This Week in Pathology — Aug 16, 2026

Welcome to This Week in Pathology. This week we're covering 10 notable papers spanning smarter test-triage strategies that replace expensive downstream assays, the persistent problem of interobserver variability and what artificial intelligence is actually doing about it, and biomarker assessment for targeted therapy in biliary tract cancer. Let's dive in.

Let's start with a theme that runs through several of this week's papers: using a cheaper, faster front-line test to decide who really needs the expensive one. In Modern Pathology, Erem and colleagues report the largest real-world experience yet with TRBC1 immunohistochemistry in cutaneous T-cell lymphoma — 665 biopsies from 566 patients, every one with paired T-cell receptor clonality testing [1]. Using the pre-specified thresholds of under fifteen percent or over eighty-five percent staining, TRBC1 was about 86 percent sensitive and 80 percent specific for clonality, with overall accuracy of 83 percent. What matters clinically is the asymmetry. A polytypic pattern was strongly associated with reactive disease and effectively excluded clonality, with a negative likelihood ratio of 0.18, and polytypic staining was uniformly seen in a separate 270-biopsy reactive cohort. Monotypic staining predicted neoplasia, with the odds rising as infiltrate density rose — roughly five-fold in patch-stage mycosis fungoides, but eighteen-fold in other cutaneous lymphoproliferative disorders. Reader agreement was near-perfect overall but dropped meaningfully in the cases that were reflexed for molecular testing, concentrated exactly where you would expect: patch-stage mycosis fungoides and CD30-positive lymphoproliferative disorders. Digital image analysis tracked molecular clonality about as well as the human read, and independently rediscovered the same cut-off. The authors argue for a TRBC1-first algorithm with reflex T-cell receptor testing reserved for borderline staining or clinicopathologic discordance.

The same logic appears in the microbiology laboratory, in Archives of Pathology and Laboratory Medicine, where Wang and colleagues asked what a paediatric hospital system would lose by reflexing urine cultures only off a positive urinalysis [10]. Across nearly seven thousand paired specimens, a negative urinalysis carried a negative predictive value of at least 96.6 percent when using culture thresholds of ten thousand colony-forming units per millilitre for catheter specimens and fifty thousand for non-sterile collections. The protocol would have cancelled about 53 percent of cultures while missing one percent of samples overall. Notably, the cultures missed by urinalysis were enriched for organisms other than Escherichia coli, which is a real caveat when you are worried about an atypical pathogen. Both papers point the same direction: front-line triage is defensible when you understand exactly which subgroups generate the false negatives.

A third variation on that theme comes from Histopathology, where Weidmann and colleagues tackled whether one tumour block is enough for molecular selection in intrahepatic cholangiocarcinoma [8]. They sequenced 117 spatially distinct blocks from 35 patients. Baseline sequencing found IDH1 mutations in just under a quarter of patients, and although a rapid cartridge assay flagged discordance in five of 117 samples, digital PCR validation showed every one of those was a technical artefact rather than true biological heterogeneity. After validation, IDH1 status was concordant across every region of every mutated tumour. The practical message is that IDH1 appears to be an early clonal event, and single-sample testing is sufficient for selecting patients for IDH1-targeted therapy — you do not need to chase multiple blocks.

That clean result stands in instructive contrast to HER2 in the biliary tree. Also in Human Pathology, Angerilli and colleagues assessed HER2 in 140 consecutively resected extrahepatic cholangiocarcinomas and gallbladder carcinomas using immunohistochemistry and dual chromogenic in situ hybridisation [4]. HER2 positivity by HERIZON-BTC criteria was around nine percent in both tumour types — clinically meaningful now that zanidatamab is approved. Amplification was present in all 3+ tumours and in a subset of 2+ and, rarely, 1+ cases, and the amplification ratio was significantly higher in 3+ than in the amplified lower-score tumours, 6.2 versus 3.9. Critically, HER2 expression frequently showed intratumoral heterogeneity, often involving fewer than half of the tumour cells. So where IDH1 is homogeneous and forgiving, HER2 in biliary cancer is patchy and score-dependent — a reminder that sampling adequacy and in situ hybridisation confirmation of equivocal cases both still matter.

Staying with tissue sampling, Histopathology also published work from Bouwmeester and colleagues on risk-reducing salpingo-oophorectomy specimens from BRCA1 and BRCA2 carriers [3]. In a small but pointed study of 19 patients across four outcome groups, deeper sections cut at 150-micrometre intervals, with a deep learning model assisting detection, revealed an isolated focus of serous tubal intraepithelial carcinoma or high-grade serous carcinoma in every patient who later developed peritoneal high-grade serous carcinoma — lesions that had been missed at initial diagnosis. Five lesions originally called serous tubal intraepithelial lesions on the basis of low Ki-67 turned out to be carcinoma on adjacent slides. The numbers are tiny, but the implication is that deeper sectioning deserves consideration when a serous tubal intraepithelial lesion is diagnosed, and that invasion is not required for progression to peritoneal disease.

Our second big theme is interobserver variability — where it persists, and where machines are starting to help. Human Pathology published an international survey by Bernhardt and colleagues of 541 pathologists from four genitourinary and general pathology societies on intraductal proliferations of the prostate [2]. Respondents broadly endorsed the Guo and Epstein criteria for intraductal carcinoma and agreed that finding it without high-grade cancer on needle biopsy usually means unsampled high-grade cancer, reinforcing its role as an exclusion for active surveillance. But when they scored 25 images, consensus was reached in only 44 percent of cases, and not a single case of the new WHO category, atypical intraductal proliferation, achieved consensus. Collapsing lesions into low-grade versus high-grade buckets lifted consensus to 88 percent, and agreement was higher for atypical intraductal proliferation grouped with intraductal carcinoma than grouped with high-grade prostatic intraepithelial neoplasia. In other words, the intermediate category may be more useful as a high-grade flag than as a distinct diagnosis — worth remembering before you write it in a report.

Against that backdrop, Kim and colleagues in Histopathology present SydneyMTL, a weakly supervised multi-task model trained on more than fifty thousand whole-slide images to grade all five attributes of the Updated Sydney System in gastric biopsies [9]. Mean lenient accuracy was 89 percent across 24 board-certified pathologists, and in a two-reader randomised crossover study artificial intelligence assistance improved interobserver agreement and cut review time by about a third. The model explicitly handles the situation where atrophy cannot be assessed because muscularis mucosae is absent, which is the kind of practical detail that determines whether a tool survives contact with routine work. Complementing this, Cancer Cytopathology carries a practical review from the American Society of Cytopathology Clinical Practice Committee on large language and vision-language models [5]. Bilal and colleagues separate applications with preliminary evidence — structured reporting, quality control, education — from those that remain hypothetical, and are direct about hallucination risk, limited explainability, bias, privacy, and validation gaps, recommending cytopathology-specific benchmarks, multi-institutional validation and incremental deployment starting with low-risk tasks.

Two papers round out the week on hard differentials. In Modern Pathology, Cabanero and colleagues developed a four-assay methylation-based droplet digital PCR panel to separate mucinous lung adenocarcinoma from gastrointestinal metastases to the lung, a differential with no reliable immunohistochemical marker [7]. Decision trees built on the assay data matched primary-versus-metastatic assignment in 83 to 90 percent of reference cases, and resolved five of ten genuinely difficult cases that went on to full clinicopathologic review — promising, but still a modest cohort. And in Archives of Pathology and Laboratory Medicine, Churg and colleagues review forty years and four thousand accessions of the United States-Canadian Mesothelioma Reference Panel [6]. The two commonest reasons for referral in 1985 — mesothelioma versus another tumour, and mesothelioma versus a benign mesothelial proliferation — are still the two commonest today, but the tools have changed completely: from electron microscopy and histochemistry to an overabundance of immunohistochemical markers, only a few of them truly specific, plus reliable surrogate immunohistochemistry and fluorescence in situ hybridisation for the benign-versus-malignant question, and newer entities including mesothelioma in situ.

If you only have time for one paper this week, make it the TRBC1 cohort in Modern Pathology [1]. It is the rare biomarker paper large enough to define exactly when you can trust the immunostain and when to reflex to molecular testing, and it could reshape your cutaneous lymphoma workup and its cost immediately.

Here are the key takeaways from this week in Pathology. First, TRBC1 immunohistochemistry is strong enough to lead a cutaneous T-cell lymphoma workup, with a polytypic result largely excluding clonality — but reflex molecular testing when staining is borderline, when the lesion is patch-stage mycosis fungoides, or when a CD30-positive process is in play. Second, reflex urine culture off a positive urinalysis in children can cancel about half of all cultures while missing about one percent of positives, with the misses skewed towards non-Escherichia coli organisms. Third, IDH1 status in intrahepatic cholangiocarcinoma is spatially homogeneous, so one block suffices — but HER2 in extrahepatic cholangiocarcinoma and gallbladder carcinoma is heterogeneous, often in fewer than half of tumour cells, and demands more care. Fourth, the new atypical intraductal proliferation category has essentially no interobserver consensus and aligns diagnostically closer to intraductal carcinoma than to high-grade prostatic intraepithelial neoplasia. And fifth, artificial intelligence is beginning to deliver measurable gains in reproducibility and turnaround for structured grading tasks, while cytopathology's professional bodies are explicitly urging validation before deployment of language-model tools.

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 if you have a paper you have been meaning to read, upload the PDF, or paste any link, at audioscholar dot C C. We will turn it into audio like this one, in any of thirty-one languages.

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