/FISH-positive. Patients were initially evaluated by local laboratory testing, and randomized if classified HER2-positive by these results. They were subsequently re-evaluated by central laboratory testing, but continued with assigned treatments regardless of results. A planned interim analysis at two years' median followup (2.4 years for B31 patients; 1.5 years for N9831 patients) for all patients randomized to the treatment arms common to both trials, pooled patients assigned to the control arms(n=1,679; AC→P) and those assigned to concurrent trastuzumab (n=1,672; AC→P+TRZ) (Romond, Perez, Bryant, et al., 2005). Trastuzumab significantly improved overall survival (OS) at four years: 91.4 percent versus 86.6 percent; hazard ratio (HR) =0.67; 95 percent CI: 0.48–0.93; p=0.015. The B31 (Paik, Kim, and Wolmark, 2008; Paik, Kim, Jeong, et al., 2007) and N9831 (Perez, Romond, Suman, et al., 2007 and Reinholz, Jenkins, Hillman et al., 2007) results included here were unplanned, post-hoc analyses. They compared outcomes of adjuvant AC→(P+/-TRZ) in subgroups found HER2 discordant or negative by central lab results, using data collected for the pooled analysis of Romond, Perez, Bryant, et al. (2005) without longer followup.

Advanced/metastatic disease. A randomized, controlled trial (CALGB 9840) that studied paclitaxel in women receiving first- or second-line therapy for metastatic breast cancer reported outcomes at two meetings (Seidman, Berry, Cirrincione, et al., 2004; Kaufman, Broadwater, Lezon-Geyda, et al., 2007) and in a published article (Seidman, Berry, Cirrincione, et al., 2008). Primary randomization in this trial compared once-weekly to every-third-week paclitaxel dosing regimens. Testing for HER2 status began after enrolling the first 171 patients, and HER2-negative patients (termed “HER2 non-overexpressors” by study authors and defined as 0 or 1+ or IHC 2+/FISH negative by local laboratory tests) were also randomized to treatment with or without trastuzumab. Seidman, Berry, Cirrincione, et al. (2004, 2008) reported outcomes for this second randomization without separating results by paclitaxel treatment frequency. HER2-positive patients (by local laboratory tests) all received trastuzumab and are excluded from the analysis for Key Question 2.

For all patients randomized (n=735), CALGB 9840 investigators first reported that response rate and time to progression (TTP) were better with weekly paclitaxel than with every third week, although the difference in median OS (24 versus 16 months; HR=1.19, p=0.17) was not statistically significant (Seidman, Berry, Cirrincione, et al., 2004). As prespecified in the CALGB 9840 protocol, the final analysis (Seidman, Berry, Cirrincione, et al., 2008) comparing paclitaxel schedules pooled additional patients (n=158) randomized to the identical dose of paclitaxel every third week (all without trastuzumab) in another trial (CALGB 9342; Winer, Berry, Woolf et al., 2004) with those randomized to this schedule in CALGB 9840. In this combined analysis, weekly paclitaxel statistically significantly improved response rate (42 percent versus 29 percent; OR=1.75, p=0.0004), TTP (median, nine versus five months; HR=1.43, p<0.0001), and OS (median, 24 versus 12 months; HR=1.28, p=0.0092), when compared with treatment every third week. Data in Table 8 on HER2 non-overexpressors exclude patients from CALGB 9342.

A post-hoc analysis on HER2 non-overexpressors randomized to paclitaxel with versus without trastuzumab in CALGB 9840 compared outcomes for subsets found FISH negative by central laboratory testing who had or did not have chromosome 17 polysomy (CALGB 150002; Kaufman, Broadwater, Lezon-Geyda, et al., 2007). This analysis was not included in the published final report (Seidman, Berry, Cirrincione, et al., 2008). It also did not include patients from CALGB 9342, none of whom were randomized to paclitaxel with or without trastuzumab.

The EGF100151 trial randomized patients with locally advanced or metastatic breast cancer to capecitabine (1 g/m² twice daily for 14 days every three weeks) plus lapatinib (1.25 g/m² daily) or to capecitabine alone (1.25 g/m² twice daily for 14 days every three weeks). Eligibility required: a T4 primary tumor and stage IIIB or IIIC disease, for those without distant metastasis; a history of progressive disease after one or more regimens that included an anthracycline, a taxane, and trastuzumab (given separately or in combinations); and local laboratory HER2 test results of IHC3+ or IHC2+/FISH positive. An interim analysis on 163 patients randomized to capecitabine plus lapatinib and 161 randomized to capecitabine monotherapy reported median TTP was 8.4 months in the combination arm and 4.4 months in the capecitabine monotherapy arm (HR=0.49; 95 percent CI: 0.34–0.71, p<0.001) (Geyer, Forster, Lindquist, et al., 2006). A second report included more patients (n=198, capecitabine plus lapatinib; n=201, capecitabine monotherapy; Cameron, Casey, Press, et al., 2008). By intent-to-treat analysis, median TTP was 6.2 months in the combined arm and 4.3 months in the monotherapy arm (HR=0.57; 95 percent CI: 0.43–0.77, p<0.001). A second interim analysis for OS found 28 percent had died (median OS, 15.6 months) in the combined therapy arm and 32 percent had died (median OS, 15.3 months) in the capecitabine monotherapy arm (HR=0.78; 95 percent CI: 0.55–1.12; p=0.177); followup for survival continues. Central laboratory IHC and FISH retesting of samples from 300 (75 percent) of the 399 randomized in this trial identified small subgroups with HER2-discordant or -negative results (Table 7).

Study Quality

Only one of four included trials (CALGB 9840) stratified randomization by HER2 status, the most informative evidence level defined in this report's study design hierarchy (see Methods, Table 3). The others are post-hoc analyses of treatment effects in HER2-discordant or -negative subgroups from larger randomized, controlled trials. One trial on adjuvant trastuzumab (NSABP B31) and both trials on patients with metastatic or advanced disease (CALGB 9840 and EGF100151) included multivariate analyses. However, neither CALGB 9840 nor EGF100151 used multivariate analysis to adjust treatment outcomes in HER2 discordant or HER2 negative subgroups. Since these subgroups from each study are small and underpowered, and since results from three of four trials are interim analyses with limited followup, we did not assess study quality using the checklist derived from REMARK and other sources (see “Methods”).

Patient Characteristics

Adjuvant therapy. Patients from B31 and N9831 were initially randomized based on positive results of local lab testing, given their assigned regimen, and followed on these randomized, controlled trials. Those in subgroups included here subsequently were reclassified HER2 discordant or HER2 negative by central laboratory results. Baseline patient characteristics and prognostic factors (Table 7) were reported for all patients randomized to each treatment arm in each trial (Romond, Perez, Bryant, et al., 2005), including those classified as HER2 positive by both local and central laboratory results. At the level of initial randomization, baseline characteristics and prognostic factors of the groups treated with versus without trastuzumab were similar. However, data were not reported to separately compare baseline characteristics and prognostic factors by treatment arm for each subgroup of HER2-discordant or -negative patients (by central laboratory results).

Data are available from B31 for two HER2-discordant groups:

• FISH positive/IHC 0, 1+, or 2+: n=56 +TRZ; n=69 -TRZ (data not reported separately for FISH-positive, IHC 0, 1+ subset)

• FISH negative/IHC 3+: n=10 +TRZ; n=21 -TRZ;

and for two (partially overlapping) HER2-negative groups:

• FISH negative/IHC 1+ or 2+: n=69 +TRZ; n=80 -TRZ

• FISH negative/IHC 0, 1+, or 2+: n=82 +TRZ; n=92 -TRZ (13 and 12 patients per arm added to the 69 and 80 in the arms above).

Data are available from N9831 for two HER2-discordant groups:

• FISH positive/IHC 0, 1+, or 2+: n=123 +TRZ; n=95 -TRZ (data not reported separately for FISH-positive, IHC 0, 1+ subset)

• FISH negative/IHC 3+: n= 23 +TRZ; n=30 -TRZ;

and for one HER2-negative group:

• FISH negative/IHC 0, 1+, or 2+: n=59 +TRZ; n=44 -TRZ.

Advanced/metastatic disease. Patients in CALGB 9840 had metastatic disease undergoing first- or second-line therapy. All were randomized to weekly or every third week paclitaxel, and those who were HER2 negative (IHC 2+/FISH negative or IHC 0 or 1+) by local laboratory results were simultaneously randomized to receive (n=113) or not receive (n=115) trastuzumab. The analysis pooled outcomes in the HER2-negative arms for patients given paclitaxel weekly or every third week. Subsequent analyses (CALGB 150002) compared outcomes separately for subgroups from CALGB 9840 who were FISH negative by central laboratory results and had (+/-TRZ, n=19 each arm) or did not have (+TRZ, n=53; -TRZ, n=50) chromosome 17 polysomy.

Patients in EGF100151 had locally advanced or metastatic disease that progressed after one or more regimens with an anthracycline, a taxane, and trastuzumab (given separately or in combinations, as adjuvant therapy or for metastasis). Women (n=399) with local laboratory HER2 test results of IHC3+ or IHC2+/FISH positive were randomized to capecitabine with or without lapatinib. Baseline characteristics and prognostic factors of the groups treated with versus without lapatinib were similar. Subsequent central laboratory reanalysis by FISH and IHC of tumor samples from 300 patients (75 percent of all randomized) identified HER2 discordant or HER2 negative subgroups (Table 7). Data were not reported to separately compare baseline characteristics or prognostic factors by treatment arm for any of these subgroups.

Results, Key Question 2

Adjuvant AC→(P±TRZ). The only available data are from post-hoc subgroup analyses, without stratification for the subgroups' defining characteristics. Neither the B31 nor the N9831 analyses reported subgroup-specific comparisons of baseline characteristics or prognostic factors by treatment arm. Furthermore, one subgroup mixed results for a discordant subgroup (IHC 0, 1+, FISH positive) with results for initially equivocal but ultimately positive (IHC 2+ but amplified by FISH) patients. Finally, data are presently unavailable from studies that classified patients using assay thresholds consistent with current guidelines (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007 see “Results and Conclusions, Key Question 1, Narrative Review”).

Neither trial reported median followup durations, or showed numbers per arm at risk over time, for the specific subgroups compared. In each subgroup from each treatment arm, failure events (e.g., death or relapse) occurred in less than 25 percent of patients (range: 5–23 percent) at the time of analysis. Therefore, length of followup was inadequate for reliable estimates of median event-free durations for any outcome reported. The interim analyses for all patients randomized in the larger trials that were sources of these subgroups (Romond, Perez, Bryant, et al., 2005) also lacked sufficient followup for reliable estimates of median overall survival or median disease-free survival (DFS).

For HER2 discrepant patients who were FISH positive and IHC 0, 1+ or 2+ by central laboratory testing, between-arm differences in outcome were not statistically significant in either trial. In B31 (n=56 +TRZ; n=69 -TRZ), the HR for failure in analysis of DFS was 0.30 (95 percent CI: 0.08–1.07; p=0.064) and the HR for failure in analysis of recurrence-free interval (RFI) was 0.35 (95 percent CI: 0.10–1.28; p=0.11). In N9831 (n=123 +TRZ; n=95 -TRZ), the HR for failure in analysis of DFS was 0.98 (95 percent CI: 0.33–2.91; p=0.97).

Few patients were FISH negative and IHC 3+ by central laboratory results (from B31: n=10 +TRZ; n=21 -TRZ; from N9831: n=23 +TRZ; n=30 -TRZ). B31 reported HR for failure was 0.91 for both DFS and RFI (for each outcome, 95 percent CI: 0.08–10.0; p=0.94), and N9831 reported hazard ratio for failure was 0.61 (95 percent CI: 0.11–3.29; p=0.57). Each between-arm subgroup comparison was not statistically significant.

Only B31 analyzed outcomes of patient subgroups that were HER2 negative by FISH but IHC 1+ or 2+ by central laboratory testing [n=69 +TRZ; n=80 -TRZ]). Between-arm differences reported by Paik, Kim, Jeong et al. (2007) were statistically significant for DFS (HR=0.30; 95 percent CI: 0.11–0.83; p=0.02) and RFI (HR=0.31; 95 percent CI: 0.10–0.95; p=0.041), and favored the subgroup given trastuzumab.

Both trials reported on patients who were FISH negative and IHC 0, 1+ or 2+ by central laboratory testing. In B31, this subgroup added FISH-negative/IHC 0 patients (13 and 12 per arm, respectively) to those in the FISH-negative/IHC 1+ or 2+ arms shown above (combined n=82 +TRZ; combined n=92 -TRZ). Between-arm differences were statistically significant for DFS (7 events, +TRZ, 20 events, -TRZ; HR=0.34; 95 percent CI: 0.14–0.80; p=0.014) and RFI (HR=0.36; 95 percent CI: 0.14–0.92; p=0.034), and again favored the subgroup given trastuzumab. One patient died in the trastuzumab arm, while 10 died in the control arm (HR=0.08; 95 percent CI: 0.01–0.64, p=0.017). In N9831 (n=59 +TRZ, n=44 -TRZ), the between-arm difference in DFS (HR=0.51; 95 percent CI: 0.21–1.2; p=0.13) was not statistically significant.

HER2 gene copy number and magnitude of benefit from trastuzumab. Additional unpublished subset analyses from the B31 trial presented at the June 2007 ASCO annual meeting (Paik, Kim, Jeong, et al., 2007), and similar analyses from the N9831 trial (Reinholz, Jenkins, Hillman, et al., 2007) and the HERA trial (McCaskill-Stevens, Proctor, Goodbrand, et al., 2007) presented at the December, 2007 San Antonio Breast Cancer Symposium, investigated the hypothesis that higher HER2 gene copy numbers, or higher HER2/CEP17 FISH ratios, were associated with a larger magnitude of relative benefit from trastuzumab. Data from the N9831 and HERA trials showed that the hazard ratio for DFS did not grow more favorable to the trastuzumab arm as average FISH ratios increased from 2.0 to 15 or greater (N9831), or from 2 to greater than 8 (HERA). Additionally, investigators found the HR for DFS did not increase as average HER2 gene copy number per cell increased from 4 to greater than 18 (HERA), or from 2 to greater than 10 (B31).

Polysomy 17 and adjuvant trastuzumab. An unpublished post-hoc analysis of data from N9831 presented at the December 2007 San Antonio Breast Cancer Symposium evaluated whether polysomy 17 influenced effects of adjuvant trastuzumab (Reinholz, Jenkins, Hillman, et al., 2007). Investigators reported that among patients with amplified HER2 genes, trastuzumab increased DFS whether or not these patients had polysomy 17. Central lab results identified very few patients without HER2 overexpression by IHC or HER2 gene amplification by FISH, but with polysomy 17. DFS was lower (79 percent versus 83 percent at 3 years; 65 percent versus 75 percent at 5 years) among those given trastuzumab than among those not given trastuzumab, although the sample size was small and few events had occurred in either arm (6 of 24 given trastuzumab, 3 of 13 controls). Investigators also analyzed slightly larger patient subsets without HER2 overexpression by IHC, HER2 gene amplification by FISH, or polysomy 17. DFS was substantially higher (94 percent versus 77 percent at 3 years; 84 percent versus 55 percent at 5 years) among those given than among those not given trastuzumab. As in the subset with polysomy 17, few events had occurred in either arm in the subset without polysomy (4 of 34 given trastuzumab, 13 of 33 controls). Additionally, unpublished data from the NSABP B31 trial showed no impact on prognosis or degree of benefit from trastuzumab (Dr. S. Paik; personal communication, May 2008).

HER2-negative patients with metastatic disease given P±TRZ for first- or second-line therapy. Patients found IHC 2+/FISH negative or IHC 0, 1+ by local laboratory results were randomized in CALGB 9840 to have or not have trastuzumab added to paclitaxel (n=113 +TRZ; n=115 -TRZ). Between-arm differences in OS (median: 21.6 versus 19.6 months, p=0.67), time to progression (TTP; median: 12 versus 6 months, p=0.088), and overall response rate (ORR; 35 percent versus 29 percent, p=0.32) were not statistically significant (Seidman, Berry, Cirrincione, et al., 2008).

CALGB 150002 reported that subgroups from CALGB 9840 found FISH negative by central laboratory results, and also found to have chromosome 17 polysomy (n=19 +TRZ; n=19 -TRZ), showed a statistically significant increase in ORR (63 percent versus 26 percent, p=0.048) among those given trastuzumab plus paclitaxel compared with those given paclitaxel alone (Kaufman, Broadwater, Lezon-Geyda, et al., 2007). In contrast, ORR did not differ between treatment arms (36 percent in each) for centrally FISH-negative patients without chromosome 17 polysomy. The ORR difference between arms for the centrally FISH-negative subgroup with polysomy 17 (+/-TRZ; n=19 each) did not yield statistically significant differences between arms for either OS (p=0.538) or TTP (p=0.88).

HER2-negative patients with advanced or metastatic disease that progressed after an anthracycline, a taxane, and trastuzumab given capecitabine ± lapatinib. Few patients randomized to capecitabine with or without lapatinib in the EGF100151 trial were HER2 discordant (Table 7). Furthermore, outcomes were not reported separately for those found FISH positive but IHC negative by central laboratory testing (with lapatinib, n=15; without lapatinib, n=7), or those found FISH negative but IHC 3+ by central lab results (with lapatinib, n=1; without lapatinib, n=2). Investigators identified a total of 74 patients (23.5 percent of 315 tested in the central laboratory) whose local results were not confirmed by the central lab as meeting HER2 eligibility criteria of IHC 3+ or FISH positive/IHC2+ (Cameron, Casey, Press et al., 2008); distribution between treatment arms was not reported. In an exploratory Kaplan-Meier analysis, investigators found no statistically significant difference between arms (capecitabine with or without lapatinib) in PFS (HR=0.772; 95 percent CI: 0.386–1.543; p=0.46).

Conclusions and Discussion, Key Question 2

Adjuvant trastuzumab. Currently available evidence is inconclusive on outcomes of trastuzumab added to adjuvant chemotherapy for resected HER2-discordant or HER2-negative patients. Evidence on each subgroup may be used to generate hypotheses, but is too weak to test hypotheses, for the following reasons. All available evidence is from post-hoc analyses on subgroups not directly randomized or stratified by the HER2 subgroups of interest. Furthermore, available reports did not show direct comparisons of baseline characteristics and prognostic factors for the specific subgroups compared. Thus, it is uncertain whether the HER2-discordant or HER2-negative subgroups were balanced by treatment arm (i.e., with or without trastuzumab; although treatment arms appeared well-balanced across all patients randomized). Finally, the data used for the two adjuvant studies are from interim analyses, with inadequate followup to estimate median survival for all patients randomized, and inadequate information on median duration of followup in the specific subgroups compared. Thus, although these were large, well-designed and well-conducted randomized, controlled trials, since the overwhelming majority of patients they randomized were unequivocally HER2-positive, only poor quality evidence is presently available on outcomes of adjuvant trastuzumab in either HER2 discordant or HER2 negative patient subgroups.

Adjuvant trastuzumab in HER2-discordant patients. Evidence is unavailable to evaluate effects of trastuzumab specifically for HER2-discordant patients who are FISH positive but IHC negative (0, 1+) by central lab results. Analyses reported from each trial pooled outcomes for these patients with outcomes for those who tested FISH positive and IHC 2+. The latter subset (initially considered equivocal if tested first by IHC) was classified HER2 positive by each trial protocol, and is ultimately classified HER2 positive by algorithms in current guidelines. A more informative analysis limited to the discordant subgroup might compare outcomes with versus without trastuzumab using data pooled from B31 and N9831 on patients who were FISH positive but IHC 0 or 1+ by central lab tests. Results from a systematic review (see Table 6, Key Question 1) estimates this subgroup as 2.4 percent (95 percent CI: 1–4.3 percent) of all breast cancer patients (Dendukuri, Khetani, McIsaac, et al., 2007).

Sample size is insufficient for conclusions from HER2-discordant B31 (total n=31) and N9831 (total n=53) subgroups that tested FISH negative but IHC 3+ by central lab results. The proportion of FISH-negative, IHC 3+ patients is 2.2 percent across both trials (total randomized: 3,822). Results of the systematic review summarized in Table 6 (Key Question 1) estimate this subgroup as 1.2 percent (95 percent CI: 0.6–2.1 percent) of all breast cancer patients (Dendukuri, Khetani, McIsaac, et al., 2007). Although at least three other randomized trials investigated adjuvant trastuzumab, they confirmed eligibility by central or reference laboratory FISH tests before randomizing patients, and have not reported on either of the HER2 discordant subgroups of interest. Thus, large database or registry analyses may be the only source of better evidence on outcomes of adjuvant trastuzumab for the two HER2 discordant subgroups, which together comprise approximately 4 percent of all breast cancer patients.

Factors influencing discordant results. Discordant results may occur if one assay is correct and the other in error, either due to preanalytic, analytic, or postanalytic factors (see Key Question 1). As with any assay, 100 percent accuracy cannot be expected even from the most careful and proficient laboratories. Proficiency testing and other quality control and quality assurance measures to minimize false-negative and false-positive results are recommended in current practice guidelines (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007). However, concordance of different methods to classify an individual as HER2 positive or negative is at least partly independent from accuracy of performing a specific assay. Even with the most careful and highly accurate laboratory techniques, discordance in classification may occur between a method that detects gene amplification (FISH in these studies, but also true with CISH or SISH) and a method that detects protein overexpression (IHC in these studies, but also true with Western blots).

By current guidelines, clinicians may categorize identical discordant patients differently with respect to HER2 status, depending on the selection and sequence of tests they order. So, for example, FISH-positive and IHC 0 or 1+ patients (1 to 4 percent of cases; see Table 6, Key Question 1) would be classified HER2 positive if tested only by FISH, but would be classified HER2 negative if tested initially by IHC, since reflex FISH would not be performed. Conversely, FISH-negative and IHC 3+ patients (1 to 2 percent of cases; see Table 6) would be considered HER2 negative if tested only by FISH, but HER2 positive if tested initially by IHC. NSABP B31 and NCCTG N9831 report the frequency of these subsets based on careful central laboratory results for FISH and IHC assays, although results are pooled across some IHC scores (see “Results and Conclusions, Key Question 1”). However, these data do not permit assessment of the subset frequencies independent of tissue fixation artifacts that may have occurred at some local hospitals and laboratories, or the margin of error that might exist even in the most proficient laboratories. Nor can the clinical consequences of such discordances be assessed from the available evidence.

Adjuvant trastuzumab in HER2-negative patients. Scant but intriguing evidence suggests the hypothesis that some patients currently classified as HER2 negative may benefit from adjuvant trastuzumab. Data reported from B31 showed significantly longer DFS and RFI in FISH-negative IHC ≤2+ patients given trastuzumab than in similar patients managed without trastuzumab, whether the analysis did or did not include those who were IHC 0. However, a similar analysis of data from N9831 did not show significant differences. Since both were interim analyses of trials in which fewer than 25 percent of subjects had reached a failure event, neither provides conclusive evidence as yet, and follow up analyses from these trials will be of great interest. Blinded review of IHC and FISH scoring would also be useful for samples from these trials, and from other adjuvant trastuzumab trials that confirmed eligibility by central lab testing before randomizing each patient. Recent guidelines conclude that present evidence does not demonstrate improved outcomes with use of adjuvant trastuzumab for patients who would be classified HER2 negative by protocols of B31, N9831, and similar studies (Wolff, Hammond, Schwartz, et al., 2007a; Carlson, Moench, Hammond, et al., 2006; Hanna, O'Malley, Barnes, et al., 2007).

Importantly, the B31 and N9831 subgroup analyses combine results for HER2-negative patients many now consider to be different: those with the so-called “triple-negative” subtype (i.e., negative for HER2, estrogen receptor, and progesterone receptor), and the luminal subtypes (luminal A or luminal B) that are negative for HER2 but positive for at least one of the hormone receptors. These subtypes were initially defined in studies using microarrays to subdivide breast cancer patients by gene expression patterns (for reviews, see Peppercorn, Perou, and Carey, 2008; Razzak, Lin, and, Winer, 2008; Kang, Martel, and Harris 2008). There is evidence that the triple negative and luminal subsets differ with respect to prognosis, chemotherapy response, and outcomes (Carey, Dees, Sawyer, et al., 2007; Liedtke, Mazouni, Hess, et al., 2008), and they clearly differ with respect to effects of endocrine therapy. Further complexity comes from reports that there is substantial but incomplete overlap between triple negative patients and those classified in the “basal-like” subset by gene expression arrays (Cheang, Voduc, Bajdik, et al., 2008). Notably, new phase III trials have recently opened (and others are planned) specifically for patients with triple negative or “basal-like” breast cancer (Kilburn, 2008). Results from these studies will likely be more conclusive than analyses that pool all HER2-negative patients to determine outcomes for subsets of HER2-negative breast cancer.

Adjuvant trastuzumab in HER2-equivocal patients. Among patients with initially equivocal HER2 test results by current clinical practice guidelines (those scored 2+ if IHC is first, or HER2 gene copy number from 4.0 to 6.0 or HER2/CEP17 ratio from 1.8 to 2.2 if ISH is first), ultimately, most are definitively categorized as HER2 positive or HER2 negative after guideline-recommended followup testing. Data are presently unavailable either to estimate effects of adjuvant trastuzumab on outcomes for the subset with initially equivocal results subsequently classified HER2 positive, or to demonstrate lack of benefit in those subsequently classified HER2 negative. For the minority who remain equivocal after followup testing, the guidelines' treatment recommendation depends on whether the patient would have been included or excluded from key randomized, controlled trials. For example, patients with HER2/CEP17 ratios 2.0 or greater but less than 2.2 were included and randomized in the adjuvant trastuzumab trials. Therefore, the guidelines consider current evidence insufficient to deny these patients trastuzumab with adjuvant chemotherapy. In contrast, patients with HER2/CEP17 ratios 1.8 or greater but less than 2.0 were excluded from these trials, and the guidelines consider current evidence insufficient to include trastuzumab in their adjuvant therapy regimens. Figures 2 and 3 (see Key Question 1) include information on trial eligibility of patients whose test results are equivocal by each HER2 assay.

Advanced or metastatic disease. No data were reported on patients with advanced or metastatic disease and discordant results from IHC and ISH HER2 testing. Evidence is available from one trial (CALGB 9840; n=226) that randomized metastatic breast cancer patients who were HER2 negative by local laboratory testing to chemotherapy with or without trastuzumab (Seidman, Berry, Cirrincione, et al., 2008). Additionally, a small subset of advanced and metastatic patients randomized to chemotherapy with or without lapatinib in another trial (EGF100151; n=74) were found by central lab confirmatory testing not to meet protocol criteria for HER2 positivity (Cameron, Casey, Press, et al., 2008). Thus, one source of good quality evidence (CALGB 9840) and one source of moderate quality evidence (EGF100151) suggest that HER2-negative patients with advanced or metastatic disease do not benefit from treatments targeting the HER2 molecule. Additional evidence supporting this conclusion comes from an analysis of data pooled from three pivotal trials of trastuzumab for metastatic breast cancer. The analysis showed that among patients found IHC 2+ by the presently unavailable “clinical trial assay,” benefit from trastuzumab was limited to those subsequently shown to have amplified HER2 genes by FISH (Mass, Press, Anderson et al., 2005).

CALGB 15002 investigators compared outcomes with versus without trastuzumab for a subgroup of FISH-negative patients who either had (n=38) or did not have (n=103) polysomy 17, (Kaufman, Broadwater, Lezon-Geyda, et al., 2007). Overall response rate was significantly higher with versus without trastuzumab for those with polysomy 17, but was identical with or without trastuzumab for those without polysomy 17. In contrast, the N9831 study on adjuvant therapy (Reinholz, Jenkins, Hillman, et al., 2007) reported no impact of polysomy 17 on benefit from trastuzumab, and unpublished data from a second study (NSABP B31; Dr. S. Paik, personal communication, May 2008)) suggested the same finding. This might be due to different definitions of polysomy 17 for CALGB 15002 (average CEP17 copy number per cell greater than 2.2) and N9831 (more than 3 CEP17 signals in more than 30% of nuclei). It might also reflect differences between adjuvant therapy and treatment for metastatic disease with respect to polysomy 17 as a predictor of benefit from trastuzumab. Note also that studies reviewed for “Results and Conclusions, Key Question 1” report conflicting data on a possible association of polysomy 17 with overexpression of HER2 protein. Thus, presently available evidence leaves unanswered questions with respect to the utility of polysomy 17 to select patients for HER2-targeted therapy.

Study Selection

The search strategy for studies on HER2 testing in breast cancer yielded 3,218 citations. Initial review of titles and abstracts selected 219 citations potentially relevant to Key Question 3 for retrieval and review as full articles. Of these, 161 were considered potentially relevant to Key Question 3a (HER2 status to guide choice of chemotherapy regimen) while 62 were considered potentially relevant to Key Question 3b (HER2 status to guide choice of hormonal therapy regimen). Four reports were considered for both question 3a and 3b.

Twenty separate studies met selection criteria and were abstracted for Key Question 3a (Table 10; Appendix Table IIIa-A ^*). Eleven studies investigated adjuvant chemotherapy for resected early stage breast cancer, including nine randomized, controlled trials, an uncontrolled series, and the standard-dose control arm of a randomized, controlled trial of high-dose chemotherapy with autologous stem-cell support (HDC/AuSCS). Six studies investigated neoadjuvant (preoperative) chemotherapy for locally advanced breast cancer; one was a randomized, controlled trial and five were uncontrolled, single-arm series. Three studies investigated first- or second-line therapy for advanced or metastatic breast cancer. Two randomized, controlled trials compared different regimens; the third randomized, controlled trial compared different doses of one drug, but pooled arms for the analysis by HER2 status.

Available Studies

Eleven studies on postsurgical adjuvant chemotherapy. The available evidence included one retrospective analysis of an uncontrolled single-arm series (Yang, Klos, Zhou, et al., 2003), and ten randomized, controlled trials. However, for one of the randomized, controlled trials, (Tanner, Isola, Wiklund, et al., 2006), one arm was excluded, since patients received HDC/AuSCS. Each randomized, controlled trial was designed to compare outcomes of treatment regimens in populations not selected or stratified for HER2 status, and most published earlier reports that compared patients, prognostic factors, and outcomes by treatment arm for all randomized patients. With only one exception (Martin, Pienkowski, Mackey, et al., 2005), reports from randomized, controlled trials included for Key Question 3a were secondary or correlative analyses on patient subgroups with archived tissue samples that permitted HER2 testing. The proportion of originally randomized patients included in the analyses by HER2 status ranged from 34 to 92 percent (see Table 10). A subset of trials compared baseline characteristics and known prognostic factors between the subgroups with known HER2 status and those with undetermined HER2 status, and a smaller subset also compared outcomes. None of these studies used trastuzumab for HER2-positive patients; studies addressing the use of trastuzumab are included in the discussion of Key Question 2.

Studies on the CMF regimen. The uncontrolled series (Yang, Klos, Zhou, et al., 2003; n=94) and one comparative randomized, controlled trial (Gusterson, Gelber, Goldhirsch, et al., 2003; n=2,504 randomized) studied the cyclophosphamide plus methotrexate plus fluorouracil (CMF) regimen. The Gusterson and co-workers trial separately randomized groups of node-negative and node-positive patients. Tissue blocks for determining HER2 status were unavailable for 515 (40 percent) of 1,275 randomized node-negative patients and for 483 (39 percent) of 1,229 randomized node-positive patients. Node-negative patients were randomized to one perioperative cycle of adjuvant CMF or to observation. Node-positive patients were randomized to multiple cycles of adjuvant CMF or to one perioperative cycle of adjuvant CMF. The relevance of these findings for current practice may be limited as taxane-based regimens have largely replaced CMF when anthracyclines are not used, particularly for hormone-receptor-negative patients.

Studies on anthracycline-based regimens. Four randomized, controlled trials (Moliterni, Menard, Valagussa, et al., 2003; Colozza, Sidoni, Mosconi, et al., 2005; Pritchard, Shepherd, O'Malley, et al., 2006; Knoop, Knudsen, Balslev, et al., 2005) compared CMF versus anthracycline-based regimens, and a fifth randomized, controlled trial compared an anthracycline-based regimen without autologous stem-cell support (AuSCS) versus a higher-dose regimen with AuSCS (Tanner, Isola, Wiklund, et al., 2006). Only the non-AuSCS arm of the Tanner and co-workers study met selection criteria for data abstraction. Moliterni, Menard, Valagussa, et al. (2003) compared CMF followed by doxorubicin (CMF→A) versus CMF alone, and included 92 percent of originally randomized patients. Colozza, Sidoni, Mosconi, et al. (2005) compared epirubicin (E) alone versus CMF, and included 76 percent of originally randomized patients. Pritchard, Shepherd, O'Malley, et al. (2006) and Knoop, Knudsen, Balslev, et al. (2005) compared cyclophosphamide plus epirubicin plus fluorouracil (CEF) versus CMF, although the Pritchard and co-workers study gave 6 cycles while the Knoop and co-workers study gave 9 cycles. Pritchard and co-workers included 89 percent of originally randomized patients while Knoop and co-workers included 79 percent. Tanner, Isola, Wiklund, et al. (2006) also gave 9 cycles of CEF in the non-AuSCS arm of their trial, although the doses administered were higher than those in the Pritchard and Knoop trials. Outcomes by HER2 status for 72 percent of those randomized to the non-AuSCS arm are considered a single-arm study in this review.

Two randomized, controlled trials with two reports each compared different doses (Dressler, Berry, Broadwater, et al., 2005; Thor, Berry, Budman, et al., 1998) or dose intensities and schedules (Del Mastro, Bruzzi, Nicolo, et al., 2005; Del Mastro, Bruzzi, Venturini, et al., 2004) for anthracycline-based regimens. The Dressler and co-workers study investigated interaction of HER2 status with dose in 524 patients from the Cancer and Leukemia Group B (CALGB) trial 8541. This trial randomized 1,549 patients to high-dose (600/60/600 mg/m² every four weeks for 16 weeks), moderate-dose (400/40/400 mg/m² every four weeks for 24 weeks) or low-dose (300/30/300 mg/m² every four weeks for 16 weeks) regimens of cyclophosphamide, doxorubicin and fluorouracil (CAF) (Budman, Berry, Cirrincione, et al., 1998). Although earlier reports (Thor, Berry, Budman, et al., 1998; Muss, Thor, Berry, et al., 1994) included different proportions of randomized patients tested for HER2 status by IHC and/or PCR, Dressler and colleagues compared outcomes separately by assay method (IHC, FISH, or PCR) for HER2 status subgroups from each dose arm (n=524, 33.8 percent of originally randomized patients).

In the GONO-MIG-1 study, Del Mastro and colleagues (2004, 2005) randomized 1,214 patients to either six cycles of CEF every three weeks (FEC21) or up to nine cycles at the same dose (600/60/600 mg/m²) every two weeks (FEC14). The analysis by HER2 status included 731 (60 percent) of originally randomized patients.

Studies on regimens with a taxane. Two randomized, controlled trials investigated effects of HER2 status on outcomes of regimens with versus without a taxane (Hayes, Thor, Dressler, et al., 2007; Martin, Pienkowski, Mackey, et al., 2005). Hayes and colleagues (2007; CALGB trial 9344) randomized 3,121 patients to doxorubicin plus cyclophosphamide (AC) followed by paclitaxel or observation. The trial used a 3 × 2 factorial design to compare three doses of doxorubicin in AC, each followed or not by paclitaxel. Since outcomes were not statistically significantly different across doxorubicin doses, the analysis of outcomes with versus without paclitaxel by HER2 status pooled patients from all three doxorubicin doses. Two groups of 750 patients each were randomly selected for this correlative analysis, but tissue blocks were available and analyzed for only 1,322 (42 percent of those originally randomized).

Martin, Pienkowski, Mackey, et al. (2005) stratified patients (n=1,491) by number of involved axillary lymph nodes and randomized them to six three-week cycles of docetaxel plus doxorubicin plus cyclophosphamide (TAC) or fluorouracil plus doxorubicin plus cyclophosphamide (FAC). The preplanned analysis by HER2 status included 1,262 (85 percent) of originally randomized patients. Patients were not stratified by HER2 status. In the TAC group, 20.8 percent were HER2 positive and 15.4 percent lacked tumor specimens for measuring HER2; in the FAC group, 22 percent were HER2 positive and 15.3 percent lacked tumor specimens. The study does not report the distribution of other prognostic factors by treatment group and HER2 status combined, which would be useful in ensuring balance in this subset of trial patients with known HER2 status.

Evidence hierarchy. The first section of Table 11 categorizes available studies on HER2 status and outcomes of adjuvant chemotherapy according to the evidence hierarchy used in this evidence report (see “Methods”). No trials stratified patients by HER2 status or randomized patients to therapy guided or not guided by HER2 status, the highest category of evidence. Only one randomized, controlled trial that compared TAC versus FAC, reported a preplanned multivariate subgroup analysis (Martin, Pienkowski, Mackey, et al., 2005). Eight randomized, controlled trials, one that compared CMF versus no or minimal CMF, four that compared CMF versus an anthracycline-based regimen, two that compared different doses or schedules of anthracycline-based regimens, and one that compared AC alone versus followed by paclitaxel, reported post-hoc multivariate subgroup analyses. Finally, single-arm data from two reports provided univariate analyses by HER2 status.

Study quality assessment. The first section of Table 12 shows that, of nine studies that analyzed the relationship of HER2 status to outcome differences in previously completed randomized, controlled trials on adjuvant chemotherapy, each was prospectively designed; included a large, well-defined and representative study population; and treated patients in each study arm homogeneously, or used rule-based selection for non-study therapies. However, only two reports (Dressler, Berry, Broadwater, et al., 2005; Martin, Pienkowski, Mackey, et al., 2005) included a prespecified hypothesis on the relationship of HER2 status to differences between regimens in treatment outcome. Each study adequately described the assays and thresholds they used to for classify patients' HER2 status, but only five (Colozza, Sidoni, Mosconi, et al., 2005; Dressler, Berry, Broadwater, et al., 2005; Del Mastro, Bruzzi, Nicolo, et al., 2005; Tanner, Isola, Wiklund, et al., 2006; Hayes, Thor, Dressler, et al., 2007) reported that individuals who assessed HER2 status were blinded to patient and tumor factors and to treatment outcomes. Only three studies from randomized, controlled trials (Moliterni, Menard, Valagussa, et al., 2003; Pritchard, Shepherd, O'Malley, et al., 2006; Martin, Pienkowski, Mackey, et al., 2005) included ≥85 percent of originally randomized patients. However, a fourth (Hayes, Thor, Dressler, et al., 2007) randomly selected two large subsets (n=750 each) and separately analyzed more than 85 percent of patients in each. Six studies from randomized, controlled trials (Moliterni, Menard, Valagussa, et al., 2003; Colozza, Sidoni, Mosconi, et al., 2005; Pritchard, Shepherd, O'Malley, et al., 2006; Knoop, Knudsen, Balslev, et al., 2005; Dressler, Berry, Broadwater, et al., 2005; Hayes, Thor, Dressler, et al., 2007) had 9 or more years' median follow-up, but in only one of these (Moliterni, Menard, Valagussa, et al., 2003) was median follow-up ~15 years. Reporting of methodologic details for multivariate analyses was inadequate in all studies.

Six studies on preoperative neoadjuvant chemotherapy. Six studies, including one randomized, controlled trial and five uncontrolled series, compared outcomes by HER2 status for patients undergoing neoadjuvant (preoperative) chemotherapy. The randomized, controlled trial (Learn, Yeh, McNutt, et al., 2005) randomized patients (n=144) to one of three arms: doxorubicin plus cyclophosphamide (AC), AC plus docetaxel (AC+D), or AC followed by docetaxel after resection (AC→D). Analysis of pathologic outcomes at resection pooled patients from the AC and AC→D arms and compared these versus the AC+D arm. The secondary, unplanned analysis by HER2 status included 104 (72 percent) of originally randomized patients.

Two uncontrolled series, one prospective (n=232, Arriola, Moreno, Varela, et al., 2006) and the other retrospective (n=67, Park, Kim, Lim, et al., 2003) reported on patients given doxorubicin alone. One uncontrolled retrospective series (n=97, Zhang, Yang, Smith, et al., 2003) reported on patients given three to six cycles of fluorouracil plus doxorubicin plus cyclophosphamide (FAC). A similar uncontrolled, retrospective series (n=77; Tinari, Lattanzio, Natoli, et al., 2006) reported on patients given three to six cycles of fluorouracil plus epirubicin plus cyclophosphamide. Finally, one uncontrolled retrospective series (n=54, Tulbah, Ibrahim, Ezzat, et al., 2002) reported on patients given three or four cycles of paclitaxel plus cisplatin. Each series reported outcomes by HER2 status for all patients (n=232 for the Arriola and co-workers series; n<100 for each of the others).

Evidence hierarchy. As shown in Section 2 of Table 11, no studies on neoadjuvant chemotherapy reported either of the two highest evidence categories. The only study from a randomized, controlled trial on neoadjuvant chemotherapy (Learn, Yeh, McNutt, et al., 2005) reported a post-hoc multivariate subgroup analysis. Two series (Park, Kim, Lim, et al., 2003; Zhang, Yang, Smith, et al., 2003) reported post-hoc multivariate subgroup analyses, while three series reported univariate analyses only.

Study quality assessment. Section 2 of Table 12 shows that only two studies (one a randomized, controlled trial) on neoadjuvant chemotherapy were prospectively designed (Learn, Yeh, McNutt, et al., 2005; Arriola, Moreno, Varela, et al., 2006), and only one reported a prespecified hypothesis for the relationship of HER2 status to outcome of neoadjuvant chemotherapy (Arriola, Moreno, Varela, et al., 2006). Only one study (Arriola, Moreno, Varela, et al., 2006) included ≥100 patients. Four of six (Arriola, Moreno, Varela, et al., 2006; Park, Kim, Lim, et al., 2003; Tulbah, Ibrahim, Ezzat, et al., 2002; Tinari, Lattanzio, Natoli, et al., 2006; but not Learn, Yeh, McNutt, et al., 2005) adequately described the assays and thresholds used to classify patients' HER2 status, but only two (Tulbah, Ibrahim, Ezzat, et al., 2002; Tinari, Lattanzio, Natoli, et al., 2006) reported HER2 assays were scored by assessors blinded to patient and tumor characteristics and treatment outcomes. Patients in each study were treated homogeneously, and each series, but not the randomized, controlled trial (Learn, Yeh, McNutt, et al., 2005), reported on all enrolled patients. Follow-up was not an issue for any study on neoadjuvant therapy, since the outcome of interest was pathologic responses at resection. Reporting of methodologic details for multivariate analyses was inadequate in all studies.

Three studies on chemotherapy for advanced or metastatic breast cancer. Each was a secondary analysis from a randomized, controlled trial designed to compare outcomes of treatment regimens in populations not selected or stratified for HER2 status, and each published earlier reports comparing outcomes by treatment arm for all randomized patients. One randomized, controlled trial (n=474, Harris, Broadwater, Lin, et al., 2006; CALGB 9342) randomized patients with stage IV or inoperable disease undergoing first- or second-line therapy to three different doses of paclitaxel. The analysis of outcomes by HER2 status included 35 percent of originally randomized patients, and pooled data across all three doses. Thus, Harris and co-workers (2006) was considered a single-arm study in this systematic review.

A second randomized, controlled trial (n=326, Di Leo, Chan, Paesmans. et al., 2004) randomized patients to doxorubicin alone (A) or docetaxel alone (T). Eligibility required patients to have metastatic disease and to have failed prior CMF (either as adjuvant therapy or for metastasis), but no prior exposure to either of the randomized drug therapies. The analysis by HER2 status included 54 percent of originally randomized patients. The third randomized, controlled trial (n=516, Konecny, Thomssen, Luck, et al., 2004) randomized patients to first-line therapy for metastatic disease with either epirubicin plus cyclophosphamide (EC) or epirubicin plus paclitaxel (ET). Up to one prior hormonal therapy for metastasis was permitted, with patients stratified by prior hormonal therapy. The analysis by HER2 status included 53 percent of originally randomized patients.

Evidence hierarchy. As shown in Section 3 of Table 11, no studies on advanced or metastatic disease reported evidence of the two highest categories. Two randomized, controlled trials (Di Leo, Chan, Paesmans. et al., 2004; Konecny, Thomssen, Luck, et al., 2004) reported post-hoc multivariate subgroup analyses. The third study, a pooled analysis across trial treatment arms (Harris, Broadwater, Lin, et al., 2006) only reported a univariate analysis.

Study quality assessment. Section 3 of Table 12 shows that each of the three included studies on HER2 status as a predictor of chemotherapy outcomes for advanced or metastatic breast cancer was designed prospectively, but none reported a prespecified hypothesis for the effect of HER2 status on outcomes. Each study included a large, well defined, and representative study population, adequately described the HER2 assays and thresholds they used to classify patients' HER2 status, and treated patients in each study arm homogeneously. Only two of three (Harris, Broadwater, Lin, et al., 2006; Di Leo, Chan, Paesmans. et al., 2004) reported blinding HER2 assessors to patient and tumor characteristics and to treatment outcomes. Each omitted 15 percent or more of enrolled patients from the analysis of outcomes by HER2 status, and each omitted key methodologic details on their multivariate analyses from the published reports. Long-term follow-up was available in only one study (Harris, Broadwater, Lin, et al., 2006), and one did not report the median duration of follow-up (Konecny, Thomssen, Luck, et al., 2004).

Patient Characteristics

Eleven studies on postsurgical adjuvant chemotherapy. Although all investigated adjuvant chemotherapy, the eleven studies varied with respect to their patient groups' distributions of baseline characteristics and risk factors for recurrent disease (Appendix Tables IIIa-B and IIIa-C ^*, Table 10). Only a subset of these studies compared the HER2 positive and negative subgroups for baseline characteristics and risk factors. Also, only a subset of the nine randomized, controlled trials compared patients included in the analysis by HER2 status with those excluded because tissue blocks were missing or unsuitable.

Studies on CMF. Of the two CMF studies, the retrospective series by Yang, Klos, Zhou, et al. (2003) pooled data for node-negative and node-positive patients, groups that Gusterson, Gelber, Goldhirsch, et al. (2003) randomized separately to different treatment arm pairs. Yang, Klos, Zhou, et al. (2003) only reported baseline characteristics and risk factors for all patients analyzed. Gusterson, Gelber, Goldhirsch, et al. (2003) compared HER2-positive versus HER2-negative patients separately for the node-positive and node-negative groups, but did not compare those with known HER2 status versus those lacking tissue blocks for HER2 assays. In node-negative patients, HER2 positivity was statistically significantly associated with larger tumor size, hormone-receptor negativity, and higher tumor grade. In node-positive patients, HER2 positivity was statistically significantly associated with menopausal status, hormone-receptor negativity, and higher tumor grade.

Studies on regimens with versus without an anthracycline. Three (Colozza, Sidoni, Mosconi, et al., 2005; Pritchard, Shepherd, O'Malley, et al., 2006; Tanner, Isola, Wiklund, et al., 2006) of five studies comparing adjuvant regimens with versus without an anthracycline compared baseline characteristics of HER2 positive and negative subgroups. Three (Colozza, Sidoni, Mosconi, et al., 2005; Knoop, Knudsen, Balslev, et al., 2005; Tanner, Isola, Wiklund, et al., 2006) explored whether subgroups tested for HER2 status were similar to the total study population or the subgroup not tested. Two trials (Moliterni, Menard, Valagussa, et al., 2003; Pritchard, Shepherd, O'Malley, et al., 2006) determined HER2 status on 92 percent or 89 percent, respectively, of the patients originally randomized and did not report comparisons to all or omitted patients. Each trial's full treatment arms were well balanced for baseline characteristics and prognostic factors.

Moliterni, Menard, Valagussa, et al. (2003) did not report data comparing baseline factors by HER2 status. All patients in this trial had one to three positive nodes, and approximately 65 percent had tumors smaller than 2.1 cm in diameter. Colozza, Sidoni, Mosconi, et al. (2005) reported that treatment arms were well balanced, whether comparing all patients randomized or only those tested for HER2 status. However, significantly more patients randomized to epirubicin than to CMF were HER2 positive (41 percent versus 28 percent, p=.03). Progesterone receptor positivity was the only factor statistically significantly associated with HER2 positivity. This trial included node-positive and node-negative patients (4 or more positive nodes in less than 25 percent), and approximately 45 percent with tumors 2 cm or smaller in diameter.

Pritchard, Shepherd, O'Malley, et al. (2006) reported baseline characteristics of patients tested for HER2 status were similar to those of all randomized patients, but did not show data for this comparison. They showed data comparing FISH-positive and FISH-negative subgroups; except for a shift toward younger age in the FISH-positive subgroup, there were no significant differences. Just over half the patients in this trial had T2 or T3 tumors, all had positive lymph nodes, with four or more positive nodes in 37 percent and 43 percent of the FISH-negative and FISH-positive groups, respectively. Knoop and co-workers (2005) reported that among all patients tested for HER2 status, treatment arms were well balanced for prognostic factors. However, they did not report comparing the HER2-positive versus HER2-negative patients, either by treatment arms or across treatments. Tumors were larger than 2 cm diameter in approximately 60 percent of patients, and approximately 30 percent had four or more positive nodes. Tanner, Isola, Wiklund, et al. (2006) reported (but did not show data) that baseline characteristics of all patients tested for HER2 status did not differ from those of the entire trial cohort. They showed that baseline characteristics were similar for HER2-tested subgroups from each arm. However, the AuSCS arm was excluded from this review, and data were not reported comparing baseline characteristics of HER2-positive versus HER2-negative patients from the FEC arm.

Studies on dose or dose intensity of anthracycline-based regimens. Studies from randomized, controlled trials that compared dose (Dressler, Berry, Broadwater, et al., 2005) or dose intensity (Del Mastro, Bruzzi, Nicolo, et al., 2005) of anthracycline-based regimens reported baseline characteristics and prognostic factors of patients with known HER2 status were similar to those of patients omitted from the analyses, since HER2 status was unknown. Dressler and co-workers (2005) did not report data comparing baseline characteristics or prognostic factors of HER2-positive versus HER2-negative patients. Del Mastro and co-workers (2005) found a greater proportion of HER2-positive than HER2-negative patients lacking expression of both estrogen and progesterone receptors (62 percent versus 32.5 percent). Other baseline characteristics and prognostic factors were similar between subgroups by HER2 status and between treatment arms.

Studies on regimens with versus without a taxane. One of two studies from randomized, controlled trials on regimens with versus without a taxane compared baseline characteristics and prognostic factors of patient with known HER2 status versus those of patients with unknown HER2 status. The trial comparing paclitaxel versus observation after AC (Hayes, Thor, Dressler, et al., 2007) showed similar baseline characteristics, prognostic factors and overall survival in the two subgroups they randomly selected and tested for HER2 status (n=643 and 679, respectively). These subgroups were also similar to all treated patients (n=3,121), and to all non-tested patients (n=1,799). Tumor diameter was 2 cm or smaller in approximately 35 percent, and approximately 54 percent had 4 or more positive nodes. The randomized, controlled trial that compared TAC versus FAC (Martin, Pienkowski, Mackey, et al., 2005) only compared patient characteristics and prognostic factors by treatment arm for all patients randomized. Neither study compared HER2-positive versus HER2-negative patients, either pooled across treatments or by treatment arm.

Six studies on preoperative neoadjuvant chemotherapy. The randomized, controlled trial on neoadjuvant therapy (Learn, Yeh, McNutt, et al., 2005) did not compare treatment arms or patient subgroups by HER2 status (neither known versus unknown nor positive versus negative) with respect to baseline characteristics or prognostic factors. This study only reported patient and tumor characteristics for all randomized patients

Only one (Tulbah, Ibrahim, Ezzat, et al., 2002) of the five included series compared baseline characteristics and prognostic factors for HER2-positive and HER2-negative subgroups. Across all five studies, approximately 55 percent to 65 percent of included patients were positive for estrogen receptors, and 45 percent to 55 percent were positive for progesterone receptors. However, their study samples varied somewhat with respect to tumor size and number of positive nodes. The series reported by Arriola, Moreno, Varela, et al. (2006) included 30 percent T2 and 70 percent T3 tumors, with 60 percent of patients node negative and 40 percent N1. Most patients (91 percent) in the series reported by Park, Kim, Lim, et al. (2003) had tumors between 5 and 10 cm in diameter. However, they did not report nodal status. Zhang, Yang, Smith, et al. (2003) include a few patients (13 percent) with T1 tumors, and approximately 33 percent node-negative patients. Most patients in the Tulbah, Ibrahim, Ezzat, et al. (2002) series had T3 or larger tumors, and approximately 55 percent had N1 disease. They reported generally well-balanced HER2-positive and HER2-negative subgroups. Finally, 75 percent of patients in the Tinari, Lattanzio, Natoli, et al. (2006) series had tumors with diameters between 2 and 5 cm; number of positive nodes was not reported.

Three studies on chemotherapy for advanced or metastatic breast cancer. Each of three included randomized, controlled trials reported that baseline characteristics and prognostic factors for the subgroup tested for HER2 status were similar to those of patients not tested. However, none compared HER2-positive versus HER2-negative subgroups, either separately by treatment arm or across arms.

Harris, Broadwater, Lin, et al. (2006) reported the only statistically significant difference between patients tested for HER2 (and other biomarkers) and those not tested was a shorter disease-free interval among those tested (19 versus 31 months, p=.0003). Investigators attributed this difference to discarding of tissue blocks after 10 years, thus a shorter interval from diagnosis to metastasis for those with blocks remaining. Hormone-receptor status (positive in 58 percent) and median number of metastatic sites (one) were the only prognostic factors reported among those tested for HER2 status. The analysis by HER2 status pooled patients across three trial arms randomized to different paclitaxel doses.

Di Leo, Chan, Paesmans, et al. (2004) showed the subgroups tested for HER2 status from each treatment arm were similar to each other and to the untested patients. Approximately half the included patients had three or more sites of disease, and more than three fourths had visceral involvement. They did not report hormone receptor status.

Konecny, Thomssen, Luck, et al. (2004) reported no statistically significant differences in baseline characteristics or prognostic factors between groups tested for HER2 and those not tested from each treatment arm compared separately. However, the HER2-positive and HER2-negative groups were not directly compared, either separately by treatment arm or pooled across arms.