Part III · Pathology and classification · Chapter 11

Receptor assessment and the measurement problem

Three questions hide inside every biomarker claim, and most of the argument comes from answering the wrong one.

1 · ER and PR testing, thresholds, and the ER-low category

Oestrogen receptor status is a percentage. It is the proportion of invasive tumour nuclei that an observer judges to be stained, read through an antibody and a detection chemistry. The 2010 ASCO and College of American Pathologists guideline set the positive threshold at 1% of tumour nuclei1. The same document estimated that up to 20% of ER and PR determinations worldwide were inaccurate, and attributed most of that to preanalytics, thresholds and interpretation rather than to the antibody.

An NCCN task force had earlier set out quality control for receptor immunohistochemistry, ahead of the first joint guideline2. The 1% threshold was chosen so that endocrine therapy would not be withheld from a patient who might benefit. It was not chosen because biology changes at 1%. The 2020 update kept 1% to 100% as positive and added a reporting category, ER Low Positive, for samples with 1% to 10% of nuclei staining3. It also required a comment with that result, and required control status to be reported whenever staining falls between 0% and 10%.

What the low band contains is now reasonably clear. Among 465 primary cancers with matched microarray data, only 24% of tumours staining 1% to 9% were also positive by ESR1 messenger RNA4. Among tumours staining above 10%, 92% were. Within that 1% to 9% group, 48% were basal-like by PAM50 and 8% were luminal B.

Outcome data point the same way. In 3055 patients with stage II or III HER2-negative disease given neoadjuvant chemotherapy, recursive partitioning placed the cut point for pathological complete response at 9.5%5. Adjuvant endocrine therapy was associated with longer time to recurrence in patients whose tumours stained 10% or above, with a hazard ratio of 0.24 against no endocrine therapy. No such association was found in the group staining 1% to under 10%.

Hold the two claims apart. The reporting threshold is 1%, and it governs eligibility. The biology of the 1% to 10% band resembles ER-negative disease in expression profile and in observed endocrine benefit.

Progesterone receptor is measured the same way and used for a different purpose. It carries prognostic weight in the setting of an ER-positive cancer, and it is not an independent basis for offering endocrine therapy3.

In practice

Treat an ER-low result as ER positive when deciding whether to offer endocrine therapy. Expect the tumour to behave like triple-negative disease in most other respects, including chemotherapy sensitivity.

Read the mandatory comment. It exists because the panel knew the number was fragile.

Check that the report states the control status. A 2% result with a failed internal control is not the same finding as a 2% result with intact internal controls.

2 · HER2 testing algorithms, IHC, ISH, and equivocal results

HER2 status is produced by an algorithm rather than by a single assay. The first joint guideline was written because about 20% of HER2 testing was thought to be inaccurate, and it found neither immunohistochemistry nor in situ hybridisation clearly superior as a predictor of benefit6. Immunohistochemistry runs first. A score of 3+ is positive and a score of 0 or 1+ is negative. A score of 2+ triggers reflex in situ hybridisation7.

The category definitions are counting rules. Immunohistochemistry 2+ is weak to moderate complete membrane staining in more than 10% of tumour cells7. That 10% exists so that two observers can reach the same category from the same slide. It is the same kind of convention examined in Definitions and thresholds, spatial, clonal, and the ten percent convention.

Dual-probe in situ hybridisation results are sorted into numbered groups. Three of them are uncommon and together account for roughly 5% of cases7. Group 2 is a HER2 to CEP17 ratio of 2.0 or above with an average HER2 copy number below 4.0. Group 3 is a ratio below 2.0 with an average copy number of 6.0 or above. Group 4 is a ratio below 2.0 with an average copy number of at least 4.0 and below 6.0. The 2018 update required concomitant immunohistochemistry review for all three, and recommended it for every single-probe result.

The abolition of the equivocal category is often read as new knowledge. It is not. An equivocal result is a statement about an assay, not a state of a tumour. The panel chose to resolve that uncertainty inside a defined workflow rather than report it, because a treatment decision needs a binary answer. Nothing about the underlying distribution of copy number changed.

Two findings the algorithm does not produce are worth naming. It does not detect activating HER2 mutations in the absence of amplification, which are a separate and targetable event8,9. It does not describe heterogeneity unless the pathologist chooses to, and HER2 heterogeneity sets out what is lost when it does not.

The 2023 update reviewed the literature and affirmed the 2018 recommendations without change10.

Caution

The algorithm converts a continuous quantity into a binary category. It does that well where the signal is strong and badly where it is weak. Reproducibility at 3+ is not evidence of reproducibility at the bottom of the scale. The two figures are routinely quoted as though one supported the other.

3 · HER2-low and HER2-ultralow, entity or assay artifact

HER2-low is not a measurement. It is a name for a range of an existing measurement, and the range was drawn by a trial entry criterion. DESTINY-Breast04 enrolled patients whose tumours scored immunohistochemistry 1+, or 2+ without amplification, and excluded immunohistochemistry 011. DESTINY-Breast06 extended the range downward. Of 866 patients randomised, 713 had HER2-low disease and 153 had HER2-ultralow disease, defined as immunohistochemistry 0 with membrane staining12. Median progression-free survival in the HER2-low group was 13.2 months with trastuzumab deruxtecan against 8.1 months with chemotherapy, a hazard ratio of 0.62.

A category invented for enrolment now governs access to an effective drug. The question is whether the boundary it rests on can be measured.

The proficiency-testing data are the least flattering and the most representative of ordinary practice. Across College of American Pathologists surveys covering 1391 to 1452 laboratories, 19% of circulated cases produced concordance of 70% or below for the distinction between 0 and 1+13.

Expert panels do better, and not by much. Sixteen expert breast pathologists in the United Kingdom and Ireland scored 50 digitised slides and reached absolute agreement in 6% of cases, all of them 3+14. Agreement was highest, at 86%, when scores were collapsed to 0 against everything else. In a separate study, 36 pathologists at four centres rescored 50 specimens previously called immunohistochemistry 015. The Fleiss kappa for separating null from ultralow was 0.230. Collapsing the two back into a single immunohistochemistry 0 category raised the kappa to 0.344, and raised the proportion of cases with high agreement from 4% to 32%.

Automation has not closed the gap. Ten independently developed computational pathology models were applied to 1124 whole-slide images from 733 patients16. Median pairwise overall percent agreement on the four-category score was 65.1%, with a kappa of 0.51. For 3+ against not 3+ the same models agreed 97.3% of the time, with a kappa of 0.86. The same analysis found that disagreement related more to interpretation of staining than to detection of tumour cells. The models fail where the observers fail, and for the same reason.

The reproducibility of this boundary among trained observers, and the guideline position on it, are set out and applied in HER2-low and ultralow reframed as a heterogeneity problem. The question this chapter carries is different, and it stays open. Either the assay is failing to measure a real difference at low expression, or there is no difference at low expression to measure. The response data in Implications for conjugate payload delivery and the bystander effect are compatible with the second reading and do not establish it.

Caution

A drug effect demonstrated within an assay-defined population is not evidence that the assay boundary is reproducible outside the trial. Those are two claims with two bodies of evidence. The first is supported for trastuzumab deruxtecan. The second is contradicted by every concordance study cited here.

4 · Intratumoral variation as a source of discordant reports

Two reports on the same patient that disagree have at least four possible explanations, and they call for different responses. The first is genuine spatial variation within the tumour. The second is sampling, meaning that two pieces of tissue contained different things. The third is preanalytics, which is the subject of Preanalytics, fixation, and the invisible sources of error. The fourth is observer variation.

Only the first is a property of the disease. The other three are properties of how the disease was handled and read. A report that says the result has changed is not saying which of the four occurred.

Discordance is common enough to be the expected finding rather than the surprising one. HER2 status has been shown to differ between primary carcinomas and synchronous axillary node metastases in a multicentre retrospective series17. Receptor conversion between primary and distant metastasis has been quantified in a systematic review and meta-analysis18. Separate foci within one breast carry genomic differences from each other19.

An arithmetic contribution is easy to overlook. A percentage estimated by counting nuclei carries an uncertainty that depends on how many nuclei were counted. Twenty nuclei give a wide interval around any proportion. A category boundary placed close to 0% or close to 100% is the least stable place to put a decision, because a small absolute change in the count moves the case across it.

That yields a usable rule. A discordance that would change treatment is worth resolving, by repeat testing or by testing a second block. A discordance that would not change treatment is worth recording and not pursuing. The clinical logic of receptor discordance between sites is developed in Receptor discordance and conversion, and the sampling problem itself in Spatial heterogeneity.

5 · Ki-67, proliferation indices, and interobserver reproducibility

Ki-67 is reported as the percentage of invasive tumour nuclei that stain. It is continuous, it has no natural discontinuity, and every cut point in use was chosen by somebody.

The reproducibility data are the most quotable numbers in this chapter. Eight laboratories, among the most experienced in the world, scored 100 breast cancers arranged on tissue microarrays, each using its own method20. Within a laboratory agreement was excellent, with an intraclass correlation coefficient of 0.94. Between laboratories it was moderate, at 0.71 with central staining and 0.59 when each laboratory stained its own slides.

The more useful figure is the spread of the laboratory means. Across the same 100 cases, the geometric mean Ki-67 reported by each laboratory ranged from 7.1% to 23.9% with central staining, and from 6.1% to 30.1% with local staining20.

Set that against the cut points in circulation. The immunohistochemical surrogate for luminal B disease was derived with a receiver operating characteristic optimum of 13.25%21. A threshold at 13.25% cannot survive a between-laboratory spread running from 7.1% to 23.9% on identical material. The same tumour sits above the line in one laboratory and below it in another.

The International Ki67 in Breast Cancer Working Group has taken the honest route twice. The 2011 recommendations set out preanalytical and scoring requirements for harmonisation22. The 2021 update reached three separate conclusions23. Analytical validity is attainable with a standardised visual scoring method. Clinical validity is established. Clinical utility is confined to prognosis in anatomically favourable ER-positive HER2-negative disease, using thresholds of 5% or below and 30% or above. Those are three different answers about one assay, and Analytical validity, clinical validity, and clinical utility defined explains why they have to be stated separately.

Ki-67 nonetheless appears inside an eligibility criterion. In monarchE, patients with one to three positive nodes qualified if they also had a tumour of 5 cm or more, histological grade 3, or a central Ki-67 of 20% or above24. The assay was run centrally, which is the condition under which this marker behaves best.

In practice

Ask how the Ki-67 was produced before using it. A counted score from a defined field is not the same measurement as a visual estimate, and only the first is reproducible enough to compare against a published threshold.

Do not compare a Ki-67 from one laboratory against a threshold derived in another.

Use the extremes and distrust the middle. A value of 3% and a value of 45% carry information. A value of 18% mostly records that somebody scored a slide.

6 · Preanalytics, fixation, and the invisible sources of error

Everything above assumes that the tissue reaching the antibody still contains what it contained in the patient. Often it does not, and the report does not say so.

Cold ischaemic time is the interval between devascularisation and fixation. Ten invasive carcinomas were divided into fragments and fixed after 0, 10 and 30 minutes, and after 1, 2, 4 and 8 hours, with one fragment held overnight in saline25. Mean oestrogen receptor scores began to fall at 2 hours and progesterone receptor scores at 1 hour. HER2 fluorescence in situ hybridisation became compromised for interpretation from 1 hour, and significantly so by 2 hours.

A second study varied both time and temperature across 25 tumours26. Significant reduction in hormone receptor and HER2 staining generally did not appear until 4 hours in refrigerated samples and 2 hours in samples left at room temperature. Refrigeration buys time. The guideline limit of under 1 hour was judged prudent.

Decalcification is the trap that catches people. Fifteen fresh primary carcinomas were divided and exposed to formic acid or hydrochloric acid for 1 or 5 hours27. One hour of formic acid changed no marker measurably. Five hours of formic acid reduced Ki-67 alone. Either 1 or 5 hours of hydrochloric acid significantly reduced oestrogen receptor, progesterone receptor, HER2, Ki-67 and GATA-3.

That finding is specific and directly actionable. A bone metastasis decalcified in hydrochloric acid can return a false triple-negative result on a tumour that is not triple negative.

None of this appears on the report. A percentage is printed without an interval, and the conditions that produced it are usually absent. The 2020 guideline pushed back on one corner of the problem by requiring that control status be reported whenever ER staining falls between 0% and 10%3.

Caution

A receptor result from a decalcified bone core is not equivalent to a result from soft tissue. If a bone biopsy is the only material and the result is negative, ask which decalcification agent was used. If it was hydrochloric acid, the negative result is not interpretable.

7 · Analytical validity, clinical validity, and clinical utility defined

Most arguments about biomarkers are arguments in which the two sides are answering different questions. There are three questions. They are ordered, and answering one says nothing about the others. The framework below is the one the Evaluation of Genomic Applications in Practice and Prevention working group set out for genomic tests, and it transfers to immunohistochemistry without modification28.

Analytical validity asks whether the assay measures what it claims to measure, accurately and reproducibly. It is a property of the assay, the laboratory and the observer. It is established by replicate measurement, by proficiency testing and by concordance studies. No patient outcome is involved at this stage. The Ki-67 interlaboratory data in Ki-67, proliferation indices, and interobserver reproducibility and the HER2 concordance data in HER2-low and HER2-ultralow, entity or assay artifact are analytical validity findings and nothing else.

Clinical validity asks whether the measurement is associated with a clinical state or an outcome. It is a property of the marker in a defined population. It is established by correlation in a cohort with follow-up. A marker can have excellent clinical validity while measuring something nobody can act on. Most prognostic factors in this book sit here and go no further.

Clinical utility asks whether using the test changes what is done, in a way that leaves patients better off. It is a property of a decision rather than of a marker. It requires a comparison between a strategy that uses the test and a strategy that does not. That is a trial design question, not a laboratory question.

The order matters because each level depends on the one before it. A measurement that cannot be reproduced cannot correlate reliably with anything. A measurement that does not correlate with outcome cannot improve a decision. The implication does not run in the other direction at any step.

One study design deserves naming, because it produced much of the usable evidence in this field. Simon and colleagues set out the conditions under which an analysis of archived specimens can support a marker claim29. The requirements are specimens from a completed randomised trial, in sufficient number and adequately representative of the enrolled population, with a fully prespecified analysis plan, and ideally confirmation in a second independent trial. An analysis meeting those conditions carries close to the weight of a prospective test. An analysis that does not meet them is hypothesis generation with a p value attached.

Three worked examples follow, and each one fails at a different point.

Ki-67 fails at the first level. Analytical validity is attainable only under a standardised scoring method, and is otherwise poor between laboratories20. Clinical validity is accepted. Clinical utility was judged present only for prognosis in anatomically favourable ER-positive HER2-negative disease23. One assay, three different answers, and a clinician who conflates them will over-read the result.

The multigene prognostic assays passed all three levels, and the way they did it is the instructive part. Clinical validity came from archived specimens. Clinical utility required randomising the decision itself. TAILORx randomised patients with an intermediate 21-gene recurrence score between endocrine therapy alone and chemoendocrine therapy30. MINDACT randomised patients whose clinical risk and genomic risk disagreed31. Neither trial tested a marker. Each tested a strategy for using one.

HER2 immunohistochemistry 0 against 1+ inverts the order, and this is the uncomfortable case. Analytical validity is poor13. Clinical validity is unestablished, and the guideline says so, holding that current data do not support the boundary as a prognostic or predictive threshold10. The boundary nonetheless governs access to treatment, because a registration trial used it to enrol11. That is regulatory utility resting on an unvalidated measurement. It is a coherent position for a regulator and an uncomfortable one for a pathologist, and naming it as such is more useful than pretending the levels were satisfied in order.

Caution

A marker with strong clinical validity and no clinical utility is the normal case rather than a failure. It means that no available action depends on the answer. Ordering such a test generates information with no decision attached to it, and that information still has to be explained to the patient.

The rest of this book applies this framework rather than rebuilding it. When a later chapter says an assay is reproducible, that is a claim about analytical validity. When it says a marker predicts outcome, that is clinical validity. When it says a test should change management, that is clinical utility, and it needs a comparison behind it.

8 · Reading a pathology report critically

A pathology report is a set of conclusions with the uncertainty removed. Reading it critically means putting some of the uncertainty back, in the places where it changes what you do.

In practice

Establish what was measured on. Ask which specimen produced each result. A core, an excision, a node and a distant metastasis are four different tumours for this purpose. Ask whether the specimen was decalcified, and with what.

Establish how each number was produced. A percentage needs a denominator. Ki-67 needs a method, counted or estimated. A HER2 result needs tracing through the algorithm, so ask whether it came from immunohistochemistry alone, and if in situ hybridisation was done, which group the case fell into.

Read the comment. Guidelines require comments in exactly the places where the number is least trustworthy. The ER Low Positive comment and the HER2 comment on distinguishing 0 from 1+ both exist because the panel knew the result was fragile3,10.

Check the internal controls. Normal breast epithelium in the same section is the built-in control for ER and PR. For a result between 0% and 10%, control status must be reported3. If it is absent, the result is not yet interpretable.

Look for the phenotype that does not fit. A grade 1 tubular or classical lobular carcinoma reported as ER negative is more likely to be an assay failure than a discovery. Morphology and receptor status usually agree, and a disagreement is a reason to repeat rather than to accept.

Ask what is not in the report. Heterogeneity is described only if the pathologist chose to describe it, and its absence from the report is not its absence from the tumour HER2 heterogeneity. Tumour cellularity governs every molecular assay run on the same block, and it is often omitted.

Know when to repeat. Repeat when the result sits close to a threshold that carries a decision. Repeat when it conflicts with morphology, when the specimen was decalcified, or when it would change management but rests on a single small core.

One habit covers most of this. When a report gives a number that will change what you do, find out what that number is a number of, and how many things were counted to produce it. The rest tends to follow.

References

All 31 references
  1. Hammond MEH, Hayes DF, Dowsett M, et al. ASCO/College of American Pathologists guideline recommendations for immunohistochemical testing of estrogen and progesterone receptors in breast cancer. J Clin Oncol 2010 28:2784-2795. PMID 20404251
  2. Allred DC, Carlson RW, Berry DA, et al. NCCN Task Force Report: Estrogen Receptor and Progesterone Receptor Testing in Breast Cancer by Immunohistochemistry. J Natl Compr Canc Netw 2009 7 Suppl 6:S1-S21; quiz S22-3. PMID 19755043
  3. Allison KH, Hammond MEH, Dowsett M, et al. Estrogen and progesterone receptor testing in breast cancer: ASCO/CAP guideline update. J Clin Oncol 2020 38:1346-1366. PMID 31928404
  4. Iwamoto T, Booser D, Valero V, et al. Estrogen receptor mRNA and ER-related gene expression in breast cancers that are 1% to 10% ER-positive by immunohistochemistry. J Clin Oncol 2012 30:729-734. PMID 22291085
  5. Fujii T, Kogawa T, Dong W, et al. Revisiting the definition of estrogen receptor positivity in HER2-negative primary breast cancer. Ann Oncol 2017 28:2420-2428. PMID 28961844
  6. Wolff AC, Hammond ME, Schwartz JN, et al. American Society of Clinical Oncology/College of American Pathologists guideline recommendations for human epidermal growth factor receptor 2 testing in breast cancer. J Clin Oncol 2007 25:118-45. PMID 17159189
  7. Wolff AC, Hammond MEH, Allison KH, et al. Human epidermal growth factor receptor 2 testing in breast cancer: ASCO/College of American Pathologists clinical practice guideline focused update. J Clin Oncol 2018 36:2105-2122. PMID 29846122
  8. Bose R, Kavuri SM, Searleman AC, et al. Activating HER2 mutations in HER2 gene amplification negative breast cancer. Cancer Discov 2013 3:224-237. PMID 23220880
  9. Hyman DM, Piha-Paul SA, Won H, et al. HER kinase inhibition in patients with HER2- and HER3-mutant cancers. Nature 2018 554:189-194. PMID 29420467
  10. Wolff AC, Somerfield MR, Dowsett M, et al. Human epidermal growth factor receptor 2 testing in breast cancer: ASCO-College of American Pathologists guideline update. J Clin Oncol 2023 41:3867-3872. PMID 37284804
  11. Modi S, Jacot W, Yamashita T, et al. Trastuzumab deruxtecan in previously treated HER2-low advanced breast cancer. N Engl J Med 2022 387:9-20. PMID 35665782
  12. Bardia A, Hu X, Dent R, et al. Trastuzumab deruxtecan after endocrine therapy in metastatic breast cancer. N Engl J Med 2024 391:2110-2122. PMID 39282896
  13. Fernandez AI, Liu M, Bellizzi A, et al. Examination of low ERBB2 protein expression in breast cancer tissue. JAMA Oncol 2022 8:1-4. PMID 35113160
  14. Zaakouk M, Quinn C, Provenzano E, et al. Concordance of HER2-low scoring in breast carcinoma among expert pathologists in the United Kingdom and the republic of Ireland. Breast 2023 70:82-91. PMID 37419078
  15. Wu S, Shang J, Li Z, et al. Interobserver consistency and diagnostic challenges in HER2-ultralow breast cancer: a multicenter study. ESMO Open 2025 10:104127. PMID 39891991
  16. McKelvey B, Torres-Saavedra PA, Li J, et al. Agreement across 10 artificial intelligence models in assessing HER2 expression in breast cancer whole-slide images. Mod Pathol 2026 39:100944. PMID 41489584
  17. Ieni A, Barresi V, Caltabiano R, et al. Discordance rate of HER2 status in primary breast carcinomas versus synchronous axillary lymph node metastases: a multicenter retrospective investigation. Onco Targets Ther 2014 7:1267-1272. PMID 25050068
  18. Schrijver WAME, Suijkerbuijk KPM, van Gils CH, et al. Receptor conversion in distant breast cancer metastases: a systematic review and meta-analysis. J Natl Cancer Inst 2018 110:568-580. PMID 29315431
  19. Desmedt C, Fumagalli D, Pietri E, et al. Uncovering the genomic heterogeneity of multifocal breast cancer. J Pathol 2015 236:457-466. PMID 25850943
  20. Polley MY, Leung SC, McShane LM, et al. An international Ki67 reproducibility study. J Natl Cancer Inst 2013 105:1897-1906. PMID 24203987
  21. Cheang MC, Chia SK, Voduc D, et al. Ki67 index, HER2 status, and prognosis of patients with luminal B breast cancer. J Natl Cancer Inst 2009 101:736-750. PMID 19436038
  22. Dowsett M, Nielsen TO, A'Hern R, et al. Assessment of Ki67 in breast cancer: recommendations from the International Ki67 in Breast Cancer working group. J Natl Cancer Inst 2011 103:1656-1664. PMID 21960707
  23. Nielsen TO, Leung SCY, Rimm DL, et al. Assessment of Ki67 in breast cancer: updated recommendations from the International Ki67 in Breast Cancer Working Group. J Natl Cancer Inst 2021 113:808-819. PMID 33369635
  24. Johnston SRD, Harbeck N, Hegg R, et al. Abemaciclib combined with endocrine therapy for the adjuvant treatment of HR+, HER2-, node-positive, high-risk, early breast cancer (monarchE). J Clin Oncol 2020 38:3987-3998. PMID 32954927
  25. Khoury T, Sait S, Hwang H, et al. Delay to formalin fixation effect on breast biomarkers. Mod Pathol 2009 22:1457-1467. PMID 19734848
  26. Yildiz-Aktas IZ, Dabbs DJ, Bhargava R. The effect of cold ischemic time on the immunohistochemical evaluation of estrogen receptor, progesterone receptor, and HER2 expression in invasive breast carcinoma. Mod Pathol 2012 25:1098-1105. PMID 22460807
  27. Clark BZ, Yoest JM, Onisko A, Dabbs DJ. Effects of hydrochloric acid and formic acid decalcification on breast tumor biomarkers and HER2 fluorescence in situ hybridization. Appl Immunohistochem Mol Morphol 2019 27:223-230. PMID 28877070
  28. Teutsch SM, Bradley LA, Palomaki GE, et al. The Evaluation of Genomic Applications in Practice and Prevention (EGAPP) Initiative: methods of the EGAPP Working Group. Genet Med 2009 11:3-14. PMID 18813139
  29. Simon RM, Paik S, Hayes DF. Use of archived specimens in evaluation of prognostic and predictive biomarkers. J Natl Cancer Inst 2009 101:1446-1452. PMID 19815849
  30. Sparano JA, Gray RJ, Makower DF, et al. Adjuvant chemotherapy guided by a 21-gene expression assay in breast cancer. N Engl J Med 2018 379:111-121. PMID 29860917
  31. Cardoso F, van 't Veer LJ, Bogaerts J, et al. 70-gene signature as an aid to treatment decisions in early-stage breast cancer. N Engl J Med 2016 375:717-729. PMID 27557300