Narrative review: clinical application and challenges of circulating tumor DNA in treatment response evaluation and prognostic prediction for esophageal cancer Authors: Rentong Gu, Tao Liu, Hai Jin, Xiaowei Wang Published: 2026 ABSTRACT AbstractBackground and ObjectiveCirculating tumor DNA (ctDNA) is a cornerstone of liquid biopsy with significant potential for non-invasive, dynamic monitoring of tumor burden and molecular characteristics in esophageal cancer (EC). This narrative review aims to synthesize current evidence on the utility of ctDNA in evaluating treatment response and predicting prognosis, and to discuss the associated technical and biological challenges.MethodsA literature search was conducted using PubMed, Web of Science, and Google Scholar for relevant studies published between January 2015 and April 2026. Keywords included “circulating tumor DNA”, “ctDNA”, “esophageal cancer”, “prognosis”, and “treatment response”. Priority was given to clinical studies, meta-analyses, and review articles.Key Content and FindingsDetection technologies, including polymerase chain reaction (PCR)- and next-generation sequencing (NGS)-based methods, continue to evolve. Evidence indicates a strong correlation between ctDNA clearance and pathological complete response (pCR) after neoadjuvant therapy. Postoperative ctDNA positivity is consistently associated with increased recurrence risk and shortened survival. Specific mutations (e.g., NFE2L2, PTEN, TP53) provide additional prognostic stratification. However, clinical implementation faces challenges such as lack of standardization, interference from clonal hematopoiesis, and high costs.ConclusionsctDNA analysis is a powerful tool for prognostication and response monitoring in EC. Future integration with multi-omics data, radiomics, and artificial intelligence (AI) promises to enhance its clinical utility. Overcoming current hurdles through standardization and technological innovation is essential for its routine clinical adoption.Keywords: Circulating tumor DNA (ctDNA), esophageal cancer (EC), treatment response evaluation, prognostic prediction, narrative review Introduction Esophageal cancer (EC) is a highly lethal gastrointestinal malignancy, ranking as the seventh most common cancer and the sixth leading cause of cancer-related mortality globally (1). Histologically, EC is primarily classified into esophageal squamous cell carcinoma (ESCC) and esophageal adenocarcinoma (EAC). China bears a disproportionately high burden of this disease, accounting for approximately 50% of global cases, with ESCC representing about 90% of all cases, making it the predominant pathological subtype in the country (2). The early stages of EC are often asymptomatic or present with non-specific symptoms, leading to a late-stage diagnosis in up to 40% of patients and a consequent 5-year survival rate of merely 5% (3,4). Although advances in comprehensive treatment strategies have improved patient survival, a significant proportion still experience recurrence or metastasis within two years after curative surgery and adjuvant therapy (5,6). This stark reality highlights the urgent need for effective tools to predict prognosis and guide management. Circulating tumor DNA (ctDNA) has emerged as one such promising biomarker. Liquid biopsy, particularly the analysis of ctDNA, has emerged as a promising tool in oncology in recent years (7). ctDNA consists of tumor-derived DNA fragments shed into the bloodstream, offering a real-time snapshot of the tumor’s molecular landscape and burden (8,9). Compared to traditional tissue biopsy, ctDNA analysis offers several advantages: it is non-invasive, allows for repeated sampling to monitor dynamic changes, and helps overcome tumor heterogeneity (10). While ctDNA has already been integrated into the clinical management of various cancers, such as colorectal, breast, and lung cancer, for applications in screening and prognostic assessment (11,12), its utility in the context of EC is still in the earlier stages of exploration and validation. The evolution of liquid biopsy represents a paradigm shift in precision oncology. As comprehensively reviewed by Bao et al., liquid biopsy—encompassing ctDNA, circulating tumor cells (CTCs), microRNAs, and extracellular vesicles—has emerged as a pivotal tool for early detection, continuous monitoring, and tailored therapy across multiple cancer types (13). Despite its notable advantages, significant translational challenges remain, including the need for improved precision, cost-effectiveness, and, most critically, standardized methodologies. These pan-cancer insights provide a valuable framework for understanding both the promise and the hurdles specific to ctDNA implementation in EC. This narrative review aims to address the following key questions: (I) what are the current technological approaches for ctDNA detection in EC? (II) How does ctDNA inform treatment response evaluation and prognostic prediction? (III) What are the major challenges limiting its clinical translation? (IV) What are the future directions for enhancing its clinical utility? We present this article in accordance with the Narrative Review reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1-0211/rc). Methods A narrative review of the literature was performed to synthesize current evidence on ctDNA in EC. The search was conducted in PubMed, Web of Science, and Google Scholar databases for articles published between January 2015 and April 2026. The search strategy utilized a combination of keywords and MeSH terms: (“circulating tumor DNA” OR “ctDNA” OR “liquid biopsy”) AND (“esophageal cancer” OR “esophageal squamous cell carcinoma” OR “esophageal adenocarcinoma”) AND (“prognosis” OR “survival” OR “recurrence” OR “treatment response” OR “neoadjuvant therapy”). The search was limited to articles published in English. Reference lists of retrieved articles and relevant reviews were also screened for additional studies (Table 1). The detailed PubMed search strategy is provided in Table S1. Table 1The search strategy summaryItems SpecificationDate of searchNovember 1, 2025–April 14, 2026Databases and other sources searchedPubMed, Web of Science, Google ScholarSearch terms usedMeSH Terms: “circulating tumor DNA”, “esophageal cancer”, “prognosis”Free-text Terms: (“circulating tumor DNA” OR “ctDNA” OR “liquid biopsy”) AND (“esophageal cancer” OR “esophageal squamous cell carcinoma” OR “esophageal adenocarcinoma”) AND (“prognosis” OR “survival” OR “recurrence” OR “treatment response” OR “neoadjuvant therapy”)Filters: publication date (January 1, 2015–April 14, 2026), English languageTimeframeJanuary 1, 2015–April 14, 2026Inclusion and exclusion criteriaInclusion: study types: clinical studies (prospective/retrospective), meta-analyses, systematic reviews, and relevant high-quality narrative reviews. Language: English. Population: patients with esophageal cancer (squamous cell carcinoma or adenocarcinoma). Topic: studies must investigate the application of ctDNA in prognosis, treatment response monitoring, or minimal residual disease detectionExclusion: (I) studies focusing solely on detection technology without clinical outcome validation; (II) studies on cancers other than esophageal; (III) conference abstracts, case reports, comments, editorials; (IV) non-English publicationsSelection processInitial screening: one reviewer screened titles and abstracts to exclude obviously irrelevant record. Full-text Screening: full texts of potentially eligible studies were retrieved and independently assessed for inclusion by two reviewers. Consensus: results were cross-checked. Any disagreements were resolved through discussion or by consultation with a third senior reviewerctDNA, circulating tumor DNA. The search strategy summary ctDNA, circulating tumor DNA. Inclusion criteria focused on clinical studies (prospective and retrospective), meta-analyses, and comprehensive reviews that investigated the role of ctDNA in prognosis, treatment monitoring, or minimal residual disease (MRD) detection in EC. Studies focusing solely on technical methodologies without clinical correlation, or those on other cancer types, were excluded. The literature selection process prioritized recent studies and those with higher levels of evidence. Quality assessment and level of evidence To enhance methodological transparency, we assessed the level of evidence for each included clinical study based on the Oxford Centre for Evidence-Based Medicine (OCEBM) Levels of Evidence framework (March 2009 version). Under this framework: ❖ Level 1: systematic reviews of randomized controlled trials (RCTs) or individual RCTs with narrow confidence intervals (CIs); ❖ Level 1: systematic reviews of randomized controlled trials (RCTs) or individual RCTs with narrow confidence intervals (CIs); ❖ Level 2: cohort studies or systematic reviews of cohort studies; ❖ Level 2: cohort studies or systematic reviews of cohort studies; ❖ Level 3: case-control studies or systematic reviews of case-control studies; ❖ Level 3: case-control studies or systematic reviews of case-control studies; ❖ Level 4: case series or poor-quality cohort/case-control studies; ❖ Level 4: case series or poor-quality cohort/case-control studies; ❖ Level 5: expert opinion without explicit critical appraisal, or based on physiology/bench research. ❖ Level 5: expert opinion without explicit critical appraisal, or based on physiology/bench research. Based on our evaluation of the cited literature: Level 1 evidence is represented by the multicenter RCT NCT05213312, which demonstrated that postoperative ctDNA clearance was significantly higher in patients receiving perioperative nivolumab plus chemotherapy (89% vs. 62.5%, P=0.01), and persistent ctDNA negativity was associated with improved disease-free survival (DFS) (P<0.001) (14). Level 2 evidence constitutes the main body of the literature. This includes multiple systematic reviews of cohort studies (e.g., 13 studies, 604 patients) showing significant associations between ctDNA positivity and survival outcomes [overall survival (OS) hazard ratio (HR) =3.65, DFS HR =6.08, progression-free survival (PFS) HR =2.84] (9), as well as numerous prospective and retrospective cohort studies on MRD detection, treatment response monitoring, gene-specific mutations [e.g., PTEN (15)], multimodal ctDNA analysis, and technical performance factors. A summary of evidence levels for key studies cited in this review is provided in Table 2. We acknowledge that Level 1 evidence directly evaluating ctDNA-guided clinical decision-making remains limited, and the majority of studies are Level 2 cohort studies. This limitation is further discussed in the “Strengths and limitations of current evidence” subsection of the Discussion. Table 2Evidence level assessment of key studies cited in the review (based on OCEBM 2009 framework)Evidence levelStudy type Examples from the review Key findings relevant to ctDNALevel 1Individual RCT NCT05213312 (multicenter phase II RCT) Postoperative ctDNA clearance: 89% (immunotherapy) vs. 62.5% (control), P=0.01; persistent ctDNA negativity associated with improved DFS (P<0.001) (14) Level 2Systematic review of cohort studies Systematic review of 13 cohort studies (604 patients) ctDNA positivity associated with OS HR =3.65, DFS HR =6.08, PFS HR =2.84 (9)Level 2Systematic review of cohort studies Systematic review of 10 studies (334 patients) Post-treatment ctDNA positivity associated with worse PFS (HR =3.81), OS (HR =3.00), lower pCR rate (OR =0.26) (16)Level 2Prospective cohort MRD monitoring study Postop ctDNA+: recurrence 66.67% vs. 21.54%; DFS HR =4.58, OS HR =5.39 (17) Level 2Prospective cohort Personalized ctDNA monitoring (2–16 weeks postop) ctDNA+ recurrence 100% vs. 30.4%; HR =30.99 (P<0.0001) (8)Level 2Prospective cohort Chemoradiotherapy monitoring (week 4) ctDNA+ predicts shorter PFS (HR =3.60) and higher local recurrence (HR =4.43) (18)Level 2Prospective cohort PTEN ctDNA mutation Preoperative PTEN ctDNA+ predicts DFS HR =7.28–7.53, OS HR =3.76–5.35 (15)Level 2Prospective cohort Multimodal ctDNA (VAF + ichorCNA) Baseline ctDNA+ HR =2.23 for progression; not cleared preoperative HR =4.58 (19)ctDNA, circulating tumor DNA; DFS, disease-free survival; HR, hazard ratio; MRD, minimal residual disease; OCEBM, Oxford Centre for Evidence-Based Medicine; OR, odds ratio; OS, overall survival; pCR, pathological complete response; PFS, progression-free survival; RCT, randomized controlled trial; VAF, variant allele frequency. Evidence level assessment of key studies cited in the review (based on OCEBM 2009 framework) ctDNA, circulating tumor DNA; DFS, disease-free survival; HR, hazard ratio; MRD, minimal residual disease; OCEBM, Oxford Centre for Evidence-Based Medicine; OR, odds ratio; OS, overall survival; pCR, pathological complete response; PFS, progression-free survival; RCT, randomized controlled trial; VAF, variant allele frequency. Discussion Technological and methodological advances in ctDNA detection Current ctDNA detection primarily relies on polymerase chain reaction (PCR)-based methods and next-generation sequencing (NGS). Quantitative PCR is characterized by its operational simplicity and low cost, yet it suffers from limited sensitivity and a narrow coverage of genomic loci (20). Digital PCR (dPCR) technologies, including BEAMing and droplet digital PCR (ddPCR), provide high sensitivity and absolute quantification via single-molecule amplification. However, their dependence on Poisson distribution, limited applicability in high-concentration samples, and low throughput (as each reaction detects only a single known mutation) restrict broader use (21). NGS enables the parallel sequencing of millions of DNA fragments, significantly improving throughput and operational convenience. A drawback is its short read lengths and suboptimal capture efficiency in high-GC regions. As demonstrated by Deveson et al. (22), NGS performs well when the variant allele frequency (VAF) in ctDNA exceeds 0.5%, but its reliability diminishes below this threshold. Third-generation sequencing (TGS) eliminates the need for PCR amplification and offers advantages of longer read lengths and greater genome coverage. However, its current high error rate limits widespread clinical application until this technical bottleneck is effectively overcome (23). NGS-based ctDNA detection strategies are mainly categorized into “tumor-informed” and “tumor-agnostic” approaches. The former requires tumor tissue sequencing to construct patient-specific mutation profiles for personalized panel design, whereas the latter uses fixed panels to detect known cancer-related mutations or epigenetic alterations, offering operational simplicity when tissue is unavailable (24). Huffman et al. proposed a hybrid strategy: they first performed whole-exome sequencing on gastroesophageal junction tumor samples to comprehensively obtain tumor-specific mutation profiles (25), then designed specific PCR primers for high-frequency or key driver mutations to track ctDNA subsequently. This method ensures high detection sensitivity and specificity while effectively reducing the cost of large-scale sequencing, achieving an optimal balance between detection efficiency and cost-effectiveness. There are notable differences in sensitivity and specificity among various ctDNA detection technologies. In terms of sensitivity, techniques like dPCR and targeted amplicon sequencing can achieve a low limit of detection (LOD), typically down to 0.1% VAF, and have demonstrated robust capabilities in detecting trace amounts of ctDNA across multiple studies (26). The integrated digital error suppression (iDES) method further enhances sensitivity. This strategy combines molecular barcoding with background polishing to significantly reduce sequencing background noise, enabling the CAPP-Seq platform to achieve an LOD of 0.02% VAF and accurately identify tumor-derived mutations even in samples with very low tumor fractions (27). Regarding specificity, while most technologies perform well—for instance, one study in EAC reported a ctDNA detection specificity exceeding 95%—special attention must be paid to false positives caused by clonal hematopoiesis of indeterminate potential (CHIP) (28), a challenge that is further discussed in the “Standardization and Biological Challenges” section (as illustrated in Figure 1). Clinical workflow of ctDNA analysis in esophageal cancer, from sample collection to clinical decision-making. ctDNA, circulating tumor DNA; MRD, minimal residual disease; NGS, next-generation sequencing; PCR, polymerase chain reaction. Application of ctDNA in treatment response evaluation for EC Assessment of response to neoadjuvant therapy ctDNA has demonstrated significant clinical value in evaluating responses to neoadjuvant therapy, particularly in ESCC. A study by Chen et al. showed that in ESCC patients receiving neoadjuvant immunochemotherapy (NICT), those achieving a pathological complete response (pCR) had significantly higher rates of ctDNA negativity and clearance during treatment compared to non-responders. The degree of ctDNA clearance was positively correlated with the pathological response grade, indicating its utility as a sensitive dynamic biomarker for treatment efficacy (29). Furthermore, Yang et al. developed a multimodal model combining positron emission tomography-computed tomography (PET-CT) radiomics with quantitative ctDNA analysis, which effectively distinguished pCR from non-pCR patients [area under the curve (AUC) =0.798 in the validation cohort] (30). This integrated approach provides an efficient, non-invasive decision-support tool for post-neoadjuvant therapy assessment. Prognostic and survival assessment Postoperative ctDNA positivity is consistently associated with a significantly elevated risk of recurrence and poorer recurrence-free survival (RFS) in EC patients (8,31). Morimoto et al. highlighted the lead-time potential of ctDNA, reporting that molecular relapse was detected a median of 5.5 months earlier than standard radiographic imaging, enabling earlier intervention (31). Multiple studies have quantitatively established the strong prognostic power of ctDNA for survival outcomes, as systematically summarized in Table 3. Table 3Prognostic value of postoperative and dynamic ctDNA monitoring in esophageal cancerAuthor (Ref.)Patient cohort & contextStudy designSample sizeDetection methodKey finding related to prognosisHR for survival (95% CI) Morimoto et al. (31)EC, postoperative (with neoadjuvant chemotherapy)Prospective cohort13Next-generation sequencingMolecular relapse detected 5.5 months earlier than imagingNot providedNg et al. (32)EC, postoperativeProspective cohort147Next-generation sequencingHigh post-op ctDNA level predicts poor OSOS: 7.84 (1.87–32.97)Post-op ctDNA dynamics predict poor OSOS: 5.71 (1.81–17.97)Post-NACRT, detectable ctDNA predicts poor PFSPFS: 3.16 (1.17–8.52)Liu et al. (33)ESCC, postoperative (no adjuvant therapy)Prospective cohort60Next-generation sequencingPost-op ctDNA positivity predicts poor OSOS: 25.8 (2.7–242.6)Post-op ctDNA positivity predicts high recurrence riskRecurrence: 184.6 (3.6–9,576.9)Wang et al. (18)Locally advanced ESCC, postoperativeProspective cohort40Next-generation sequencingMolecular relapse detected 4.4 months earlier than imagingNot providedvan Velzen et al. (24)mGEC, on-treatmentProspective cohort72Next-generation sequencingDetectable ctDNA at 9 weeks of treatment predicts poor OSOS: 4.95 (1.53–16.04)Detectable ctDNA at 9 weeks of treatment predicts poor PFSPFS: 4.08 (1.31–12.75)CI, confidence interval; ctDNA, circulating tumor DNA; EC, esophageal cancer; ESCC, esophageal squamous cell carcinoma; HR, hazard ratio; mGEC, metastatic gastroesophageal cancer; OS, overall survival; PFS, progression-free survival; Ref., reference. Prognostic value of postoperative and dynamic ctDNA monitoring in esophageal cancer CI, confidence interval; ctDNA, circulating tumor DNA; EC, esophageal cancer; ESCC, esophageal squamous cell carcinoma; HR, hazard ratio; mGEC, metastatic gastroesophageal cancer; OS, overall survival; PFS, progression-free survival; Ref., reference. Beyond resectable disease, ctDNA monitoring holds value in the management of unresectable locally advanced ESCC. In the phase II EC-CRT-001 trial investigating toripalimab with definitive chemoradiotherapy, patients with negative ctDNA during and after treatment exhibited significantly higher rates of clinical complete response (cCR) and significantly longer PFS and OS, firmly establishing ctDNA’s role in prognostication for advanced disease (34). Impact of specific gene mutations The detection of specific gene mutations in ctDNA provides an additional layer of prognostic stratification. Mutations in genes such as NFE2L2 and PTEN have been identified as key independent prognostic factors. Ng et al. reported that NFE2L2 alterations detected in pre-treatment ctDNA were associated with higher tumor burden and more aggressive disease (32). Ko et al. further demonstrated that the persistence of oncogenic NFE2L2 mutations in ctDNA after neoadjuvant chemoradiotherapy was an independent predictor of tumor recurrence and poor survival (35). Similarly, Gu et al. found that preoperative detection of PTEN mutations in ctDNA was associated with significantly shorter PFS and OS (15). Additionally, Yang et al. indicated that TP53 mutation status, particularly in exon regions, contributed significantly to predicting pCR (30). Prognostic value of other ctDNA features Other quantitative and qualitative features of ctDNA also carry prognostic significance. In a study of metastatic gastroesophageal cancer (mGEC), a high detection rate (75%) and strong concordance (86%) with metastatic tissue were observed. The VAF in ctDNA showed a significant positive correlation with baseline tumor volume (Pearson’s r=0.53, P<0.0001), establishing it as a surrogate for tumor burden (24). This study also found that detecting multiple gene mutations at baseline was a marker of poor prognosis. Crucially, the kinetics of ctDNA during therapy were highly informative; patients with detectable ctDNA at 9 weeks post-treatment had significantly worse OS and PFS (24). van den Ende et al. enhanced prognostic accuracy by integrating ctDNA mutation profiles with copy number alteration (CNA) data from shallow whole-genome sequencing (sWGS). They found that baseline ctDNA positivity was associated with higher T-stage, and a VAF >1% or an inferred tumor fraction >3% were significant predictors of disease progression (19). Standardization and biological challenges in ctDNA analysis The clinical translation of ctDNA detection in ESCC is confronted by several technical and biological challenges that necessitate resolution. A primary obstacle is the lack of technical standardization. Detection platforms employed across different laboratories and institutions vary significantly in critical parameters, including sequencing depth, target region coverage, and bioinformatic analysis pipelines. The absence of unified standard operating procedures (SOPs) directly compromises the comparability and reproducibility of results, thereby hindering its widespread adoption in clinical decision-making (36). The pre-analytical phase presents another major challenge, centering on sample quality control. The stability of ctDNA is highly susceptible to conditions during sample collection, transport, and storage. Furthermore, the release of genomic DNA (gDNA) from leukocyte lysis during processing can lead to contamination, which severely compromises the accuracy of detecting low-frequency mutations (36). A significant biological confounder is the difficulty in distinguishing somatic mutations associated with CHIP from genuine tumor-derived variants. This issue is particularly pronounced in elderly patients, who have a higher prevalence of CHIP, and is a notable source of false-positive results (36). From a biological perspective, the inherent characteristics of ctDNA pose substantial detection hurdles. ctDNA molecules are short and exist at an extremely low concentration in the bloodstream. In patients with early-stage ESCC, ctDNA may constitute only about 0.01% of the total cell-free DNA (cfDNA). This low abundance often necessitates whole-genome amplification prior to detection, a process that can introduce amplification biases and affect quantitative accuracy (37). The highly fragmented nature of ctDNA also means that target sequences can be lost during extraction and library construction, further complicating analysis. Additionally, ctDNA exists in heterogeneous forms in the blood—bound to exosomes or wrapped in nucleosomes—each with distinct stability and degradation kinetics. The final detection outcome is also influenced by individual nuclease activity and the clearance function of the liver and spleen. The clinical implications of these biological features, including false-negative risks in low tumor burden settings, are discussed in the “Strengths and limitations” subsection below. Finally, practical barriers remain. The current cost of ctDNA analysis is still high, and a consensus on the selection and design of optimal gene panels specific to EC has not been established. The lack of either fixed panels or personalized approaches that have been validated in large-scale clinical trials collectively constrains the large-scale application and integration of ctDNA testing into routine clinical practice (38). Strengths and limitations of current evidence and methodological heterogeneity The overall strength of the evidence supporting the prognostic utility of ctDNA in EC is derived primarily from Level 2 cohort studies or systematic reviews of cohort studies (see Methods for the OCEBM framework). Although several well-designed prospective cohort studies provide robust evidence, Level 1 evidence from RCTs directly testing ctDNA-guided interventions remains limited. Furthermore, critical appraisal of the literature reveals substantial variability in assay sensitivity, notable false-negative risks in the setting of low tumor burden, and considerable heterogeneity across study designs, cohort sizes, and detection methodologies. Given these limitations in both evidence level and methodological consistency, ctDNA should currently be considered a strong prognostic biomarker rather than a standalone decision-making tool. Readers and investigators should interpret the existing findings with an understanding of these constraints, which also warrant careful consideration in the design of future studies. Variability in assay sensitivity Different ctDNA detection platforms exhibit markedly different sensitivity profiles. In a study of 111 patients with resectable EAC, the baseline ctDNA positivity rate was only 44.3% using amplicon sequencing, but dropped to 10.5% using ichorCNA (sWGS); combining both methods increased the detection rate to 50.5% (19). Furthermore, tumor-informed personalized panels achieve substantially higher sensitivity than fixed panels. In 132 patients with locally advanced ESCC, a personalized ctDNA assay detected mutations in 99.2% of patients at baseline, whereas fixed panels had considerably lower sensitivity. Importantly, tumor stage profoundly influences detection rates: preoperative ctDNA positivity in locally advanced ESCC ranged from 50% in stage I to 100% in stages II–IV, reflecting a strong correlation between tumor burden and assay sensitivity (8). False-negative risks in low tumor burden settings False negatives are a major concern, particularly in the postoperative MRD window and during early recurrence phases. The literature explicitly notes that ctDNA may remain undetectable in low tumor burden settings, especially when using single-method assays (19). In a real-world study of recurrent/metastatic gastroesophageal cancers, ctDNA testing occasionally failed to detect disease progression even when imaging showed clear evidence of relapse (25). Moreover, during dynamic monitoring after chemoradiotherapy, ctDNA sensitivity at week 4 was only 69%, and some patients remained ctDNA-negative despite subsequent radiographic progression (18). These observations underscore the risk of false negatives and the need for multimodal strategies. Heterogeneity in study designs, cohorts, and methodologies The published studies vary widely in design (prospective cohorts vs. retrospective analyses vs. real-world data), cohort size (e.g., n=28, n=40, n=54 in several key studies), detection platforms (personalized mPCR-NGS, fixed gene panels such as 61-gene or 474-gene panels, Signatera, amplicon sequencing combined with ichorCNA), and definitions of key endpoints. For example, the MRD time window was defined as 2–16 weeks post-surgery in some studies but as within 16 weeks in others. Dynamic metrics such as “ctDNA clearance” and “conversion” lack standardized thresholds. Such heterogeneity limits cross-study comparability and generalizability. A systematic review and meta-analysis explicitly identified heterogeneity as a source of concern and called for standardization of detection methods to enhance clinical utility (9). Potential biases and clinical implications Small sample sizes, single-center designs, retrospective selection bias, and subjective threshold settings are frequently mentioned as potential confounders. Despite these limitations, ctDNA has demonstrated clear value in prognostic stratification, early relapse warning, and treatment response monitoring, as detailed in the previous sections. However, these limitations suggest that: (I) single-method assays may underestimate risk, and multimodal strategies (e.g., combining amplicon sequencing with ichorCNA) are recommended; (II) in low-burden settings, ctDNA results should be interpreted alongside conventional modalities such as imaging; and (III) large-scale, prospective studies with standardized protocols—including unified time points, thresholds, and panel designs—are urgently needed to further validate clinical utility. Toward clinical implementation: practical considerations for ctDNA-guided management Based on the evidence synthesized in this review, we offer the following practical considerations for integrating ctDNA into clinical decision-making across treatment phases. These suggestions are hypothesis-generating and require prospective interventional validation. Pre-treatment: baseline ctDNA profiling for high-risk mutations may aid patient stratification in clinical trials or identify candidates for intensified monitoring. This is not yet standard of care. During neoadjuvant therapy: persistent ctDNA positivity despite treatment suggests a high likelihood of residual disease, warranting closer radiological surveillance or consideration of alternative perioperative strategies (within trials). Early ctDNA clearance supports continued planned therapy. Postoperative MRD: ctDNA positivity after curative surgery signals a high risk of recurrence. We suggest considering adjuvant therapy escalation (e.g., adding immunotherapy) and intensive imaging surveillance. ctDNA-negative patients have lower risk; de-escalation may be considered only with caution, and repeat testing is recommended due to potential false negatives. Longitudinal surveillance: new detection of ctDNA positivity during follow-up should prompt timely imaging and multidisciplinary discussion to assess for occult relapse. Advanced/metastatic disease: persistent ctDNA positivity during systemic therapy indicates poor prognosis and may prompt consideration of treatment change or clinical trial enrollment. ctDNA clearance supports continuation of the current regimen. These recommendations are based on observational (Level 2) evidence. Standardization of assays, thresholds, testing frequency, and integration with imaging remain unresolved. ctDNA should complement, not replace, conventional clinical and radiological assessment. Future directions Multi-omics integration and technological innovation The clinical application of ctDNA in prognostic prediction for EC continues to face challenges, primarily limited sensitivity due to low tumor burden in early-stage or MRD settings (9,19). To break through this bottleneck, multi-omics integration strategies have emerged as a pivotal direction. This includes combining ctDNA mutation profiles with CNA information derived from sWGS, utilizing algorithms like ichorCNA to improve the accuracy of tumor fraction quantification. Research indicates that such integrated strategies can significantly enhance the predictive power for pCR and long-term survival outcomes in EC (19). Beyond genomic integration, multimodal strategies also show immense potential. For instance, fusing quantitative ctDNA data with PET/CT radiomic features enables a more accurate and non-invasive assessment of pCR following NICT in ESCC, providing a more comprehensive basis for clinical decision-making (39,40). Simultaneously, exosome-based liquid biopsy technologies have made significant progress. Techniques like the ultra-fast isolation system (EXODUS) for efficient purification of circulating exosomes can provide crucial information about the tumor microenvironment, effectively complementing ctDNA in reflecting tumor heterogeneity and evolutionary dynamics (41). The exploration of novel epigenetic biomarkers represents another major frontier. Integrating the analysis of ctDNA mutation profiles with methylation signatures and fragmentomics patterns can provide a more comprehensive molecular portrait. For example, the non-destructive TAPS (TET-assisted pyridine borane sequencing) technology enables simultaneous analysis of genomic and methylation data from a single assay, achieving a sensitivity of 94.9% and specificity of 88.8% in multi-cancer early detection, showing great promise for overcoming the low abundance of ctDNA (42). Furthermore, tools like “Oncoder”, an interpretable deep learning (DL) tool, efficiently deconvolutes tumor signals from cfDNA methylation maps, achieving an AUC of 0.95 in early-stage hepatocellular carcinoma diagnosis, demonstrating the power of integrating AI with epigenetic markers (43). The combination of novel biomarker classes is also boosting detection performance. Studies indicate that combining exosomal RNA (exoRNA) with ctDNA can significantly improve the sensitivity of detecting key mutations like EGFR. A study in NSCLC found that in patients with intrathoracic metastasis (M0/M1a), exoRNA increased the detection sensitivity for EGFR activating mutations from 26% to 74% (P=0.003), forming a vital supplement and challenge to traditional ctDNA-only mutation detection strategies (44). The challenges of biomarker validation, standardization, and clinical integration highlighted in other malignancies are directly relevant to EC. As noted by Yu et al. in their analysis of hepatocellular carcinoma, the successful translation of liquid biopsy biomarkers such as ctDNA requires rigorous validation, harmonized protocols, and integration into practical clinical workflows (45). These principles should guide ongoing efforts to establish ctDNA as a routine biomarker for EC management. Artificial intelligence (AI) and big data analytics The application of AI in ctDNA data analysis is profoundly advancing the field of precise cancer prognosis. To address the high genomic heterogeneity of EC, sophisticated AI algorithms can integrate multi-dimensional data—including dynamic ctDNA monitoring, CHIP features, and tumor microenvironment characteristics (e.g., tumor-infiltrating lymphocyte subsets)—to construct more accurate prognostic models. AI and machine learning are directly enhancing the analytical sensitivity and specificity of ctDNA assays. For instance, ensemble machine learning models, such as Random Forest, have been developed to predict high-confidence somatic ctDNA variants. By integrating multiple features from different variant callers, these models outperform traditional rule-based filtering, effectively distinguishing true low-frequency variants from sequencing artifacts, which is crucial for MRD detection (46). Platforms like “MRD-EDGE” leverage machine learning to guide whole-genome sequencing (WGS) data analysis, achieving ultra-sensitive monitoring of tumor burden dynamics at very low tumor fractions. This platform can not only accurately identify postoperative MRD in various cancer types but also dynamically monitor changes in tumor fraction during neoadjuvant immunotherapy, providing critical real-time evidence for timely adjustments to clinical treatment strategies (47). A forward-looking perspective is provided by Xu et al., who proposed a standardized, stepwise diagnostic-therapeutic framework integrating tissue- and blood-based biomarkers with AI-enabled imaging for personalized cancer management (48). In the context of EC, such a framework would combine ctDNA analysis with AI-assisted imaging and pathology to enable dynamic risk stratification and tailored adjuvant therapy decisions. This integrated, multimodal approach represents the logical next step for translating ctDNA from a prognostic tool to a clinical decision-making engine. Furthermore, DL is pioneering new paradigms for risk stratification. Loeffler et al. combined DL with ctDNA analysis in colorectal cancer. Their model extracts prognostic morphological features directly from routine hematoxylin and eosin (H&E)-stained whole-slide images to predict DFS (49). The study showed that combining the DL score with ctDNA-based MRD status significantly improved risk re-stratification within both MRD-positive and MRD-negative groups. Notably, this strategy has crucial clinical implications: among MRD-negative patients, those predicted as high-risk by the DL model significantly benefited from adjuvant chemotherapy (HR =0.49, P=0.01), while those predicted as low-risk did not. This finding demonstrates that AI-driven multimodal analysis can identify high-risk patients who might be missed by traditional ctDNA testing alone, enabling more precise individualized adjuvant therapy decisions. Moving forward, there is a need to build AI-assisted decision-support platforms that integrate clinicopathological parameters, multi-modal ctDNA features (mutations, methylation, fragmentomics), and radiomic data. Such platforms are essential to overcome current bottlenecks in early relapse warning and the identification of hyper-progression, thereby advancing individualized therapy for EC. Through multi-center collaboration and data sharing, the establishment of large-scale, standardized databases and specialized biobanks will be instrumental in accelerating the clinical translation and validation of AI technologies in ctDNA analysis. Conclusions In summary, ctDNA analysis has firmly established its clinical relevance in the management of EC, particularly in prognostication and treatment response monitoring. A substantial body of evidence consistently demonstrates that the presence and dynamics of ctDNA—whether following curative surgery, during neoadjuvant therapy, or in the context of advanced disease—are powerfully associated with recurrence risk and survival outcomes. The ability of ctDNA to detect molecular relapse months before radiographic evidence provides a critical window for early intervention, underscoring its potential as a sensitive biomarker for MRD. Furthermore, the integration of specific mutational information, such as alterations in NFE2L2 and PTEN, adds a layer of molecular stratification, enabling more personalized risk assessment. However, the translation of ctDNA into routine clinical practice faces significant hurdles. The lack of technical standardization, challenges in pre-analytical sample handling, and biological confounders like clonal hematopoiesis necessitate the development of unified protocols and rigorous validation. The inherently low abundance and fragmented nature of ctDNA in early-stage disease further demand continuous technological innovation to enhance detection sensitivity and specificity. Prospective interventional trials are urgently needed to validate ctDNA-guided decision-making frameworks before they can be routinely applied to direct adjuvant therapy escalation or de-escalation. Looking ahead, the future of ctDNA in EC lies in multi-modal integration and sophisticated data analytics. The combination of mutational analysis with CNA profiling, methylation signatures, and fragmentomics promises a more comprehensive liquid biopsy. Simultaneously, the fusion of quantitative ctDNA data with radiomic features from PET/CT and pathological images can create robust, non-invasive predictive models. The application of AI and machine learning is poised to revolutionize this field, not only by refining variant calling and overcoming technical noise but also by deconvoluting complex datasets to build powerful prognostic tools. These integrated approaches will be crucial for optimizing adjuvant therapy decisions, guiding treatment escalation or de-escalation, and ultimately improving survival rates for patients with this lethal malignancy. Concerted efforts in multi-center collaboration, large-scale prospective trials, and the establishment of standardized biobanks are imperative to validate these advanced strategies and fully realize the promise of precision oncology in EC. Supplementary The article’s supplementary files as Acknowledgments We thank all study participants for their involvement. Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. Footnotes Reporting Checklist: The authors have completed the Narrative Review reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1-0211/rc Funding: None. Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1-0211/coif). The authors have no conflicts of interest to declare. References Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1-0211/prf Supplementary Materials The article’s supplementary files as