Introduction

Additional surgery is recommended for endoscopically resected non-curative T1 colorectal cancers (CRC) as there is a 10% risk of recurrence. However, little is known about the long-term prognosis of non-curative T1 CRC followed up without additional surgery, especially for non-curative resection with negative deep margins.

Aims & Methods

This study aimed to identify major risk factors for recurrence and to clarify the long-term prognosis of non-curative T1 CRC without additional surgery.
Retrospective analysis of consecutive patients with T1 CRC who underwent ESD or EMR at two centres between January 2006 and December 2021. The cohort was divided into the presence or absence of non-curative resection factors as defined by the Japanese Society for Cancer of the Colon and Rectum guidelines (2019) and additional surgery. The recurrence rate and major risk factors for recurrence were analysed.

Results

The median age of 190 patients was 70 years, 63% were male, and endoscopic treatment was 64% EMR and 36% ESD. 82 patients were in the curative resection group and 108 in the non-curative resection group, comprising 79 patients with additional surgery and 29 patients without additional surgery. The five-year recurrence-free survival rates were 98.4% (95% CI 89.3-99.8) for curative T1 CRC, 98.3% (95% CI 88.8-99.8) for non-curative T1 CRC with additional surgery, and 73.7% (95% CI 46.5-88.5) for non-curative T1 CRC without additional surgery. Lymphatic invasion and budding grade 2/3 were major risk factors for recurrence, with hazard ratios of 40.7 and 23.1, respectively. Stratifying non-curative T1 CRC without additional surgery by the presence or absence of major risk factors, the five-year recurrence-free survival and five-year disease-specific survival rates were 85.6% (95% CI 52.5-96.3) and 91.7% (95% CI 53.9-98.8) in the group without major risk factors (n=21). However, the prognosis was poor for the group with no additional surgery and major risk factors and did not reach 5 years, with a median recurrence-free survival and disease-specific survival of 2.5 and 3.1 years, respectively (n=8).

Conclusion

Additional surgery is strongly recommended for non-curative T1 CRC with lymphatic invasion or budding grade 2/3 after endoscopic resection with negative deep margins. For patients who are intolerant to surgery and do not have major risk factors for recurrence, follow-up may be considered.

Introduction

Inflammatory bowel disease (IBD), including Crohn’s disease (CD) and ulcerative colitis (UC), is known to be associated with various extra-intestinal manifestations, impacting organ systems beyond the gastrointestinal tract. Identifying comorbidities in IBD and the timing of their development can provide valuable insight into the mechanisms underlying IBD development.

Aims & Methods

We conducted the first population- and disease-wide phenomic association study in IBD, using >6 million ICD-10 coded healthcare contacts from 10 years before and up to 17 years after IBD diagnosis to investigate associations with 1583 diseases. In order to explore co-morbidities with potential aetiological significance with IBD, we additionally assessed the strength of association with all diseases in the pre-diagnosis compared with the post-diagnosis period. To correct for multiple testing, we adjust our significance threshold (p<0.05) with the Bonferroni correction, resulting in an adjusted p-value of 7.90×10^-6, which we refer to as disease-wide statistical significance.

Results

We identified 312 disease associations with disease-wide statisitical significance and 125 of these diseases appear up to 10 years before diagnosis.The risk of immune-mediated diseases and extra-intestinal manifestations are among those diseases increased up to 10 years prior to IBD diagnosis (psoriasis: RR_CD: 2.57, 95% CI: 2.00-3.29; RR_UC: 1.54, 95% CI: 1.25-1.87; enteropathic arthropathies: RR_CD: 3.57, 95% CI: 2.65-4.78; RR_UC: 1.8, 95% CI: 1.38-2.32). This was also the case for gastroenterological and liver disorders (gall stones: RR_CD: 1.82, 95% CI: 1.62-2.04; RR_UC: 1.26, 95% CI: 1.15-1.37; acute pancreatitis: RR_CD: 1.83, 95% CI: 1.30-2.53; RR_UC: 2.27, 95% CI: 1.84-2.79). The risk of cardiometabolic diseases and neuropsychological disorders had increased disease-wide statistical significance both pre- and post-diagnostically, whereas potential sequelae of treatment, such as osteoporosis (HR_CD: 2.56, 95% CI: 2.30-2.86; HR_UC: 1.92, 95% CI: 1.79-2.07) or herpes simplex infections (HR_CD: 4.04, 95% CI: 2.76-5.91; HR_UC: 1.69, 95% CI: 1.2-2.38) were primarily seen post-diagnostically. Of potential aetiological importance to CD, diagnosis with infectious mononucleosis (RR: 1.87, 95% CI: 1.37-2.52) was limited to the pre-diagnostic period.

Conclusion

Our results demonstrate IBD as an essentially multisystemic disease, particularly manifesting as gastrointestinal, metabolic, immune, and neuropsychological disorders, present up to 10 years prior to IBD diagnosis.

Disclosure

AE, GP, and MB report no competing interests.
BV reports research support from AbbVie, Biora Therapeutics, Landos, Pfizer, Sossei Heptares and Takeda; speaker fees from Abbvie, Biogen, Bristol Myers Squibb, Celltrion, Chiesi, Falk, Ferring, Galapagos, Janssen, MSD, Pfizer, R-Biopharm, Takeda, Truvion and Viatris; consultancy fees from Abbvie, Alimentiv, Applied Strategic, Atheneum, Biora Therapeutics, Bristol Myers Squibb, Galapagos, Guidepont, Landos, Mylan, Inotrem, Ipsos, Janssen, Pfizer, Progenity, Sandoz, Sosei Heptares, Takeda, Tillots Pharma and Viatris.
RE reports support from Novo Nordisk.
CWL has acted as a consultant to Abbvie, Janssen, Takeda, Pfizer, Galapagos, Bristol 468 Myers Squibb, B.I., Sandoz, Novartis, GSK, Gilead, ViforPharma, Dr Falk, Trellus Health 469 and Iterative Scopes; he has received speaking fees and travel support from Pfizer, 470 Janssen, Abbvie, Galapagos, MSD, Takeda, Shire, Ferring, Hospira and Dr Falk.

Introduction

For training deep learning algorithms in medical imaging, application-specific data is often scarce. Deep learning systems are therefore generally pretrained on large publicly available labeled data sets of general imagery, unrelated to the envisioned application, to have the algorithm learn basic features from these widely available data followed by refinement training on the generally scarce application-specific images. Pretraining might be more effective if the images for pretraining resemble the envisioned application, i.e., domain-specific pretraining.

Aims & Methods

We investigated if pretraining on general endoscopic imagery results in a better performance of five existing AI systems with an application in gastro-intestinal endoscopy, compared to current state-of-the art pretraining approaches (i.e., supervised pretraining with ImageNet and semi-weakly supervised pretraining with the Billion-scale data set). Our group has created an endoscopy-specific dataset called GastroNet for pretraining deep learning systems in endoscopy. GastroNet consists of 5,084,494 endoscopic images retrospectively collected between 2012 and 2020 in seven Dutch hospitals. We created four pretrained models: one using GastroNet and three using ImageNet and/or the Billion-scale data set. The pretraining method was either supervised, self-supervised, or semi-weakly supervised. The pretrained models were subsequently trained towards five independent, commonly used applications in GI endoscopy, using their original application-specific datasets. Outcome parameters were: 1) classification and/or localization performance of the five trained applications; 2) change in performance when the number of available application-specific training data was reduced, to investigate a possible difference in performance drop for the different pretrained models.

Results

Overall, the domain-specific pretrained model resulted in a statistically superior performance for the five different GI applications. More detailed results are presented in Table 1. The superiority was also reflected in a smaller drop in performance when the number of application-specific training data were reduced artificially.

Table 1. Overview of performance of the five different application-specific data sets using four different pretrained models. Cells highlighted in green represent the highest scoring pretrained model per application-specific data set.

Conclusion

Domain-specific pretraining, using unlabeled general endoscopic images, is superior to current state-of-the-art pretraining approaches for developing deep learning algorithms in GI endoscopy. It also allows more effective use of the generally scarce application-specific endoscopy images. These findings might cause a paradigm shift in the development of AI systems in endoscopy.