Artificial intelligence (AI) can predict the presence of molecular alterations directly from routine histopathology slides. However, training robust AI systems requires large datasets for which data collection faces practical, ethical and legal obstacles. These obstacles could be overcome with swarm learning (SL), in which partners jointly train AI models while avoiding data transfer and monopolistic data governance. Here, we demonstrate the successful use of SL in large, multicentric datasets of gigapixel histopathology images from over 5,000 patients. We show that AI models trained using SL can predict BRAF mutational status and microsatellite instability directly from hematoxylin and eosin (H&E)-stained pathology slides of colorectal cancer. We trained AI models on three patient cohorts from Northern Ireland, Germany and the United States, and validated the prediction performance in two independent datasets from the United Kingdom. Our data show that SL-trained AI models outperform most locally trained models, and perform on par with models that are trained on the merged datasets. In addition, we show that SL-based AI models are data efficient. In the future, SL can be used to train distributed AI models for any histopathology image analysis task, eliminating the need for data transfer.

The proportion of epithelial and stromal cells in tumours is thought to have an important role in the progression of epithelial malignancy. We aimed to determine whether the relative proportion of tumour (PoT) was related to survival in colorectal cancer. The PoT at the luminal surface was measured by point counting using virtual tissue sections in a series of 145 colorectal cancer cases. The relationship of PoT to clinicopathological parameters including cancer-specific survival was analysed. Modified receiver operating characteristic curves were used to determine the optimum cut off points to dichotomise the data for survival analyses. Tumours with PoT-low (47%) were associated with significantly lower cancer-specific survival when compared to PoT-high (hazard ratio (HR)=2.087, 95% CI=1.088–4.003, P=0.024). On sub-analysis, the prognostic effect remained significant in colonic tumours (HR=2.474, 95% CI=1.132–5.408, P=0.019) and tumour, node, metastasis stage III disease (HR=3.480, 95% CI=0.325–9.136, P=0.007). Multivariate Cox regression analysis demonstrated that PoT was an independent prognostic marker when adjusted for age, T stage, N stage and extramural vascular invasion (P=0.017). This study suggests that a low proportion of tumour cells in colorectal cancer is related to poor cancer-specific survival. A relatively quick, inexpensive and well-established method such as point counting on diagnostic tissue sections could be used to identify a subset of patients who may benefit from adjuvant therapy.

Although radical surgery remains the cornerstone of cure in resectable gastric cancer, survival remains poor. Current evidence-based (neo)adjuvant strategies have shown to improve outcome, including perioperative chemotherapy, postoperative chemoradiotherapy and postoperative chemotherapy. However, these regimens suffer from poor patient compliance, particularly in the postoperative phase of treatment. The CRITICS-II trial aims to optimize preoperative treatment by comparing three treatment regimens: (1) chemotherapy, (2) chemotherapy followed by chemoradiotherapy and (3) chemoradiotherapy. In this multicentre phase II non-comparative study, patients with clinical stage IB-IIIC (TNM 8th edition) resectable gastric adenocarcinoma are randomised between: (1) 4 cycles of docetaxel+oxaliplatin+capecitabine (DOC), (2) 2 cycles of DOC followed by chemoradiotherapy (45Gy in combination with weekly paclitaxel and carboplatin) or (3) chemoradiotherapy. Primary endpoint is event-free survival, 1 year after randomisation (events are local and/or regional recurrence or progression, distant recurrence, or death from any cause). Secondary endpoints include: toxicity, surgical outcomes, percentage radical (R0) resections, pathological tumour response, disease recurrence, overall survival, and health related quality of life. Exploratory endpoints include translational studies on predictive and prognostic biomarkers. The aim of this study is to select the most promising among three preoperative treatment arms in patients with resectable gastric adenocarcinoma. This treatment regimen will subsequently be compared with the standard therapy in a phase III trial. clinicaltrials.gov NCT02931890 ; registered 13 October 2016. Date of first enrolment: 21 December 2017.

The development of gastric adenocarcinoma is a complex multistep process involving multiple genetic alterations. Based on pathology, four different macroscopic types and at least two major histological types, intestinal and diffuse, have been described. Most gastric cancer (GC) show genetic instability, either microsatellite instability or chromosomal instability, which is considered an early event in gastric carcinogenesis. Molecular studies of alterations of single genes have provided evidence that intestinal and diffuse type GC evolve via different genetic pathways. Recent results from high-throughput whole-genome expression or copy number studies have demonstrated extensive genetic diversity between cases and within individual GC. Sets of commonly up- or downregulated microRNAs have been identified in GC and might be useful in the near future to identify pathways of GC progression. Results from detailed molecular and/or pathological GC studies, although promising, still have limited clinical utility in predicting survival and stratifying GC patients for appropriate treatment.