This Week in Oncology — Jun 11, 2026

Welcome to This Week in Oncology. This week we're covering 8 notable papers spanning the rise of artificial intelligence in pathology, refinements in immunotherapy, and new insights from molecular biomarkers. Let's dive in.

We begin with a major emerging theme: the use of artificial intelligence to extract deeper meaning from routine pathology slides. Two papers this week showcase how computational pathology is moving from concept to clinical application. First, in Cancer Research, a large international study analyzed over 8,000 cases of gastrointestinal stromal tumors, or GIST [6]. Researchers trained deep learning models on whole-slide images to predict key molecular features without sequencing. The models were highly accurate, achieving an area under the curve of 0.87 for predicting KIT mutations and 0.96 for PDGFRA mutations. They could even identify specific subtypes, like KIT exon 11 deletions, and predict sensitivity to tyrosine kinase inhibitors like imatinib and avapritinib. Furthermore, the AI-derived risk score for recurrence-free survival performed comparably to established pathology-based risk models. This work provides a powerful proof of concept that standard histology slides may contain a wealth of molecular and prognostic information that can be unlocked with AI, potentially speeding up diagnosis and treatment selection. Taking this concept a step further, a study in Annals of Oncology applied a multimodal AI model to data from the STAMPEDE platform trials in prostate cancer [8]. The goal was to predict which men with very high-risk, non-metastatic disease would benefit most from the addition of abiraterone to long-term androgen deprivation therapy. The model, which incorporates digital pathology images along with clinical data like PSA and tumor stage, stratified patients into a 'very high-risk' group and a 'standard high-risk' group. The benefit of adding abiraterone was concentrated almost entirely in the AI-defined very high-risk group, where it cut the risk of metastasis by more than half, with a hazard ratio of 0.47. In contrast, there was limited benefit in the standard high-risk group. This post-hoc analysis suggests the AI tool functions as a predictive biomarker, identifying patients who will derive substantial benefit from treatment intensification while helping to spare others the toxicity and cost of a drug they are unlikely to benefit from.

Next, we turn to immunotherapy, where several trials are exploring new combinations and settings. A phase 2 trial in Clinical Cancer Research investigated a dual immunotherapy approach in recurrent or metastatic nasopharyngeal carcinoma, combining the TIM-3 inhibitor TQB2618 with the PD-1 inhibitor penpulimab [1]. The study had two cohorts with starkly different results. In patients whose disease had already progressed on prior immunotherapy, the chemotherapy-free dual regimen showed limited efficacy, with no objective responses and a median progression-free survival of only 1.6 months. However, in treatment-naïve patients, the same immunotherapy duo combined with standard gemcitabine and cisplatin chemotherapy demonstrated encouraging activity. The objective response rate was over 86%, with a median progression-free survival of 10.8 months. This highlights that while this dual immunotherapy approach may not be enough to overcome established resistance, it could be a potent addition to first-line chemotherapy. In breast cancer, the BEGONIA platform trial, published in Nature Cancer, evaluated another novel combination: the PD-L1 inhibitor durvalumab plus the antibody-drug conjugate trastuzumab deruxtecan, or T-DXd [3]. This arm of the trial enrolled women with first-line metastatic, hormone receptor-negative, HER2-low disease. The combination produced an objective response rate of 62.1%, with a median progression-free survival of 12.6 months and an impressive median overall survival of 30.3 months. While the safety profile was manageable, it's important to note that the trial narrowly missed its pre-specified primary endpoint for objective response. Additionally, drug-related interstitial lung disease or pneumonitis occurred in about one-fifth of patients. So, while the results are clinically relevant, they also underscore the ongoing challenge of optimizing efficacy and safety with these powerful combinations. While these trials test new combinations, two basic science papers offer a glimpse into the future of immunotherapy by dissecting mechanisms of immune control and resistance. A paper in Cancer Cell uncovers a metabolic strategy that tumor cells use to resist immune checkpoint blockade [5]. Researchers found that in resistant tumors, specialized macrophages, identified by the marker TREM2, essentially feed tumor cells. These macrophages engage in efferocytosis—the cleanup of dead cells—and in the process, recycle fatty acids to nearby tumor cells via extracellular vesicles. This fuel supports tumor cell survival and promotes signaling pathways that lead to immune evasion. Blocking this pathway with an anti-TREM2 antibody resensitized tumors to checkpoint inhibitors in preclinical models, identifying a new metabolic vulnerability to target. And finally in this section, a fundamental discovery published in Nature identifies a specific subset of regulatory CD8 T-cells guided by the receptor GPR15 that are critical for controlling intestinal inflammation [4]. These cells home to the colon and eliminate inflammatory macrophages. This finding has direct implications for inflammatory bowel disease, which is a major risk factor for colon cancer, and points to GPR15 as a potential future therapeutic target to restore immune balance in the gut.

One of the most pressing questions in oncology today is how to act on the detection of molecular residual disease, or MRD. A landmark trial in Nature Medicine provides a sobering answer for one specific strategy. The ALTAIR trial was a randomized, phase 3 study for patients with resected colorectal cancer who became ctDNA-positive after standard therapy, indicating MRD, but had no radiological evidence of disease [2]. These patients were randomized to receive six months of trifluridine/tipiracil or a placebo. The primary endpoint was disease-free survival. The trial was negative. Median disease-free survival was 9.3 months with the active drug versus 5.55 months with placebo, a difference that was not statistically significant, with a p-value of 0.107. The intervention also came with a significant toxicity burden, with 73% of patients on trifluridine/tipiracil experiencing grade 3 or higher hematologic adverse events compared to just 3% in the placebo group. This is a critical finding, demonstrating that simply detecting MRD and intervening with this oral chemotherapy does not significantly improve outcomes and adds toxicity.

Finally, a study in Cancer Cell pushes beyond genomics to better understand the heterogeneity of diffuse large B cell lymphoma, or DLBCL [9]. By integrating proteomic, transcriptomic, and genomic data from nearly 500 tumors, investigators identified seven distinct 'proteogenotypes' that reflect unique biological states. One of these, named PG4, was associated with poor outcomes independent of all established risk factors, including cell-of-origin and the international prognostic index. This high-risk subtype is characterized by enhanced MYC activity and an exhausted T-cell microenvironment. This work provides a new, more nuanced framework for classifying DLBCL and identifies a high-risk group of patients for whom novel therapeutic strategies targeting MYC or the tumor microenvironment may be needed.

If you only have time for one paper this week, make it the multimodal AI study in high-risk prostate cancer from Annals of Oncology [8]. It represents a tangible step toward using AI not just for prognosis, but as a predictive tool to guide complex treatment decisions, in this case, identifying who truly benefits from the addition of abiraterone.

Here are the key takeaways from this week in Oncology: First, artificial intelligence models applied to standard pathology slides can now predict not only molecular alterations and prognosis in GIST, but can also function as a predictive biomarker for abiraterone benefit in very high-risk prostate cancer, paving the way for more personalized treatment. Second, in nasopharyngeal cancer, adding a TIM-3/PD-1 inhibitor combination to first-line chemotherapy appears promising, but the dual immunotherapy alone is not effective for patients with immunotherapy-refractory disease. Third, the ALTAIR trial in colorectal cancer delivered a key negative result: early intervention with trifluridine/tipiracil based on a positive ctDNA test for molecular residual disease does not significantly improve disease-free survival. Fourth, a new proteogenomic classification of DLBCL has identified a high-risk subtype, PG4, that is independent of current risk factors and is defined by MYC activation and an exhausted T-cell microenvironment, suggesting new avenues for targeted therapy.

That's your roundup for This Week in Oncology. 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.

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