Overview of studies evaluating MRD in NSCLC

Numerous retrospective and prospective studies have evaluated the clinical utility of ctDNA-based MRD detection in patients with early to locally advanced NSCLC. These investigations, summarized in Table 1, span a range of study designs, stages, treatment contexts, and MRD detection methodologies [6, 7, 15–17, 18–31]. Despite their heterogeneity, a consistent theme has emerged: post-treatment ctDNA positivity is strongly associated with increased recurrence risk, often identifying disease relapse months prior to radiographic detection. Across studies, ctDNA detection lead times—the interval between MRD positivity and radiographic recurrence—range from approximately 88 to over 200 days, with a median around 3 to 6 months in landmark trials [7, 17–19]. This early detection window highlights the potential of ctDNA analysis to enable preemptive interventions. Most trials employ personalized or tumor-informed NGS approaches, with limits of detection (LoD) ranging from 0.01% to 0.1% variant allele frequency. Importantly, both sensitivity and specificity are generally high, especially when sampling occurs 1–4 weeks post-treatment and is followed by longitudinal monitoring. While certain studies focus exclusively on recurrence prediction, others explore how MRD status may inform the benefit of adjuvant therapies. The emerging evidence supports a paradigm shift toward risk-adapted strategies, wherein MRD-positive patients may derive greater benefit from adjuvant interventions, while MRD-negative patients may be spared unnecessary toxicity.

LUNGCA-1 trial: prospective validation of ctDNA-based MRD in resected NSCLC

The LUNGCA-1 trial [17] was a multicenter prospective cohort study involving 330 patients with resectable stage I–III NSCLC, designed to evaluate the clinical utility of ctDNA-based MRD detection. Plasma samples were collected at three perioperative time points: preoperatively, 3 days after surgery, and 1 month after surgery. MRD was defined as ctDNA positivity at either of the two postoperative time points, based on tumor-informed sequencing using a 425-gene targeted NGS panel. The study found that preoperative ctDNA positivity was significantly associated with shorter recurrence-free survival (RFS), with a hazard ratio (HR) of 4.2 (P < 0.001). Importantly, detecting ctDNA postoperatively (i.e., MRD positivity) independently predicted a markedly elevated risk of relapse, with an HR estimated at 11.1 (P < 0.001). MRD status demonstrated greater prognostic power for RFS than traditional clinicopathologic variables such as TNM stage or lymphovascular invasion. In evaluating the impact of adjuvant therapies, a key interaction between MRD status and treatment benefit was observed. Among MRD-positive patients, those who received adjuvant therapy had significantly improved RFS compared to those who did not (HR = 0.3; P = 0.008). Conversely, in the MRD-negative group, receipt of adjuvant therapy was paradoxically associated with worse RFS (HR = 3.1; P < 0.001). Multivariable analysis confirmed that adjuvant therapy remained an independent predictor of RFS only in the MRD-positive population (P = 0.002), but not in MRD-negative patients (P = 0.283). These findings suggest that ctDNA-based MRD status can not only predict recurrence with high accuracy but also identify patients most likely to benefit from adjuvant treatment. The LUNGCA-1 study supports the clinical utility of perioperative ctDNA monitoring as both a prognostic and predictive biomarker, paving the way for MRD-guided postoperative management in early-stage NSCLC [17].

TRACERx study: insights into tumor evolution and MRD detection

The TRACERx (TRAcking Cancer Evolution through therapy) study represents a landmark prospective effort to investigate tumor evolution, intratumoral heterogeneity, and recurrence mechanisms in early-stage NSCLC. Through the integration of multi-region whole-exome sequencing of resected tumors and serial plasma ctDNA profiling, TRACERx has provided essential insights into how residual disease evolves and escapes detection. In the foundational TRACERx report, Abbosh et al. [6] demonstrated that postoperative ctDNA detection could anticipate disease recurrence before radiologic confirmation. They showed that ctDNA levels rose several months prior to imaging-based relapse in a substantial fraction of patients, supporting the role of molecular monitoring as a sensitive early indicator of MRD [6]. This study also revealed that subclonal mutations—those not captured by single-region tumor sampling—frequently emerged at relapse, underscoring the limitations of conventional tissue-based profiling and the need for multi-region approaches. Building on this work, Abbosh et al. [21] later expanded the cohort and applied personalized ctDNA assays to trace early metastatic dissemination events. Their 2023 study revealed that ctDNA could capture the emergence of new metastatic subclones long before clinical relapse, highlighting ctDNA’s utility not only for recurrence detection but also for mapping evolutionary trajectories under selective pressure [21]. Further analysis by Al Bakir et al. [32] focused on the evolutionary dynamics shaping recurrence. Using paired primary and recurrent tumors, they identified patterns of immune escape, including neoantigen depletion and immune evasion mechanisms, that were linked to therapeutic resistance. Longitudinal ctDNA analysis confirmed that immune-driven clonal sweeps could be detected noninvasively, reinforcing the value of ctDNA for capturing biological processes that govern relapse [32]. Together, findings from TRACERx illustrate that ctDNA-based MRD detection not only provides a sensitive measure of recurrence risk but also serves as a window into the evolutionary biology of NSCLC. The evidence supports incorporating personalized, longitudinal ctDNA monitoring into perioperative management to detect subclinical relapse, guide adjuvant treatment, and ultimately refine risk stratification in early-stage lung cancer.

Retrospective single-center study by Chaudhuri et al.

The study “Early Detection of Molecular Residual Disease in Localized Lung Cancer by Circulating Tumor DNA Profiling,” conducted by Chaudhuri et al. [7] and published in Cancer Discovery in 2017, was a single-center retrospective study conducted at Stanford University. This study investigated the use of ctDNA to detect MRD in patients with localized lung cancer following curative-intent treatment. The study enrolled 40 patients with stage I–III lung cancer, analyzing 255 plasma samples collected before and after treatment. The researchers employed CAncer Personalized Profiling by Deep Sequencing (CAPP-Seq), an NGS assay targeting 128 genes commonly mutated in lung cancer, to detect ctDNA mutations. They defined ctDNA positivity as the presence of at least one somatic mutation in plasma corresponding to mutations identified in the primary tumor tissue. The assay demonstrated a sensitivity to detect mutant allele fractions as low as 0.003%. During a median follow-up of 775 days, 14 patients experienced disease relapse. Notably, 13 of these 14 patients (93%) had detectable ctDNA before or at the time of clinical relapse. However, only 36% of patients who eventually relapsed had detectable ctDNA MRD at the first post-surgical timepoint, highlighting the importance of longitudinal monitoring. Conversely, 90% of patients with undetectable ctDNA MRD at the initial timepoint did not develop disease relapse, demonstrating the high negative predictive value (NPV) of the assay. The study also evaluated the impact of adjuvant chemotherapy in MRD-positive and MRD-negative patients. Although median RFS values were not explicitly reported, ctDNA positivity after curative treatment was significantly associated with shorter RFS based on Kaplan-Meier analysis (log-rank P < 0.05).

Prospective study by Gale et al.: prognostic utility of MRD detection in early-stage NSCLC

Gale et al. [19] conducted a prospective study to evaluate the role of ctDNA-based MRD detection in early-stage NSCLC patients undergoing curative treatment. This study included 88 patients with stage I to III NSCLC and employed a personalized RaDaR™ assay to track up to 48 tumor-specific mutations in plasma samples. The study demonstrated a clear correlation between ctDNA detection and recurrence risk, underscoring the clinical utility of MRD monitoring in NSCLC management. Pre-treatment ctDNA detection rates varied by disease stage, with 24% in stage I, 77% in stage II, and 87% in stage III patients. Post-treatment ctDNA analysis revealed that among 28 patients who experienced recurrence, 64.3% had detectable ctDNA before clinical relapse, with a median lead time of 212.5 days before radiologic evidence of recurrence. Notably, patients with detectable ctDNA within two weeks to four months after treatment completion had a significantly higher risk of recurrence, with an RFS HR of 14.8 and an overall survival HR of 5.48. These findings highlight the predictive value of ctDNA analysis in post-treatment surveillance and early intervention strategies.

The study also emphasized the complexity of interpreting post-surgical ctDNA dynamics. While ctDNA detection shortly after surgery was strongly associated with recurrence, 25% of patients with ctDNA detected within 1–3 days post-surgery did not develop recurrence. This suggests that immediate post-surgical ctDNA detection may not reliably predict long-term outcomes and requires careful interpretation. The findings support the integration of longitudinal ctDNA monitoring into routine clinical workflows to enable personalized, risk-adapted post-surgical management strategies in NSCLC patients. The clinical implications of these results are significant. By identifying high-risk patients before clinical relapse, ctDNA-based MRD detection enables earlier therapeutic interventions, potentially improving long-term survival. This study reinforces the growing body of evidence supporting ctDNA as a prognostic biomarker for recurrence risk assessment and treatment stratification in NSCLC. Further large-scale prospective trials are warranted to validate these findings and facilitate the clinical adoption of MRD-based decision-making in oncology [19].

Prospective study by Waldeck et al.: early post-surgical ctDNA detection for recurrence prediction in NSCLC

Waldeck et al. [29] conducted a prospective study to evaluate the utility of early postoperative ctDNA detection in predicting tumor recurrence in patients with early and locally advanced NSCLC undergoing curative-intent surgical resection. The study enrolled 33 patients with stage IA–IIIB NSCLC, and serial plasma samples were collected at multiple time points, including pre-surgery, intraoperatively, and postoperatively at 1–2 weeks, followed by additional follow-up samples. A total of 96 plasma samples were collected from 21 patients for longitudinal analysis. The study utilized a highly sensitive NGS-based assay with a personalized tumor-informed approach to detect ctDNA mutations. MRD positivity was defined as the presence of at least one tumor-specific somatic mutation in postoperative plasma, corresponding to mutations identified in the primary tumor tissue. The assay employed had a LoD of 0.01% variant allele frequency, ensuring high sensitivity for detecting MRD. At the early postoperative time point (1–2 weeks post-surgery), 4 out of 21 patients (19%) were ctDNA-positive, and all subsequently experienced disease progression. In contrast, of the 12 patients with undetectable ctDNA at the same time point, only 4 (33%) experienced recurrence during follow-up. The presence of postoperative ctDNA was significantly associated with shorter PFS (P = 0.013) and overall survival (P = 0.004). These findings highlight the potential role of early ctDNA monitoring in identifying high-risk patients who may benefit from adjuvant therapies or intensified surveillance. This study underscores the importance of intraoperative and early postoperative ctDNA assessment for detecting MRD in NSCLC. By incorporating serial ctDNA testing, clinicians may refine post-surgical management strategies and implement personalized treatment approaches to improve patient outcomes. These findings further emphasize the need for future studies to validate the prognostic value of early ctDNA monitoring in larger patient cohorts and guide the integration of liquid biopsy into routine NSCLC clinical practice [29].

Assay technologies, input requirements, and pre-analytical challenges in MRD detection

Accurate detection of MRD through ctDNA requires not only highly sensitive sequencing platforms but also consistent and standardized handling of pre-analytical variables such as plasma volume, DNA yield, and input amounts. As shown in Table 2, the volume of plasma used for cfDNA extraction across studies ranged from 2 mL to 10 mL, depending on the assay requirements and institutional protocols. Larger volumes generally enhance sensitivity, particularly in early-stage or MRD-negative patients where ctDNA levels are expected to be low.

The most commonly used extraction kits were the QIAamp Circulating Nucleic Acid Kit (Qiagen) and the MagMAX Cell-Free DNA Isolation Kit (ThermoFisher or Applied Biosystems), each differing in yield efficiency, input tolerance, and compatibility with downstream workflows. Some studies reported the purified cfDNA amount directly in nanograms, while others quantified DNA input using human genome equivalents (hGE), assuming approximately 3.3 pg per diploid human genome. Normalizing ctDNA levels by hGE allows for more standardized comparisons across samples and platforms, and several studies adopted this approach to calculate the percentage of ctDNA relative to total input. For example, Chaudhuri et al. [7] and Moding et al. [20] expressed ctDNA quantification as %ctDNA per hGE, while others such as Qiu et al. [16] and Gale et al. [19] employed variant allele frequency or estimated mutant molecules per mL of plasma. Depending on the assay design, some studies measured only high-confidence driver mutations, while others—particularly those using tumor-informed approaches—tracked a broader panel of both clonal and subclonal variants to enhance sensitivity.

The diversity in ctDNA quantification reflects differing biological assumptions. While driver mutations may have strong predictive value, subclonal variants often precede clinical relapse and are useful for early detection. Assays such as RaDaR and PROPHET incorporated personalized mutation lists of up to several dozen variants per patient to improve performance, with reported LoD as low as 0.004% in some platforms. Meanwhile, studies using broader panels without personalization often reported LoDs in the range of 0.01–0.1%. These differences influence not only assay sensitivity but also how MRD positivity is defined, which varies across trials from detecting any shared mutation to requiring multiple variant calls.

An important technical confounder identified in several studies is clonal hematopoiesis of indeterminate potential (CHIP), wherein age-related somatic mutations in hematopoietic stem and progenitor cells can be released into the plasma and mistakenly interpreted as tumor-derived ctDNA. Commonly mutated genes in CHIP include DNMT3A, TET2, ASXL1, and TP53, which frequently overlap with mutations found in solid tumors, complicating interpretation in MRD settings. Without appropriate filtering, these CHIP-associated variants may lead to false-positive MRD calls, particularly in older patients in whom CHIP prevalence is estimated to exceed 10–20% in individuals over 65 years [33, 34]. To mitigate this issue, recent MRD studies have implemented bioinformatic filtering strategies, often involving paired sequencing of peripheral blood mononuclear cells (PBMCs) or white blood cells (WBCs) to distinguish hematopoietic-derived mutations from true tumor-specific alterations. Alternatively, databases of recurrent CHIP variants or allele frequency thresholds are used to flag likely CHIP-related events [35, 36]. This filtering step is increasingly regarded as a critical quality control measure in ctDNA-based MRD pipelines, especially in populations at higher risk for CHIP-associated confounding. Collectively, these findings underscore that successful clinical implementation of ctDNA-based MRD testing in NSCLC relies not only on assay sensitivity but also on pre-analytical standardization, accurate quantification of tumor-derived signals, and rigorous computational error correction. Harmonization of input volumes, extraction protocols, and reporting standards will be essential for ensuring reproducibility and clinical confidence as MRD-guided strategies move toward broader adoption in practice.

Predictive accuracy of ctDNA-based MRD detection: sensitivity, specificity, and clinical implications

As summarized in Table 3, multiple prospective and retrospective studies have reported the diagnostic performance of ctDNA-based MRD assays for predicting recurrence in NSCLC, with varying degrees of sensitivity, specificity, positive predictive value (PPV), and NPV. Across studies, sensitivity values ranged widely from approximately 30% to over 90%, depending on factors such as assay type, timing of sampling, stage distribution, and the definition of MRD positivity. In landmark studies such as LUNGCA-1 and PROPHET, sensitivity values for detecting recurrence were reported around 30–84%, while specificity was consistently high, reaching 90–100% in many cohorts [15, 17, 19]. Importantly, the PPV of MRD positivity—that is, the probability of clinical recurrence following a ctDNA-positive result—was also high in most studies, typically exceeding 70–80%, indicating that ctDNA detection is a reliable indicator of impending relapse. Conversely, NPV values were more variable, often influenced by follow-up duration and sampling frequency, but generally fell within the range of 60–90%, supporting the use of ctDNA negativity as a reassuring marker in postoperative surveillance.

The lead time between ctDNA detection and radiographic or clinical recurrence averaged between 3 and 7 months, offering a potential therapeutic window for early intervention. Notably, studies that employed longitudinal monitoring strategies and personalized, tumor-informed assays tended to achieve better performance across all metrics, underscoring the value of repeated sampling and molecular tracking over time. These findings collectively support the use of ctDNA-based MRD assessment not only as a prognostic tool but also as a clinically actionable biomarker to inform surveillance intensity and adjuvant therapy decisions. However, the heterogeneity in assay platforms, timepoints, and positivity thresholds highlights the need for further standardization and prospective validation before these metrics can be uniformly integrated into clinical decision-making frameworks.

Ongoing prospective NSCLC MRD studies

In parallel with completed studies, several large-scale prospective trials are actively evaluating the clinical utility of ctDNA-based MRD monitoring in NSCLC (summarized in Table 4). These trials aim to integrate MRD detection into postoperative treatment decision-making, particularly to guide adjuvant therapy strategies and personalize surveillance intensity based on molecular relapse risk. The MERMAID-1/2 trial is a randomized phase III study designed to assess whether the addition of durvalumab (anti-PD-L1) to platinum-based chemotherapy improves DFS in MRD-positive NSCLC patients following curative-intent surgery. MRD is assessed using the tumor-informed Signatera™ assay, with an LoD of approximately 0.01% variant allele frequency. Interim findings have demonstrated improved DFS with the addition of immunotherapy in the MRD-positive population [37, 38]. The LC-SCRUM-Advantage/MRD trial is a prospective observational study enrolling patients with resectable stage I–III NSCLC in Japan. This study aims to evaluate perioperative ctDNA dynamics using a tumor-informed sequencing approach, with the goal of stratifying patients for adjuvant therapy based on MRD status. While the study is ongoing, its results are expected to clarify the clinical relevance of postoperative ctDNA detection and its potential role in guiding treatment decisions [39]. The ALINA trial is a randomized phase III study evaluating adjuvant alectinib versus platinum-based chemotherapy in patients with resected ALK-positive NSCLC, with planned exploratory analyses of ctDNA for MRD assessment [40]. Similarly, the neoADAURA trial is investigating neoadjuvant osimertinib with or without chemotherapy in resectable EGFR-mutant NSCLC, incorporating serial ctDNA testing to evaluate MRD dynamics and guide treatment strategies [41]. Other notable phase III trials include BR.31 and ANVIL, each of which is assessing the benefit of adjuvant systemic therapy in resected NSCLC patients. The BR.31 trial (NCT02273375) is evaluating the efficacy of atezolizumab (anti-PD-L1) compared to placebo in patients with stage IB–IIIA NSCLC following surgical resection [42]. This Canadian Cancer Trials Group—led study includes DFS and overall survival as primary endpoints and is also collecting biospecimens for exploratory MRD and biomarker analyses using ctDNA and PD-L1 expression as stratification factors. The ANVIL trial (NCT02595944), part of the ALCHEMIST umbrella protocol sponsored by the U.S. NCI, is investigating nivolumab (anti-PD-1) in a similar adjuvant context, with DFS and OS as co-primary endpoints [43]. While MRD is not explicitly defined as a study endpoint, blood samples are being collected for exploratory biomarker analyses, which may include ctDNA-based investigations under correlative research components.

Remaining challenges and future directions

Collectively, these ongoing trials are designed to test whether MRD-guided treatment escalation for MRD-positive patients and de-escalation for MRD-negative patients can improve outcomes and reduce overtreatment. They also aim to establish standardized protocols for MRD testing and determine its added value compared to conventional clinicopathologic risk factors. Technological innovation continues to enhance the sensitivity and clinical utility of MRD assays. Advances in NGS and digital PCR have enabled detection of ctDNA at VAF below 0.01%. Emerging strategies such as DNA methylation profiling offer complementary insights that may improve specificity, particularly in early-stage disease. Multi-omic approaches that combine ctDNA with proteomic and transcriptomic signals are being explored to increase diagnostic accuracy. Despite progress, challenges remain in translating MRD testing into routine practice. There is still no consensus on thresholds for MRD positivity, and CHIP remains a source of potential false positives, especially in older individuals. The high cost of sequencing-based assays and limited access in resource-constrained settings further complicate widespread adoption. Cost-effectiveness analyses and payer policy development will be essential components of future implementation. Looking forward, artificial intelligence (AI)-based models hold promise for refining MRD interpretation and predicting recurrence risk with higher accuracy. Although not conducted in NSCLC, trials such as CIRCULATE-Japan (colorectal), PEGASUS (colorectal), and IMvigor011 (urothelial) offer early validation of MRD-guided treatment frameworks that could be extrapolated to lung cancer and warrant similar prospective efforts in the NSCLC setting. These efforts represent a critical step toward the broader integration of MRD-guided strategies into precision oncology.

While ctDNA-based MRD detection offers earlier recurrence signals than radiologic imaging—often with a lead time of several months—whether this molecular lead time can be translated into overall survival benefit through earlier therapeutic intervention remains an unresolved question. In addition to technical and logistical challenges, critical clinical questions remain regarding the interpretation and actionability of MRD positivity. While ctDNA-based assays can detect molecular relapse several months before radiographic evidence, it is still unclear what constitutes the optimal therapeutic intervention during this lead time. Potential strategies may include early initiation or intensification of adjuvant chemotherapy, targeted therapy, or immunotherapy. However, it remains to be proven whether such early interventions improve overall survival compared to standard approaches initiated upon radiologic progression. Furthermore, MRD positivity represents a probabilistic risk signal rather than direct evidence of viable malignant cells, raising ethical and clinical dilemmas regarding treatment escalation in patients without overt disease. Future prospective trials must address not only whether earlier treatment improves outcomes, but also how to define actionable thresholds for intervention that balance benefit with the risk of overtreatment. Most existing studies focus on prognostic associations, and although MRD positivity is consistently linked with higher recurrence risk, there is still a lack of definitive prospective evidence demonstrating that MRD-guided escalation or de-escalation of therapy improves long-term outcomes. Ongoing phase III trials such as MERMAID-1 and NCT04585490, along with prospective observational studies like LC-SCRUM-Advantage/MRD in NSCLC are anticipated to address this gap. However, until such data become available, the clinical utility of ctDNA-based MRD detection—particularly its ability to guide actionable decisions that lead to survival benefit—remains investigational. The next wave of MRD research must focus not only on technical optimization but also on establishing the predictive value of MRD through well-powered, prospective interventional studies.