Clinical performance study of a fecal bacterial signature test for colorectal cancer screening.

The fecal immunochemical test (FIT) is the most widely used test for colorectal cancer (CRC) screening. RAID-CRC Screen is a new non-invasive test based on fecal bacterial markers, developed to complement FIT by increasing its specificity. The test was previously clinically evaluated in FIT-positive patients (>20 µg of hemoglobin/g of feces, "FIT20"), in which it reduced the proportion of false positive results by 16.3% while maintaining most of FIT20's sensitivity. The aim of this study was to compare the sensitivity and specificity of a CRC screening program using RAID-CRC Screen in addition to FIT20 as a triage test in a European screening population undergoing screening colonoscopy with a CRC screening program with FIT20 alone in the same cohort. A cohort of 2481 subjects aged > 55 years from the German screening colonoscopy program was included. The colonoscopy findings were used as the gold standard in calculating the diagnostic capacity of the tests and included 15 CRC and 257 advanced neoplasia cases. RAID-CRC Screen added to FIT20 provided the same sensitivity as FIT20 alone (66.7%) in detecting CRC and a significantly higher specificity (97.0% vs. 96.1%, p<0.0001). The positive predictive value was 11.9% when using RAID-CRC Screen and 9.5% with FIT20 alone, and the negative predictive value was 99.8% in the two scenarios. For advanced neoplasia detection, the use of RAID-CRC Screen yielded significantly lower sensitivity than with FIT20 alone (17.5% vs. 21.8%, p = 0.0009), and the overall specificity was significantly higher when using RAID-CRC Screen compared with FIT20 alone (98.2% vs. 97.8%, p = 0.0039). Our findings confirm the results obtained in previous clinical studies in a CRC screening setting, showing the potential of RAID-CRC Screen to increase the overall specificity of FIT-based screening.

Machine learning in the identification of prognostic DNA methylation biomarkers among patients with cancer: A systematic review of epigenome-wide studies.

BACKGROUND: DNA methylation biomarkers have great potential in improving prognostic classification systems for patients with cancer. Machine learning (ML)-based analytic techniques might help overcome the challenges of analyzing high-dimensional data in relatively small sample sizes. This systematic review summarizes the current use of ML-based methods in epigenome-wide studies for the identification of DNA methylation signatures associated with cancer prognosis. METHODS: We searched three electronic databases including PubMed, EMBASE, and Web of Science for articles published until 2 January 2023. ML-based methods and workflows used to identify DNA methylation signatures associated with cancer prognosis were extracted and summarized. Two authors independently assessed the methodological quality of included studies by a seven-item checklist adapted from 'A Tool to Assess Risk of Bias and Applicability of Prediction Model Studies (PROBAST)' and from the 'Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK). Different ML methods and workflows used in included studies were summarized and visualized by a sunburst chart, a bubble chart, and Sankey diagrams, respectively. RESULTS: Eighty-three studies were included in this review. Three major types of ML-based workflows were identified. 1) unsupervised clustering, 2) supervised feature selection, and 3) deep learning-based feature transformation. For the three workflows, the most frequently used ML techniques were consensus clustering, least absolute shrinkage and selection operator (LASSO), and autoencoder, respectively. The systematic review revealed that the performance of these approaches has not been adequately evaluated yet and that methodological and reporting flaws were common in the identified studies using ML techniques. CONCLUSIONS: There is great heterogeneity in ML-based methodological strategies used by epigenome-wide studies to identify DNA methylation markers associated with cancer prognosis. In theory, most existing workflows could not handle the high multi-collinearity and potentially non-linearity interactions in epigenome-wide DNA methylation data. Benchmarking studies are needed to compare the relative performance of various approaches for specific cancer types. Adherence to relevant methodological and reporting guidelines are urgently needed.

Bovine meat and milk factor protein expression in tumor-free mucosa of colorectal cancer patients coincides with macrophages and might interfere with patient survival.

Bovine milk and meat factors (BMMFs) are plasmid-like DNA molecules isolated from bovine milk and serum, as well as the peritumor of colorectal cancer (CRC) patients. BMMFs have been proposed as zoonotic infectious agents and drivers of indirect carcinogenesis of CRC, inducing chronic tissue inflammation, radical formation and increased levels of DNA damage. Data on expression of BMMFs in large clinical cohorts to test an association with co-markers and clinical parameters were not previously available and were therefore assessed in this study. Tissue sections with paired tumor-adjacent mucosa and tumor tissues of CRC patients [individual cohorts and tissue microarrays (TMAs) (n = 246)], low-/high-grade dysplasia (LGD/HGD) and mucosa of healthy donors were used for immunohistochemical quantification of the expression of BMMF replication protein (Rep) and CD68/CD163 (macrophages) by co-immunofluorescence microscopy and immunohistochemical scoring (TMA). Rep was expressed in the tumor-adjacent mucosa of 99% of CRC patients (TMA), was histologically associated with CD68+ /CD163+ macrophages and was increased in CRC patients when compared to healthy controls. Tumor tissues showed only low stromal Rep expression. Rep was expressed in LGD and less in HGD but was strongly expressed in LGD/HGD-adjacent tissues. Albeit not reaching statistical significance, incidence curves for CRC-specific death were increased for higher Rep expression (TMA), with high tumor-adjacent Rep expression being linked to the highest incidence of death. BMMF Rep expression might represent a marker and early risk factor for CRC. The correlation between Rep and CD68 expression supports a previous hypothesis that BMMF-specific inflammatory regulations, including macrophages, are involved in the pathogenesis of CRC.

Prognostic role of detailed colorectal location and tumor molecular features: analyses of 13,101 colorectal cancer patients including 2994 early-onset cases.

BACKGROUND: The pathogenic effect of colorectal tumor molecular features may be influenced by several factors, including those related to microbiota, inflammation, metabolism, and epigenetics, which may change along colorectal segments. We hypothesized that the prognostic association of colon cancer location might differ by tumor molecular characteristics. METHODS: Utilizing a consortium dataset of 13,101 colorectal cancer cases, including 2994 early-onset cases, we conducted survival analyses of detailed tumor location stratified by statuses of microsatellite instability (MSI), CpG island methylator phenotype (CIMP), and KRAS and BRAF oncogenic mutation. RESULTS: There was a statistically significant trend for better colon cancer-specific survival in relation to tumor location from the cecum to sigmoid colon (Ptrend = 0.002), excluding the rectum. The prognostic association of colon location differed by MSI status (Pinteraction = 0.001). Non-MSI-high tumors exhibited the cecum-to-sigmoid trend for better colon cancer-specific survival [Ptrend < 0.001; multivariable hazard ratio (HR) for the sigmoid colon (vs. cecum), 0.80; 95% confidence interval (CI) 0.70-0.92], whereas MSI-high tumors demonstrated a suggestive cecum-to-sigmoid trend for worse survival (Ptrend = 0.020; the corresponding HR, 2.13; 95% CI 1.15-3.92). The prognostic association of colon tumor location also differed by CIMP status (Pinteraction = 0.003) but not significantly by age, stage, or other features. Furthermore, MSI-high status was a favorable prognostic indicator in all stages. CONCLUSIONS: Both detailed colonic location and tumor molecular features need to be accounted for colon cancer prognostication to advance precision medicine. Our study indicates the important role of large-scale studies to robustly examine detailed colonic subsites in molecular oncology research.

Molecular Characteristics of Early-Onset Colorectal Cancer According to Detailed Anatomical Locations: Comparison With Later-Onset Cases.

INTRODUCTION: Early-onset colorectal cancer diagnosed before the age of 50 years has been increasing. Likely reflecting the pathogenic role of the intestinal microbiome, which gradually changes across the entire colorectal length, the prevalence of certain tumor molecular characteristics gradually changes along colorectal subsites. Understanding how colorectal tumor molecular features differ by age and tumor location is important in personalized patient management. METHODS: Using 14,004 cases with colorectal cancer including 3,089 early-onset cases, we examined microsatellite instability (MSI), CpG island methylator phenotype (CIMP), and KRAS and BRAF mutations in carcinomas of the cecum, ascending colon, transverse colon, descending colon, sigmoid colon, and rectum and compared early-onset cases with later-onset cases. RESULTS: The proportions of MSI-high, CIMP-high, and BRAF -mutated early-onset tumors were lowest in the rectum (8.8%, 3.4%, and 3.5%, respectively) and highest in the ascending colon (46% MSI-high; 15% CIMP-high) or transverse colon (8.6% BRAF -mutated) (all Ptrend <0.001 across the rectum to ascending colon). Compared with later-onset tumors, early-onset tumors showed a higher prevalence of MSI-high status and a lower prevalence of CIMP-high status and BRAF mutations in most subsites. KRAS mutation prevalence was higher in the cecum compared with that in the other subsites in both early-onset and later-onset tumors ( P < 0.001). Notably, later-onset MSI-high tumors showed a continuous decrease in KRAS mutation prevalence from the rectum (36%) to ascending colon (9%; Ptrend <0.001), followed by an increase in the cecum (14%), while early-onset MSI-high cancers showed no such trend. DISCUSSION: Our findings support biogeographical and pathogenic heterogeneity of colorectal carcinomas in different colorectal subsites and age groups.

Validation of the prognostic value of CD3 and CD8 cell densities analogous to the Immunoscore® by stage and location of colorectal cancer: an independent patient cohort study.

In addition to the traditional staging system in colorectal cancer (CRC), the Immunoscore® has been proposed to characterize the level of immune infiltration in tumor tissue and as a potential prognostic marker. The aim of this study was to examine and validate associations of an immune cell score analogous to the Immunoscore® with established molecular tumor markers and with CRC patient survival in a routine setting. Patients from a population-based cohort study with available CRC tumor tissue blocks were included in this analysis. CD3+ and CD8+ tumor infiltrating lymphocytes in the tumor center and invasive margin were determined in stained tumor tissue slides. Based on the T-cell density in each region, an  immune cell score closely analogous to the concept of the Immunoscore® was calculated and tumors categorized into IS-low, IS-intermediate, or IS-high. Logistic regression models were used to assess associations between clinicopathological characteristics with the immune cell score, and Cox proportional hazards models to analyze associations with cancer-specific, relapse-free, and overall survival. From 1,535 patients with CRC, 411 (27%) had IS-high tumors. Microsatellite instability (MSI-high) was strongly associated with higher immune cell score levels (p < 0.001). Stage I-III patients with IS-high had better CRC-specific and relapse-free survival compared to patients with IS-low (hazard ratio [HR] = 0.42 [0.27-0.66] and HR = 0.45 [0.31-0.67], respectively). Patients with microsatellite stable (MSS) tumors and IS-high had better survival (HRCSS  = 0.60 [0.42-0.88]) compared to MSS/IS-low patients. In this population-based cohort of CRC patients, the immune cell score was significantly associated with better patient survival. It was a similarly strong prognostic marker in patients with MSI-high tumors and in the larger group of patients with MSS tumors. Additionally, this study showed that it is possible to implement an analogous immune cell score approach and validate the Immunoscore® using open source software in an academic setting. Thus, the Immunoscore® could be useful to improve the traditional staging system in colon and rectal cancer used in clinical practice.

Reproductive Factors and Colorectal Cancer Risk: A Population-Based Case-Control Study.

BACKGROUND: Hormone-replacement therapy (HRT) is associated with lower colorectal cancer (CRC) risk among postmenopausal women. However, little is known about the effects of lifetime exposure of women to varying levels of estrogen and progesterone through reproductive factors such as parity, use of oral contraceptives (OC), breastfeeding, and menstruation on CRC risk. METHODS: We assessed associations between reproductive factors and CRC risk among 2650 female CRC patients aged 30+ years and 2175 matched controls in a population-based study in Germany, adjusting for potential confounders by multiple logistic regression. RESULTS: Inverse associations with CRC risk were found for numbers of pregnancies (odds ratio [OR] per pregnancy = 0.91, 95% confidence interval [CI] = 0.86 to 0.97), breastfeeding for 12 months and longer (OR = 0.74, 95% CI = 0.61 to 0.90), and use of either OC or HRT (OR = 0.75, 95% CI = 0.64 to 0.87) or both (OR = 0.58, 95% CI = 0.48 to 0.70). Similar results were found for postmenopausal women only and when adjusting for number of pregnancies and for all reproductive factors analyzed together. Breastfeeding duration of 12 months and longer was associated with lower risk of cancer only in the proximal colon (OR = 0.58, 95% CI = 0.45 to 0.74). CONCLUSIONS: Several reproductive factors were associated with lower CRC risk in women, including number of pregnancies, breastfeeding duration, and use of OC and HRT. This suggests that women's exposure to female reproductive hormones plays a key role in the difference in CRC risk between women and men and in site-specific CRC risk.

Benchmarking weakly-supervised deep learning pipelines for whole slide classification in computational pathology.

Artificial intelligence (AI) can extract visual information from histopathological slides and yield biological insight and clinical biomarkers. Whole slide images are cut into thousands of tiles and classification problems are often weakly-supervised: the ground truth is only known for the slide, not for every single tile. In classical weakly-supervised analysis pipelines, all tiles inherit the slide label while in multiple-instance learning (MIL), only bags of tiles inherit the label. However, it is still unclear how these widely used but markedly different approaches perform relative to each other. We implemented and systematically compared six methods in six clinically relevant end-to-end prediction tasks using data from N=2980 patients for training with rigorous external validation. We tested three classical weakly-supervised approaches with convolutional neural networks and vision transformers (ViT) and three MIL-based approaches with and without an additional attention module. Our results empirically demonstrate that histological tumor subtyping of renal cell carcinoma is an easy task in which all approaches achieve an area under the receiver operating curve (AUROC) of above 0.9. In contrast, we report significant performance differences for clinically relevant tasks of mutation prediction in colorectal, gastric, and bladder cancer. In these mutation prediction tasks, classical weakly-supervised workflows outperformed MIL-based weakly-supervised methods for mutation prediction, which is surprising given their simplicity. This shows that new end-to-end image analysis pipelines in computational pathology should be compared to classical weakly-supervised methods. Also, these findings motivate the development of new methods which combine the elegant assumptions of MIL with the empirically observed higher performance of classical weakly-supervised approaches. We make all source codes publicly available at https://github.com/KatherLab/HIA, allowing easy application of all methods to any similar task.

Swarm learning for decentralized artificial intelligence in cancer histopathology.

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.

Variation of Positive Predictive Values of Fecal Immunochemical Tests by Polygenic Risk Score in a Large Screening Cohort.

INTRODUCTION: Prevalence of colorectal neoplasms varies by polygenic risk scores (PRS). We aimed to assess to what extent a PRS might be relevant for defining personalized cutoff values for fecal immunochemical tests (FITs) in colorectal cancer screening. METHODS: Among 5,306 participants of screening colonoscopy who provided a stool sample for a quantitative FIT (Ridascreen Hemoglobin or FOB Gold) before colonoscopy, a PRS was determined, based on the number of risk alleles in 140 single nucleotide polymorphisms. Subjects were classified into low, medium, and high genetic risk of colorectal neoplasms according to PRS tertiles. We calculated positive predictive values (PPVs) and numbers needed to scope (NNS) to detect 1 advanced neoplasm (AN) by the risk group, and cutoff variation needed to achieve comparable PPVs across risk groups in the samples tested with Ridascreen (N = 1,271) and FOB Gold (N = 4,035) independently, using cutoffs yielding 85%, 90%, or 95% specificity. RESULTS: Performance of both FITs was very similar within each PRS group. For a given cutoff, PPVs were consistently higher by 11%-15% units in the high-risk PRS group compared with the low-risk group (all P values < 0.05). Correspondingly, NNS to detect 1 advanced neoplasm varied from 2 (high PRS, high cutoff) to 5 (low PRS, low cutoff). Conversely, very different FIT cutoffs would be needed to ensure comparable PPVs across PRS groups. DISCUSSION: PPVs and NNS of FITs varied widely across people with high and low genetic risk score. Further research should evaluate the relevance of these differences for personalized colorectal cancer screening.

Weakly supervised annotation-free cancer detection and prediction of genotype in routine histopathology.

Deep learning is a powerful tool in computational pathology: it can be used for tumor detection and for predicting genetic alterations based on histopathology images alone. Conventionally, tumor detection and prediction of genetic alterations are two separate workflows. Newer methods have combined them, but require complex, manually engineered computational pipelines, restricting reproducibility and robustness. To address these issues, we present a new method for simultaneous tumor detection and prediction of genetic alterations: The Slide-Level Assessment Model (SLAM) uses a single off-the-shelf neural network to predict molecular alterations directly from routine pathology slides without any manual annotations, improving upon previous methods by automatically excluding normal and non-informative tissue regions. SLAM requires only standard programming libraries and is conceptually simpler than previous approaches. We have extensively validated SLAM for clinically relevant tasks using two large multicentric cohorts of colorectal cancer patients, Darmkrebs: Chancen der Verhütung durch Screening (DACHS) from Germany and Yorkshire Cancer Research Bowel Cancer Improvement Programme (YCR-BCIP) from the UK. We show that SLAM yields reliable slide-level classification of tumor presence with an area under the receiver operating curve (AUROC) of 0.980 (confidence interval 0.975, 0.984; n = 2,297 tumor and n = 1,281 normal slides). In addition, SLAM can detect microsatellite instability (MSI)/mismatch repair deficiency (dMMR) or microsatellite stability/mismatch repair proficiency with an AUROC of 0.909 (0.888, 0.929; n = 2,039 patients) and BRAF mutational status with an AUROC of 0.821 (0.786, 0.852; n = 2,075 patients). The improvement with respect to previous methods was validated in a large external testing cohort in which MSI/dMMR status was detected with an AUROC of 0.900 (0.864, 0.931; n = 805 patients). In addition, SLAM provides human-interpretable visualization maps, enabling the analysis of multiplexed network predictions by human experts. In summary, SLAM is a new simple and powerful method for computational pathology that could be applied to multiple disease contexts. © 2021 The Authors. The Journal of Pathology published by John Wiley & Sons, Ltd. on behalf of The Pathological Society of Great Britain and Ireland.

Validation of Genetic Markers Associated with Survival in Colorectal Cancer Patients Treated with Oxaliplatin-Based Chemotherapy.

BACKGROUND: Associations between candidate genetic variants and treatment outcomes of oxaliplatin, a drug commonly used for colorectal cancer patients, have been reported but not robustly established. This study aimed to validate previously reported prognostic and predictive genetic markers for oxaliplatin treatment outcomes and evaluate additional putative functional variants. METHODS: Fifty-three SNPs were selected based on previous reports (40 SNPs) or putative function in candidate genes (13 SNPs). We used data from 1,502 patients with stage II-IV colorectal cancer who received primary adjuvant chemotherapy, 37% of whom received oxaliplatin treatment. Multivariable Cox proportional hazards models for overall survival and progression-free survival were applied separately in stage II-III and stage IV patients. For predictive SNPs, differential outcomes according to the type of chemotherapy (oxaliplatin-based vs. others) were evaluated using an interaction term. For prognostic SNPs, the association was assessed solely in patients with oxaliplatin-based treatment. RESULTS: Twelve SNPs were predictive and/or prognostic at P < 0.05 with differential survival based on the type of treatment, in patients with stage II-III (GSTM5-rs11807, ERCC2-rs13181, ERCC2-rs1799793, ERCC5-rs2016073, XPC-rs2228000, P2RX7-rs208294, HMGB1-rs1360485) and in patients with stage IV (GSTM5-rs11807, MNAT1-rs3783819, MNAT1-rs4151330, CXCR1-rs2234671, VEGFA-rs833061, P2RX7-rs2234671). In addition, five novel putative functional SNPs were identified to be predictive (ATP8B3-rs7250872, P2RX7-rs2230911, RPA1-rs5030755, MGMT-rs12917, P2RX7-rs2227963). CONCLUSIONS: Some SNPs yielded prognostic and/or predictive associations significant at P < 0.05, however, none of the associations remained significant after correction for multiple testing. IMPACT: We did not robustly confirm previously reported SNPs despite some suggestive findings but identified further potential predictive SNPs, which warrant further investigation in well-powered studies.

Salicylic Acid and Risk of Colorectal Cancer: A Two-Sample Mendelian Randomization Study.

Salicylic acid (SA) has observationally been shown to decrease colorectal cancer (CRC) risk. Aspirin (acetylsalicylic acid, that rapidly deacetylates to SA) is an effective primary and secondary chemopreventive agent. Through a Mendelian randomization (MR) approach, we aimed to address whether levels of SA affected CRC risk, stratifying by aspirin use. A two-sample MR analysis was performed using GWAS summary statistics of SA (INTERVAL and EPIC-Norfolk, N = 14,149) and CRC (CCFR, CORECT, GECCO and UK Biobank, 55,168 cases and 65,160 controls). The DACHS study (4410 cases and 3441 controls) was used for replication and stratification of aspirin-use. SNPs proxying SA were selected via three methods: (1) functional SNPs that influence the activity of aspirin-metabolising enzymes; (2) pathway SNPs present in enzymes' coding regions; and (3) genome-wide significant SNPs. We found no association between functional SNPs and SA levels. The pathway and genome-wide SNPs showed no association between SA and CRC risk (OR: 1.03, 95% CI: 0.84-1.27 and OR: 1.08, 95% CI: 0.86-1.34, respectively). Results remained unchanged upon aspirin use stratification. We found little evidence to suggest that an SD increase in genetically predicted SA protects against CRC risk in the general population and upon stratification by aspirin use.

Smoking Behavior and Prognosis After Colorectal Cancer Diagnosis: A Pooled Analysis of 11 Studies.

Background: Smoking has been associated with colorectal cancer (CRC) incidence and mortality in previous studies, but current evidence on smoking in association with survival after CRC diagnosis is limited. Methods: We pooled data from 12 345 patients with stage I-IV CRC from 11 epidemiologic studies in the International Survival Analysis in Colorectal Cancer Consortium. Cox proportional hazards regression models were used to evaluate the associations of prediagnostic smoking behavior with overall, CRC-specific, and non-CRC-specific survival. Results: Among 12 345 patients with CRC, 4379 (35.5%) died (2515 from CRC) over a median follow-up time of 7.5 years. Smoking was strongly associated with worse survival in stage I-III patients, whereas no association was observed among stage IV patients. Among stage I-III patients, clear dose-response relationships with all survival outcomes were seen for current smokers. For example, current smokers with 40 or more pack-years had statistically significantly worse overall, CRC-specific, and non-CRC-specific survival compared with never smokers (hazard ratio [HR] =1.94, 95% confidence interval [CI] =1.68 to 2.25; HR = 1.41, 95% CI = 1.12 to 1.78; and HR = 2.67, 95% CI = 2.19 to 3.26, respectively). Similar associations with all survival outcomes were observed for former smokers who had quit for less than 10 years, but only a weak association with non-CRC-specific survival was seen among former smokers who had quit for more than 10 years. Conclusions: This large consortium of CRC patient studies provides compelling evidence that smoking is strongly associated with worse survival of stage I-III CRC patients in a clear dose-response manner. The detrimental effect of smoking was primarily related to noncolorectal cancer events, but current heavy smoking also showed an association with CRC-specific survival.

Deep learning can predict lymph node status directly from histology in colorectal cancer.

BACKGROUND: Lymph node status is a prognostic marker and strongly influences therapeutic decisions in colorectal cancer (CRC). OBJECTIVES: The objective of the study is to investigate whether image features extracted by a deep learning model from routine histological slides and/or clinical data can be used to predict CRC lymph node metastasis (LNM). METHODS: Using histological whole slide images (WSIs) of primary tumours of 2431 patients in the DACHS cohort, we trained a convolutional neural network to predict LNM. In parallel, we used clinical data derived from the same cases in logistic regression analyses. Subsequently, the slide-based artificial intelligence predictor (SBAIP) score was included in the regression. WSIs and data from 582 patients of the TCGA cohort were used as the external test set. RESULTS: On the internal test set, the SBAIP achieved an area under receiver operating characteristic (AUROC) of 71.0%, the clinical classifier achieved an AUROC of 67.0% and a combination of the two classifiers yielded an improvement to 74.1%. Whereas the clinical classifier's performance remained stable on the TCGA set, performance of the SBAIP dropped to an AUROC of 61.2%. Performance of the clinical classifier depended strongly on the T stage. CONCLUSION: Deep learning-based image analysis may help predict LNM of patients with CRC using routine histological slides. Combination with clinical data such as T stage might be useful. Strategies to increase performance of the SBAIP on external images should be investigated.

Gastrointestinal cancer classification and prognostication from histology using deep learning: Systematic review.

BACKGROUND: Gastrointestinal cancers account for approximately 20% of all cancer diagnoses and are responsible for 22.5% of cancer deaths worldwide. Artificial intelligence-based diagnostic support systems, in particular convolutional neural network (CNN)-based image analysis tools, have shown great potential in medical computer vision. In this systematic review, we summarise recent studies reporting CNN-based approaches for digital biomarkers for characterization and prognostication of gastrointestinal cancer pathology. METHODS: Pubmed and Medline were screened for peer-reviewed papers dealing with CNN-based gastrointestinal cancer analyses from histological slides, published between 2015 and 2020.Seven hundred and ninety titles and abstracts were screened, and 58 full-text articles were assessed for eligibility. RESULTS: Sixteen publications fulfilled our inclusion criteria dealing with tumor or precursor lesion characterization or prognostic and predictive biomarkers: 14 studies on colorectal or rectal cancer, three studies on gastric cancer and none on esophageal cancer. These studies were categorised according to their end-points: polyp characterization, tumor characterization and patient outcome. Regarding the translation into clinical practice, we identified several studies demonstrating generalization of the classifier with external tests and comparisons with pathologists, but none presenting clinical implementation. CONCLUSIONS: Results of recent studies on CNN-based image analysis in gastrointestinal cancer pathology are promising, but studies were conducted in observational and retrospective settings. Large-scale trials are needed to assess performance and predict clinical usefulness. Furthermore, large-scale trials are required for approval of CNN-based prediction models as medical devices.

DNA Methylation-Based Estimates of Circulating Leukocyte Composition for Predicting Colorectal Cancer Survival: A Prospective Cohort Study.

Leukocytes are involved in the progression of colorectal cancer (CRC). The proportion of six major leukocyte subtypes can be estimated using epigenome-wide DNA methylation (DNAm) data from stored blood samples. Whether the composition of circulating leukocytes can be used as a prognostic factor is unclear. DNAm-based leukocyte proportions were obtained from a prospective cohort of 2206 CRC patients. Multivariate Cox regression models and survival curves were applied to assess associations between leukocyte composition and survival outcomes. A higher proportion of lymphocytes, including CD4+ T cells, CD8+ T cells, B cells, and NK cells, was associated with better survival, while a higher proportion of neutrophils was associated with poorer survival. CD4+ T cells outperformed other leukocytes in estimating the patients' prognosis. Comparing the highest quantile to the lowest quantile of CD4+ T cells, hazard ratios (95% confidence intervals) of all-cause and CRC-specific mortality were 0.59 (0.48, 0.72) and 0.59 (0.45, 0.77), respectively. Furthermore, the association of CD4+ T cells and prognosis was stronger among patients with early or intermediate CRC or patients with colon cancer. In conclusion, the composition of circulating leukocytes estimated from DNAm, particularly the proportions of CD4+ T cells, could be used as promising independent predictors of CRC survival.

Consistent Major Differences in Sex- and Age-Specific Diagnostic Performance among Nine Faecal Immunochemical Tests Used for Colorectal Cancer Screening.

Evidence on diagnostic performance of faecal immunochemical tests (FITs) by sex and age is scarce. We aimed to evaluate FIT performance for detection of advanced colorectal neoplasia (AN) by sex and age across nine different FIT brands in a colonoscopy-controlled setting. The faecal samples were obtained from 2042 participants of colonoscopy screening. All eligible cases with AN (n = 216) and 300 randomly selected participants without AN were included. Diagnostic performance for detection of AN was assessed by sex and age (50-64 vs. 65-79 years for each of the nine FITs individually and for all FITs combined. Sensitivity was consistently lower, and specificity was consistently higher for females as compared with males (pooled values at original FIT cutoffs, 25.7% vs. 34.6%, p = 0.12 and 96.2% vs. 90.8%, p < 0.01, respectively). Positive predictive values (PPVs) were similar between both sexes, but negative predictive values (NPVs) were consistently higher for females (pooled values, 91.8% vs. 86.6%, p < 0.01). Sex-specific cutoffs attenuated differences in sensitivities but increased differences in predictive values. According to age, sensitivities and specificities were similar, whereas PPVs were consistently lower and NPVs were consistently higher for the younger participants. A negative FIT is less reliable in ruling out AN among men than among women and among older than among younger participants. Comparisons of measures of diagnostic performance among studies with different sex or age distributions should be interpreted with caution.