Introduction

While histo-endoscopic improvement is achieved with Upadacitinib (Upa) treatment of ulcerative colitis (UC), its effect on the inflammatory microenvironment (IME) has yet to be fully resolved. Unlocking quantitative insights of histologic features of the IME has the potential to reveal biomarkers for treatment selection. Here, we assessed machine learning (ML)-derived histologic IME features associated with endoscopic response (ER) using UC biopsies at baseline and induction in two Upa clinical trials.

Aims & Methods

IBD Explore¹ (PathAI, Boston, version 1.3), a suite of ML models previously trained to quantify tissue regions (e.g., epithelium) and cells (e.g., goblet cells), was deployed on hematoxylin and eosin (H&E)-stained whole-slide images (WSI) of baseline and induction biopsies from patients with UC (N=732) enrolled in two clinical trials (NCT02819635, NCT03653026). Human interpretable features (HIFs) were extracted from each WSI and associated with Upa treatment and ER measured at induction (week 8; Mayo Endoscopic score≤1). Analysis of Covariance (ANCOVA) was used to associate change from baseline (CBL) of HIFs with Upa treatment, and logistic regression was used to associate baseline HIFs with ER, using baseline clinical data and total Geboes scores as covariates. Multivariable classification models, using baseline HIFs, were also trained to predict ER in Upa treated patients. Models were trained using 5-fold cross-validation and evaluated on the held-out test set using average predictions across these folds.

Results

ML-derived HIFs (N=402) displayed a significant CBL after Upa treatment, and 193 HIFs were prognostic of ER. These HIFs included fraction of goblet cell mucin area, fraction of normal epithelium, density of goblet cells in epithelium, fraction of infiltrated epithelium relative to all epithelium, fraction of crypts with abscess (CA) + complete granulation tissue (CGT) area relative to mucosa, and density of neutrophils (Table 1). HIFs quantifying the fraction of eosinophils and neutrophils within 20 μm of goblet cells were both independently prognostic and Upa- dependent predictive of ER. Baseline HIFs were also prognostic, as a group, of ER in the treatment group (mean validation AUROC = 0.78, test-set AUROC = 0.84).

Conclusion

Here, using an ML-based approach to assess the UC IME, we identified a significant association between features relating to goblet cells, goblet cell mucin, and immune cells near goblet cells and Upa ER. These findings underscore the importance of granular IME characterization to better understand treatment mechanisms. These results also highlight the potential of histological biomarkers for predicting therapeutic outcomes. Further validation of these potential biomarkers in larger cohorts is necessary.

References

¹IBD Explore is for research use only. Not for use in diagnostic procedures.

Disclosure

PathAI authors are employees at PathAI and holders of company stock. ​
AbbVie authors are employees at AbbVie and holders of company stock

Introduction

Colorectal cancer (CRC) screening has been widely implemented, and recent advances in fecal immunochemical testing (FIT) have led to a shift in screening modalities from guaiac fecal occult blood testing (gFOBT) to FIT. In Japan, FIT was introduced into the national screening program in 1992, and a two-sample method has been offered to individuals aged 40 years and older. However, there is no standardized use of testing devices or cutoff values. Although hemoglobin concentrations of 100-150 ng/g are commonly used, few studies have evaluated cut-off values in population-based screening. An optimal cut-off value should be defined for adequate quality assurance by considering the benefits and harms of CRC screening.

Aims & Methods

To evaluate the optimal cutoff value for population-based screening, we compared the test performance of FIT according to different cutoff values. A total of 383,793 individuals participated in population-based screening in Ibaraki Prefecture from April 2021 to March 2023. The OC-sensor was used as the testing modality, with a default cut-off value of 100 ng/g. Test performance indicators, including positive rate, detection rate, positive predictive value (PPV), cancer miss rate, and number needed to scope (NNS), were assessed across cut-off values ranging from 100 to 250 ng/g. Detection rate, PPV, and NNS were explicitly calculated for CRC/ Advanced neoplasia (AN). Since the colonoscopy completion rate is typically around 70％, PPV and NNS were calculated on a per-protocol basis.

Results

Subjects who submitted two samples were 370419. Among them, 863 cases of CRC and 1944 cases of advanced adenomas were diagnosed. At a cutoff value of 100 ng/g at the baseline, the positive rate was 6.7% (95%CI: 6.7-6.8) and the colonoscopy completion rate was 76.0% (95%CI: 75.4-76.5) (Table). For AN, the detection rate was 0.76％ (95%CI: 0.73-0.79), 14.8% for PPV (95%CI: 14.3-15.3), and 6.8% (95%CI: 6.5-7.0) for NNS. For CRC detection, the detection rate was 0.23％ (95%CI: 0.22-0.25), 4.5% for PPV (95%CI: 4.3-4.8), and 22.0% (95%CI: 22.0-23.5) for NNS. The NNS decreased as the cutoff value increased, but the CRC miss rate also increased. The cutoff value reached 180 ng/g, the CRC miss rate became 19.5% (95%CI:16.8-22.1), whereas the positive rate was 4.2% (95%CI: 4.2-4.3), 0.19％ (95%CI: 0.17-0.20) for CRC detection rate, 5.9% (95%CI: 5.5-6.3) for PPV, and 16.9 (95%CI: 15.8-18.2) for NNS.

Conclusion

The current cutoff value led to high detection rates for both CRC and AN, but it also required a greater number of colonoscopies. The availability of colonoscopy services varies across local municipalities and is often insufficient in rural areas. Adjusting the cutoff value is frequently used to address disparity in access to colonoscopy. However, the trade-off among benefits, harms, and resources is influenced by the social context, including individual preference and value. Therefore, an acceptable level of cancer miss rates should be carefully evaluated. Since the optimal cutoff value cannot be determined unequivocally, a comprehensive decision should be made by balancing benefits, harms, and resources, while also considering the preference of local residents.

References

Young GP, et al. Detection of advanced colorectal neoplasia and relative colonoscopy workloads using quantitative faecal immunochemical tests: an observational study exploring the effects of simultaneous adjustment of both sample number and test positivity threshold. BMJ Open Gastroenterol. 2020 ;7: e000517

Introduction

Hepatocellular carcinoma (HCC) is characterized by high invasiveness and metastatic potential, with persistently low five-year survival rates. While vitamin D has demonstrated anti-tumor effects in various cancers, its precise molecular mechanism in HCC remains unclear. This study aimed to investigate the inhibitory role of vitamin D in HCC progression, focusing on the regulatory function of acetyl-CoA acetyltransferase 2 (ACAT2).

Aims & Methods

Huh-7 cells were treated with increasing concentrations of vitamin D (0, 15, 20, 30, and 40 nM) for 48 hours. Cell migration was assessed using Transwell assays, and apoptosis was quantified via Annexin V/PI staining followed by flow cytometry. ACAT2 mRNA and protein levels were measured by qPCR and Western blot, respectively. Prognostic relevance of ACAT2 expression was analyzed using Kaplan–Meier survival curves based on the GSE76427 dataset. Mechanistically, single-cell transcriptomic data from 26 HCC patients were integrated to identify ACAT2⁺ malignant cell populations and their enriched pathways via KEGG analysis. Mendelian randomization (MR) using the inverse-variance weighted (IVW) method was employed to assess the causal relationship between genetically predicted vitamin D levels and HCC risk, and to quantify the mediation effect of ACAT2.

Results

Vitamin D treatment significantly suppressed Huh-7 cell migration and induced apoptosis in a dose-dependent manner. ACAT2 expression was consistently downregulated at both mRNA and protein levels. Clinical analysis revealed that high ACAT2 expression was associated with poor prognosis in HCC patients. Single-cell analysis showed that ACAT2 expression was markedly elevated in tumor hepatocytes, and ACAT2⁺ cells were enriched in cytochrome P450-related xenobiotic metabolism pathways. MR analysis demonstrated a significant inverse association between genetically predicted vitamin D levels and HCC risk, and ACAT2 was identified as a significant mediator, accounting for 9.6%.

Conclusion

This study provides novel evidence that vitamin D suppresses HCC progression by downregulating ACAT2. ACAT2 may serve as a potential prognostic biomarker and therapeutic target. These findings support the translational potential of vitamin D as an adjunctive strategy in HCC prevention and treatment.

Introduction

Cholangiocarcinoma (CCa) is a complex neoplasm with poor prognosis, classified as intrahepatic (iCCa), perihilar (pCCa), or distal (dCCa) in the bile duct. Its variable phenotypes pose diagnostic challenges. Digital single-operator cholangioscopy (DSOC) aids diagnosis but has limitations, including suboptimal biopsy sensitivity, lack of a standardized morphology classification, as well as technical issues. Emerging evidence highlights AI’s potential to enhance DSOC diagnostic capacity, but the differential performance according to tumor characteristics, namely its topography, remains unassessed.

Aims & Methods

This study aimed to evaluate an AI algorithm’s effectiveness in detecting iCCa, pCCa, and dCCa. A deep learning model based on a convolutional neural network (CNN) was designed to automatically predict the malignancy status of biliary strictures. The model was built using DSOC images from four international high-volume centers: Centro Hospitalar Universitário de São João (Porto, Portugal), Hospital Universitario Puerta de Hierro Majadahonda (Madrid, Spain), New York University Langone Hospital (New York, USA) and University of Chicago Medicine (Chicago, USA). Images showing strictures due to cholangiocarcinoma or benign conditions were included. The model was trained with a learning rate of 0.0001, a batch size of 32, and a number of epochs of 10, using PyTorch framework. The detection rate of the algorithm was compared for each location category.

Results

A total of 18166 images from 32 patients from 4 centers. A total of 22 patients were male (69%) and the mean age was 63 years (SD: 11 years). Common bile duct (CBD) strictures represented 44% of the total (n=14) while intrahepatic and perihilar accounted for 22% (n=7) and 34% (n=11), respectively. The algorithm’s DR of strictures representing CCa was 91.6% for iCCa, 89.4% for pCCa and 92.4% for dCCa. Pairwise comparisons revealed a significant difference in the DR for iCCa and pCCa (p<0.05) and dCCa and pCCa (p<0.001).

Conclusion

Topographic characteristics of biliary strictures influence the diagnostic performance of DSOC. While AI shows promise in improving the diagnostic capacity of DSOC, its performance is impacted by DSOC technical limitations. To our knowledge, this is the first multicentric and transatlantic algorithm, capable of providing detection and differentiation of morphologic features of biliary strictures. Developing models should address and control for these issues before widespread clinical integration. Additionally, this evidence stresses the importance of keeping a “human on the loop”, providing clinical reasoning to guide the application of AI algorithms.