This Week in Pathology — Sep 26, 2026

This AudioScholar briefing is generated by artificial intelligence for healthcare professionals and trainees. It is not medical advice.

Welcome to This Week in Pathology. This week we're covering 10 notable papers spanning reproducibility and grading in surgical pathology, the shifting role of molecular testing and artificial intelligence in thyroid nodules, and ancillary immunohistochemical and sequencing tools that resolve diagnostic ambiguity. Let's dive in.

We start with the perennial problem of observer agreement, where three papers this week ask whether structured rules can make subjective morphology more reproducible. In Histopathology, Loneman and colleagues report a real-world validation of the 2022 Banff consensus algorithm for grading macrovesicular steatosis on frozen donor liver biopsies [1]. Fifteen pathologists of varying subspecialty background and experience assessed 27 frozen biopsies twice, first by their own usual method and then by the Banff algorithm. For precise percentage estimates, agreement was only moderate either way, and in fact numerically slightly lower with the algorithm. But when those estimates were collapsed into the grades that actually drive organ acceptance, agreement improved from fair to substantial, with the biggest gains among gastrointestinal subspecialists and among pathologists with more than ten years of experience. Notably, the Banff method also pushed estimates downward, with no grade 3 calls at all, including on biopsies with abundant azonal steatosis that are classically difficult. This is a single-institution study of a small biopsy set, so it does not tell us whether lower grades mean more accurate grades or simply more conservative ones, but it does support the algorithm as a way of harmonising calls at the thresholds that matter clinically.

The same reproducibility question surfaces in pancreatic grading. In Virchows Archiv, Sano and colleagues propose a modified College of American Pathologists grading scheme for resected pancreatic ductal adenocarcinoma, built on 139 tumours without neoadjuvant therapy, including 41 pT1 cases [2]. Their system grades by the overall proportion of non-tubular components and adds a category for dense non-tubular clusters, defined as at least thirty clusters within a field of about 0.8 square millimetres. In multivariable analysis of pT1c through pT3 tumours, higher grade by this system was independently associated with roughly a fivefold worse disease-free survival and about a threefold worse overall survival compared with the lowest grade. Interobserver reproducibility was modest across the board but best for the modified system, with a kappa just above 0.5 against 0.5 for standard College of American Pathologists grading and under 0.3 for the World Health Organization scheme. This is a single retrospective cohort and needs external validation before anyone rewrites a synoptic report, but it is a serious attempt to make grade carry prognostic weight in early-stage disease. Also in Virchows Archiv, Dundr and colleagues tackle the hardest call in ovarian mucinous tumours, separating borderline tumour from carcinoma with expansile invasion, across 181 cases [3]. Abnormal p53 staining was present in about a fifth of borderline tumours, in seven out of ten carcinomas, and in two thirds of morphologically equivocal cases. Their receiver operating characteristic analysis found that lower cut-offs performed slightly better overall, but the previously proposed threshold of strong nuclear expression in at least thirty percent of tumour cells gave substantially higher specificity, which the authors argue is what matters in practice because it limits overdiagnosis of borderline tumours as carcinoma. This independently validates an earlier proposal, though p53 here is a supportive surrogate, not a standalone arbiter.

Turning to thyroid, two papers in Cancer Cytopathology examine how we triage indeterminate nodules, one through molecular testing and one through artificial intelligence. Rana and colleagues pooled 127 studies, 100 Western and 27 Asian, and found that molecular testing reduces surgery in both settings but plays fundamentally different roles [4]. Without molecular testing, resection rates were around two thirds of indeterminate nodules in Western cohorts and just over half in Asian cohorts, a difference that was not statistically significant; with testing, rates fell to roughly four in ten in both. What differed significantly was the risk of malignancy, which was substantially higher in Asian cohorts both with and without molecular testing, persisting after excluding non-invasive follicular thyroid neoplasm with papillary-like nuclear features. Western practice leans on comprehensive commercial panels as a rule-out tool, while Asian practice uses focused mutation assays for confirmatory risk stratification. Within Western studies, second-generation platforms produced lower resection rates but a higher risk of malignancy among those resected than first-generation tests. The authors conclude that regionally tailored guidelines are warranted, which is a pointed observation given how much of the world applies guidance derived from low-prevalence populations. Alongside that, Satturwar and colleagues evaluated a thyroid-specific deep learning algorithm integrated into a digital cytology platform, with two cytopathologists and one cytologist reading 100 archival ThinPrep slides three ways, by microscope, by artificial-intelligence-assisted single-layer whole-slide imaging, and by seven-layer Z-stack imaging, with two-week washouts [5]. Against expert consensus, artificial intelligence assistance raised sensitivity for Bethesda category three and above by roughly twenty points, to somewhere near eighty percent, and raised accuracy about ten points, but specificity fell because of more false positives, concentrated precisely among indeterminate cases. Review time dropped by between forty and sixty-five percent. Z-stack imaging added speed but no diagnostic gain over single-layer. So this is a feasibility study in an archival set of a hundred slides, showing a real efficiency signal and a real specificity cost that its authors say needs further refinement.

The third theme is ancillary testing that resolves genomic ambiguity, and here the message is consistently that protein-level and functional assays rescue calls that sequencing leaves uncertain. In Modern Pathology, Thibaudin and colleagues analysed MTAP copy number status across 2,409 tumours profiled by whole-exome sequencing, then validated a subset by immunohistochemistry with three antibody clones, with fluorescence in situ hybridisation and RNA sequencing in selected cases [6]. Concordance between immunohistochemistry and sequencing was high for clear wild-type and clear homozygous deletion calls. The interesting group was the probable heterozygous calls, a substantial proportion of which immunohistochemistry reclassified, with loss of protein expression matching homozygous deletion on fluorescence in situ hybridisation. Importantly, those ambiguous calls were not explained by low tumour cellularity. With PRMT5 and MAT2A inhibitors moving through trials, this supports immunohistochemistry as an accessible, fast way to firm up patient selection rather than relying on sequencing alone. Also in Modern Pathology, Bosch-Schips and colleagues applied a dual kappa and lambda messenger RNA in situ hybridisation assay across 53 lymph node biopsies spanning the nodular lymphocyte-predominant to classic Hodgkin lymphoma spectrum [7]. Strong diffuse cytoplasmic light chain expression, the canonical pattern, was present in all thirteen typical nodular lymphocyte-predominant cases and in twelve of fifteen overlap cases, while lymphocyte-rich classic Hodgkin lymphoma showed only faint non-canonical patterns or nothing. The authors argue this places morphologically ambiguous overlap cases within the nodular lymphocyte-predominant category, with aberrant features. The cohort is small and the assay is not yet widely available, but it is a biologically coherent tiebreaker for one of the hardest differentials in lymphoma. On the functional side, in The Journal of Pathology, Pavlova and colleagues assembled sixty leukaemia samples carrying in-frame TP53 insertions and deletions from thirteen laboratories in the chronic lymphocytic leukaemia ERIC network [8]. Standard interpretation classified most as variants of uncertain significance, but yeast-based transactivation assays, supported by work in a TP53-knockout cell line, showed loss of function in the vast majority, with all 49 variants within the core DNA-binding domain showing complete loss of transactivation. Second-allele inactivation and clonal expansion at relapse supported the same conclusion. The five variants downstream of codon 286 behaved heterogeneously and still require case-by-case interpretation. Given that TP53 status drives treatment decisions in chronic lymphocytic leukaemia, this converts a large bucket of uncertainty into evidence for pathogenicity.

Finally, two papers connect morphology to outcome. In Histopathology, Koh and colleagues reviewed 203 patients with early triple-negative breast cancer treated with neoadjuvant pembrolizumab-based chemoimmunotherapy [9]. Pathological complete response occurred in just under two thirds of patients, and higher grade, higher tumour-infiltrating lymphocytes and higher Ki-67 were each independently associated with it. Residual disease was most often a single confined focus, fibrotic tumour beds clustered in non-responders while stromal elastosis clustered in responders, and nodal responses were heterogeneous, including complete, absent and mixed patterns. Recurrence was more frequent without complete response, and also with multifocal baseline tumours, higher clinical stage, low lymphocytes and residual nodal disease. This is real-world and retrospective, but it argues that pathological complete response and ypN status alone under-describe what is happening in these specimens. And in Human Pathology, Chan and colleagues assembled 44 prostate adenocarcinomas with RAS, RAF or MAPK pathway alterations across three institutions, about four percent of sequenced cases [10]. BRAF was the commonest altered gene, two thirds were grade group 5, aggressive histology appeared in about seven in ten, and metastatic disease developed in 84 percent. Many cases lacked co-occurring driver alterations, suggesting these may be genuine drivers in a small subset. It is a selected, sequencing-enriched cohort, so prevalence figures should be read cautiously.

If you only have time for one paper this week, make it the pan-cancer MTAP concordance study in Modern Pathology [6]. It settles a practical question that is about to land on a lot of benches, namely what to do with an equivocal MTAP copy number call when a PRMT5-directed therapy decision depends on it.

Here is what this week's evidence adds up to in Pathology. First, structured grading rules improve agreement where it counts even when they do not improve precision, with the Banff steatosis algorithm and the modified pancreatic grading scheme both showing gains at clinically meaningful thresholds, though each rests on a single institution's data. Second, protein-level and functional assays are outperforming sequencing at the margins, with immunohistochemistry resolving ambiguous MTAP calls and functional assays reclassifying most in-frame TP53 variants as deleterious, which strengthens the case for orthogonal confirmation rather than sequencing alone. Third, in thyroid, molecular testing reduces surgery in both Western and Asian practice but does so for different reasons and against different baseline malignancy risks, which is an argument the authors make for regionally tailored guidance. Fourth, artificial intelligence assistance in thyroid cytology bought substantial time savings and better sensitivity at the cost of specificity in exactly the indeterminate cases that are hardest, so the trade-off is not yet resolved. And fifth, detailed pathological characterisation beyond complete response in triple-negative breast cancer, and morphologic recognition of RAS and RAF-altered prostate cancer, both point to prognostic information that current reporting conventions do not capture.

That's your roundup for This Week in Pathology. The full transcript and references are available on the episode page in your AudioScholar library. This is an AI-curated summary — for clinical decisions, always consult primary sources and current guidelines. See you next week.

If this weekly briefing is useful, follow the show in your podcast app so new episodes arrive automatically. And think of one colleague — in any specialty — who never has time to keep up with the literature. Tell them about AudioScholar: a free ten-minute weekly for every specialty, to listen to in any podcast app, or to read at audioscholar dot C C.

This is an automated summary generated by artificial intelligence, which can make mistakes. Always review the original source materials.