The 2018 ASCO/CAP HER2 testing guidelines represent a significant refinement over their 2013 predecessor — tightened criteria for the 3+ positive call, revised membrane completeness requirements, and clarified rules for heterogeneous expression patterns. Yet despite this framework, published inter-observer concordance studies continue to show measurable disagreement at the boundary cases. The problem is not that pathologists are inattentive. It is that HER2 IHC sits at an inherently difficult perceptual threshold.
The Semiquantitative Scale Is Not Evenly Difficult
When pathologists discuss HER2 IHC reproducibility in the abstract, the numbers sound manageable. Overall concordance between experienced pathologists reviewing clear-cut 0, 1+, or 3+ cases routinely exceeds 90%. The challenge concentrates in one category: the 2+ equivocal band.
The 2+ designation applies when membrane staining is complete and circumferential but weak-to-moderate in intensity, in more than 10% of invasive tumor cells. The word "moderate" does exactly the work you would expect in a staining assay — it spans a continuous range of chromogenic signal that any human viewer must compress into a discrete category boundary. Published studies of inter-observer agreement for 2+ cases specifically, rather than for all HER2 cases in aggregate, show weighted kappa values ranging from approximately 0.55 to 0.72 depending on the cohort composition, staining platform, and reader experience. That range describes what most statisticians call moderate to substantial agreement — which is not the same as clinical precision.
What Disagreement Looks Like in Practice
Consider a scenario familiar to any breast pathology subspecialist: a core biopsy arrives at a cancer center pathology department, IHC-stained with the SP3 antibody, showing circumferential membranous staining in 30% of invasive cells at what the attending pathologist reads as moderate intensity. A second pathologist reviewing the same slide on the same day calls it 1+. Neither read is unreasonable — both are consistent with their respective mental calibrations of the same visual scale.
The downstream consequence is not trivial. A 2+ call triggers reflex in situ hybridization (ISH) testing — usually FISH or DISH — to resolve HER2 gene amplification status. At many academic centers this adds 3–5 days to the turnaround time and roughly $300–600 in additional assay costs per case. A 1+ call does not. The patient is either progressed toward or kept off a HER2-targeted treatment arm based on which pathologist reviewed the slide on which day.
We are not saying that individual pathologists are making errors — the 2018 guidelines acknowledge exactly this interpretive space. What we are saying is that the current manual IHC workflow builds in a reproducibility ceiling that has real clinical consequences, and the ceiling is structural, not the product of insufficient expertise.
Why the 3+ / 2+ Boundary Is Particularly Unstable
The ASCO/CAP 2018 update introduced the critical concept that strong, complete, circumferential staining in more than 10% of cells constitutes 3+. The word "strong" is calibrated against antibody-positive control cells on the same slide, not against an absolute chromogenic threshold. This is epistemically necessary — DAB development time, antibody lot variability, tissue fixation duration, and scanner exposure settings all affect the absolute brown color intensity a viewer perceives. Relative comparison to an internal control is the right answer scientifically. It is also an answer that requires consistent mental calibration across readers, institutions, and time.
Several groups have studied what happens when pathologists use the same scoring rubric but have been trained in different lab environments. Calibration drift is observable: pathologists from high-staining-intensity labs tend to perceive moderate staining as lower intensity than colleagues trained in labs running the same antibody at lower DAB concentrations. This is not laziness or sloppiness. It is a well-characterized perceptual anchoring effect applied to a visual task that lacks an absolute chromatic reference standard.
The ISH Reflex Rate as a Proxy for Uncertainty
Across academic cancer centers in the United States, ISH reflex testing rates for HER2 vary considerably — estimates from pathology department quality metrics suggest that the proportion of invasive breast cancer IHC cases triaged to reflex testing ranges from roughly 12% to 28% depending on the institution. This spread is not explained by patient population differences or varying disease stages alone. A meaningful portion reflects genuine boundary uncertainty at the 2+ call, which itself reflects underlying inter-observer variability in the IHC score that precedes it.
Reducing unnecessary ISH referrals without increasing the rate of missed 3+ positives requires not just better scoring, but better characterization of uncertainty at the point of the IHC review. A pathologist who has reviewed 15 cases before lunch and is mentally fatigued is likely to have less consistent 2+/3+ boundary discrimination than the same pathologist fresh in the morning. This is a human physiological reality, not a character deficiency, and no amount of training or rubric revision makes it go away.
What Algorithmic Analysis Offers — and Doesn't
Automated IHC scoring approaches the 2+/3+ boundary from a different direction. A convolutional model trained on annotated whole-slide images learns to quantify several features simultaneously: the proportion of cells with circumferential versus incomplete membrane staining, the continuous DAB optical density at the membrane versus cytoplasm versus background, and the spatial distribution of staining intensity across the tumor region. None of these measurements are bounded by the perceptual fatigue, anchoring, or calibration effects that affect human reviewers.
But we should be precise about what this means and what it does not mean. A well-validated algorithm with an internal concordance ICC of 0.94 for the continuous H-score underlying HER2 grading does not eliminate diagnostic uncertainty — it changes its character. The algorithm's uncertainty is quantifiable, reproducible, and stable across time of day, staining batch, and examiner experience. That is a different kind of uncertainty than what manual scoring produces, and in most clinical decision contexts it is a preferable kind. An algorithmic 2+ call accompanied by a confidence interval derived from the continuous underlying score gives the reviewing pathologist materially more information than a 2+ call with no supporting data.
The goal, properly understood, is not for software to replace the pathologist's sign-off. It is to replace the most unreliable part of the manual workflow — the unaided perceptual judgment of a continuous chromogenic signal compressed into a four-category scale — while preserving the pathologist's role in clinical interpretation and final report authorization.
The Role of Digital Pathology Infrastructure
One underappreciated driver of inter-pathologist variability is the heterogeneity of viewing conditions. A slide reviewed on a backlit glass stage under a 20x objective looks meaningfully different from the same slide scanned at 40x on an Aperio GT 450 and viewed on a calibrated 4K display. Color calibration profiles vary by scanner manufacturer. Monitor gamut varies by lab. USCAP and the Digital Pathology Association have both emphasized that standardized viewing conditions are prerequisite to any serious reproducibility program, but widespread adoption of calibrated digital review workflows at the slide-scanning level remains incomplete.
Digital pathology does not automatically solve the reproducibility problem; it relocates it. A whole-slide image is more reproducible in the sense that the same scan can be reviewed by multiple pathologists under defined and documentable conditions. But if the scanning parameters and display calibration are not standardized, the underlying perceptual task remains variable. Any algorithmic tool for IHC scoring is only as reproducible as the imaging pipeline feeding it.
Where This Leaves the Practicing Pathologist
HER2 IHC scoring variability is a documented, structurally embedded feature of the current diagnostic workflow — not a fixable anomaly that better guidelines alone will eliminate. Published concordance studies, quality audits from CAP proficiency testing programs, and the observable variation in ISH reflex rates across institutions all point to the same conclusion: the manual IHC scoring step introduces reproducibility variance that is clinically meaningful, particularly at the 2+ equivocal boundary.
The practical question for oncology pathology departments is how to systematically characterize and reduce that variance without adding workflow friction. Algorithmic analysis of the continuous underlying staining signal — not just the discrete grade — is one well-characterized approach. Structured second-read protocols for 2+ cases are another. Neither replaces the clinical judgment of the reviewing pathologist. Both reduce the dependence of the final score on which pathologist happened to review a particular slide on a particular day.
Reproducibility is not a luxury in cancer diagnostics. When the treatment decision pivots on the distance between 2+ and 3+, the measurement that drives it should be as stable and as well-characterized as anything else in the clinical workflow.