The field of artificial intelligence/machine learning applications has been growing robustly over the past five years in both pathology and radiology (see my March 2022 blog here). A few months ago, FDA released validation documents for its 2021 Paige clearance (here). One area that's very interesting has been predicting estrogen receptor status from H&E slides (here, here). But I've been unsure how much that call varies with known worse- and better-differentiated cancers on H&E (which pathologists already see) or how its utility compares with a standard of care estrogen receptor immunohistochemistry (which is standard of care and required in quality metrics plus being pivotal to many drug labels.)
But the new issue in 2022 has been measurement of low-Her2 cases. Publications back in March 2022 validated that trastuzumab-deruxtecan (Enhertu - get it?) was a successful drug in conventional, Her2-positive cases (see Cortes, NEJM 2022 here). But the big news out of ASCO this year was the big success of Enhertu also in very low Her2 cases, which would conventionally fall below Her2-drug guidelines. See Modi et al. in NEJM (387:9) here, and op ed by Hurvitz (387:75) here. NCCN has already acted to endorse Enhertu for low-Her2 cases, see June 23 notice here.
But wait, there's more. How can we define the low Her2 cases?
Pair the NCCN guideline and Modi clinical trial with an excellent cover story and deep dive article in June 2022 CAP TODAY by Karen Titus - here.
Like the Hurvitz op-ed in NEJM, this article notes that immunohistochemistry for Her2 was always designed to pick up the Her2-2+ and 3+ cases, and not sift through the 0's and 1's precisely. When you turn to 0's and 1's, pathologist interobserver reliability falls through the floor (Titus article). See also a one-hour expert panel just posted on YouTube by CAP Today here.
PAIGE and Others See AI for Accurate Low-Her2
PAIGE and others see AI as providing a solution, and potentially a rapid one, for reading low-Her2 cases accurately.
On July 11, Molika Ashford at Precision Oncology News provided an excellent article (subscription), on where companies are positioning themselves around the cutting edge of this technology. Owkin is another company in this space.
As noted by Ashford, in June, PAIGE already announced "CE-IVD and UK Conformity Assessed designations for a new, artificial intelligence-driven digital pathology assay called HER2Complete, intended to identify patients with breast cancers that express HER2, but at low enough levels that they might otherwise be missed by standard-of-care tests."
AI Studies Even Easier, Faster, than FFPE RNA Studies
For years, follow-on studies have used FFPE (paraffin blocks) from trials to study new mutations or new RNA signatures. But AI/ML studies are even more rapid and non-invasive of the archive, because only digitized slide images are required, and no original banked material (including slides) are affected in any way.* This allows research studies to progress very quickly (although it also potentially lowers barriers to entry for competitors).
Digital Pathology Reimbursement?
As I noted recently, on July 1, 2022, AMA released a long set of Category III codes as add-on codes for pathology case digital imaging (with a remark, if used clinical and not primarily for teaching or research). With the exception of just one or two PLA codes, other reimbursement codes and more important, coverage policies, for digital pathology are works in progress.
The FDA-cleared application for PAIGE Prostate slide software is basically scanning for missed neoplastic pathology. Reading a current opinion column by an experienced prostate pathologist just reminded me that this may be a quite important use case (here). See an open-access one hour video by Digital Pathology Alliance and Pathology Innovation Consortium PICC alliance on PAIGE's FDA documents, YouTube post June 22, here.
* "No slides were damaged in the making of this FDA regulatory submission."