GI Oncology Daily Digest

May 12, 2026 — Cancer Screening Special Edition
Curated by Dr. Allan Pereira — Moffitt Cancer Center

Top 5 Papers

#1
NORCCAP 23-year follow-up — Single sigmoidoscopy screening sustains CRC incidence and mortality reductions in men two decades later
Source: Annals of Internal Medicine  |  Authors: Botteri E, Holme Ø, Løberg M, Bretthauer M, Kalager M, Randel KR, Hoff G, et al.  |  Published: May 12, 2026
Score: 13/20 — Base 7 (Annals of Internal Medicine) + 3 (Phase III RCT) + 2 (mortality benefit in men) + 1 (longest CRC screening RCT ever reported, anchor evidence for sigmoidoscopy) = 13
Longest follow-up of any CRC screening RCT to date. NORCCAP randomized 100,210 Norwegian adults aged 50–64 (98,654 in ITT analysis; 20,552 screening vs 78,102 no screening) to once-only flexible sigmoidoscopy ± one fecal immunochemical test, or no screening. Screening participation was 61.4% in men and 64.7% in women. At 23 years, cumulative CRC incidence was 4.3% vs 6.0% in men (risk difference −1.7 percentage points, 95% CI −2.2 to −1.2) and 4.2% vs 4.7% in women (−0.5 pp, 95% CI −1.0 to −0.01). CRC mortality dropped from 2.2% to 1.4% in men (−0.8 pp, 95% CI −1.1 to −0.5) but did not change in women (1.3% vs 1.4%). The effect was strongest for rectosigmoid cancer, and adding FIT to sigmoidoscopy did not augment screening benefit. The result reframes the contemporary post-NORDICC debate by showing that sigmoidoscopy benefit is durable, sex-asymmetric, and concentrated where the scope actually reaches.
Post angle: Longest CRC screening RCT ever: 23-year NORCCAP shows a single sigmoidoscopy keeps cutting CRC incidence and mortality two decades later — but only meaningfully in men. Distal-anchored benefit. #CRC #Screening #GIOnc
#2
NORDICC 13-year follow-up — Colonoscopy screening RCT in 84,583 adults reduces CRC incidence but mortality benefit remains elusive
Source: Lancet  |  Authors: Kaminski MF, Kalager M, Løberg M, Bretthauer M, et al.  |  Published: May 5, 2026
Score: 14/20 — Base 9 (Lancet) + 3 (Phase III RCT) + 2 (large population endpoint dataset) = 14
Multi-country NORDICC trial — 84,583 men and women aged 55–64 across Norway, Poland, and Sweden randomized 1:2 to single-screening colonoscopy or no screening. At 13 years, CRC incidence was 1.46% in the screening arm vs 1.80% in controls (intention-to-screen RR 0.81, 95% CI 0.71–0.90; per-protocol RR 0.55, 0.33–0.81). Effect was driven by distal lesions (RR 0.79, p<0.0001 for distal vs proximal interaction) and more pronounced in men (RR 0.77) than women (RR 0.87). CRC mortality was not statistically reduced (ITT RR 0.88, 95% CI 0.68–1.08; per-protocol RR 0.70, 0.26–1.25). Crucially, baseline CRC mortality in the no-screen group (0.47%) was far below the 0.82% assumed when the trial was designed — leaving the trial underpowered for mortality. Read alongside NORCCAP: the modality you choose matters less than the anatomy it reaches.
Post angle: NORDICC at 13 years: one colonoscopy cuts CRC incidence (RR 0.81 ITT, 0.55 per-protocol) but mortality benefit still not statistically significant. Background mortality fell harder than expected — interpret carefully. #CRC #Screening #GIOnc
#3
REDMOD — Next-generation AI radiomics on standard-of-care CT detects visually occult pancreatic cancer ~475 days before clinical diagnosis
Source: Gut  |  Authors: Mukherjee S, Antony A, Patnam NG, Trivedi KH, Karbhari A, Fletcher JG, Truty M, Chari ST, Goenka AH, et al. (Mayo Clinic + MD Anderson)  |  Published: April 28, 2026
Score: 9/20 — Base 6 (Gut, Tier-1 GI specialty journal) + 2 (large multi-institutional validation dataset) + 1 (biomarker-guided / AI-driven precision screening) = 9
REDMOD (Radiomics-based Early Detection MODel) is a fully automated AI framework that identifies subvisual radiomic signatures of pre-diagnostic pancreatic ductal adenocarcinoma (PDAC) on standard-of-care CT. Trained on a multi-institutional cohort (n=969; 156 pre-diagnostic, 813 controls) and tested on an independent low-prevalence (~1:6) cohort (n=493; 63 pre-diagnostic, 430 controls). On the independent set REDMOD achieved AUC 0.82 with 73.0% sensitivity at a median 475-day lead time — nearly twofold higher sensitivity than expert radiologists (38.9%; p<0.001), growing to nearly threefold (68.0% vs 23.0%) at >24-month lead time. The model demonstrated 90–92% longitudinal test–retest stability and externally validated specificity (81.3% in n=539; 87.5% in n=80). Predictive power derived principally from multi-scale wavelet-filtered textural features (AUC 0.82 vs 0.74 for unfiltered, p=0.007). The proof-of-principle that AI can rescue early-detection signal from imaging that the human eye already called negative.
Post angle: Pancreatic cancer's biggest screening barrier is that early disease is invisible on CT. REDMOD turns radiomic noise into signal: AUC 0.82, 73% sensitivity, ~475-day lead time, nearly 2× the sensitivity of expert radiologists. #PDAC #PancreaticCancer #AI #GIOnc
#4
Non-contrast abbreviated MRI outperforms ultrasound for HCC surveillance — Prospective single-center diagnostic accuracy trial in cirrhotic patients (NCT05716620)
Source: Hepatology  |  Authors: Gupta P, Singh S, Gulati A, Mudgil P, Kalra N, Premkumar M, Taneja S, Duseja A, et al. (PGIMER Chandigarh)  |  Published: May 11, 2026
Score: 9/20 — Base 6 (Hepatology, Tier-1 hepatology journal) + 2 (prospective paired diagnostic accuracy study) + 1 (early-stage detection / biomarker-guided surveillance) = 9
Prospective single-center study enrolling 404 cirrhotic patients with annual HCC risk >5% who underwent paired surveillance with ultrasound (US) and rapid non-contrast abbreviated MRI (AMRI) across two rounds 6 months apart (614 paired exams). Multiphasic contrast-enhanced MRI was the reference standard, triggered by any positive screen. AMRI substantially outperformed US: per-patient sensitivity 94.6% (95% CI 83.3–98.9) vs 51.4% (34.7–67.8), specificity 96.6% vs 69.5% (both p<0.001), AUROC 0.956 vs 0.604 (p<0.001), per-lesion detection 37/39 (94.9%) vs 20/39 (51.3%). Critically, 97.3% of AMRI-detected HCCs were BCLC stage 0 or A — the curable window. Interobserver agreement was 'almost perfect' for AMRI (κ=0.929) vs 'moderate' for US (κ=0.631). If confirmed in outcome trials, a 10-minute non-contrast MRI could meaningfully reshape HCC surveillance, particularly in MASLD cirrhosis where US sensitivity is known to be poor.
Post angle: HCC surveillance may be due for a regime change. AMRI sensitivity 94.6% vs ultrasound 51.4% (p<0.001), and 97% of MRI-found HCCs are early-stage (BCLC 0/A). Time to ask whether US is still the right default in MASLD cirrhosis. #HCC #LiverCancer #GIOnc
#5
Liquid-biopsy multi-methylation panel achieves AUC 0.96 for minimally invasive gastric cancer screening
Source: Clinical Epigenetics  |  Authors: Long F, Xu Y, Wu K, Fu X, et al.  |  Published: April 25, 2026
Score: 6/20 — Base 5 (Clinical Epigenetics) + 1 (biomarker-guided precision screening) = 6
A 5-gene blood-based methylation panel (ELMO1, FGF12, NPY, SEPTIN9, ZNF671) was developed for gastric cancer screening using Random Forest modeling. In a training cohort of 605 subjects the model achieved AUC 0.9585, sensitivity 81.85%, and specificity 92.49%. In an independent validation cohort of 152 subjects, AUC was 0.8868 with 82.19% sensitivity and 81.01% specificity. The model performed consistently across pathological stages with AUCs ranging from 0.82 (stage 0–IB) to 0.94 (stage IV), demonstrating real potential for early-stage detection through a peripheral blood test. In a disease where population-level gastroscopy is logistically impossible in most countries, a credible blood-based screen is the missing piece — and the cleanest complement to the imaging-and-endoscopy axis covered by the other four papers in this edition.
Post angle: A simple blood test with AUC 0.96 for gastric cancer screening? A 5-gene methylation panel that could complement (not yet replace) gastroscopy for population screening. Validation at scale is the next step. #GastricCancer #LiquidBiopsy #EarlyDetection #GIOnc

Additional Papers of Interest

  1. Why this edition matters — Five papers, four modalities, one organizing question
    Editorial framing — Sigmoidoscopy (NORCCAP) and colonoscopy (NORDICC) keep teaching us that anatomy and population dose matter as much as the scope; AI radiomics (REDMOD) suggests the CT scans we already do contain a pre-diagnostic signal we have been missing; abbreviated non-contrast MRI may unseat ultrasound as the surveillance modality of choice in the MASLD-cirrhosis era; and a 5-gene blood methylation panel hints that blood-based gastric screening is finally tractable. Read together, they argue that 'cancer screening' in 2026 is no longer one decision — it is an organ-by-organ optimization across modality, biology, and lead time.
  2. NORDICC vs NORCCAP — A side-by-side reading
    Methodological note — NORCCAP (n=98,654, 23 yr, sigmoidoscopy) shows a durable mortality benefit in men; NORDICC (n=84,583, 13 yr, colonoscopy) does not. The likely explanation is not modality superiority but trial design: NORDICC's background CRC mortality fell to roughly half of what was assumed at design, leaving the trial badly underpowered for the mortality endpoint, while NORCCAP's longer follow-up and anatomy-of-reach effect allow distal-lesion mortality to surface. Distinguishing 'no effect' from 'undetectable effect' is the whole game.
  3. Companion editorial — Seeing the unseen: AI radiomics unmasking occult pancreatic cancer?
    Gut editorial (Michl P, Roth L, May 8, 2026) — Companion piece to REDMOD reflecting on what AI early-detection on routine CT means for surveillance of high-risk PDAC populations (familial pancreatic cancer kindreds, BRCA carriers, IPMN follow-up cohorts) and how prospective interception trials should be designed to convert a 475-day lead time into improved resectability and survival.
Back to all digests