Thursday, August 24, 2023

More Updates on Low Her2 in Breast Pathology: Feisty Debate

Updated 2023 guideline for low HER2 staining for new breast cancer drugs.  Op Ed says AI holds promise.


Big news last year was that a new class of breast cancer drugs was indicated for Her2 1+ cases, a big change because previous drugs were not used in 1+ cases.   1+ cases are borderline and difficult to call. 

One of my biggest blogs during 2022 was whether low Her2 diagnosis might be a "killer app" for digital pathology or AI pathology - something that digital might do better than people.  (By which I mean, the digital scores might correlate r=0.8 with drug response while visual scores correlated r=0.7 - just a hypothetical).

Archives of Pathology and Lab Medicine have now published a new nine page article on how to rate 1+ cases, Wolff et al., and it comes with an op ed by Schnitt et al.   (And if these are of interest, see also a new review in the same journal by Sun et al. of IHC uses in breast cancer more generally - here.)

There's also a detailed review of Wolff et al., with comments by other experts, in CAP TODAY this month - by Karen Titus; find it here.  CAP TODAY doesn't shy away from pretty sharp differences of opinion among the experts interviewed.  

See also an additional, unrelated article on trends in Her2 in Genomeweb this month by Alison Kanski.

Register for a ROCHE seminar on Her2 on September 27 here.

See a JAMA Oncology letter on the topic, Robbins et al, July 2023, here.


Schnitt et al. close their op ed with a remark akin to my June 2022 speculation - 

  • While the guideline update also does not endorse the use of newer methods to quantify HER2 protein levels at the present time, newer technology such as quantitative immunofluorescence assays and the use of artificial intelligence algorithms applied to digitized slides may be particularly well suited for HER2 quantification in a consistent and reproducible manner if a clinical need persists.


Like some of the experts in CAP Today, I find the positioning of Wolff et al a little perplexing.   (Basically, it's like, Her2 negative (0,1+) now includes Her2 low (1+).)  See also a second op ed in APLM by Jaffer on the "perplexingness" of the field.  

I extract a summary from Wolff et al here:

"While it is premature to change reporting terminology for lower levels of HER2 IHC expression (eg, HER2-Low), pathology labs should include a footnote in their HER2 testing reports (IHC and in situ hybridization [ISH]) with the following recommended comment:

  • “Patients with breast cancers that are HER2 IHC 3þor IHC 2þ/ISH amplified may be eligible for several therapies that disrupt HER2 signaling pathways. Invasive breast cancers that test ‘HER2-negative’ (IHC 0, 1þor 2þ/ISH not-amplified) are more specifically considered ‘HER2-negative for protein overexpression/gene amplification’ since non-overexpressed levels of the HER2 protein may be present in these cases. Patients with breast cancers that are HER2 IHC 1þor IHC 2þ/ISH not amplified may be eligible for a treatment that targets non-amplified/non-overexpressed levels of HER2 expression for cytotoxic drug delivery (IHC 0 results do not result in eligibility currently).”

• HER2 IHC 1+ or 0 results are still both interpreted as HER2-negative (HER2 is not overexpressed) using the previously recommended scoring criteria (Figure). Importantly, the semiquantitative IHC score [bq range 0 to 3] must always be reported as well to ensure patients who meet eligibility criteria for trastuzumab deruxtecan [bq 1,2,3] can be identified. 

Example: HER2-negative for protein overexpression (1+staining present)."


AI Corner.  

The Schnitt Op Ed doesn't have an abstract, and I've clipped a GPT4 summary here:

The editorial by Schnitt et al. delves into the challenges faced by pathologists in identifying HER2-low breast cancers. These challenges emerged following the publication of the DESTINY-Breast04 (DB-04) phase 3 trial in June 2022, which indicated the efficacy of trastuzumab-deruxtecan (T-DXd), an anti-HER2 antibody-drug conjugate (ADC), in treating patients with HER2-low metastatic breast cancer.

Historically, pathology communities utilized IHC assays to detect tumors with high HER2 protein overexpression. These assays, however, were not designed to discern tumors with low HER2 protein levels. The three previous versions of the ASCO-CAP HER2 guidelines did provide a distinction between HER2=0  and  HER2=1+ cases. Still, this differentiation was deemed clinically irrelevant --  until the revelation of the DB-04 results.

The identification of HER2-low breast cancers is made more complex due to several factors:

  • Preanalytic and analytic variables such as cold ischemic time, fixation time, decalcification procedures, epitope retrieval protocols, and more.
  • High interobserver variability when differentiating between HER2 0 and 1+ cases.
  • Variability in the HER2-low phenotype over time within patients, with roughly 40% of cases showing conversions between primary tumors and subsequent recurrences.

Furthermore, uncertainty remains about the response of HER2 0 tumor patients to ADCs like T-DXd, as these patients were not part of the DB-04 trial.

In response to these challenges, the ASCO-CAP 2023 guideline update offers best practices for pathologists, which include adhering to the 2018 ASCO-CAP guidelines and considering the review of a second pathologist for borderline cases. However, the guidelines do not advise adopting the "HER2-low" terminology, cautioning against its potential misapplication.

In the face of these challenges, emerging technologies such as quantitative immunofluorescence assays and artificial intelligence algorithms applied to digitized slides hold promise for a more consistent and reproducible assessment of HER2 levels in the future.


The Jaffer Op Ed also refers to progress in Her2 automated image analysis. and in her CAP interview, she discusses this as promising, as does Yale's Dr. Rimm.

See a ChatGPT analysis of the CAP Today article here, including, asking it to highlight expert disagreements, and, to pull out info about AI.  See a ChatGPT summary of Jaffer here.