Real-world analysis of ctDNA and other biomarkers in patients with curatively resected stage I-III biliary tract cancer Authors: M Malla, A Esmail, A Tin, MM Khayat, L Bathla, C Bridges, CM Brewer, J Ferguson, JB Ortiz, J Ensor, K Magee, J Yu, AR He, T Cannon, A Winer, GG King, A Mahipal, A Nagarajan, JW Franses, A Fungtammasan, A Jurdi, MC Liu, S Kodali, AA Connor, RM Ghobrial, RD Kim, M Abdelrahim Published: 2026 ABSTRACT AbstractBackgroundGrowing evidence supports the prognostic and predictive value of circulating tumor DNA (ctDNA) in gastrointestinal cancers. This study evaluated ctDNA as a prognostic biomarker for detecting molecular residual disease (MRD) and recurrence in patients with resected biliary tract cancer (BTC).Materials and methodsWe retrospectively analyzed real-world data from patients (N = 167) with stage I-III resectable BTC who underwent ctDNA analysis using a personalized, tumor-informed 16-plex multiplex PCR-coupled next-generation sequencing assay (SignateraTM; Natera, Inc., Austin, TX) between July 2020 and February 2024. Plasma samples (n = 751) were collected preoperatively, postsurgically (2-12 weeks; MRD window), and longitudinally (postdefinitive treatment surveillance). ctDNA results were compared with traditional biomarkers carbohydrate antigen 19-9 and carcinoembryonic antigen.ResultsThe median follow-up was 21 months (range 2-97). ctDNA detection rates during MRD and postdefinitive treatment surveillance windows were 23% (19 of 82) and 38% (31 of 82), respectively. ctDNA positivity during MRD and postdefinitive treatment surveillance was significantly associated with inferior relapse-free survival (RFS) and overall survival (OS). ctDNA positivity was the most significant prognostic factor associated with RFS during the MRD window [hazard ratio (HR) 15.86, 95% CI 4.69-53.6, P < 0.001] and during postdefinitive treatment surveillance (HR 14.93, 95% CI 5.33-41.9, P < 0.001) by multivariate analysis. In contrast, carbohydrate antigen 19-9 and carcinoembryonic antigen were not significantly prognostic in either setting.ConclusionsPatients with ctDNA positivity during the MRD and postdefinitive treatment surveillance windows demonstrated significantly inferior RFS and OS, and ctDNA was more accurate than current clinical biomarkers. These findings highlight the value of ctDNA monitoring in improving prognostication in BTC, with the potential to enhance post-operative risk stratification and inform surveillance strategies through earlier detection of recurrence.Key words: biliary tract cancers, biomarkers, molecular residual disease, recurrence monitoring Introduction Biliary tract cancer (BTC) is a rare and highly aggressive malignancy, encompassing intrahepatic and extrahepatic cholangiocarcinoma, gallbladder carcinoma, and ampullary cancer.1 Despite curative-intent surgery and the integration of adjuvant chemotherapy, long-term outcomes remain poor, with recurrence rates exceeding 60% across stages I-III disease.2, 3, 4 Outcomes after radiologic relapse remain poor, underscoring the need for more precise tools to stratify recurrence risk and enable timely therapeutic intervention. Currently, post-operative surveillance strategies rely on cross-sectional imaging and serum biomarkers such as carbohydrate antigen 19-9 (CA19-9) and carcinoembryonic antigen (CEA).5,6 However, these modalities suffer from suboptimal sensitivity and specificity, often leading to delayed or ambiguous detection of recurrence. In contrast, circulating tumor DNA (ctDNA) has emerged as a transformative biomarker in several solid tumors,7, 8, 9, 10, 11, 12 offering a noninvasive approach to detect molecular residual disease (MRD) and predict relapse months before radiographic or clinical progression. In colorectal and gastroesophageal cancers, longitudinal ctDNA monitoring has demonstrated high sensitivity and prognostic value for recurrence, informing risk-adapted strategies in both adjuvant and surveillance settings.11,13,14 Yet, the prognostic value of ctDNA in BTC remains unclear, with limited data evaluating its utility. However, early studies suggest that ctDNA may outperform conventional markers like CA19-9 in predicting recurrence and guiding follow-up.15,16 Given the aggressive nature of BTC and the unmet clinical need for reliable recurrence monitoring, this study investigated the role of ctDNA as a prognostic biomarker of relapse in patients with resected BTC. We present data from a retrospective cohort evaluating the association between ctDNA dynamics and recurrence risk and explore the potential of ctDNA-guided surveillance to enhance early detection of relapse beyond conventional imaging and clinical assessment. Materials and methods Subjects and study design This study retrospectively analyzed real-world data from prospectively collected longitudinal plasma samples from a multi-institutional cohort of 223 patients with stage I-III resectable BTC treated between July 2020 and February 2024. After excluding 56 patients due to either stage IV disease at diagnosis (n = 7), unresectable tumors (n = 4), <2 months of follow-up (n = 40), or missing ctDNA data (n = 5), 167 patients were assessable and included in the final analysis (Figure 1). For all patients, clinical ctDNA testing on peripheral blood samples (two 10 mL Streck tubes) was carried out at the discretion of the ordering physician, with an average of 4.5 samples per patient (range 1-18). The median time between two consecutive ctDNA tests was 90.5 days (range 29-643), and the median frequency of radiographic imaging was 97 days (range 0-357). The study was conducted in compliance with Natera protocol 21-058, the Declaration of Helsinki, Title 21 of the United States Code of Federal Regulations (CFR) as applicable, Good Clinical Practice guidelines, and International Conference on Harmonization guidelines. A waiver of the consent process and of the requirement for documentation of informed consent was granted according to 45 CFR 46.116(d) and 45 CFR 46.117 (c)(2), respectively. Flow diagram of patients included in the study and ctDNA detection rate by BTC subtype.BTC, biliary tract cancer; ctDNA, circulating tumor DNA; MRD, molecular residual disease. Personalized, tumor-informed ctDNA assay All biological specimens were processed following a Clinical Laboratory Improvement Amendments-validated standard operating procedure at Natera, Inc. (Austin, USA). ctDNA analysis was carried out using a clinically validated, personalized, and tumor-informed 16-plex multiplex PCR (mPCR) assay (SignateraTM; Natera, Inc.) in patients undergoing commercial testing as previously described.17 Briefly, whole-exome sequencing (WES) was carried out on formalin-fixed, paraffin-embedded tumor tissue, along with matched normal DNA blood samples from each patient. Based on the WES results, mPCR primers were designed for 16 high-ranked, patient-specific, somatic single-nucleotide variants (SNVs) for each patient. The mPCR primers targeting the personalized SNVs were used to track ctDNA in the respective patients’ plasma cell-free DNA (cfDNA). Plasma cfDNA samples with two or more SNVs detected above a predefined algorithm’s confidence threshold were considered ctDNA-positive. ctDNA concentration (levels) was quantitatively reported as mean tumor molecules per milliliter of plasma. Statistical analysis The primary objective was to assess relapse-free survival (RFS), measured between the date of surgical resection and the date of radiologic/clinical disease progression or death by any cause. Overall survival (OS) and RFS were estimated using the Kaplan–Meier method, and comparisons were accomplished using log-rank tests (R version 4.3.1, R Foundation for Statistical Computing, Vienna, Austria). For ctDNA analysis, the MRD window was defined as the ctDNA time points within 2-12 weeks after surgery. The postdefinitive treatment surveillance window consisted of post-MRD time points if no adjuvant therapy was given or 2 weeks post-adjuvant chemotherapy (ACT), until the last follow-up or recurrence. For survival analysis, any ctDNA measurements after clinical recurrence were excluded. To account for immortal time bias, patients who were alive and event-free until at least 12 weeks after surgery were included in the survival analysis, thus excluding patients who were effectively stage IV at the time of surgery. A multivariable Cox proportional hazards model was used to assess the most significant prognostic factor associated with RFS. All P values were based on two-sided testing; differences were considered significant at P ≤ 0.05. The Cox proportional hazards model was used with a time-dependent covariate, applying Firth’s penalized partial likelihood to address separation due to rare events. The model assumed that no intrasubject correlation was used for serial ctDNA analysis during the postdefinitive treatment surveillance window. Results Patient cohort A total of 223 patients with BTC were included in the analysis. After excluding 56 patients due to stage IV disease at diagnosis, unresectable tumors, <2 months of follow-up, or missing ctDNA data, 167 patients were assessable and included in the final analysis (Figure 1). Of these, 53 patients (31.7%) had been included in a previously published smaller cohort,16 representing continued follow-up and expanded multicenter validation in the present study. When assessing BTC subtypes, the cohort included 74 (44.3%) patients with intrahepatic cholangiocarcinoma (iCCA), 35 (21.0%) patients with extrahepatic cholangiocarcinoma (eCCA), 30 (18.0%) patients with gallbladder cancer, and 28 (16.8%) patients with ampullary carcinoma. Patient and tumor characteristics are shown in Table 1. The clinical course, including ctDNA analysis for each patient, is depicted in Supplementary Figure S1, available at https://doi.org/10.1016/j.esmogo.2026.100344. Table 1Patient demographic and tumor characteristicsCharacteristicN = 167CharacteristicN = 167Age (years)68 (27-92)MSI statusSexMSI-High3 (1.8) Male92 (55.1)MSS147 (88.0) Female75 (44.9)Not specified17 (10.2)SubtypeAdjuvant therapy Intrahepatic74 (44.3)Received115 (68.9) Extrahepatic35 (21.0)Omitted52 (31.1) Gallbladder30 (17.9)Lymphovascular invasion Ampullary28 (16.8)No81 (48.5)StageYes74 (44.3) I35 (21.0)Not specified12 (7.2) II64 (38.3)Perineural invasion III67 (40.1)No78 (46.7) Not specified1 (0.6)Yes69 (41.3)MarginsNot specified20 (12.0) R0136 (81.4)Relapse status R124 (14.4)No relapse97 (58.1) Not specified7 (4.2)Relapse70 (41.9)GradeVital status G112 (7.2)Alive149 (89.2) G2103 (61.7)Deceased18 (10.8) G349 (29.3)RFS (months)11 (0-97) Not specified3 (1.8)Follow-up (months)21 (2-97)Values are presented as n (%) or median (range) unless otherwise indicated.MSI, microsatellite instability; MSS, microsatellite stable; RFS, relapse-free survival. Patient demographic and tumor characteristics Values are presented as n (%) or median (range) unless otherwise indicated. MSI, microsatellite instability; MSS, microsatellite stable; RFS, relapse-free survival. ctDNA detection rates by time point and subtype Among patients with available ctDNA data, ctDNA positivity rates varied by time point and BTC subtype. At baseline, defined as samples collected within 30 days before surgery and up to the day of surgery, ctDNA positivity was observed in 100% (13 of 13) of patients, followed by 23% (19 of 82) at the MRD time point, 25% (15 of 61) at the on-treatment time point, and 38% (31 of 82) during postdefinitive treatment surveillance. Furthermore, subtype-specific ctDNA-positivity rates at the MRD time point were 22% in iCCA, 35% in eCCA, 13% in gallbladder cancer, and 23% in ampullary cancer. ctDNA positivity during postdefinitive treatment surveillance ranged from 27% in ampullary cancer to 50% in gallbladder cancer, highlighting heterogeneity in ctDNA detection across BTC subtypes (Figure 1). Prognostic value of ctDNA positivity during the MRD window and postdefinitive treatment surveillance Of the 167 patients who met the study inclusion criteria, 82 had ctDNA results available within the MRD window (2-12 weeks after surgery), and ctDNA positivity was significantly associated with inferior RFS (HR 9.33, 95% CI 3.40-25.64, P < 0.0001; Figure 2A) and OS (HR 5.29, 95% CI 1.28-25.19, P = 0.02; Figure 2C) compared with ctDNA-negative status. Additionally, 12-month RFS for ctDNA-positive and ctDNA-negative patients during the MRD window was 12.90% (95% CI 2.47% to 67.40%) and 75.30% (95% CI 62.30% to 90.90%), respectively. Furthermore, the 12-month OS for ctDNA-positive patients was 55.70% (95% CI 28.50% to 100%) compared with 95.80% (95% CI 90.10% to 100%) for ctDNA-negative patients. ctDNA status during the MRD window and during postdefinitive treatment surveillance is prognostic of RFS and OS. ctDNA status and RFS and OS in patients with BTC. Kaplan–Meier estimates for RFS (A, B) and OS (C, D) stratified by ctDNA status during the MRD window (A, C) and postdefinitive treatment surveillance (B, D). Forest plots (E, F) depicting the multivariate analysis, including ctDNA status during the MRD window (E) and postdefinitive treatment surveillance (F), and other clinicopathologic features for RFS in patients with BTC. The unadjusted HRs (squares) and 95% CIs (horizontal lines) are shown for each prognostic factor; the vertical dotted line represents the null hypothesis.BTC, biliary tract cancer; CI, confidence interval; ctDNA, circulating tumor DNA; HR, hazard ratio; mOS, median overall survival; MRD, molecular residual disease; mRFS, median relapse-free survival; NR, not reached; OS, overall survival; RFS, relapse-free survival. In the postdefinitive treatment surveillance setting, ctDNA positivity remained strongly associated with poor outcomes. To ensure adequate clinical follow-up in accordance with National Comprehensive Cancer Network surveillance guidelines (imaging every 3-6 months), only ctDNA plasma samples from nonrecurrent patients with ≥6 months of follow-up were included in the analysis, resulting in a sensitivity of 80.6% and specificity of 95.7% for detecting relapse. Compared with patients who were serially ctDNA-negative, patients who tested positive during postdefinitive treatment surveillance were found to have significantly shorter RFS (HR 18.12, 95% CI 8.21-40.01, P < 0.0001), inferior 12-month RFS [32.26% (95% CI 19.37-53.70) versus 91.80% (95%CI 84.30-99.90); Figure 2B] and shorter OS [HR 47.54, 95% CI 2.31-976.40, P < 0.0001; 12-month OS: 85.10% (95% CI 72.60% to 99.80%) versus 98.0% (95% CI 94.20% to 100%); Figure 2D]. Furthermore, to evaluate the prognostic value of ctDNA when compared with other clinicopathologic risk factors, we carried out a multivariate Cox regression model incorporating ctDNA outcomes, BTC cancer subtype, stage, grade, and margin. We found that ctDNA positivity during the MRD window (HR 15.86; 95% CI 4.69-53.60, P < 0.001; Figure 2E) and during postdefinitive treatment surveillance (HR 14.93, 95% CI 5.33-41.90, P < 0.001; Figure 2F) remained the most statistically significant predictor of relapse. Additionally, we found that among the 39 patients who had a clinical recurrence and a ctDNA-positive result before recurrence, the median lead time for ctDNA positivity over radiographic relapse was 3.70 months (range 0.164-18.3). Recurrence site information was available for 24 patients. Among these, distant recurrence predominated in both ctDNA-positive patients (14 of 18, 77.8%) and ctDNA-negative patients (5 of 6, 83.3%), with liver and lymph nodes representing the most common metastatic sites in the ctDNA-positive cohort; local-only recurrence was observed in one ctDNA-negative and three ctDNA-positive patients, and one ctDNA-positive patient developed both local and distant disease (Supplementary Figure S2, available at https://doi.org/10.1016/j.esmogo.2026.100344). Association of ctDNA status and dynamics with ACT administration We next assessed the prognostic value of ctDNA positivity during postdefinitive treatment surveillance in patients who received ACT and in those who did not. In total, 46 patients who received ACT and 36 patients who did not receive ACT had ctDNA time points available during postdefinitive treatment surveillance. In both the ACT and no ACT groups, ctDNA positivity during postdefinitive treatment surveillance was associated with poor RFS compared with serial ctDNA negativity [ACT group: HR 16.40, 95% CI 5.46-49.28, P < 0.0001; clinical relapse rate: 88.23% (15 of 17); 12-month RFS: 41.20% (95% CI 23.30% to 72.70%); Figure 3A; did not receive ACT group: HR 28.72, 95% CI 6.16-133.98, P < 0.0001; clinical relapse rate: 100% (14 of 14); 12-month RFS: 21.43% (95% CI 7.86% to 58.40%); Figure 3B]. Additionally, we evaluated the association of anytime ctDNA dynamics on RFS according to ctDNA status at the last available on-treatment time point. Patients with persistent ctDNA positivity (HR 41.38, 95% CI 9.62-177.97, P < 0.001) or who were ctDNA-positive at the last time point (HR 19.23, 95% CI 3.65-101.39, P < 0.001) experienced inferior RFS compared with those who were persistently ctDNA-negative (Figure 3C). Prognostic value of posttreatment ctDNA status and anytime ctDNA dynamics during ACT. Kaplan–Meier estimates for RFS stratified by ctDNA status during postdefinitive treatment surveillance in (A) patients who received ACT, in (B) patients who did not receive ACT, and (C) anytime ctDNA dynamics during ACT.ACT, adjuvant chemotherapy; CI, confidence interval; ctDNA, circulating tumor DNA; HR, hazard ratio; mRFS, median relapse-free survival; NR, not reached; RFS, relapse-free survival. Comparison with CA19-9 and CEA Next, we assessed the association of traditional biomarkers, specifically CA19-9 and CEA, with the RFS of patients in our cohort. In contrast to ctDNA status, CA19-9 and CEA levels at the MRD and postdefinitive treatment surveillance time points were not significantly associated with outcomes. During the MRD window, 50 patients had available CA19-9 data, and of these, 15 showed elevated CA19-9 levels, of whom 33.33% (5 of 15) relapsed. Moreover, 35 patients showed CA19-9 levels within the normal range, and of these, 42.86% (15 of 35) relapsed (Figure 4A). Similarly, during postdefinitive treatment surveillance, among 62 patients having available CA19-9 data, 48.0% (12 of 25) of the patients with elevated CA19-9 levels relapsed, whereas 43.24% (16 of 37) of patients with CA19-9 levels within the normal range relapsed (Figure 4B). Finally, CA19-9 status was not prognostic of relapse in the MRD window [HR 1.22, 95% CI 0.41-3.65, P = 0.72; 12-month RFS for CA19-9 normal: 58.70% (95% CI 42.90% to 80.20%) compared with CA19-9 elevated: 65.50% (95% CI 44.90% to 95.40%); Figure 4A] or during postdefinitive treatment surveillance [HR 1.32, 95% CI 0.58-3.0, P = 0.5; 12-month RFS for CA19-9 normal: 63.70% (95% CI 49.60% to 81.80%) compared with CA19-9 elevated: 66.50% (95% CI 49.90% to 88.70%); Figure 4B]. Finally, CA19-9 demonstrated a sensitivity of 42.90% and specificity of 61.80% for detecting relapse. CA19-9 and CEA status during the MRD window and postdefinitive treatment surveillance are not prognostic of RFS. CA19-9 and CEA status and RFS in patients with BTC. (A, B) Kaplan–Meier estimates for RFS stratified by CA19-9 status during the MRD window (A) and postdefinitive treatment surveillance (B). (C, D) RFS stratified by CEA status during the MRD window (C) and postdefinitive treatment surveillance (D). Patients were included in the mixed results category if they had two or more ctDNA timepoints during treatment, with at least one ctDNA result being positive and at least one ctDNA result being negative.CA19-9, carbohydrate antigen 19-9; CEA, carcinoembryonic antigen; CI, confidence interval; ctDNA, circulating tumor DNA; HR, hazard ratio; MRD, molecular residual disease; mRFS, median relapse-free survival; NR, not reached; RFS, relapse-free survival. CEA data were available for 18 patients within the MRD window and for 15 patients after definitive treatment. During the MRD window, six patients showed elevated CEA, and 33.33% (2 of 6) of these patients relapsed. During postdefinitive treatment surveillance, six patients showed elevated CEA, and 33.33% (2 of 6) relapsed. CEA was not prognostic for RFS in either setting [MRD: HR 2.01, 95% CI 0.31-13.17, P = 0.47; 12-month RFS for CEA normal: 64.80% (95% CI 39.31% to 100%) compared with CEA elevated: 53.30% (95% CI 21.40% to 100%); Figure 4C; and postdefinitive treatment surveillance: HR 0.62, 95% CI 0.08-4.57, P = 0.64; 12-month RFS for CEA normal: 58.30% (95% CI 30.0% to 100%) compared with CEA elevated: 66.70% (95% CI 37.90% to 100%); Figure 4D]. CEA was found to have a sensitivity of 33.33% and specificity of 55.60% for relapse detection. Discussion In this study, ctDNA positivity during both the MRD window and postdefinitive treatment surveillance was strongly associated with inferior RFS and OS, underscoring the value of ctDNA as a dynamic biomarker of residual and recurrent disease. These findings support the role of ctDNA as a clinically informative tool for post-operative risk stratification and prognostication in BTC, a disease marked by high recurrence rates and limited sensitivity of conventional surveillance modalities. Our findings contribute to the growing body of literature affirming the value of longitudinal ctDNA testing in gastrointestinal malignancies. BTC remains an aggressive and heterogeneous disease, with more than half of the patients experiencing recurrence after curative-intent surgery despite adjuvant therapy.3,4 Current available surveillance tools, imaging, and serum biomarkers, CA19-9 and CEA, lack the sensitivity and specificity to reliably identify molecular recurrence in an early and potentially more actionable phase. In this context, ctDNA offers a tumor-specific and highly sensitive complementary tool capable of capturing residual or recurrent disease, often preceding radiographic relapse by several months. In our study, ctDNA positivity during the MRD window was associated with a significant increase in the risk of relapse and a similar magnitude of risk for overall mortality, confirming its prognostic significance early in the post-operative course. Consistent with recent real-world and prospective studies, our study also demonstrated that ctDNA positivity during postdefinitive treatment surveillance was highly predictive of RFS (HR 18.12, P < 0.0001) and OS (HR 47.54, P < 0.0001), with a median lead time of over 3.7 months. Additionally, ctDNA demonstrated high sensitivity (80.6%) and specificity (95.7%) for recurrence detection, underscoring its clinical reliability as a molecular test that complements imaging and is more sensitive and specific than serum markers (CA19-9: 42.9% and 61.8%; CEA: 33.33% and 55.6%, respectively). Additionally, by restricting the analysis to patients who did have ctDNA testing within 12 months of treatment completion, the performance of the assay in the postdefinitive surveillance window remained highly consistent, with a sensitivity of 80% (28 of 35 patients) and specificity of 97.5% (39 of 40). Similar to findings in colorectal and esophagogastric cancers,10,11,18 our results highlight the value of ctDNA monitoring in providing clinically relevant risk stratification that may inform surveillance strategies, including imaging intervals and the timing of clinical evaluation. Notably, CA19-9 and CEA did not provide equivalent prognostic value in our cohort, nor were they independently associated with survival in time-dependent analyses, supporting ctDNA as a more informative surveillance biomarker in BTC. Several prior studies have documented the feasibility and biological relevance of ctDNA in BTC. For example, in a previously published retrospective study by Yu et al.,15 which included a subset of this cohort, ctDNA identified 94% of recurrences with a median lead time of 3.7 months and significantly outperformed CA19-9. Hwang et al.19 also observed high concordance between ctDNA and tumor tissue profiling in patients with advanced BTC, particularly for therapeutically actionable mutations such as IDH1, PIK3CA, and BRCA1/2, with a ctDNA sensitivity of 84.8% and predictive value of 79.4%. Our findings expand on this work by evaluating ctDNA in a large, multicenter resected cohort, stratified by MRD and surveillance windows, and by characterizing treatment-associated ctDNA dynamics. Notably, persistently ctDNA-positive patients had markedly inferior survival during adjuvant therapy, whereas patients with ctDNA negativity exhibited favorable outcomes, suggesting that molecular response monitoring may inform real-time therapeutic decisions. These results align with earlier studies in hepatocellular carcinoma and BTC, demonstrating that ctDNA status during treatment reflects treatment efficacy and disease progression. Importantly, we also demonstrated that ctDNA positivity was the strongest independent predictor of relapse, outperforming CA19-9, CEA, and clinicopathologic features, including tumor stage, subtype, and resection margin status. This reinforces the role of ctDNA, not simply as a correlative biomarker, but as a clinically actionable tool with potential to stratify patients for risk-adapted surveillance and adjuvant treatment intensification. This is especially important in the BTC field, where heterogeneity and delayed detection of clinical recurrence undermine curative strategies. ctDNA offers a unified, dynamic biomarker that can overcome both the biological and logistical limitations of tissue-based profiling. Although prospective data are lacking, emerging case-level evidence in BTC suggests that ctDNA-guided therapeutic modification may be feasible in selected molecular contexts. For example, a recent case report described initiation of immunotherapy in a patient with microsatellite instability-high cholangiocarcinoma based on rising ctDNA levels before radiographic recurrence, resulting in sustained molecular and radiographic remission.20 Although limited, such observations illustrate the potential clinical scenarios in which ctDNA-informed decision making may be explored in future prospective trials. Recent prospective trials illustrate both the promise and challenges of translating ctDNA-based risk stratification into treatment-directed benefit. In the DYNAMIC-III trial,21 in stage III colorectal cancer, ctDNA-guided (Safe-SeqS) escalation and de-escalation strategies did not improve RFS compared with standard management, despite the strong prognostic performance of ctDNA, highlighting that effective therapeutic options are required to successfully act on ctDNA positivity. In contrast, ctDNA-guided (Signatera; Natera, Inc.) intervention has demonstrated clear clinical benefit in settings where active therapies are available, as exemplified by the phase III IMvigor011 trial in muscle-invasive bladder cancer,22 in which ctDNA-guided adjuvant atezolizumab substantially improved disease-free survival and OS. In BTC, where recurrence rates are high and therapeutic options remain limited, our findings establish ctDNA as a robust prognostic and risk-stratification tool and provide a foundation for future prospective studies evaluating ctDNA-guided strategies as the treatment landscape evolves. The incorporation of ctDNA into clinical practice for BTC must also acknowledge its current limitations. The retrospective nature of our analysis, variability in ctDNA collection intervals, and lack of uniform imaging protocols could have introduced bias, especially in assessing lead time. Of note, the ctDNA positivity rate during the MRD window (23%) in our study was lower than the cumulative recurrence rates reported in historical BTC series; however, MRD positivity reflects molecular detection within a defined early post-operative window rather than long-term recurrence risk. Importantly, ctDNA detection at a single post-operative timepoint is inherently constrained by tumor shedding kinetics and low disease burden, and a negative result does not necessarily indicate the absence of residual disease but may instead reflect delayed molecular detectability. As such, conventional diagnostic performance metrics cannot be applied to a single MRD time point, as these measures assume a static test against a uniformly ascertained ground truth, which is not applicable to a dynamic, time-dependent biomarker such as ctDNA. In our overall cohort, 42% of patients developed recurrence, consistent with published data. The real-world nature of this dataset, including nonuniform MRD sampling, heterogeneous follow-up, and variable imaging schedules, further limits the interpretability of landmark diagnostic metrics, as it may introduce interval censoring and misclassification. Additionally, the inclusion of heterogeneous BTC subtypes and stages (including 17% ampullary carcinoma and 21% stage I disease, which are associated with more favorable prognosis and lower ctDNA shedding, respectively) may also have influenced the observed MRD detection rate. In addition, surveillance analyses may be subject to guarantee-time (immortal time) bias, as patients must remain recurrence-free long enough to undergo serial ctDNA testing; although ctDNA status was modeled as a time-dependent covariate to mitigate this effect, residual bias cannot be entirely excluded in this real-world setting where testing intervals were physician-directed. Although our cohort included patients with all major BTC subtypes, the sample size limited our ability to perform robust subtype-specific or molecular subgroup analyses. Nonetheless, the consistency of our findings across MRD, postdefinitive treatment surveillance, and treatment windows highlights the broad applicability of ctDNA testing across clinical scenarios in resected BTC. Our study provides compelling evidence that ctDNA positivity is a highly prognostic biomarker for recurrence and survival in resected BTC and can serve as an early signal of molecular relapse during postdefinitive treatment surveillance. When integrated with clinical parameters and imaging, ctDNA has the potential to personalize monitoring, inform therapeutic escalation, and improve long-term outcomes for patients with BTC. Future prospective studies are needed to validate ctDNA-guided treatment strategies and to determine whether preemptive intervention based on molecular recurrence can translate into improved survival. Acknowledgments Data availability The authors declare that all relevant, nonproprietary data used to conduct the analyses are available within the article. To protect the privacy and confidentiality of patients in this study, clinical data are not made publicly available in a repository or the supplementary material of the article, but can be requested at any time from the corresponding author. All data shared will be deidentified. Funding None declared. Disclosure AT, MMK, CB, CMB, JF, JBO, KM, AF, AJ, and MCL are employees of Natera, Inc., each receiving a salary and may own stock and/or stock options. AE, LB, JY, AN, AAC, and MA have nothing to disclose. MM reports receiving consulting fees from Incyte, Ipsen, AstraZeneca, and BMS, and honoraria/payment for lectures or educational events from Natera. JE reports serving on the Data Safety Monitoring Board (DSMB) for the Neal Cancer Center at Houston Methodist Hospital and holding stock or stock options in Natera. ARH reports consulting or advisory roles with AstraZeneca, Boston Scientific, and Genentech/Roche; speakers’ bureau membership with AstraZeneca and Eisai; and research funding from Genentech and Merck. TC reports consulting fees from Deciphera and Revolution Medicine, honoraria for a presentation at Best of GI ASCO 2025, and participation on the ARPA/ADAPT advisory board. AW reports receiving consulting fees and serving on an advisory board for Gilead and AstraZeneca, has received honoraria for lectures/presentations from Total Health and Reusch Symposium, and holds stock/stock options in J&J, Baxter, and Abbott Labs. GGK discloses institutional grants from Iterion and Omegat Therapeutics, consulting fees from AstraZeneca, Ipsen, Tempus, Jazz, and Pfizer, honoraria for educational events and travel support from Binaytara Foundation, and additional support for attending meetings from the Cholangiocarcinoma Summit. AM reports receiving honoraria from AstraZeneca, Taiho Oncology, and Incyte for speakers bureau participation. JWF reports consulting fees from Boehringer-Ingelheim, Bristol Myers-Squibb, Genentech, and Guardant; Institutional research funding from AbbVie, Alnylam, AstraZeneca, Genentech, Iteration, and Omega Therapeutics; and research funding from NIH/NCI, Cancer Research Foundation, Genentech, GI Research Foundation, and UCCCC Immuno-Oncology pilot program. SK reports receiving honoraria and payment for lectures/presentations from Astrazeneca, Gilead, Sirtex, Madrigal, and Boston Scientific, and support for attending meetings and travel from ACG and a DEI conference. RMG reports consulting fees from Sanofi, TransMedics, and LyGenesis; honoraria for presentations and support for travel/meeting attendance from Sanofi and TransMedics; service on paid advisory boards for Sirtex, TransMedics, and LyGenesis; the holding of stock or stock options in TransMedics and Eli Lilly; and unpaid leadership roles as Treasurer, President, and Past-President of the International Liver Transplant Society (ILTS) and as a Member of the American Society of Transplant Surgeons (ASTS) Foundation. RDK reports receiving consulting fees from Taiho, Bayer, Exelixis, Eisai, Roche, Ipsen, AbbVie, Servier, Jazz, AstraZeneca, Incyte, Elevar, and Takeda. Supplementary data References Associated Data Data Availability Statement The authors declare that all relevant, nonproprietary data used to conduct the analyses are available within the article. To protect the privacy and confidentiality of patients in this study, clinical data are not made publicly available in a repository or the supplementary material of the article, but can be requested at any time from the corresponding author. All data shared will be deidentified.