Every two weeks, a critical review of the most significant colorectal cancer research indexed in PubMed, with key findings, methodological appraisal, and clinical implications.
1. Interventions to improve follow-up colonoscopy after an abnormal FIT or FOBT result: systematic review and meta-analysis (JAMA Internal Medicine)
Wei Yu Chua (Yong Loo Lin School of Medicine, National University of Singapore), Luke Xavier Chew, Claire K. M. Chan, Cheng Ean Chee, Ravindran Kanesvaran, Michael Pignone, et al.; last author Kennedy Y. Y. Ng (Harvard T.H. Chan School of Public Health)
Key points
- Systematic review and random-effects meta-analysis of 21 randomized and nonrandomized interventional studies (19,613 patients aged 50 to 75) testing ways to raise colonoscopy completion after an abnormal fecal immunochemical test (FIT) or fecal occult blood test (FOBT).
- Pooled follow-up colonoscopy completion was 69.1% with interventions (6,272 of 9,350) versus 47.8% in control groups (4,755 of 10,263).
- Patient navigation: risk ratio (RR) 1.24 (95% CI 1.15-1.33; I² = 57.1%). Patient reminders: RR 1.32 (95% CI 1.04-1.67; I² = 89.9%).
- Clinician reminders: RR 1.36 (95% CI 1.12-1.65; I² = 94.4%). Automated referral to gastroenterology consult: RR 1.68 (95% CI 1.09-2.60; I² = 91.9%).
- Heterogeneity fell to low levels for navigation and both reminder types after excluding outliers, but stayed moderate for automated referrals.
Key limitations
- Moderate to high between-study heterogeneity, particularly for automated referrals and clinician reminders.
- Mix of randomized and nonrandomized designs, with English-language studies only.
- The incremental contribution of each component in multicomponent programs could not be separated.
Future directions
- Randomized evaluation of automated gastroenterology referral, the strategy with the largest but least consistent effect.
- Component-level (factorial) trials to identify which elements of multicomponent programs drive completion.
- Linking higher completion rates to stage at diagnosis and colorectal cancer mortality.
Stool-based screening with the fecal immunochemical test (FIT) or guaiac fecal occult blood test (FOBT) is a cost-effective, widely used route into colorectal cancer screening, and colorectal cancer remains the second leading cause of cancer death worldwide. A stool test is only the first step, however: its benefit depends on prompt diagnostic colonoscopy after an abnormal result. Rates of follow-up colonoscopy after a positive stool test remain suboptimal in many health systems, and every patient who does not complete it represents a missed chance to diagnose cancer early or remove an advanced precursor lesion.
Health systems have tried many ways to close this gap, from patient navigators who guide people through bowel preparation and scheduling, to letters, calls and text reminders, electronic alerts to clinicians, and automatic referral to gastroenterology once a positive result is filed. Individual studies of these interventions have differed widely in setting, population and design, and it has been unclear which approaches work best and how large their effects are. Chua and colleagues set out to quantify the pooled completion rate and the relative effect of each type of intervention.
The investigators, from the National University of Singapore, the National Cancer Centre Singapore, Singapore General Hospital, the Chinese University of Hong Kong, Duke University and the Harvard T.H. Chan School of Public Health, among others, searched MEDLINE, Embase and Cochrane from inception to May 9, 2026, and screened bibliographies. Eligible studies were English-language randomized trials and nonrandomized interventional studies that compared strategies to improve follow-up colonoscopy after an abnormal FIT or FOBT. Two authors extracted data independently, with a third resolving discrepancies, following PRISMA guidance. Prevalence and risk ratios with 95% confidence intervals were pooled using random-effects models.
Twenty-one studies with 19,613 patients (41.3% female, 58.7% male; aged 50 to 75) were included. Across control groups, 4,755 of 10,263 patients completed follow-up colonoscopy (pooled prevalence 47.8%), compared with 6,272 of 9,350 after interventions (pooled prevalence 69.1%). Each major intervention type was associated with higher completion: patient navigation (RR 1.24; 95% CI 1.15-1.33), patient reminders (RR 1.32; 95% CI 1.04-1.67), clinician reminders (RR 1.36; 95% CI 1.12-1.65) and automated referrals to gastroenterology consult (RR 1.68; 95% CI 1.09-2.60).
Heterogeneity varied considerably by intervention. It was moderate for patient navigation (I² = 57.1%) and high for patient reminders (89.9%), clinician reminders (94.4%) and automated referrals (91.9%). In sensitivity analyses excluding outlying studies, heterogeneity became low for navigation and both forms of reminder, which strengthens confidence in those estimates. For automated referrals it remained moderate, so the largest point estimate in the analysis is also the least certain. The consistency of benefit across patient-directed and system-directed strategies suggests that the gap has several causes, including patient barriers and failures of clinical follow-through.
The authors point to two main gaps. First, the effect of automated referrals is uncertain because of between-study heterogeneity. Second, many programs combined several components, and the incremental contribution of each could not be isolated. Further cautions follow from the design: the meta-analysis pooled randomized and nonrandomized studies, was restricted to English-language reports, and measured colonoscopy completion rather than downstream outcomes such as stage at diagnosis or cancer mortality. Health-system context, including insurance, scheduling capacity and existing tracking infrastructure, probably modifies how well any single intervention transfers to a new setting.
For screening programs and primary care practices, the findings support building follow-up into the screening pathway rather than leaving it to the patient. Navigation, reminders and clinician alerts each have reasonably consistent evidence, and automated referral is a promising but less well-characterized option. The authors call for more research on automated referrals and on the incremental effect of individual components in multicomponent programs. Pragmatic trials that compare combinations head to head, and that follow patients through to stage at diagnosis, would help programs decide where to invest limited resources.
Source: Chua et al., "Evaluation of Intervention Strategies to Improve Follow-Up Colonoscopy After Abnormal FIT/FOBT Result: A Systematic Review and Meta-Analysis," JAMA Internal Medicine, published online October 5, 2026 (doi:10.1001/jamainternmed.2026.4827). Read the paper
2. An updated blood-based colorectal cancer screening test: clinical validation in average-risk adults from the PREEMPT CRC study (ACG 2026)
Presenting author Aasma Shaukat (NYU Grossman School of Medicine), on behalf of the investigators
Key points
- Assay and algorithm enhancements to the SimpleScreen CRC cell-free DNA blood test (Freenome) were evaluated in 9,001 previously untested samples from the prospective, multicenter PREEMPT CRC study in average-risk adults.
- At specificity fixed at 90%, sensitivity for advanced precancerous lesions (APL), a coprimary endpoint, rose from 13.7% (95% CI 12.4%-15%) with the original test to 18.2% (95% CI 16.3%-20.4%).
- Colorectal cancer sensitivity, the other coprimary endpoint, was 80.4% (95% CI 70.2%-87.7%) with the updated test versus 81.1% with the original; by stage it was 52% (I), 100% (II), 97.3% (III) and 100% (IV).
- Sensitivity for high-grade dysplasia (secondary endpoint) was 41.9% (95% CI 34%-50.3%), an 11.5-percentage-point gain; microsimulation modeling projected a 9.5% reduction in both CRC incidence and mortality.
- Presented at the ACG 2026 Annual Scientific Meeting (Nashville, October 2026); not yet published in a peer-reviewed journal, and the updated version is not yet FDA approved.
Key limitations
- APL sensitivity remained low (about 18%), and stage I cancer sensitivity was 52%.
- The enhancements were evaluated in samples from the same trial program used to validate the original test, which may limit generalizability.
- The incidence and mortality benefit is modeled, not observed; the study is tied to the test's developer and has not been peer reviewed.
Future directions
- Regulatory submission of the updated assay, which the investigators plan.
- Further development (a planned version three) aimed at stage I cancer and precancerous lesion detection.
- Real-world studies of adherence to follow-up colonoscopy after a positive blood test.
Blood-based colorectal cancer screening tests that analyze cell-free DNA are attractive because a blood draw can be added to a routine clinic visit and avoids stool handling and bowel preparation. Their central weakness has been limited detection of advanced precancerous lesions (APL), the adenomas and serrated lesions whose removal prevents cancer, as well as reduced sensitivity for early-stage cancer. Because much of the benefit of colonoscopy and stool-based screening comes from prevention rather than early detection alone, improving APL sensitivity is the main technical goal for this class of tests.
The first version of the SimpleScreen CRC test, developed by Freenome, received US Food and Drug Administration approval in July 2026 on the basis of the prospective, multicenter PREEMPT CRC validation study. As NYU's Aasma Shaukat explained at the meeting, the update was developed because the team wanted to maximize sensitivity. Iterative improvement of assays and classification algorithms is common for molecular diagnostics, but each version requires its own clinical validation, since gains in sensitivity can come at the cost of specificity or fail to replicate in independent samples.
The investigators applied assay and algorithm enhancements to the existing test and evaluated performance using previously untested samples from PREEMPT CRC, collected both before and after the April 2021 cutoff of the original validation study. The study population comprised average-risk adults, and sensitivity estimates were weighted and adjusted to US census age and weight distributions. Coprimary endpoints were sensitivity for colorectal cancer and for APL; sensitivity for high-grade dysplasia was secondary. Specificity was fixed at 90%. In all, 9,001 participants were analyzed (mean age 59.9 years; 51.3% women; 76.3% White), including 89 with CRC, 1,570 with APL and 7,342 with neither.
At 90% specificity, APL sensitivity improved from 13.7% (95% CI 12.4%-15%) with the original test to 18.2% (95% CI 16.3%-20.4%) with the updated version. Sensitivity rose across all APL subtypes, with the largest gain, 11.5 percentage points, for high-grade dysplasia, for which the updated test reached 41.9% (95% CI 34%-50.3%). Colorectal cancer sensitivity was essentially unchanged: 80.4% (95% CI 70.2%-87.7%) with the updated test versus 81.1% with the original.
Stage-specific cancer sensitivity with the updated test was 52% for stage I, 100% for stage II, 97.3% for stage III and 100% for stage IV, underscoring that detection of the earliest cancers remains the weakest point. Using a validated microsimulation model, the investigators estimated that the updated test would reduce both colorectal cancer incidence and mortality by 9.5%. Shaukat described the improvements as incremental gains and said the team planned to submit the updated version to regulators while continuing work toward a third version.
Several limitations temper the findings. APL sensitivity, although improved, remains below 20%, and stage I sensitivity of about half means many early cancers would still be missed. Confidence intervals for specificity and for the original test's cancer sensitivity were not reported in the coverage, and the cancer sensitivity estimate rests on only 89 cancers, giving a wide interval. The incidence and mortality projections come from modeling rather than observed outcomes. The enhancements were evaluated in samples from the same trial program used for the original approval, the study is linked to the test's developer, and the lead investigator reports consulting relationships with several screening companies.
The study illustrates both the progress and the ceiling of current blood-based screening. Better APL detection would bring blood tests closer to the preventive performance of stool DNA testing and colonoscopy, but the present gains are modest. For clinicians, blood tests remain an option for patients who decline other modalities rather than a substitute for them, and a positive result must be followed by colonoscopy. Independent validation, peer-reviewed publication and real-world data on follow-up adherence will be needed before the updated test's place in screening programs is clear.
Source: Shaukat et al., "Improved sensitivity for advanced precancerous lesions demonstrated in clinical validation of an updated colorectal cancer screening blood test in an average-risk population," presented at the ACG 2026 Annual Scientific Meeting, Nashville, October 2026; not yet published in a peer-reviewed journal. Conference coverage (Healio)
3. Sex differences in fecal immunochemical test sensitivity and sex-specific positivity thresholds: analysis of the DeeP-C and BLUE-C validation studies (ACG 2026)
Presenting author Thomas F. Imperiale (Indiana University School of Medicine), on behalf of the investigators
Key points
- Retrospective analysis of cross-sectional data from 30,165 participants (mean age 63.4 years; 46.6% men) in the DeeP-C and BLUE-C multitarget stool DNA validation studies, both of which used the same quantitative comparator FIT at a threshold of 100 ng/mL (about 20 µg/g feces).
- Among 163 colorectal cancers, FIT sensitivity was 77.4% in men versus 62% in women (P = .03); among 2,901 advanced precancerous lesions, 25.8% versus 20.2% (P < .001).
- Median fecal hemoglobin was higher in men than women with CRC (1,541 vs 955 ng/mL; P = .03) and with APL (230 vs 174 ng/mL; P < .001).
- At 95% specificity, the CRC threshold would be 120 ng/mL for men and 94 ng/mL for women. Male sex (OR 1.22; 95% CI 1.02-1.47), smoking (OR 1.36; 1.14-1.63), BMI ≥30 (OR 1.39; 1.17-1.67) and distal or rectal location (OR 2.41; 2.01-2.89) were independently associated with FIT sensitivity for advanced neoplasia.
- Presented at the ACG 2026 Annual Scientific Meeting (Nashville, October 2026); not yet published in a peer-reviewed journal.
Key limitations
- Only one FIT brand, evaluated at a single threshold, was studied.
- A single application of FIT was assessed, whereas programs use repeated annual testing.
- Retrospective analysis of trial populations; the presenter reports research support from Exact Sciences.
Future directions
- Validation of sex-specific thresholds for other FIT brands, since optimal cutoffs vary by assay.
- Modeling of colonoscopy demand and harms if thresholds were lowered for women.
- Evaluation of sensitivity over repeated rounds of annual FIT.
The fecal immunochemical test (FIT) is the most widely used stool-based screening test for colorectal cancer worldwide, and its performance is governed by a single number: the fecal hemoglobin concentration above which a result is called positive. Most programs apply one threshold to everyone. Women, however, have long been reported to have lower FIT sensitivity than men, which raises the question of whether a uniform cutoff provides equal protection for both sexes. If sensitivity differs systematically, a single threshold could leave more cancers undetected in women.
Explanations proposed for lower sensitivity in women include differences in tumor location, with right-sided lesions that bleed less detectably, lower circulating hemoglobin levels and differences in transit time. Testing these explanations and estimating sex-specific thresholds requires large datasets in which every participant has both a quantitative FIT result and a colonoscopy reference standard. The DeeP-C and BLUE-C studies, which validated the original and next-generation multitarget stool DNA tests, provide such data because each participant also completed the same comparator FIT.
Imperiale and colleagues at Indiana University School of Medicine performed a retrospective analysis of cross-sectional data from both studies, which used the same independent quantitative FIT with a positivity threshold of 100 ng/mL, roughly 20 µg of hemoglobin per gram of feces. The combined population comprised 30,165 individuals with a mean age of 63.4 years, of whom 46.6% were men. Colonoscopy identified 163 colorectal cancers and 2,901 advanced precancerous lesions (APL). The investigators compared fecal hemoglobin concentrations and FIT sensitivity by sex and modeled the thresholds needed to achieve 95% specificity in each sex.
Fecal hemoglobin was higher in men than in women both among patients with colorectal cancer (1,541 vs 955 ng/mL; P = .03) and among those with APL (230 vs 174 ng/mL; P < .001). At the standard 100 ng/mL threshold, FIT sensitivity for colorectal cancer was 77.4% in men and 62% in women (P = .03), and sensitivity for APL was 25.8% versus 20.2% (P < .001). To achieve 95% specificity for cancer, the threshold would be 120 ng/mL in men and 94 ng/mL in women.
Distal colon and rectal cancers and APL were more common in men, whereas the size distribution of lesions did not differ by sex. In multivariable analysis, factors independently associated with FIT sensitivity for advanced neoplasia were male sex (OR 1.22; 95% CI 1.02-1.47), current or former smoking (OR 1.36; 95% CI 1.14-1.63), BMI of 30 or more (OR 1.39; 95% CI 1.17-1.67) and distal or rectal location (OR 2.41; 95% CI 2.01-2.89). Male sex therefore remained associated with sensitivity after accounting for lesion location, suggesting that location alone does not fully explain the gap.
The researchers acknowledged that the analysis used a single FIT brand, a single threshold and a single application of the test. Programmatic screening relies on repeated annual or biennial FIT, and cumulative sensitivity over several rounds may narrow sex differences. The populations were enrolled in validation trials of a commercial stool DNA test, which may not reflect program participants, and the presenter reports research support from Exact Sciences. The coverage did not report confidence intervals for the sensitivity estimates or specificity at the 100 ng/mL threshold, which limits assessment of precision.
The results add to evidence that a sex-neutral FIT threshold does not give equal cancer sensitivity. Imperiale noted that to match the cancer sensitivity in men, the threshold in women would have to be lowered, and that the exact value would depend on the test used. Lower thresholds for women would increase positivity and colonoscopy demand, so programs considering sex-specific cutoffs will need assay-specific validation and modeling of capacity and harms. In the meantime, the findings reinforce the importance of adherence to repeated FIT rounds, particularly for women.
Source: Imperiale et al., retrospective analysis of the DeeP-C and BLUE-C studies, presented at the ACG 2026 Annual Scientific Meeting, Nashville, October 2026; not yet published in a peer-reviewed journal. Conference coverage (Healio)
4. Multilevel mailed FIT outreach versus usual care for colorectal cancer screening in Brooklyn community health centers (randomized trial, ACG 2026)
Presenting author Aasma Shaukat (NYU Grossman School of Medicine), on behalf of the investigators
Key points
- Randomized trial at five federally qualified health centers (FQHCs) in Brooklyn of 2,007 adults aged 45-75 (FIT outreach n = 891; usual care n = 1,116); median age 57.5 years, 66.5% women, 69.4% Black, 40.7% on Medicaid.
- Intervention: explanatory letter with FIT kit and instructions, a reminder call or voice message, a follow-up letter if needed, and vouchers for bowel preparation and transportation after an abnormal FIT.
- Primary outcome, FIT completion within 6 months: 35.9% vs 23.6% (P = .001), an absolute increase of 12 percentage points.
- Ages 45-49: 31.9% vs 20.1% (P = .02). Colonoscopy completion after abnormal FIT: 92% vs 54% (P = .03).
- Presented at the ACG 2026 Annual Scientific Meeting (Nashville, October 2026); not yet published in a peer-reviewed journal. Distinct from the San Diego cluster-randomized mailed FIT trial published later in October.
Key limitations
- Confidence intervals for the main comparisons and the size of the abnormal-FIT subgroup were not reported.
- Single urban setting of Brooklyn FQHCs, which may limit generalizability.
- Six-month outcome; sustained participation over repeated screening rounds was not reported.
Future directions
- Peer-reviewed publication with full methods, confidence intervals and the planned cost analysis.
- Testing of approaches that address unemployment and other social determinants associated with lower uptake.
- Longer follow-up of repeat screening and diagnostic completion.
Colorectal cancer screening uptake remains lowest among low-income, uninsured and Medicaid-insured adults, many of whom receive care at federally qualified health centers (FQHCs). Black Americans bear a disproportionate burden of colorectal cancer, and the extension of screening recommendations to adults aged 45-49 added a younger group with historically low participation. Clinic-based screening depends on patients attending visits, which is a substantial barrier for people with unstable employment, transport problems or limited time off work.
Mailed FIT outreach, in which a stool test kit is sent directly to the patient's home, is an established way to raise screening rates, but the size of its effect varies across health systems and populations. Two weaknesses recur: a substantial proportion of recipients never return the kit, and patients with an abnormal FIT often fail to complete diagnostic colonoscopy. Shaukat and colleagues at NYU Grossman School of Medicine designed a multilevel intervention intended to address both steps for underserved adults in New York City.
The randomized trial enrolled 2,007 adults aged 45-75 at five FQHCs in Brooklyn, with 891 assigned to FIT outreach and 1,116 to usual care. Participants had a median age of 57.5 years; 66.5% were women, 69.4% were Black, 40.4% were unemployed and 40.7% were insured through Medicaid. The intervention combined an explanatory letter with FIT instructions, a reminder phone call or voice message, a follow-up letter when needed and, for patients with an abnormal FIT, vouchers covering colonoscopy preparation and transportation. The primary outcome was FIT completion within 6 months.
FIT completion within 6 months was 35.9% in the outreach group versus 23.6% with usual care (P = .001), an absolute increase of 12 percentage points. The effect was similar among adults aged 45-49, the group newly eligible for screening, in whom completion was 31.9% versus 20.1% (P = .02). Shaukat emphasized the practical lesson that clinicians cannot simply wait for patients to come to clinic but need to reach out to them.
Among patients with an abnormal FIT, colonoscopy completion was 92% in the outreach group versus 54% with usual care (P = .03), suggesting that the vouchers and navigation-style supports addressed a major point of loss. In adjusted analyses, unemployment was associated with lower odds of screening (OR 0.77; 95% CI 0.63-0.95) and older age with modestly higher odds (OR 1.024; 95% CI 1.008-1.039). The odds ratio reported for Black race was 0.708 (95% CI 0.528-0.944), although the accompanying coverage described this association inconsistently, so it should be interpreted cautiously until the full report is available.
The findings come from a conference presentation, and the full methods, randomization details and confidence intervals for the primary comparisons were not available. The number of patients with an abnormal FIT was not reported, so the large difference in diagnostic colonoscopy rests on an unknown and probably small denominator. Projected costs and qualitative barriers were examined but not reported in detail. The population was restricted to Brooklyn FQHCs, and the 6-month outcome does not show whether participation is sustained over repeated annual rounds. The presenter reports consulting relationships with several screening companies.
Even with outreach, almost two-thirds of participants did not complete FIT, a reminder that mailed outreach raises but does not solve the problem of low screening in underserved communities. The intervention's apparent effect on diagnostic colonoscopy after an abnormal FIT may be its most important element, because an unresolved positive test offers little protection. The association between unemployment and lower uptake points to social determinants that kit mailing alone cannot address. Peer-reviewed publication, cost results and longer follow-up will determine whether this multilevel model can be scaled across safety-net systems.
Source: Shaukat et al., randomized trial of multilevel mailed FIT outreach, presented at the ACG 2026 Annual Scientific Meeting, Nashville, October 2026; not yet published in a peer-reviewed journal. Conference coverage (Healio)
5. Liver transplantation for colorectal liver metastases in the United States: transplant registry analysis and meta-analysis (ACG 2026)
Presenting author Ragesh B. Thandassery (University of Arkansas for Medical Sciences), on behalf of the investigators
Key points
- Analysis of 203 adults in the Scientific Registry of Transplant Recipients who underwent liver transplantation for colorectal liver metastases (CRLM) from 2017 to 2025, plus a meta-analysis of published studies.
- Recipients were highly selected: mean age 49.6 years, 64% men, median MELD score 8, and 94.1% with Karnofsky performance status above 70.
- Overall survival after transplant was 90.9% at 1 year, 69.8% at 3 years and 48.9% at 5 years.
- In the meta-analysis, pooled 5-year overall survival was 57% (95% CI 46%-67.3%). Of the 16.3% of recipients who died, 45.5% of deaths were malignancy-related and 3% were due to graft failure.
- Presented at the ACG 2026 Annual Scientific Meeting (Nashville, October 2026); not yet published in a peer-reviewed journal.
Key limitations
- Registry analysis without a comparison group receiving chemotherapy alone or other locoregional therapy.
- Highly selected recipients with excellent performance status; results cannot be generalized to all patients with unresectable CRLM.
- Registry data lack detail on tumor biology, selection criteria and recurrence patterns.
Future directions
- Standardized US selection criteria and prospective registries for transplant in CRLM.
- Comparison with evidence from TRANSMET and with chemotherapy-alone outcomes.
- Strategies to expand graft supply, including living donation, without disadvantaging other waitlisted patients.
The liver is the most common site of metastasis from colorectal cancer, and most patients with colorectal liver metastases (CRLM) have disease that cannot be removed by partial hepatectomy. For these patients, systemic therapy is usually palliative. Liver transplantation offers a way to remove all intrahepatic disease when resection is impossible, and it has moved from a contraindicated indication toward selective use in carefully chosen patients with liver-only metastases that have responded to chemotherapy.
The renewed interest stems largely from European experience, including the TRANSMET trial of chemotherapy plus liver transplantation in unresectable CRLM. In the United States, however, transplant for CRLM has remained uncommon and practice has been concentrated in a small number of centers, and national outcome data have been limited. Thandassery and colleagues at the University of Arkansas for Medical Sciences examined how the indication is being used in the US and how outcomes compare with international experience.
The investigators analyzed data from the Scientific Registry of Transplant Recipients on adults who received liver transplants for CRLM from 2017 to 2025, with follow-up through March, and compared survival trends with international data. They also performed a meta-analysis of published studies of liver transplantation for CRLM. The US cohort comprised 203 recipients with a mean age of 49.6 years; 64% were men, 82.3% were White and 76.8% had private insurance. The median MELD score was 8, and 94.1% had a Karnofsky performance status above 70.
Overall survival after transplantation was 90.9% at 1 year, 69.8% at 3 years and 48.9% at 5 years. In the meta-analysis, pooled 5-year overall survival was 57% (95% CI 46%-67.3%). By the end of follow-up, 16.3% of recipients had died; 45.5% of those deaths were malignancy-related and 3% were attributed to graft failure, indicating that recurrent cancer rather than transplant complications drives mortality.
The transplant course reflected substantial use of nonstandard grafts: 41.4% of recipients received organs from living donors, 26.6% from extended-criteria donors and 22.2% from donors after cardiac death. Median waitlist time was 118 days and median hospital stay 8 days. The high proportion of living donation is notable because the use of deceased-donor livers for an oncologic indication competes with other waitlisted patients, so living donation may be one way to extend access without increasing pressure on the deceased-donor pool.
Thandassery stressed that this was a highly selected population with excellent functional status and low MELD scores, and that the favorable outcomes should not be generalized to all patients with CRLM. The registry analysis has no comparator receiving chemotherapy alone, and it lacks granular data on tumor biology, response to chemotherapy, selection criteria and patterns of recurrence. The coverage did not describe the meta-analysis methods, the number of included studies or heterogeneity, and the results have not been peer reviewed. Survival estimates at 5 years rest on the subset of patients transplanted early in the study period.
The data suggest that, in carefully selected patients, US outcomes are broadly in line with international experience, with roughly half of recipients alive at 5 years. Thandassery described the indication as emerging but highly selective in the United States. Wider adoption will require agreed selection criteria, multidisciplinary assessment of tumor biology and response, prospective outcome tracking and careful attention to organ allocation. For oncologists, the practical message is that patients with chemotherapy-responsive, liver-only, unresectable disease may warrant early referral to a transplant center with experience in this indication.
Source: Thandassery et al., presented at the ACG 2026 Annual Scientific Meeting, Nashville, October 2026; not yet published in a peer-reviewed journal. Conference coverage (Healio)
6. The implementation gap in aspirin chemoprevention for Lynch syndrome: the LYNX survey (The Lancet Gastroenterology & Hepatology)
Anne-Sophie Layritz (University Hospital Bonn) and Nicola Reents (SemiColon self-help organization), co-first authors; corresponding author Robert Hüneburg (University Hospital Bonn), with colleagues from the Center for Integrated Oncology Bonn and the German Consortium for Familial Colorectal Cancer
Key points
- Nationwide survey (LYNX) of 539 people with Lynch syndrome in Germany, conducted December 2025 to February 2026 and published as a correspondence on October 8, 2026.
- About 97% knew the specific colonoscopy surveillance recommendations, and nearly all reported regular colonoscopies.
- Only 10.4% reported currently taking aspirin, and just 6.9% took it specifically for cancer prevention.
- Reasons for non-use included concern about side effects and uncertainty about the evidence.
- About three-quarters were open in principle to a preventive mRNA cancer vaccine study (about 80% after more information), and more than 92% would want such a vaccine if approved.
Key limitations
- Self-reported survey data from Germany only.
- Fielded before publication of the CaPP3 dose results, so it does not capture the effect of newer evidence on use.
- Published as a letter, with limited methodological detail.
Future directions
- Studies of whether lower-dose (100 mg) aspirin increases uptake.
- Integration of chemoprevention counseling into Lynch syndrome surveillance clinics.
- Engagement of patient organizations in recruitment for cancer vaccine trials.
Lynch syndrome is the most common inherited cause of colorectal cancer, caused by germline variants in DNA mismatch repair genes. Carriers face a high lifetime risk of colorectal and other cancers, and management relies on regular colonoscopic surveillance. Surveillance reduces mortality but does not eliminate interval cancers, which has driven interest in chemoprevention. Aspirin is the leading candidate, because trial evidence in Lynch syndrome suggests it lowers long-term colorectal cancer risk.
The CAPP2 trial showed that 600 mg of aspirin daily reduced long-term colorectal cancer risk in Lynch syndrome carriers, and the subsequent CaPP3 study suggested that 100 mg daily may offer a similar preventive profile with fewer side effects and bleeding complications, although the comparison did not meet formal equivalence criteria in all analyses. How often carriers actually take it has been unclear. Investigators at University Hospital Bonn set out to measure uptake and attitudes in a national sample.
The LYNX study surveyed 539 people with Lynch syndrome across Germany between December 2025 and February 2026, with involvement of the SemiColon self-help organization, the Center for Integrated Oncology Bonn and the German Consortium for Familial Colorectal Cancer. The questionnaire covered knowledge of surveillance recommendations, colonoscopy attendance, aspirin use and its purpose, reasons for non-use, and willingness to take part in a study of a preventive mRNA cancer vaccine. The survey was designed and fielded before the CaPP3 results were published, so responses reflect the evidence available at that time.
Awareness of surveillance was high: about 97% of respondents knew the specific screening recommendations for Lynch syndrome, and nearly all reported having regular colonoscopies. Aspirin use, by contrast, was rare. Only 10.4% reported currently taking aspirin, and only 6.9% took it specifically to prevent cancer. Co-first author Anne-Sophie Layritz reported that the main reasons for non-use were concerns about side effects and uncertainty about the scientific evidence.
Attitudes toward a newer prevention strategy were strikingly positive. About three-quarters of respondents were open in principle to joining a study of a preventive mRNA cancer vaccine, rising to about 80% after they received more information, and a majority accepted the possibility of placebo assignment and the risk of side effects. More than 92% said they would want such a vaccine if it were approved. The contrast between high willingness to try an investigational vaccine and low uptake of an inexpensive oral drug with trial evidence underlines that the barrier to aspirin lies in perceived risk and the clarity of recommendations rather than in a general reluctance to take preventive measures.
The study has clear limitations. It is a self-reported survey limited to Germany, and respondents recruited with the help of a patient organization may be more engaged than the wider carrier population, which would, if anything, overestimate uptake. The survey predates CaPP3 publication, so it cannot show whether evidence for a lower, better-tolerated dose changes behavior. The work is published as correspondence, with limited methodological detail, and corresponding author Robert Hüneburg cautioned that the lower-dose results should still be interpreted with some caution.
The findings reveal a substantial implementation gap between evidence and practice in hereditary colorectal cancer prevention. Hüneburg suggested that 100 mg daily could be a practical and better-tolerated option for many patients, and the authors call for studies of whether lower doses increase uptake. Practical steps include routine discussion of aspirin at surveillance visits, clearer guideline language on dose and duration, and shared decision-making about bleeding risk. The strong interest in vaccine trials also suggests that Lynch syndrome carriers are a motivated population for prevention research.
Source: Layritz, Reents, Hüneburg et al., "The implementation gap in aspirin chemoprevention for Lynch syndrome," The Lancet Gastroenterology & Hepatology (correspondence), October 8, 2026. Read the paper · Press release
7. Random forest machine learning versus a large language model for automated detection of colorectal neoplasia in 5,000 colonoscopy reports (ACG 2026)
Presenting author Vignesh Krishnan Nagesh (Aiken Regional Medical Centers; research collaborator with Mayo Clinic Florida), on behalf of the investigators
Key points
- Retrospective comparison of two previously validated models, a random forest model and a large language model (LLM), for automated adenoma detection rate (ADR) reporting in 5,000 colonoscopy reports from 2023 to 2025.
- The models agreed in 97.5% of cases; the 123 discordant cases (2.5%) were reviewed manually.
- Among discordant cases, the random forest model missed 2 of 16 tubular adenomas and the LLM missed 7; both missed 4 of 47 sessile serrated lesions.
- The random forest model produced many more false positives: 92 versus 8 for tubular adenomas, and 59 versus 8 for mixed lesions.
- Presented at the ACG 2026 Annual Scientific Meeting (Nashville, October 2026); not yet published in a peer-reviewed journal.
Key limitations
- Retrospective design, with performance reported as counts in discordant cases rather than sensitivity, specificity or AUC with confidence intervals.
- Small numbers for some lesion categories (for example, 16 tubular adenomas and 7 mixed cases among discordant reports).
- Single dataset; generalizability across health systems and over time is unknown.
Future directions
- Prospective, multisystem studies of automated ADR reporting.
- Evaluation of combined or hybrid approaches rather than a single model.
- Cost and implementation analyses, given the higher cost of LLMs.
The adenoma detection rate (ADR), the proportion of screening colonoscopies in which at least one adenoma is found, is the principal quality indicator for colonoscopy, and higher ADR is associated with lower risk of post-colonoscopy colorectal cancer. Calculating ADR requires linking each endoscopy report to its pathology result, a task that many units still perform manually. Manual abstraction is labor-intensive and slow, which limits how often endoscopists receive feedback on their performance.
Natural language processing has been used to automate ADR measurement, and newer tools include classic machine learning approaches, such as random forest classifiers, and general-purpose large language models (LLMs). Both families of models can extract findings from free-text reports, but they differ in how they err, how they are maintained and what they cost to run. Nagesh and colleagues, working with Mayo Clinic, compared two models that had each been validated previously, with reported detection rates of at least 99% for adenomas and serrated lesions.
The investigators applied both models retrospectively to 5,000 colonoscopy reports from 2023 to 2025 and identified cases in which the models disagreed. These discordant reports were reviewed manually to determine which model was correct. Outcomes included missed lesions (false negatives) and falsely identified lesions (false positives) for tubular adenomas, sessile serrated lesions and cases containing both lesion types. Because concordant reports were not reported as separately audited, the comparison characterizes the models' differences rather than their absolute accuracy.
The two models agreed in 97.5% of cases, leaving 123 discordant reports (2.5%). Among discordant cases, the random forest model missed 2 of 16 tubular adenomas, whereas the LLM missed 7. For sessile serrated lesions, both models missed 4 of 47. In the 7 discordant cases with both lesion types, the random forest model missed 2 and the LLM omitted 5. The random forest model therefore showed higher sensitivity in the cases where the two disagreed.
That sensitivity came at the cost of many more false positives. For tubular adenomas, the random forest model falsely identified 92 of 107 cases, compared with 8 for the LLM. False positives for sessile serrated lesions were uncommon with both (2 versus 1 among 76 cases), but for mixed-lesion cases the random forest model produced 59 false positives versus 8 for the LLM among 67 cases. If false positives were counted as adenomas, the random forest approach could overestimate ADR, while the LLM carries higher implementation cost.
The study has important limitations. It is retrospective and appears to draw on a single dataset, and the coverage did not specify whether reports came from one institution. Results were presented as raw counts in discordant cases rather than as sensitivity, specificity or area under the curve with confidence intervals, and several lesion categories involved very small numbers. Because only discordant cases were adjudicated, errors shared by both models would not be detected. The findings have not yet been peer reviewed.
Nagesh concluded that automated approaches may offer an efficient way to facilitate ADR reporting in clinical practice, but that additional studies are needed before choosing the optimal model. The trade-off observed here, higher sensitivity with more false positives for the random forest and greater specificity at higher cost for the LLM, suggests that a combined approach, or a model that flags uncertain cases for human review, may be most reliable. Prospective studies across health systems and over time are needed before automated ADR reporting can be relied on for quality benchmarking.
Source: Nagesh et al., "Comparison of random forest approach machine learning and large language models for automated detection of colorectal neoplasia: A retrospective study of 5,000 colonoscopies," presented at the ACG 2026 Annual Scientific Meeting, Nashville, October 2026; not yet published in a peer-reviewed journal. Conference coverage (Healio)