ESMO 2025: Advancing precision in adjuvant treatment of colorectal cancer

Although the ESMO Congress 2025 in Berlin did not present practice-changing studies in colorectal cancer (CRC), several findings offered valuable insights.

Standard management of high-risk stage II (pT4) and stage III (pN+) colon cancer involves surgical resection followed by adjuvant chemotherapy, which typically consists of a fluoropyrimidine such as 5-fluorouracil or capecitabine, administered either alone or in combination with oxaliplatin for three to six months.

However, nearly half of the patients may be cured with surgery alone and could therefore avoid adjuvant chemotherapy, whereas a substantial proportion will relapse despite receiving chemotherapy. Recent efforts to improve risk stratification, including the DYNAMIC-III trial using circulating tumor DNA (ctDNA) and approaches integrating artificial intelligence with digital pathology, highlight ongoing strategies to individualize adjuvant therapy in CRC.

DYNAMIC-III Trial: ctDNA-guided adjuvant therapy in stage III CRC

ctDNA is a strong prognostic marker for relapse in colorectal cancer.¹ The DYNAMIC trial, published in 2022 with recent updates, randomized stage II colon cancer patients to ctDNA-guided adjuvant chemotherapy or standard management. Only ctDNA-positive patients received chemotherapy, reducing treatment rates from 28% to 15% without compromising 2-year recurrence-free survival (RFS). At a median follow-up of 60 months, 5-year RFS was similar between groups (88% vs. 87%, 95% CI: −5.8% to 8.0%). The trial met its primary endpoint, supporting ctDNA-guided therapy as a strategy to safely limit overtreatment in stage II CRC.²

This year, from the same group, the ctDNA-negative arm of the DYNAMIC-III trial was presented at ESMO, and the ctDNA-positive arm at ASCO.³ This multicenter, randomized, phase 2/3 study enrolled 968 patients with stage III colon cancer who underwent ctDNA testing 5–6 weeks post surgery and were randomized to ctDNA-guided or standard management. A tumor-informed personalized approach for ctDNA analysis was used (SaferSeqS-targeted CRC panel). In the ctDNA-guided arm, ctDNA-negative patients received deescalated therapy, whereas ctDNA-posi- tive patients received intensified therapy. Primary endpoints were 3-year RFS for ctDNA-negative patients and 2-year RFS for ctDNA-positive patients, with secondary endpoints including hospitalization and ctDNA clearance.

Among 702 ctDNA-negative patients (72.5%), deescalation reduced oxaliplatin use (34.8% vs. 88.6%) and hospitalizations (8.5% vs. 13.2%), with slightly lower 3-year RFS (85.3% vs. 88.1%, 95% CI: −8% to +2.5%), though non-inferiority criteria were not met (non-inferiority was defined as the lower bound of the one-sided 97.5% CI for the difference at 3 years not crossing −7.5%). In ctDNA-positive patients, higher ctDNA levels were associated with increased recurrence risk (3-year RFS: 77% to 23%; p<0.001). Escalated therapy did not improve outcomes (2-year RFS: 51% vs. 61%), and persistent ctDNA after treatment predicted markedly worse prognosis (3-year RFS: 14% vs. 79%).

Overall, these findings confirm ctDNA as a strong prognostic tool in colon cancer; however, the trial was formally negative for the primary endpoints. Considering that the non-inferiority design is the appropriate approach to assess these questions, and that the subgroup analysis did not substantially contribute to hypothesis generation, the study leaves open questions regarding clinician-driven therapy selection, the choice of deescalation strategies, and whether the SaferSeqS-targeted CRC panel used was the most optimal and precise assay.

High-risk stage II–III colon cancer without chemotherapy: CAPAI

In the Netherlands, the Combined Analysis of Pathologists and Artificial Intelligence (CAPAI) was developed to refine prognostic assessment and guide ACT in high-risk stage II and III colon cancer. CAPAI combines histopathologic features from H&E slides, analyzed using the DoMore-v1-CRC deep learning biomarker, with tumor stage (pT/pN) and lymph node count to classify patients into low-, intermediate-, or high-risk groups. DoMore-v1-CRC was previously optimized and validated in large European cohorts, demonstrating superior prognostic performance compared with traditional morphological and molecular markers, stratifying patients into good, uncertain, or poor prognosis categories with 3-year cancer-specific survival (CSS) up to 97% in low-risk patients and hazard ratios exceeding 10 for high-versus low-risk groups.⁴

Leveraging the Netherlands’ restrictive ACT practices, a nationwide cohort of 453 patients under 70 years, with good performance status and R0 resection, who did not receive neoadjuvant or adjuvant therapy, was analyzed. CAPAI effectively stratified patients, identifying nearly half as low-risk (3-year CSS: 93.7%), 34% as intermediate-risk (87.5%), and 18% as high-risk (60.4%), with statistically significant differences (log-rank p<0.001). These findings suggest that integrating DoMore-v1-CRC with conventional staging enables precise identification of patients who may safely avoid ACT while highlighting those who might benefit from treatment intensification.⁵

Outlook

Taken together, these findings underscore the need to reassess clinical trial designs and refine diagnostic and prognostic tools, with the ultimate goal of delivering adjuvant therapy only to patients most likely to benefit. The rapidly evolving landscape of emerging biomarkers presents an opportunity for coordinated evaluation.

For ctDNA, the proliferation of assays in clinical investigation brings several key questions to the forefront, including whether systematic cross-assay comparisons are needed to establish analytical validity and clinical utility, and whether sequential measurements can provide additional insight into residual disease and relapse risk. Concurrently, advances in artificial intelligence applied to digital pathology suggest that AI-derived biomarkers may play an important role, particularly when integrated with established molecular and clinical markers.

Incorporating these complementary, question-oriented approaches into future trials holds strong promise for improving patient stratification and advancing toward true precision in adjuvant colorectal cancer therapy.