Ki-67 Proliferation Index
The percentage of tumour cells staining for Ki-67 — a prognostic and predictive breast biomarker whose clinical value is limited less by biology than by how irreproducibly it is stained and scored.
What it is
Ki-67 is a nuclear antigen expressed in proliferating cells, and the Ki-67 index is the proportion of invasive tumour cells with positive nuclear staining, read off an immunohistochemical stain and reported as a percentage. In breast cancer it is used as a prognostic and predictive marker — a proliferation readout that sits alongside the mitotic component of grade rather than replacing it.
The number looks simple and is not, for reasons that are almost entirely measurement rather than biology. Three decisions sit inside the single percentage, and each is a documented source of variation:
- Staining. Ki-67 is an IHC stain, so fixation, antibody clone, and protocol all move the result before anyone counts anything. This is between-laboratory variation of the kind Scanner and Stain Variability describes and the kind external quality assessment schemes exist to catch.
- Where to score. Global average across the section, versus the most proliferative hotspot, versus the invasive edge — these can give materially different indices on the same slide, and heterogeneous expression makes the choice consequential rather than cosmetic.
- How many cells, and the threshold. How many tumour nuclei are counted sets the precision; which cut-point (if any) converts the percentage into a category sets whether two laboratories agree on the clinical call even when their raw counts are close.
The consequence is that Ki-67’s headline weakness is poor reproducibility — between readers and between laboratories — which is exactly why it has attracted repeated national and international standardisation efforts rather than being left to local habit.
Why it matters for my work
It is the concrete marker behind the Biomarker Cut Points page. That page already uses “Ki-67 percentage” as its running example of a continuous marker that gets dichotomised. Ki-67 is where the optimal-cutpoint hazard stops being abstract: different published thresholds for the same marker are a recurring source of irreproducibility, so a pre-specified guideline cut-point is the defensible alternative — and the UK recommendation (sources/papers/shaaban-2026-uk-ki67-recommendations.md) is the national guideline that supplies one. [unverified] whether the paper endorses a specific threshold and how it derived it — that detail was not readable from the abstract.
It is the most likely readout of Aiforia Breast. The endpoint of the Aiforia evaluation is undocumented, but Ki-67 / IHC quantification is the named candidate given the vendor’s scope. If that is the readout, then the UK recommendation is the human reference standard the algorithm should be measured against, and the “AI improves reproducibility over manual scoring” claim is precisely what a routine-workflow evaluation would be testing locally. [unverified]
Its reproducibility problem is an Interobserver Agreement problem. Ki-67 is a canonical low-agreement IHC marker, so any claim that digital or AI scoring improves it is a claim about raising the agreement ceiling — the exact quantity that governs whether an AI comparison against a single human reader can mean anything.
How it connects
Biomarker Cut Points — Ki-67 percentage is that page’s canonical dichotomised marker; the optimal-cutpoint trap is why a national guideline threshold matters here rather than a lab-derived one.
Interobserver Agreement — poor between-reader and between-lab agreement is Ki-67’s defining weakness, and the reproducibility claim in the guidance is a claim about this ceiling.
Mitotic Count — the other proliferation measure in breast pathology; both estimate how fast the tumour divides, both degrade on a screen, and both are near-threshold-sensitive, but Ki-67 fails through staining and denominator choice where mitotic count fails through hotspot, area calibration and z-axis.
Nottingham Grading — the mitotic component of grade is the proliferation signal Ki-67 parallels, which is why the two are often discussed and occasionally conflated.
Scanner and Stain Variability — Ki-67 is an IHC stain, so between-laboratory staining spread is a first-order confounder before any scoring rule is applied.
Intratumoral Heterogeneity — hotspot-versus-global scoring is the same sampling problem that makes any single score a summary of a spatial distribution.
Aiforia Breast — the group’s breast AI evaluation, whose likely readout this marker is and whose reference standard the guidance governs.
Turnaround Time — automated Ki-67 scoring removing a manual counting step is a segment-level TAT effect, and the guidance’s “reduced turnaround time” claim lands here.
Spatial Proteomics — the fifty-marker version of the same measurement problem, and a reason for caution: scoring one marker reproducibly is already hard enough to need national guidance.
Shaaban 2026 — UK Recommendations for Ki-67 Immunohistochemical Staining and Interpretation in Breast Cancer — the national guidance behind this page: how to stain, score and report Ki-67, with digital pathology and AI positioned explicitly as the route to the reproducibility the marker currently lacks.
Open questions
- Which scoring convention does the UK recommendation endorse — global, hotspot, or a fixed tumour-cell count — and which antibody clone and cut-point? All
[unverified]; the abstract does not say and the full text is paywalled. - Is the Aiforia Breast readout actually Ki-67? If so, this concept governs its reference standard directly; if not, the link weakens to background.
- Does the department have a house Ki-67 protocol, and is it audited against an EQA scheme the way the UK guidance assumes? Not recorded.
[unverified] - Is there an in-house agreement number for Ki-67 (reader-vs-reader or human-vs-AI) that would let the guidance’s reproducibility claim be checked locally rather than imported?