Minimal Residual Disease Enhances Prognostic Stratification beyond Pathologic Response in Resectable Non–Small Cell Lung Cancer

Abstract

Purpose:

Perioperative chemoimmunotherapy is the standard of care for resectable, locally advanced non–small cell lung cancer (NSCLC). Although pathologic complete response (pCR) correlates with excellent survival outcomes, some patients without pCR still exhibit long-term survival. In this study, we evaluate the added value of minimal residual disease (MRD).

Experimental Design:

MRD was assessed in 60 patients from the NADIM II trial (NCT03838159) using the Guardant Reveal assay. In NADIM II, patients with NSCLC without EGFR or ALK alterations were randomly assigned to receive neoadjuvant nivolumab plus chemotherapy (experimental arm) or chemotherapy alone, followed by surgery. Patients in the experimental arm with R0 resection received adjuvant nivolumab.

Results:

The MRD detection rate was 9.6%. MRD after surgery or during adjuvant treatment was associated with inferior event-free survival (EFS) and overall survival [OS; hazard ratio (HR): 10.2; 95% confidence interval (CI), 3.7–28.3 and HR: 10.0; 95% CI, 2.0–49.9, respectively]. All patients with MRD-negative plasma samples in at least two time points were alive [HR: not estimable (NE), P < 0.001], with only one relapse (HR: 41.6; 95% CI, 5.0–348.8), corresponding to a patient relapsing with a single brain metastasis. MRD enhanced the prognostic value of pCR for both EFS (P < 0.001) and OS (P = 0.015). Among non-pCR patients, MRD remained a significant prognostic marker (HR: 6.2; 95% CI, 2.2–17.1 and HR: 6.5; 95% CI, 1.3–32.5, for EFS and OS respectively). All non-pCR patients with MRD-negative results in at least two time points were alive (HR: NE, P = 0.025), with one relapse (HR: 19.9; 95% CI, 2.4–165.6), corresponding to the aforementioned case.

Conclusions:

MRD may refine prognostic evaluation beyond pCR in resectable NSCLC undergoing perioperative chemoimmunotherapy.

Translational Relevance.

Pathologic complete response (pCR) is an established early endpoint in resectable non–small cell lung cancer treated with perioperative chemoimmunotherapy; however, some patients without pCR still experience long-term survival. Active research is currently focused on determining whether minimal residual disease (MRD)-guided monitoring can inform individualized postsurgical treatment decisions. In this study, we evaluated the prognostic value of a tumor-agnostic, methylation-based MRD assay in the NADIM II trial. MRD status provided significant prognostic value beyond pCR alone, enabling more accurate risk stratification. Prognostic discrimination improved when multiple samples were analyzed. Notably, MRD negativity in at least two consecutive plasma samples identified a subgroup of non-pCR patients with improved long-term survival outcomes. Our results suggest that further investigation is warranted to safely identify candidates for treatment de-escalation.

Introduction

Lung cancer remains the leading cause of cancer-related mortality worldwide (1). However, the advent of neoadjuvant or perioperative chemoimmunotherapy has marked a paradigm shift in the management of potentially resectable non–small cell lung cancer (NSCLC), significantly improving survival outcomes. Landmark studies such as NADIM and CheckMate 816 demonstrated the superiority of nivolumab combined with chemotherapy, showing enhanced pathologic complete response (pCR) and progression-free survival (PFS; refs. 2, 3) Similarly, CheckMate 77T and the NADIM II trials further reinforced these findings (4, 5). The KEYNOTE-671 study demonstrated pembrolizumab’s perioperative benefit, offering additional validation for this approach (6). Notably, in the NADIM trial, the 5-year PFS rate in the intention-to-treat (ITT) population was 65% [95% confidence interval (CI), 49.4–76.9], and the overall survival (OS) rate was 69.3% (95% CI, 53.7–80.6; ref. 7). These advancements have redefined standard care for locally advanced potentially resectable NSCLC, providing new hope for long-term survival in a disease historically associated with fatal outcomes. Nonetheless, approximately 30% to 35% of patients still decease within 5 years (8). The ability to identify patients at highest risk of relapse remains limited, underscoring the need for sensitive, dynamic biomarkers to guide adjuvant treatment or surveillance strategies.

Minimal residual disease (MRD) measurement by means of circulating tumor DNA (ctDNA) is a promising minimally invasive method to capture subclinical disease before overt relapse (9). Currently, two primary approaches are used: tumor-informed, which uses personalized sequencing based on tumor-specific mutations, and tumor-agnostic which relies on preestablished panels (10, 11). Whereas tumor-informed assays often offer greater sensitivity, they require tumor tissue, longer turnaround times, and may miss emerging subclones. Tumor-agnostic approaches are more scalable and avoid dependence on tissue samples. In this study, we assess the clinical utility of a tumor-agnostic MRD assay in the NADIM II trial for identifying early relapses.

Materials and Methods

Study cohort

MRD was measured in samples obtained from patients of the NADIM II trial (NADIM II ClinicalTrials.gov identifier, NCT03838159; EudraCT number, 2018-004515-45.). This is a multicenter, randomized phase II trial. The study was conducted in accordance with the Declaration of Helsinki, Good Clinical Practice guidelines, and all applicable regulatory requirements. The study protocol was approved by the Ethics Committee of Hospital Puerta de Hierro Majadahonda and the Spanish Agency of Medicines and Medical Devices. Written informed consent from all patients was obtained. Clinical outcomes of the NADIM II trial have been already published (5).

Briefly, eligible patients were adults (≥18 years) with untreated, pathologically confirmed NSCLC, stage IIIA/B (American Joint Committee on Cancer eighth edition criteria), and an Eastern Cooperative Oncology Group (ECOG) performance status of 0 or 1. Tumor resectability was assessed by a multidisciplinary tumor board. Patients whose tumors were positive for EGFR sensitizing mutations or ALK rearrangements were excluded.

Patients were randomly assigned in a 2:1 ratio to receive neoadjuvant nivolumab (360 mg) plus chemotherapy (paclitaxel 200 mg/m² and carboplatin area under the curve 5, 5 mg/mL per minutes), once on day 1 of each 21-day cycle, for three cycles (experimental group) or chemotherapy alone (control group), followed by surgery. Patients in the experimental group who had R0 resections also received adjuvant treatment with nivolumab at a dose of 480 mg once every 4 weeks for 6 months. Patients in the control group who underwent surgery were followed by three observation visits.

The primary end point was the pCR rate, defined as a complete absence of viable tumor cells in the primary tumor and surgically removed lymph nodes after neoadjuvant treatment, as determined by blinded independent central review.

Exploratory objectives included ctDNA analysis and evaluation of its potential as a prognostic factor.

Plasma samples were collected before and after neoadjuvant treatment. ctDNA analysis of plasma samples before and after neoadjuvant treatment has been published elsewhere (5). In addition, plasma samples were collected after surgery as well as at 3 and 6 months of adjuvant treatment (experimental arm) or follow-up (control arm).

Tumor response was evaluated with positron emission tomography–computed tomography (CT) and CT, according to RECIST, version 1.1, before and after neoadjuvant treatment, after surgery, at 3 and 6 months during adjuvant treatment, every 3 months during the first 2 years of follow-up, and every 6 months thereafter.

Laboratory procedures

A total of 83 samples from 60 patients were valid for MRD assessment, with 58 samples collected after surgery and 25 samples collected during adjuvant treatment (experimental arm) or follow-up (control arm; Supplementary Fig. S1). Peripheral whole-blood samples were collected in two 8.5-mL PPT tubes (Becton Dickinson) after surgery at 3 and 6 months of adjuvant treatment or follow-up. Plasma was separated from the cellular fraction by two consecutive centrifugations at 1,600 g for 10 minutes and at 6,000 g for 10 minutes. Samples were then divided into 4 aliquots of 2.0 mL. Samples were stored at −80°C until analysis.

Frozen samples were sent to Guardant Health, Inc. in three batches. MRD was analyzed using the Guardant Reveal assay, a tissue-free epigenomic assay intended for detection and quantification of ctDNA. As previously published (12), up to 30 ng of cell-free DNA was physically partitioned based on methylation status using a nondestructive method, tagged with molecular identifiers, PCR-amplified, and sequenced on the Illumina NovaSeq platform. More than 20,000 differentially methylated regions (DMR) were covered, including those in both lung and other cancers along with control regions. A proprietary bioinformatics software pipeline was trained to detect the presence or absence of ctDNA based on these DMRs, and the resultant output was either “ctDNA detected” or “ctDNA not detected.” Tumor fraction (TF) was estimated based on DMRs. The limit of quantitation for this assay is 0.01%. As these samples were run on the Guardant Reveal assay powered by the Guardant Infinity platform, samples with ctDNA detected were then queried for the presence of somatic alterations. Variants passing quality control (QC) filters in a predefined panel of genes were reported.

Statistical analysis

The median follow-up time was estimated using the reverse Kaplan–Meier (KM) method. The clinical database was locked on November 30, 2022. Event-free survival (EFS) and OS were calculated using the KM method, along with their corresponding 95% CIs, in the ITT population, defined as all randomized patients who received at least one cycle of neoadjuvant treatment.

OS was defined as the time from the initiation of neoadjuvant treatment to death from any cause. EFS was defined as the time from the start of neoadjuvant treatment to the occurrence of disease relapse, assessed per RECIST criteria version 1.1, or death from any cause, whichever occurred first. As the main objective of the present study was to evaluate the potential of MRD measurements to predict tumor relapses and because there were events unrelated to cancer, such as COVID-19–related deaths, EFS and OS were analyzed considering only cancer-related events by censoring patients at the time of non–cancer-related deaths. Events not related to cancer are listed in Supplementary Table S1.

Associations between MRD status (positive vs. negative) and categorical clinical variables, including pathologic response, were assessed using the χ² test or Fisher exact test, depending on the number of individuals per category.

The associations between MRD detection and survival outcomes were assessed using Cox proportional hazards models. The proportional hazards assumption was tested using Schoenfeld residuals. KM curves and log-rank tests were used to evaluate statistical differences between groups (MRD-positive and -negative).

To assess the added prognostic value of MRD beyond pCR, we calculated likelihood ratio (LR) statistics. First, a univariate Cox model including only pathologic response as a covariate was fitted. Then, MRD status was added as a second variable, and the significance of the improvement in model fit was tested using the LR test. A significant increase in the LR statistic indicated that MRD provided additional predictive information beyond pCR alone. This approach was applied separately for both EFS and OS endpoints.

All P values of <0.05 were considered statistically significant (two-sided).

Results

MRD detection

Among the 73 patients who underwent surgical resection in the NADIM II trial, 60 patients (82.2%) had at least one sample available for MRD analysis. A total of 86 samples were analyzed across three independent molecular batches, of which 97.5% (83/86) samples met the quality criteria and were deemed valid for downstream analysis. MRD was measured in 58 plasma samples collected after surgery (baseline) and in 25 cases during adjuvant treatment or observation (Supplementary Fig. S1). The clinical characteristics of the patient cohort are summarized in Table 1, comprising 45 patients in the experimental arm and 15 patients in the control arm.

Table 1.

Baseline clinical and demographic characteristics of patients included in the study (n = 60).

Baseline characteristics	All patients (n = 60)
Age median (range) years	64 (28–82)
Sex (%)	​
Male	34 (56.7%)
Female	26 (43.3%)
BMI category	​
Overweight (25–30)	30 (50%)
Normal (18.5–25)	17 (28.3%)
Obesity (>30)	12 (20%)
Underweight (<18.5)	1 (1.7%)
Smoking status	​
Smoker	37 (61.7%)
Former smoker	19 (31.7%)
Never smoker	4 (6.7%)
Unio Internationale Contra Cancrum (UICC) stage	​
IIIA	48 (80%)
IIIB	12 (20%)
Histology	​
Adenocarcinoma	27 (45%)
Squamous	23 (38.3%)
Other	10 (16.7%)
Treatment arm	​
Experimental	45 (75%)
Control	15 (25.0%)

MRD positivity was assessed using an epigenomic, methylation-based assay. MRD results for all samples are provided in Supplementary Data S1. MRD was detected in 8 of 83 valid samples, corresponding to seven individual patients, yielding an overall positivity rate of 9.6%. Detection rates were consistent across batches: 10.0% in batch 1, 10.4% in batch 2, and 8% in batch 3. Six of 58 (10.3%) samples collected after surgery were MRD-positive, whereas 2 of 25 (8%) were MRD-positive during adjuvant treatment. The median TF was 0.08% (minimum–maximum: 0.05%–2.99%). The median TF was 0.07% at baseline and 0.45% during follow-up.

Somatic variants were identified in seven the eight samples. The molecular alterations detected on ctDNA are presented in Table 2, with TP53 being the most frequently mutated gene—altered in all samples harboring somatic mutations. Two of these samples presented comutations: one in ERBB2, and the other in TERT, NFE2L2, and PIK3CA. Notably, among these seven patients, three had matched tumor tissue and blood samples collected prior to initiation of neoadjuvant therapy. In one of these cases, the same TP53 variant [c.480_481delinsCT; p.(Met160_Ala161delinsIleSer)] was identified in both pretreatment tumor tissue and the ctDNA sample obtained after surgery, supporting the tumor origin of the detected mutation.

Table 2.

Molecular characteristics of MRD-positive cases.

Case	Arm	Pathologic response	Time point for MRD +	Methylation profile	Somatic mutation	Gene	Transcript	Mutation
Case A	Experimental	Major	After surgery	Yes	No	​	—	​
Case B	Control	Incomplete	6 months	Yes	Yes	TP53	NM_000546.6	c.461G>T; p.(Gly154Val)
Case C	Experimental	Incomplete	After surgery	Yes	Yes	TP53	NM_000546.6	c.839G>C; p.(Arg280Thr)
​	​	​	​	​	​	ERBB2	NM_004448.4	c.2324_2325insCTCCGTGATGGC; p.(Ala775_Gly776insSerValMetAla)
Case D	Control	Incomplete	After surgery + 6 months	Yes	Yes	TERT	NM_198253.3	c.-124C>T
​	​	​	​	​	​	TP53	NM_000546.6	c.535C>T; p.(His179Tyr)
​	​	​	​	​	​	NFE2L2	NM_001313902.2	c.101G>C; p.(Arg34Pro)
​	​	​	​	​	​	PIK3CA	NM_006218.4	c.1633G>A; p.(Glu545Lys)
Case E	Experimental	Incomplete	After surgery	Yes	Yes	TP53	NM_000546.6	c.480_481delinsCT; p.(Met160_Ala161delinsIleSer)
Case F	Experimental	Major	After surgery	Yes	Yes	TP53	NM_000546.6	c.722C>G; p.(Ser241Cys)
​	​	​	​	​	​	TP53	NM_000546.6	c.532del; p.(His178fs)
Case G	Control	Incomplete	After surgery	Yes	Yes	TP53	NM_000546.6	c.722C>T; p.(Ser241Phe)

MRD-positive samples corresponded to three patients (20%) in the control arm and four patients (8.9%) in the experimental arm. There were no statistically significant differences in MRD status according to clinical characteristics at baseline, except for tumor stage at diagnosis (P = 0.025). Specifically, only 6.2% of patients with stage IIIA were MRD-positive, whereas this percentage increased to 33.3% in patients diagnosed with stage IIIB disease.

Prognostic value of MRD

None of the patients with MRD-positive detection at any time point were diagnosed with a pCR after neoadjuvant treatment. On the other hand, all patients with pCR were MRD-negative (Fig. 1). Detection of MRD at any time point (either after surgery or during adjuvant treatment/observation) was significantly associated with inferior EFS and OS, with hazard ratios (HR) of 10.2 (95% CI, 3.7–28.3) for EFS and 10 (95% CI, 2–49.9) for OS (Fig. 2). KM curves for each treatment arm are presented in Supplementary Fig. S2.

Figure 1.

Association of pCR with MRD detection in the whole cohort (A) and in the subset including only cases in which MRD was called negative in at least two consecutive plasma samples (B).

Figure 2.

KM curves for EFS (left) and OS (right), according to MRD status (n = 60).

All patients who were identified as MRD-positive at any time point eventually experienced relapse. Specifically, three patients developed distant relapse (involving the bone and suprarenal gland), three others had local relapse, and one patient experienced both distant (bone) and local relapses. The median EFS for MRD-positive patients was 14.1 (95% CI, 12.4–Not reached) months, whereas it was not reached in the MRD-negative population. The median time from MRD detection to disease relapse was 7.8 months (range, 1.1–10 months).

When restricting the analysis to MRD-negative cases with at least two consecutive MRD-negative plasma samples (n = 25), we found that all MRD-negative patients were alive at the data cutoff [HR: not estimable (NE), P < 0.001], with only one relapse reported (HR: 41.6; 95% CI, 5–348.8; Fig. 3). This single relapse case involved a 48.6-year-old male smoker diagnosed with stage IIIA adenocarcinoma NSCLC, PD-L1 <1%, and ECOG performance status 0. The patient received neoadjuvant chemoimmunotherapy, followed by surgical resection and 6 months of adjuvant immunotherapy (experimental arm). He had an incomplete pathologic response, and he ultimately experienced disease relapse after 15.1 months, consisting of a single brain metastasis. The patient remained alive at the data cutoff, with an OS of 26.9 months. For this case, MRD analysis was performed on two plasma samples: the first was collected 8.3 months prior to the diagnosis of relapse, and the second 2.3 months before the relapse event.

Figure 3.

KM curves for EFS (left) and OS (right) based on MRD status, restricted to patients with at least two consecutive MRD-negative plasma samples (n = 25).

We previously showed in the NADIM II trial that patients who had a pCR had excellent survival outcomes, as they all were alive and with no evidence of disease (5). However, some patients with major or incomplete pathologic responses also demonstrated long-term survival. To evaluate whether MRD status could provide additional prognostic information beyond pathologic response alone, we compare the independent contribution of each variable (pathologic response and MRD status) using the LR statistic. This approach allowed us to evaluate whether adding MRD status to a model already including pathologic response improved its ability to predict survival outcomes. MRD status (measured in at least two samples) significantly enhanced model performance, contributing a substantial increase in the LR statistic beyond pCR for both EFS (P < 0.001) and OS (P = 0.015; Supplementary Fig. S3). Among non-pCR patients, MRD measured after surgery or during adjuvant therapy remained a significant prognostic marker (HR: 6.2; 95% CI, 2.2–17.1 and HR: 6.5; 95% CI, 1.3–32.5, for EFS and OS respectively; Supplementary Fig. S4). All non-pCR patients with MRD-negative results at both time points were alive (HR: NE, P = 0.025), with only one relapse (HR: 19.9; 95% CI, 2.4–165.6), corresponding to the aforementioned patient with a solitary brain metastasis (Fig. 4).

Figure 4.

KM curves for EFS (left) and overall survival (OS) (right) based on MRD status, restricted to patients with at least two consecutive MRD-negative plasma samples in the non-PCR population.

Discussion

A growing body of literature has evidenced the potential of ctDNA-based MRD assessment in early-stage and locally advanced NSCLC. Currently, two main approaches are used: tumor-informed and tumor-naïve (or agnostic) strategies (10, 13). Several studies have demonstrated that tumor-informed MRD detection can identify recurrence months before radiographic progression, providing a valuable window for early therapeutic intervention (11, 14, 15). Although tumor-informed assays generally offer enhanced sensitivity by tracking patient-specific mutations, they require prior sequencing of the tumor tissue, which may increase turnaround time and raise costs. Additionally, it has been reported that MRD panels can be developed for most—but not all—patients, generally due to insufficient/inadequate material of the solid tumor sample (16). Moreover, tumor-informed approaches may not fully capture tumor heterogeneity (10, 13). On the other hand, tumor-naïve assays can identify clonal variants that emerge over time, do not require tissue biopsies, and generally offer shorter turnaround times, but they are potentially limited by sensitivity.

We previously demonstrated the prognostic value of pretreatment ctDNA and ctDNA clearance following neoadjuvant chemoimmunotherapy but before surgery in the NADIM clinical trials (5, 7, 17). In the present study, we expand these findings by demonstrating that MRD detection, using a tumor-naïve, epigenomic ctDNA assay, is a significant prognostic marker in resectable stage III NSCLC. MRD detection after surgery or during adjuvant therapy was strongly associated with inferior EFS and OS, supporting its potential as a dynamic biomarker of recurrence risk.

One of the most promising clinical applications of MRD assessment in NSCLC is the de-escalation of adjuvant therapy. However, although the prognostic value of several MRD assays has been demonstrated in the context of clinical trials (18, 19), first-generation MRD assays lack the analytic sensitivity required to confidently support this approach (20). To overcome this limitation, phased variant enrichment and detection sequencing, which identifies multiple somatic mutations in individual DNA fragments, has shown detection limits below 1 part per million (21, 22), significantly enhancing analytic sensitivity and potentially enabling MRD-guided treatment de-intensification strategies. Nevertheless, despite the improvements in the limit of detection, a subset of patients with undetectable MRD still experience relapse, as reported by Isbell and colleagues (22). Likewise, minor allele enrichment sequencing through recognition oligonucleotides (MAESTRO), which combines massively parallel mutation enrichment with duplex sequencing, has shown potential for accurate MRD testing. Unlike conventional hybrid-capture duplex sequencing, which uses long probes to capture mutant and wild-type alleles equally, MAESTRO uses short probes to enrich for tumor-specific mutant alleles and detects the same mutant duplexes using up to 100-fold fewer reads (23, 24). In the same way, integrated predictive models that combine tumor genomic features, radiomics, and ctDNA dynamics have demonstrated superior performance in predicting survival and recurrence, offering a more comprehensive framework for personalized treatment strategies (25).

Another straightforward strategy to improve the sensitivity of MRD detection is to increase the number of plasma samples analyzed per patient (20). In our study, the inclusion of at least two consecutive MRD-negative samples significantly enhanced the prognostic value of MRD, despite decreasing the overall sample size. Notably, all patients with two MRD-negative samples remained alive, and only one patient relapsed, which manifested as a single brain metastasis, underscoring the potential value of serial MRD monitoring in refining risk assessment.

In the NADIM II trial, all patients who had a pCR (defined as the absence of any viable tumor cells in both the primary tumor bed and resected lymph nodes) were alive and with no evidence of disease (5). However, in the previous NADIM trial, eight non-pCR patients—three with incomplete and five with major pathologic responses—also remained alive and without evidence of disease at 5 years of follow-up (7). These findings suggest that whereas pCR is a strong prognostic marker, it does not fully capture the spectrum of long-term outcomes, and some patients with residual viable tumor may still have durable survival. In this way, a standardized scoring system to assess pathologic response has shown that each 1% residual viable tumor in the primary tumor and lymph nodes was associated with a 0.017 HR increase for EFS (26), showing that a cutoff of 5% correlated with favorable outcomes using samples for CheckMate 816 trial (3). These findings suggest that pathologic response is best viewed as a continuum, rather than a binary outcome. Nonetheless, standardization and concordance for these assessments across institutions remains challenging (27). In this context, our study demonstrates that MRD assessment adds significant prognostic value for both EFS and OS. Importantly, MRD negativity identified a subset of non-pCR patients who remained relapse-free and alive, suggesting that MRD may serve as a complementary biomarker capable of refining risk stratification beyond pCR alone.

It is important to acknowledge that NADIM II was not originally designed to evaluate MRD, and the results presented here should be interpreted with caution. The sample size for MRD-evaluable patients was modest, and subgroup analyses—particularly those restricted to non-pCR or serially tested patients—may be underpowered. In addition, although our MRD assay was tumor-agnostic and epigenome-based, its performance may not be generalizable to other platforms with different methodologic characteristics. Despite these limitations, this study benefits from a prospective, multicenter design with high-quality clinical annotation and longitudinal follow-up.

In our analysis, serial MRD testing enhanced risk stratification beyond pCR. Validation of MRD-guided therapeutic strategies in non-pCR patients is therefore of particular interest. In advanced NSCLC, a recent nonrandomized controlled trial evaluated ctDNA-guided de-escalation of tyrosine kinase inhibitor therapy following local consolidative treatment. This adaptive approach achieved substantial treatment-free intervals, supporting the feasibility of MRD-guided de-escalation in selected patients (28). Although these findings are encouraging, further studies are required to determine the optimal timing and frequency of MRD assessment needed to safely guide treatment de-escalation.

It is important to acknowledge that in the NADIM trial (2), adjuvant treatment was administered for 1 year, whereas in NADIM II (5) and other trials, the duration was reduced to 6 months. Thus far, treatment duration has been empirically determined; however, our data suggest that MRD testing—at least in two plasma samples—may help identify patients with a favorable prognosis who may not need to extend treatment for 1 year. Nevertheless, requiring multiple consecutive negative MRD assays for accurate prognostication may limit the feasibility of using these tests for adaptive trials. Beyond improving treatment-related toxicity, de-escalation could also reduce economic toxicity. Noninferiority studies would also be of particular interest in the non-pCR population. In this regard, the DYNAMIC study in colorectal cancer demonstrated that MRD-guided de-escalation significantly reduced chemotherapy use without compromising recurrence-free survival, thereby meeting its noninferiority endpoint (29). However, such trials would require substantially larger cohorts, potentially involving hundreds of patients.

In summary, MRD detection using a tumor-agnostic assay provides prognostic information beyond pathologic response in resectable stage III NSCLC. Serial MRD negativity identifies patients without pCR who experience long-term outcomes comparable with those with tumors demonstrating pCR. These findings warrant prospective validation of MRD-guided strategies to optimize postoperative management in locally advanced NSCLC undergoing perioperative chemoimmunotherapy.

Supplementary Material

Supplementary Data S1

MRD results for all samples

Supplementary Figure S1

Supplementary Figure 1. Study design and MRD sampling scheme of the NADIM II trial. Patient numbers shown for each treatment arm indicate the number of patients for whom MRD analysis was performed.

Supplementary Figure S2

Supplementary Figure 2. Kaplan–Meier curves for event-free survival (EFS) and overall survival (OS) according to minimal residual disease (MRD) status. (A) EFS in the experimental arm. (B) OS in the experimental arm. (C) EFS in the control arm. (D) OS in the control arm. MRD positivity was defined as MRD detection at any time point. Hazard ratios (HR) with 95% confidence intervals were estimated using Cox proportional hazards models, and P values were calculated with the log-rank test.

Supplementary Figure S3

Supplementary figure 3. Incremental prognostic value of MRD status beyond pathological response in multivariable models. Likelihood ratio (LR) tests show the added predictive contribution of MRD (defined as detection in at least two samples) when incorporated into models including pathological response for (A) event-free survival (EFS) and (B) overall survival (OS). P values from likelihood ratio tests are indicated above each plot.

Supplementary Figure S4

Supplementary figure 4. Kaplan–Meier curves for (A) event-free survival (EFS) and (B) overall survival (OS) in non-pCR patients according to MRD status. MRD positivity was defined as detection at any time point. Hazard ratios (HRs) and 95% confidence intervals were estimated with Cox proportional hazards models, and P values were derived from log-rank tests, as indicated in each plot. Abbreviations: pCR, pathological complete response; MRD, minimal residual disease.

Supplementary Table S1

Supplementary table 1. Non–cancer-related deaths, including treatment arm, surgical resection status, pathological response, and reported cause of death.

Data Availability

MRD results for all samples are provided in Supplementary Data S1. De-identified clinical data will be made available upon reasonable request through the corresponding authors, requiring the approval of the Spanish Lung Cancer Group and the institutional Ethics Committee. Data behind the figures are available upon request to the corresponding author.

Authors’ Disclosures

M. Provencio reports grants, personal fees, and nonfinancial support from Bristol Myers Squibb, Roche, AstraZeneca, and Takeda outside the submitted work. R. Serna-Blasco reports other support from MSD and grants from Johnson & Johnson outside the submitted work. E. Nadal reports personal fees from AbbVie, Amgen, AstraZeneca, Apollomics, BeOne Medicines, Daiichi-Sankyo, Genmab, GSK, Boehringer Ingelheim, Illumina, Lilly, Johnson & Johnson, MSD, PharmaMar, Pierre Fabre, Qiagen, Regeneron, Sanofi, and Takeda and grants and personal fees from Bristol Myers Squibb, Merck Serono, Pfizer, and Roche outside the submitted work, as well as travel support from Roche, Takeda, Johnson & Johnson, and MSD. A. Martinez-Martí reports personal fees and other support from AstraZeneca, Bristol Myers Squibb, F. Hoffmann–La Roche, Merck Sharp & Dohme, MSD Oncology, Pfizer, Amgen, and Lilly outside the submitted work. J. Bosch-Barrera reports personal fees from Regeneron, AstraZeneca, MSD, Bristol Myers Squibb, Pierre Fabre, Roche, Pfizer, Merck, and Takeda outside the submitted work. V. Calvo reports personal fees and other support from Roche and personal fees from AstraZeneca, MSD, Bristol Myers Squibb, Takeda, Regeneron, Amgen, GSK, Boehringer Ingelheim, Johnson & Johnson, BeOne Medicines, and Pierre Fabre outside the submitted work. A. Insa Molla reports other support from Amgen, Regeneron, Roche, Takeda, AstraZeneca, MSD, and Pfizer outside the submitted work. N. Reguart reports personal fees from AbbVie, Amgen, ArriVent Biopharma, AstraZeneca, Bayer, Bristol Myers Squibb, Boehringer Ingelheim, Gilead, Harpoon, Janssen, Johnson & Johnson, Merck, MSD, Novartis, Pharmamar, Regeneron, Revolution Medicine, Roche, Summit Therapeutics, BeOne Medicines, and Tubulis outside the submitted work. J. de Castro Carpeño reports grants and personal fees from AstraZeneca, Bristol Myers Squibb, Merck Sharp & Dohme, and F. Hoffmann–La Roche and personal fees from Bayer, Boehringer Ingelheim, GSK, Janssen–Cilag, Lilly, Novartis, Pfizer, Takeda, and Sanofi outside the submitted work. C. Aguado de la Rosa reports personal fees and nonfinancial support from MSD and Pierre Fabre; nonfinancial support from Roche and Pharmamar; and personal fees from AstraZeneca, Pfizer, Regeneron, Janssen, Takeda, Bristol Myers Squibb, Amgen, and Daichi Sankyo outside the submitted work. R. Palmero Sanchez reports personal fees from AstraZeneca, Guardant Health, and Pfizer and nonfinancial support from Merck Sharp & Dohme outside the submitted work. J. Martín-López reports grants from Roche outside the submitted work. P. Mediavilla-Medel reports grants from Fellowship Insitituto de Salud Carlos III during the conduct of the study. M. Lázaro reports personal fees and nonfinancial support from Pierre Fabre, Ipsen, AstraZeneca, Roche, Merck, and MSD; nonfinancial support from Bayer; and personal fees from BeOne Medicines and Bristol Myers Squibb outside the submitted work. J. Hayes reports personal fees from Guardant Health during the conduct of the study and outside the submitted work. B. Massutí reports personal fees from Bristol Myers Squibb, Johnson & Johnson, and Roche; nonfinancial support from MSD; and personal fees and nonfinancial support from AstraZeneca outside the submitted work. A. Romero reports nonfinancial support from Guardant Health, Inc. and other support from Bristol Myers Squibb during the conduct of the study, as well as other support from Takeda; personal fees from AstraZeneca, Johnson & Johnson, Menarini, and Thermo Fisher Scientific; and grants from EMQN, Johnson & Johnson, and Health In Code outside the submitted work. No disclosures were reported by the other authors.

Authors’ Contributions

M. Provencio: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing. R. Serna-Blasco: Data curation, software, formal analysis, methodology. E. Nadal: Resources, data curation, validation, investigation. J.L. González Larriba: Resources, validation, visualization, writing–review and editing. A. Martínez-Martí: Resources, validation, writing–review and editing. R. Bernabé Caro: Validation, visualization, writing–review and editing. J. Bosch-Barrera: Data curation, validation, investigation, visualization. V. Calvo: Resources, data curation, validation, investigation, visualization, writing–review and editing. A. Insa Molla: Validation, investigation, methodology. N. Reguart: Validation, investigation, writing–review and editing. J. de Castro Carpeño: Validation, investigation, writing–review and editing. C. Aguado de la Rosa: Validation, investigation, writing–review and editing. R. Palmero Sanchez: Validation, Investigation, writing–review and editing. F. Hernando Trancho: Validation, investigation, writing–review and editing. J. Martín-López: Validation, investigation, writing–review and editing. A. Rodríguez-Festa: Validation, investigation, writing–review and editing. P. Mediavilla-Medel: Validation, investigation, writing–review and editing. M. Lázaro: Validation, investigation, writing–review and editing. J. Hayes: Resources, formal analysis. A. Cruz-Bermúdez: Resources, data curation, writing–review and editing. B. Massutí: Formal analysis, validation, investigation, methodology, writing–review and editing. A. Romero: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, visualization, writing–original draft, project administration, writing–review and editing.