Prognostic performance of a novel multiplex circulating tumor DNA assay for lung cancer outcomes

Abstract

Background:

Lung cancer is a heterogeneous disease with unpredictable trajectories. Risk stratification in lung cancer remains imprecise within tumor-node-metastasis (TNM) stages.

Objectives:

This work aims to evaluate whether methylated circulating tumor DNA (ctDNA) can improve prognostication at the time of diagnosis.

Methods:

We enrolled 213 participants with recently diagnosed lung cancer, of whom 100 underwent curative treatment and 113 underwent palliative treatment. All blood samples were collected prior to lung cancer diagnosis and analyzed using multiplex methylation-specific digital droplet polymerase chain reaction targeting five methylated ctDNA markers. ctDNA output was interpreted in two models: a three-tiered (negative/Low-level/High-level ctDNA) and a binary (negative/positive).

Results:

We observed significantly increased overall survival (OS) among all participants with negative or low-level ctDNA results compared with those with high-level ctDNA. The adjusted hazard ratio (HR) for OS was higher when comparing high-level ctDNA to negative results (HR = 1.96, 95% confidence interval (CI): 1.09–3.51; p = 0.024). In the palliative treatment subgroup, HR for progression-free survival was higher comparing high-level ctDNA to negative results (adjusted HR = 2.32, 95% CI: 1.10–4.92; p = 0.028). In the curative treatment subgroup, the association between ctDNA levels and recurrence-free survival was not statistically significant (adjusted HR = 1.81, 95% CI: 0.73–4.48; p = 0.200).

Conclusion:

Detection of, and risk stratification based on, levels of methylated ctDNA at diagnosis appears to provide valuable prognostic information in lung cancer, both in early and late stages of the disease. Such information may have implications for the choice of follow-up regimens and treatment strategies.

Design:

Retrospective observational cohort study using prospectively collected blood samples.

Introduction

Lung cancer remains a leading cause of cancer-related mortality worldwide. According to GLOBOCAN 2022 estimates, there were approximately 2.5 million new cases and 1.8 million deaths attributed to lung cancer, accounting for 18% of all cancer deaths globally. ¹ Prognosis in lung cancer is closely linked to the tumor-node-metastasis (TNM) stage at diagnosis. The 5-year survival rate for localized lung cancer is approximately 64%, but this rate declines to about 9% for patients with distant metastases. ² Although TNM staging remains the most important prognostic factor, lung cancer is a heterogeneous disease with unpredictable clinical trajectories. This variability persists even among patients in the same stage and histopathological subgroup, despite accounting for actionable driver mutations, predictive biomarkers such as PD-L1 expression, or clinical markers such as comorbidity and Eastern Cooperative Oncology Group (ECOG) performance status.^3–6 Treatment strategies for lung cancer are guided by TNM stage and overall clinical status at diagnosis. Localized disease is typically managed with curative intent through surgical resection or stereotactic body radiotherapy, sometimes combined with adjuvant systemic therapies. ⁷ Regional disease is treated with definitive chemoradiotherapy, sometimes followed by adjuvant immunotherapy. ⁷ In contrast, metastatic disease is typically managed with systemic therapy, with palliative intent, using chemotherapy, immunotherapy, or targeted agents.^8,9

In recent years, liquid biopsies have emerged as a promising tool for the detection and prognostication of lung cancer.^10,11 One such approach is to analyze circulating tumor DNA (ctDNA) in the bloodstream to identify genetic and epigenetic alterations associated with cancer.

Detection of ctDNA generally falls into two categories: tumor-informed methods, which are guided by known genomic alterations, and tumor-agnostic methods, which do not require prior tumor profiling. The tumor-informed method involves identifying patient-specific alterations and using them to design a personalized assay for ctDNA detection in plasma. ¹² While these assays offer high sensitivity and specificity, they require access to tumor tissue and are resource-intensive, limiting their applicability in early detection and screening settings. In lung cancer, this approach has been studied for minimal residual disease (MRD) detection and disease monitoring.^13–16

Conversely, the tumor-agnostic approach employs standardized panels that target common genomic and epigenomic alterations, without requiring prior knowledge of a patient’s tumor. ¹² These assays are faster and more cost-effective than tumor-informed approaches, making them suitable for cancer screening, diagnostic work-up, and early prognostic assessment. Moreover, because they do not depend on tumor tissue, they are ideal when biopsy samples are limited or inaccessible. However, tumor-agnostic assays may have lower sensitivity and specificity and a higher risk of false positives compared with the tumor-informed approaches. ¹⁶

Aberrant DNA methylations are epigenetic modifications occurring early in carcinogenesis. These alterations can be detected in ctDNA, making them promising biomarkers for early lung cancer detection. ¹⁰ Moreover, several studies have highlighted the potential of methylated ctDNA detection in liquid biopsies as a prognostic biomarker.^17–20 Accordingly, we recently developed a novel tumor-agnostic multiplex assay for detecting methylated ctDNA. This assay targets five regions identified as being hypermethylated in lung cancer tissue (HOXA9, OTX1, MCIDAS, SP9, and TFAP2B) and is intended for application across disease stages, from early detection to patient monitoring. ²¹

In this study, we evaluated the prognostic value of our tumor-agnostic ctDNA multiplex assay in patients with lung cancer, using blood samples collected immediately prior to diagnosis. Specifically, we performed survival analyses comparing overall survival (OS), recurrence-free survival (RFS), and progression-free survival (PFS) by ctDNA status in individuals diagnosed with lung cancer across stages I–IV.

Materials and methods

Study design and participants

The present study consecutively enrolled patients evaluated for and diagnosed with lung cancer at the fast-track lung cancer clinic at Lillebaelt Hospital, Vejle, University Hospital of Southern Denmark, between May 2022 and April 2023. The study was approved by the Regional Committee on Health Research Ethics of Southern Denmark (S-20220014), ensuring compliance with relevant laws and institutional guidelines. The study was conducted in accordance with the 1964 Declaration of Helsinki. All participants provided written and oral informed consent to enroll in this study. The inclusion criteria were aged 18 years or over, a histologically confirmed lung cancer diagnosis after diagnostic work-up, and informed written consent for study enrollment. Patients were excluded if they were diagnosed with other malignancies during the period from 2 years before to 1 year after study enrollment. Exceptions were made for non-melanoma skin cancer and cervical carcinoma in situ. Patients with severe comorbidities prohibiting diagnostic procedures were also excluded. Cancer status before enrollment and during follow-up was assessed using the Danish National Pathology Data Bank, which records all cancer diagnoses in Denmark. ²²

Two hundred and fifty participants met the inclusion criteria. Thirty-seven were excluded, resulting in a study cohort of 213 participants (Figure 1). For subgroup analyses, patients were classified by treatment intent: curative or palliative.

Figure 1.

Enrollment of study population.

This prognostic biomarker cohort study was reported in accordance with the REMARK guideline (Table S1). ²³

Diagnosis and follow-up

At the first hospital visit of the diagnostic work-up, blood samples were collected in BD Vacutainer^® K2E tubes (BD, Franklin Lakes, NJ, USA) prior to invasive diagnostic procedures. Plasma was isolated within 4 h of sampling and stored at −80°C until analysis. Diagnostic work-up was done in accordance with Danish national guidelines; CT scan of the thorax and abdomen, supplemented by whole-body positron emission computed tomography. ²⁴ Thoracentesis, endobronchial ultrasound-guided transbronchial needle aspiration, and ultrasound- or CT-guided needle biopsy were performed as indicated. A multidisciplinary team assessed clinical, radiological, and histopathological findings. Lung cancer staging was performed according to the eighth edition of the TNM classification of the International Association for the Study of Lung Cancer. After diagnosis, participants were treated in accordance with national guidelines. Participants were followed for at least 23.5 months, with the longest follow-up being 34.4 months. The primary endpoint was a composite of death, disease recurrence, and disease progression. RFS and OS were analyzed in participants receiving curative treatment (curative treatment subgroup). PFS and OS were analyzed in participants receiving palliative treatment (palliative treatment subgroup).

Multiplex droplet digital polymerase chain reaction

The methylation-specific droplet digital polymerase chain reaction (ddPCR) assay was performed as previously described. ²¹ In brief, 4 mL of plasma was thawed, an exogenous spike-in DNA fragment was added as an extraction control, and processed for cfDNA extraction using the DSP Circulating DNA Kit on the QIAsymphony SP (Qiagen, Hilden, Germany). Extraction efficiency, total cfDNA yield, and genomic DNA contamination were assessed as described in do Canto et al. ²⁵ Extracted DNA was concentrated, bisulfite-converted using the EZ DNA Methylation-Lightning Kit (Zymo Research, Irvine, CA, USA), and analyzed in a 6-plex assay using primers and probes specific for methylated loci located in relation to HOXA9, OTX1, MCIDAS, SP9, and TFAP2B with ALB as a reference gene, on the Bio-Rad QX600 droplet digital PCR system (Bio-Rad, Hercules, CA, USA). Detailed information on the CpG islands near OTX1, MCIDAS, SP9, and TFAP2B is provided in Table S2. Positive and negative controls were included as described, ²¹ and all analyses were performed in duplicates. Samples with fewer than 15,000 droplets per reaction were excluded from the final analysis.

The limit of blank (LoB) was set to a fixed specificity of ⩾95% for each of the five methylated ctDNA markers, based on an external cohort of 60 cancer-free, age- and sex-matched individuals undergoing hip or knee replacement surgery or with benign pulmonary conditions. ²¹ A methylated ctDNA marker detected above its LoB was considered positive; if it was equal to or below the LoB, it was considered negative. The multiplex assay was classified as positive if at least one of the five methylated ctDNA markers was positive.

Statistical analysis

Baseline characteristics of the participants were reported using descriptive statistics. Continuous variables were reported as medians with interquartile ranges, and categorical variables as absolute numbers and proportions.

Survival analyses were performed to evaluate the prognostic impact of the ctDNA assay. Patients were stratified by their ctDNA results using two approaches: (1) a binary categorization (negative/positive), where a positive result was defined by the detection of at least one methylated ctDNA marker, ²¹ and (2) a three-tiered categorization based on the ratio of the concentration of methylated ctDNA to that of the reference gene, ALB. For the three-tiered categorization, individuals with a negative ctDNA assay result retained their status, while individuals with a positive ctDNA signal were categorized as low-level or high-level. The 75th percentile of the methylated ctDNA/ALB ratio was used as a cut-off to identify individuals with particularly elevated ctDNA levels. The ratio was defined as:

$$\mathrm{ctDNA}-\mathrm{ratio}=\frac{\left(\begin{array}{l}\left[\mathrm{HOXA}9\right]+\left[\mathrm{MCIDAS}\right]+\\ \left[\mathrm{OTX}1\right]+\left[\mathrm{SP}9\right]+\left[\mathrm{TFAP}2\mathrm{B}\right]\end{array}\right)}{[\mathrm{ALB}]}$$

Survival functions were estimated using Kaplan–Meier curves and compared with the log-rank test, with time at risk commencing on the date of diagnosis and ending on the date of recurrence, progression, death, or censoring, as appropriate. Cox proportional hazards regression models were used to estimate hazard ratios (HRs) with corresponding 95% confidence intervals (CIs) for the association between ctDNA strata and survival outcomes. Potential confounders were identified using directed acyclic graphs. Multivariable Cox regression models included lung cancer stage and previous lung cancer as covariates. The proportional hazards assumption was tested using Schoenfeld residuals and log–log plots. The assumption was violated for the variable “previous lung cancer” in the model using the binary categorization, and this was addressed in the model using stratification. Sensitivity analyses were conducted to evaluate the robustness of the findings. These included adjustments for additional variables suspected to influence survival, specifically performance status and comorbidity. To assess a trend across ctDNA tiers, a test for trend was performed using the multivariable regression model. Statistical significance was set at p < 0.05. Statistical analysis was performed using Stata/BE 18.0 (StataCorp LLC, College Station, TX, USA) and graphs were created in GraphPad Prism 11 (GraphPad Software, San Diego, CA, USA).

Results

Baseline characteristics

Baseline characteristics are presented in Table 1. Two hundred and thirteen patients were enrolled in the study. Following ctDNA analysis, 57 participants were negative, while 107 and 49 participants were low-level and high-level, respectively. The negative group comprised more females than the low-level and high-level groups. ECOG performance status was marginally better in the negative group compared with the other two groups. A history of lung cancer was more common in the negative group, less common in the low-level group, and rare in the high-level group. The distribution of age, body mass index, smoking history, Medical Research Council scores, and comorbidity appeared similar across the three groups.

Table 1.

Baseline characteristics (N = 213).

Clinical characteristic	All participants	High-level a	Low-level a	Negative a
	N = 213	N = 49	N = 107	N = 57
Participants characteristics
Age, median years (IQR)	72 (66–77)	71 (64–77)	72 (67–77)	72 (64–77)
Sex, male	115 (54.0%)	28 (57.1%)	64 (59.8%)	23 (40.4%)
BMI, median (IQR)	25.2 (22.5–29.0)	25.1 (22.6–28.8)	25.2 (22.5–28.4)	25.2 (22.4–29.5)
Smoking status
Ever	200	48 (98.0%)	100 (93.5%)	52 (91.2%)
Never	13	1 (2.0%)	7 (6.5%)	5 (8.8%)
Pack years, median (IQR)	50 (25–52)	40 (30–55)	40 (25–55)	34 (23–45)
ECOG performance status
0	104 (48.8%)	21 (42.9%)	51 (47.7%)	32 (56.1%)
1	68 (31.9%)	15 (30.6%)	36 (33.6%)	17 (29.8%)
⩾2	41 (19.2%)	13 (26.5%)	20 (18.7%)	8 (14.0%)
MRC-score
1	90 (42.2%)	20 (40.8%)	45 (42.1%)	25 (42.9%)
2	73 (34.3%)	19 (38.8%)	34 (31.8%)	20 (35.1%)
⩾3	50 (23.5%)	10 (20.4%)	28 (26.2%)	12 (21.1%)
Any comorbidity	160 (75.5%)	32 (65.3%)	86 (80.4%)	42 (73.7%)
Previous lung cancer	14 (6.6%)	1 (2.0%)	13 (12.3%)	16 (28.1%)
Years since lung cancer (IQR)	4 (2–5)	4	3 (2–5)	4 (3–6)
Tumor characteristics
Largest lung tumor, median cm (IQR)	3.5 (2.0–5.6)	6.3 (2.9–7.5)	3.7 (2.0–5.5)	2.3 (1.7–4.0)
Histopathology
Squamous cell carcinoma	46 (21.6%)	10 (20.4%)	25 (23.4%)	11 (19.3%)
Adenocarcinoma	131 (61.5%)	21 (42.9%)	72 (67.3%)	38 (66.7%)
Small cell carcinoma	19 (8.9%)	17 (34.7%)	2 (1.9%)	0 (0%)
Other b	17 (8.0%)	1 (2.0%)	8 (7.5%)	8 (14.0%)
Stage
I	47 (22.1%)	1 (2.0%)	22 (20.6%)	24 (42.1%)
II	19 (8.9%)	1 (2.0%)	10 (9.4%)	8 (14.0%)
III	54 (25.4%)	15 (30.6%)	22 (20.6%)	17 (29.8%)
IV	93 (43.7%)	32 (65.5%)	53 (49.5%)	8 (14.0%)

Continuous variables are reported with median and IQR, and categorical variables are reported as numbers and percentages. Not all percentages add up to 100% due to rounding.

^aHigh-level, ⩾75th percentile; low-level, <75th percentile; negative, no methylated ctDNA markers detected.

^bNine low differentiated carcinomas, five carcinoid tumors, one large cell neuroendocrine carcinoma, one sarcomatoid, one pleomorphic carcinoma.

BMI, body mass index; ECOG, Eastern Cooperative Oncology Group; IQR, interquartile range; MRC, Medical Research Council dyspnea scale.

Tumors tended to be larger, and lung cancer stage was more advanced in the high-level group. Small cell lung cancer was observed most frequently in the high-level group, while other histopathological subtypes were most common in the negative and low-level groups. One hundred patients received curative treatment, while 113 were in the palliative treatment subgroup. One participant in the curative treatment subgroup had stage IV lung cancer (curatively treated oligometastatic disease), while in the palliative treatment subgroup, 1 patient had stage II lung cancer, and 20 patients had stage III lung cancer. The latter participants were in too poor health to receive curative treatment, and thus only less intensive, palliative treatment was available.

Survival analysis

We assessed the prognostic impact of the ctDNA assay based on the three-tiered categorization: negative, low-level, and high-level. Among all participants, Kaplan–Meier analysis with log-rank test demonstrated a statistically significant difference in OS across ctDNA tiers (p < 0.001, Figure 2). The median OS was 24.9 months in the low-level group, 9.1 months in the high-level group, while median OS was not reached in the negative group. The log-rank test results between tiers are reported in Table S3.

Figure 2.

Kaplan–Meier curves for OS (118 events) in all participants. ctDNA levels: high-level (green), ⩾75th percentile; low-level (red), <75th percentile; negative (blue), no methylated ctDNA markers detected.

ctDNA, circulating tumor DNA; OS, overall survival.

In the curative treatment subgroup, a statistically significant difference in RFS was observed across ctDNA tiers (p < 0.001, Figure 3(a)). The median RFS was not reached in either the negative or low-level groups, whereas it was 14.2 months in the high-level group. OS analysis in this subgroup also showed a statistically significant difference, with the median OS being 18.7 months in the high-level group and not reached in the negative or low-level groups (p = 0.003, Figure 3(b)).

Figure 3.

Survival according to ctDNA level in curative and palliative treatment subgroups. Kaplan–Meier curves with numbers at risk. (a) RFS (38 events) and (b) OS (30 events) in the curative treatment subgroup; (c) PFS (103 events) and (d) OS in the palliative treatment subgroup (88 events). ctDNA levels were defined as high-level (green), ⩾75th percentile; low-level (red), <75th percentile; and negative (blue), no methylated ctDNA detected.

ctDNA, circulating tumor DNA; OS, overall survival; PFS, progression-free survival; RFS, recurrence-free survival.

Among patients receiving palliative treatment, PFS differed significantly across ctDNA tiers (p = 0.036, Figure 3(c)). The median PFS was 9.6 months in the negative group, 6.2 months in the low-level group, and 3.6 months in the high-level group. A similar pattern was observed for OS, with median values of 17.9, 10.2, and 6.0 months for the negative, low-level, and high-level groups, respectively (p = 0.005, Figure 3(d)). The log-rank test results between tiers in the subgroups are reported in Table S3.

The prognostic impact of the ctDNA assay based on the binary categorization was also assessed. Using this approach, the Kaplan–Meier survival analysis and a log-rank test demonstrated a statistically significant difference in OS between negative and positive ctDNA status among all participants (p < 0.001, Figure S1). In the treatment subgroups, however, log-rank tests for OS, RFS, and PFS did not reach statistical significance (Figure S2).

Multivariable Cox regression analyses

To adjust for potential confounders, multivariable Cox proportional hazards models were fitted for each survival outcome. The models included ctDNA status (binary or three-tiered, respectively), lung cancer stage, and previous lung cancer status as covariates.

In the three-tiered model, the adjusted HR for OS was 1.96 (95% CI: 1.09–3.51; p = 0.024) when comparing the high-level group with the negative group among all participants. For PFS in the palliative treatment subgroup, the adjusted HR was 2.32 (95% CI: 1.10–4.92; p = 0.028) for the high-level versus negative group. For RFS in the curative treatment group, the adjusted HR comparing the high-level group with the negative group was 1.81 (95% CI: 0.73–4.48; p = 0.200), which was not statistically significant. Results are summarized in Table 2.

Table 2.

Multivariable Cox proportional hazards regression model for the three-tiered ctDNA categorization.

ctDNA status	Crude HR (95% CI)	Adjusted HR (95% CI) a	p-Value b
All participants, N = 213
OS
Negative	1	1
Low-level	1.67 (1.01–2.76)	1.03 (0.60–1.77)	0.91
High-level	4.11 (2.42–6.96)	1.96 (1.09–3.51)	0.024
Curative treatment subgroup, N = 100
OS
Negative	1	1
Low-level	0.74 (0.31–1.75)	0.86 (0.34–2.14)	0.739
High-level	3.09 (1.30–7.37)	2.09 (0.75–5.80)	0.158
RFS
Negative	1	1
Low-level	0.76 (0.35–1.62)	0.74 (0.34–1.64)	0.464
High-level	3.49 (1.59–7.64)	1.81 (0.73–4.48)	0.200
Palliative treatment subgroup, N = 113
OS
Negative	1	1
Low-level	1.25 (0.61–2.55)	1.11 (0.52–2.37)	0.786
High-level	2.43 (1.16–5.10)	1.99 (0.88–4.50)	0.099
PFS
Negative	1	1
Low-level	1.57 (0.83–2.99)	1.59 (0.82–3.12)	0.173
High-level	2.31 (1.16–4.59)	2.32 (1.10–4.92)	0.028

^aAdjusted for stage and previous lung cancer status.

^bp-Value from comparison of low-level and high-level HR to negative in the adjusted Cox regression model.

CI, confidence interval; ctDNA, circulating tumor DNA; HR, hazard ratio; OS, overall survival; PFS, progression-free survival; RFS, recurrence-free survival.

A statistically significant trend across increasing ctDNA tiers was observed for OS in the full cohort (p = 0.005). Similar trends were seen in the palliative treatments subgroup for both OS (p = 0.021) and PFS (p = 0.020), whereas no significant trends were observed in the curative subgroup.

In the binary model, no statistically significant differences in adjusted HRs were observed between negative and positive ctDNA status for OS among all participants, nor within treatment subgroups for OS, RFS, and PFS. HRs with corresponding 95% CIs and p-values are reported in Table S4.

Discussion

In this study, we found that methylated ctDNA measured in plasma at diagnosis provided valuable prognostic information for patients with newly diagnosed lung cancer, independent of confounders. Across the study population, higher ctDNA levels were associated with significantly increased HRs and shorter survival times, with trend analysis suggesting a dose-response relationship. In the palliative treatment subgroup, higher ctDNA levels pointed toward worse outcomes, with statistical significance limited to PFS. The same tendency was observed in the curative treatment subgroup; however, HRs were non-significant and associated with wide CIs, reflecting reduced statistical power in the subgroups.

The present study builds on the work of Wen et al., ²⁶ who assessed methylated HOXA9 in advanced NSCLC. The common denominator across the studies is the analysis of lung cancer patients receiving palliative treatment using the same methodology; however, Wen et al. used a singleplex assay, whereas we applied a multiplex assay with HOXA9 and four additional ctDNA markers. The multiplex assay appeared to improve the detection rate of stage IV disease (90% vs 79%), whereas detection rates in stage III disease were comparable (69% vs 71%). We observed longer OS and PFS in the palliative subgroup compared with Wen et al., likely due to the widespread adoption and use of immune checkpoint inhibitors and targeted agents between the two study periods. In Wen et al., median OS and PFS among ctDNA-positive individuals were 7.4 and 4.2 months, respectively. When separating the positive cases in the palliative treatment group into low- and high-level groups, we observed a more nuanced risk stratification, identifying intermediate- and high-risk groups with median OS of 10.2 and 6.0 months. Similarly, stratifying positive cases into low- and high-level tiers identified additional risk strata with median PFS of 6.2 and 3.6 months. These findings represent a meaningful step toward more precise prognostic classification in advanced lung cancer using a ddPCR platform.

Several other studies have focused on detecting single-gene methylated ctDNA markers using PCR methods. Notably, aberrant methylation of promoter regions of genes such as SOX17, DCLK1, BRMS1, NALCN, and KMT2C has been associated with poorer prognosis when detected as single-gene markers across various stages of lung cancer.^17–20,27 In operable or early-stage NSCLC, detection rates ranged from 5.1% to 56.2%, and in advanced disease from 10.2% to 63.5%. These detection rates were typically lower than those obtained with the multiplex approach evaluated in the present study. In turn, the measure of single-gene methylated ctDNA markers may increase the risk of false-negative classification in patients with an unfavorable prognosis. By aggregating signals across multiple loci, multiplexed methylation assays may offer more consistent detection and could, if applied in disease monitoring, provide a more reliable basis for follow-up. However, prospective validation is still required to define their clinical utility. Likewise, further studies on larger cohorts are needed to establish and validate the most clinically relevant cut-off for risk stratification, as the threshold applied in the present study was selected for exploratory purposes.

While Ooki et al. ²⁸ studied individuals with stage I lung cancer, they also performed prognostic stratification, defining low-, moderate-, and high-risk groups for OS based on specific methylation patterns in four promoter region markers. In contrast with the present study’s approach, where overall ctDNA levels are quantified as a methylated ctDNA/ALB ratio, Ooki et al.’s method retained marker-specific information and directly incorporated it into their risk model. A potential limitation of our approach is that such marker-specific information is lost. In our assay, only the oncogenic role of HOXA9 has been characterized, with established links to proliferation, survival, metabolism, invasion, and epigenetic control. ²⁹ The four additional markers, named according to the nearest gene (OTX1, MCIDAS, SP9, and TFAP2B), were identified through a data-driven genome-wide screen and validated as lung cancer-specific in tissue samples. ²¹ The physiological and pathological roles of these CpG islands are not yet fully understood, though aberrant methylation is presumed to influence gene transcription and expression. ³⁰ Thus, our assay reflects general methylation patterns as a biomarker for lung cancer, rather than specific pathological or carcinogenic processes for each selected marker. For this reason, we hypothesize that any loss of marker-specific information is less relevant for this assay, and that the quantification of ctDNA levels is more important in our assay.

Although our analyses only evaluated ctDNA at diagnosis, we observed that higher baseline ctDNA levels were associated with poorer prognosis. It is important to emphasize that the analyses within the palliative and curative subgroups were exploratory in nature and based on relatively small sample sizes. Consequently, the resulting estimates are associated with considerable uncertainty and should be interpreted with caution. Nevertheless, if confirmed and validated in independent cohorts, risk stratification based on baseline ctDNA levels could support both escalation and de-escalation of adjuvant therapy and follow-up strategies: patients with higher ctDNA levels might be considered for adjuvant approaches and closer surveillance, whereas those with low or undetectable ctDNA might avoid overtreatment and undergo less intensive follow-up. These implications are hypothesis-generating and require prospective evaluation to define thresholds, timing, and decision rules. Related evidence comes from Rosenlund et al., ³¹ who conducted a prospective MRD study using the tumor-agnostic CAPP-seq assay. They found that post-treatment mutational ctDNA, particularly at 4.5–7.5 months, was associated with higher recurrence risk and shorter RFS, with timing effects varying by treatment modality, such as radiotherapy or chemoradiotherapy. ³¹

This study has some limitations. Sample size in ctDNA tiers and treatment subgroups is small, which may have limited statistical power and contributed to findings that were not statistically significant in some analyses. This applies specifically to the subset of participants with high-level ctDNA in the curative treatment subgroup, and, conversely, participants with low-level ctDNA in the palliative treatment subgroup. Moreover, the sample size precludes performing sub-analyses on histological subtypes, despite their potential relevance. The generalizability of this single-center study is also limited to the Danish population, where the vast majority of lung cancer patients are ever-smokers. This may make the results less valid for populations with different demographic compositions.^1,32 Thus, evaluation in independent multicenter cohorts covering a broader range of patient populations is needed to assess the generalizability of the findings. Furthermore, by including only patients with lung cancer at diagnosis and excluding those with other malignancies, the findings do not fully reflect the broader lung cancer population, which often includes more complex clinical cases. Finally, tumor burden was not systematically quantified. Therefore, although the prognostic association of ctDNA was assessed after adjustment for disease stage, this analysis cannot determine whether ctDNA was an independent prognostic biomarker beyond tumor burden or partly reflected tumor burden itself. Direct quantification of tumor burden, for example, by volumetric assessment of baseline CT scans, would be a valuable addition to future studies.

A strength of this study is the prospective, consecutive enrollment of patients, undertaken without prior knowledge of their final diagnoses, staging, general health status, or treatment plans. This reduces the risk of selection bias and enhances the generalizability of the findings. We excluded patients with a history of other cancer types, or who developed other cancer types during the follow-up period. While this slightly narrows the study population, it also ensures that the measured signal derives from lung cancer rather than other sources. Another strength worth noting is that methylation analyses using ddPCR have characteristics that may facilitate future clinical implementation. The clinical implementation of molecular assays depends on several factors. They must be highly robust analyses with low analytical variation to ensure comparable results over time. Turnaround time from sampling to result must be short (<1 week) to be relevant for clinical decision-making, and costs must be sufficiently low. Methylation analyses using ddPCR fulfill these criteria and are already being evaluated in clinical trials.

Conclusion

In conclusion, the results suggest that the multiplex assay measuring five methylated ctDNA markers in plasma may have greater prognostic value than equivalent singleplex assays in newly diagnosed lung cancer patients. Larger, prospective studies designed to account for the histological subtypes and for the clinical complexity often associated with lung cancer patients are essential to determine the validity and potential of the assay and of methylated HOXA9, OTX1, MCIDAS, SP9, and TFAP2B ctDNA as biomarkers.