Integrative modeling of longitudinal cell-free DNA and tumor volume dynamics: a multimodal quantitative prognostic framework

Introduction

Liquid biopsy based on cell-free DNA (cfDNA) in oncology has been recognized as a promising technique for tracking cancer dynamics (1-3). This approach offers several advantages such as its non-invasiveness, the ability to capture integrated information from multiple tumor sites, and the feasibility of longitudinal monitoring (4,5). Given these advantages, cfDNA analysis has found applications in early detection, therapeutic-response evaluation, and disease-progression identification, making it a cornerstone for modern personalized medicine (6-8).

Among its clinical applications, the identification of minimal residual disease (MRD), defined as the presence of micro-metastases remaining after definitive treatment, using cfDNA analysis has been most widely studied (9-12). Detection and characterization of MRD through cfDNA analysis helps to identify molecular relapses prior to clinical or radiographic signs (13). This opportunity allows prompt intervention and improves patient outcomes. To further increase the sensitivity, a limited number of studies have attempted to temporarily amplify cfDNA release through treatment-induced cell death (14-16).

In contrast to recent advances in spatial targeting, a temporal approach to cfDNA analysis has yet to be established. Most current cfDNA applications are primarily restricted to detecting the presence of cfDNA before or after treatment (17-19). While this method provides valuable insights to support its clinical potential, it only provides a static snapshot of tumor dynamics and overlooks the full spectrum of tumor response over time. To fully comprehend the role of cfDNA in oncology, a more refined, temporal evaluation of cfDNA dynamics, using longitudinal liquid biopsy samples, is essential.

To tackle this, we introduce a novel approach that considers velocity and acceleration of radiotherapy (RT)-induced cfDNA levels from the early (pre-day 3) to late phase (post-day 3) after treatment initiation to capture their dynamic changes in tumor response. In particular, cfDNA samples in our study were mostly collected within first 3 days to capture early treatment response in detail. With wider intervals between samplings in later periods due to practical limitations, we further refined our model based on our hypothesis that regularly acquired computed tomography (CT) scans could provide complementary information to missing cfDNA data on later days. To evaluate its performance, we have compared our combined model against the Response Evaluation Criteria in Solid Tumors (RECIST)-based model and cfDNA-only model. Finally, to assess the potential of model-selected variables, we compare them with conventional clinical features across the identified subgroups. We present this article in accordance with the TRIPOD reporting checklist (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-940/rc).

Methods

Study population

This prospective, single-center study included 25 patients treated with curative-intent RT for localized non-small cell lung cancer (NSCLC) at Samsung Medical Center between June 2019 and November 2020. A subset of these subjects were reported in a previous study investigating targeted cfDNA amplification of NSCLC patients undergoing definitive RT (15). All participants participated in serial peripheral blood sampling for cfDNA analysis before and after RT. Eligible patients were included in this study if they (I) had either histologically confirmed World Health Organization-defined NSCLC with American Joint Committee on Cancer v9 stage I–IIIA disease or a clinical diagnosis of lung cancer without pathological confirmation due to diagnostic procedural risk; (II) received curative-intent RT; and (III) had serial cfDNA samples and chest CT scans collected before, during, and after RT. Patients were excluded if they had (I) incomplete clinical data, (II) unavailable or insufficient cfDNA samples for analysis, or (III) insufficient post-RT CT images for evaluation. Demographic and clinical data, such as age, sex, smoking history, body mass index, histological type, tumor size, tumor (T) stage, node (N) stage, and radiation doses, were retrieved from the medical records.

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This prospective study was approved by institutional review board of Samsung Medical Center (Nos. 2017-09-120, 2018-05155). Informed consent was taken from all the patients.

Sample preparation and cfDNA extraction

Whole-blood samples (10–20 mL) were collected in cfDNA BCT tubes (Streck, La Vista, NE, USA). Plasma was prepared through the following three centrifugation steps at room temperature with increasing centrifugal force: 840 ×g for 10 min, 1,040 ×g for 10 min, and 5,000 ×g for 10 min. All collected plasma samples were stored at 70 ℃ until use. Circulating cfDNA was extracted from plasma using the QIAamp Circulating Nucleic Acid Kit (Qiagen, Hilden, Germany). For validation of variants in plasma samples, LiquidSCAN (GENINUS, Seoul, South Korea) was used, targeting cancer-related genes such as EGFR, TP53, and PIK3CA, among others. The quantitative levels of cfDNA for all targeted genes were summed up and measured as genome-equivalents per mL of plasma. This is calculated as the product of mean variant allele frequency and the plasma cfDNA concentration in a subject divided by 0.0033 ng (20,21).

Longitudinal cfDNA sampling

Longitudinal peripheral blood samples for cfDNA analysis were collected prospectively at multiple times, including before RT (baseline); at 24, 48, and 72 h after baseline; weekly during the first month (weeks 1–4); and during post-treatment follow-up months 1, 4, 7, 10, 13, 16, 19, and 22. A total of 214 plasma samples from 25 patients were analyzed. Figure 1 provides a conceptual demonstration of the collection method.

Figure 1 Conceptual demonstration of the collection method. Serial blood samples and chest CT scans were collected before and after initial RT for each patient. CT, computed tomography; RT, radiotherapy.

CT imaging acquisition and volumetric feature extraction

Chest CT imaging was performed at pre-treatment baseline and every 2–3 months after completion of RT as part of routine surveillance. All CT images were obtained with the following parameters: detector collimation, 1.25 or 0.625 mm; scans performed at 120 kVp with 150–200 mA; and reconstruction interval, 1–2.5 mm.

A region of interest (ROI) was obtained using commercial deep learning software (Aview, version 1.0.23, 2018; Coreline Soft, Seoul, South Korea) for each target lesion and corrected manually when the lesion delineation was incorrect. An additional correction was performed to exclude ground-glass opacities or bronchovascular bundles. Each ROI was confirmed by an experienced thoracic radiologist (H.Y.L.). From the ROIs, 14 quantitative CT volumetric features related to size and volume were automatically extracted. Extraction processes were conducted using the PyRadiomics package implemented in Python (https://pyradiomics.readthedocs.io/en/latest/) (22).

Parametrization of cfDNA dynamics

To describe the longitudinal dynamics of cfDNA over time, we obtained several temporal features that capture both magnitude- and trend-related characteristics. The entire observation period was divided into two phases—an early phase (baseline to day 3) and a late phase (after day 3)—to reflect early and later dynamic changes in cfDNA. All longitudinal parameters of cfDNA were calculated separately for each period, and only patients with at least three observations both within and beyond day 3 were included. Figure 2 presents a schematic representation of this approach.

Figure 2 Conceptual graph illustrating longitudinal cfDNA and CT volume after RT. Kinetic parameters, such as velocity and acceleration, were introduced to analyze the temporal dynamics of cfDNA and CT volume. Note that the distinction between early and late phases is defined by day 3. cfDNA, cell-free DNA; CT, computed tomography; RT, radiotherapy.

First, we calculated basic summary statistics, including the average, minimum, and maximum values throughout the study period. To quantify the overall burden, we calculated the area under the curve with respective to the x-axis divided by days to obtain an average area per day (Area). Additionally, we measured a signed area, reflecting whether trends mostly increased or decreased over time, which was also normalized per day (auc). To depict fluctuation patterns, we defined the vertex feature, defined as the frequency of increases and decreases over time, scaled between 0 and 1. To evaluate dynamic trajectories, we computed the average rate of change (velocity) and average rate of change in velocity (acceleration), both per day. Table S1 summarizes the definitions of all parameters.

Parametrization of CT volume

CT scans were obtained every 2–3 months after treatment. Their longitudinal dynamics of volume over time were described using the same temporal parameters as those for cfDNA. The volumetric parameters were obtained across the entire period, so at least three observations within the entire period were required for calculating the dynamic parameters.

Statistical analysis

Based on these longitudinal parameters, we applied K-means clustering to patients, with a focus on their association with progression-free survival (PFS). Clustering was performed to divide patients into two groups using (I) cfDNA-related parameters and (II) a combination of cfDNA- and volume-related parameters. In both clustering cases, patients with higher PFS values were classified as responders. For comparison, we performed clustering based on RECIST v1.1, using the longest tumor diameters at baseline and the first follow-up CT imaging session, recorded around 2 months after treatment, as input variables. Patients with complete or partial responses were classified as responders, while those with progressive or stable disease were classified as non-responders. Kaplan-Meier curves were drawn, and log-rank tests were conducted to compare PFS between the two groups in each clustering. For each clustering, the result with the lowest P value was selected.

For three clustering models, clinical and volumetric variables with P<0.1 were selected, and multivariate logistic regression was conducted. To assess model performance, the area under the receiver operating characteristic curve (AUC) was calculated. For the model integrating cfDNA- and volume-related parameters, clinical and CT volumetric features from initial CT scans were compared using Fisher’s exact test for categorical variables and the non-parametric Mann-Whitney U test for continuous variables. The overall research flow is shown in Figure 3.

Figure 3 Overall research flow. Clinical data, serial cfDNA data, and longitudinal CT volumetric features were acquired. Longitudinal features of cfDNA and CT volume were incorporated for K-means clustering, with patients divided into two groups. Clinical features, cfDNA amount, and mode-selected variables of these two groups were compared. cfDNA, cell-free DNA; CT, computed tomography.

All statistical analyses were performed using R software (version 4.4.2; R Foundation for Statistical Computing, Vienna, Austria). All reported P values are two-sided, and P<0.05 was considered statistically significant.

Results

Among the initially enrolled 25 patients, 1 patient was lost to follow up, 1 patient had no follow-up CT images, and 6 patients had insufficient cfDNA samples for analysis. As a result, a total of 17 patients were included in the final analysis.

Table 1 summarizes the three groups formed by K-means clustering with different sets of input variables. For clustering based on cfDNA parameters from early and late phases, the responder group exhibited a higher mean PFS of 36 months compared to that of 14.5 months in the non-responder group (P=0.04, AUC =0.848). Clustering based on cfDNA and volume parameters showed the greatest significance in separating patients. The responder group had a mean PFS value of 44.2 months, while the non-responder group had a mean PFS of 16.6 months (P=0.02, AUC =0.955). The Kaplan-Meier curve for this comparison is shown in Figure 4. Clustering based on RECIST v1.1 showed a non-significant difference, with a mean PFS of 25 months for responder and 23 months for non-responders (P=0.62).

Figure 4 Kaplan-Meier survival curve comparing two groups, defined by K-means clustering. PFS, progression-free survival.

The combination of cfDNA and volume parameters—specifically, two cfDNA parameters in the early phase (auc and acceleration) and two volume parameters (vertex and velocity)—was further analyzed. Table 2 summarizes the baseline characteristics of the responders and non-responders based on this clustering approach. The two groups did not differ significantly in age, sex, smoking status, body mass index, histological type, T-stage, lymph node metastasis, daily radiation dose, or total radiation dose. There was a borderline significant difference in tumor size (P=0.05).

Figure 5 illustrates a visual comparison of selected clinical features, cfDNA amount, and four model-selected variables between responders and non-responders. Average cfDNA amount during the observation period, denoted as cfDNA amount, showed no statistical significance. Among four variables in our model, auc and acceleration were statistically significant (P<0.05), while vertex and velocity were not. A detailed comparison of these features is summarized in Table S2.

Figure 5 Radial graphs comparing clinical features, cfDNA amount, and model-selected variables between responder and non-responder group. All numerical variables are min-max normalized. Sex is shown as the percentage of female patients. ‘cfDNA amount’ represents average amount of detected cfDNA during the observation period. ‘auc’ and ‘acceleration’ are calculated from early phase of cfDNA kinetics, while ‘vertex’ and ‘velocity’ from volumetric kinetics. auc, a signed area reflecting whether trends mostly increased or decreased over time, normalized by day; cfDNA, cell-free DNA; T, tumor.

Discussion

Previous studies have predominantly focused on the clinical utility of cfDNA in identifying MRD, based on the detectability of cfDNA before or after treatment. Bossé et al. (23) assessed the prognostic significance of preoperative cfDNA detection in patients with resectable stage I NSCLC. Pan et al. (24) measured cfDNA levels of patients with locally advanced NSCLC undergoing chemoradiotherapy reductions as treatment continued and found that undetectable cfDNA levels on-RT and after-RT time points were associated with better prognosis. However, their method of binary detection capture provided only a static evaluation of tumor response and may not fully capture its dynamic nature.

This study explored whether longitudinal cfDNA kinetics can provide insights beyond the currently known role of cfDNA in MRD detection, by introducing a novel approach. First, we divided our longitudinal cfDNA dataset into two timeframes to reflect both early and late changes in cfDNA kinetics after RT. By collecting more samples within first three days than previous studies, we were able to capture early treatment response that may have been missed (25,26). This idea was motivated by previous studies that short-term fluctuations in ctDNA levels following the start of RT are related to early tumor cell death, while medium- and long-term ctDNA trends reflect overall tumor burden (24,27,28). Second, we calculated kinetics-related parameters such as velocity and acceleration to capture temporal shifts in cfDNA levels and obtain a more comprehensive evaluation of tumor burden changes.

Due to practical limitations, most cfDNA samples in our study were collected in an early phase, with wider intervals of collection in a later phase. To overcome the potential effects of this discrepancy, we hypothesized that incorporation of CT volume data could bridge the discontinuity of cfDNA sampling intervals in later days and enhance the overall accuracy of treatment response evaluation. The validity of this approach is supported by findings from Dawood et al. (29), who compared different diagnostic modalities and found that cfDNA had high specificity for detecting recurrence, while imaging assessments such as CT have demonstrated high sensitivity.

Our findings suggest that longitudinal cfDNA kinetics better differentiate patients between responders and non-responders than conventional RECIST-based evaluation. Furthermore, the incorporation of CT volumetric parameters along with cfDNA kinetics further enhanced model performance, implying better stratification of patients by treatment response. This model included cfDNA parameters from the early phase and volumetric parameters from the late phase. Assaf et al. (30) previously showed both that early cfDNA fluctuation provides strong prognostic information and that machine learning models using cfDNA trends are superior to radiographic assessment. In accordance with these findings, our model suggests that short-term cfDNA spikes capture the early treatment response, while CT volume better suggests the tumor burden in the long term. In other words, CT imaging may serve as a complementary tool to cfDNA analysis during a late phase and integrating both modalities at different timepoints can depict tumor dynamics more precisely.

Our model uncovered four potential markers that showed discriminative potential between responders and non-responders. While conventional features such as sex, age, and T stage, size, and cfDNA amount were not statistically different between two groups, auc and acceleration, two of the selected markers, did. Although vertex and velocity were not statistically significant, they were identified as important predictors in our model, implying their potential in treatment response prediction.

Despite its strengths, our study also has several limitations. First, it was a single-center study with a small number of patients, which limits the generalizability of its results. Further validation is required in larger populations; however, the consistency with previous studies is promising. Second, our model does not take into account the tumor heterogeneity; for example, patients with KRAS-mutant NSCLC receiving KRAS^G12C inhibitors, compared to those with EGFR- or ALK-mutant forms, have much lower treatment response rates (31-33). Genetic variability should be considered in future work.

Conclusions

In conclusion, our study demonstrated that longitudinal cfDNA kinetics, especially when combined with CT volume, has potential as an alternative to conventional imaging-based assessment for monitoring treatment response and predicting prognosis in NSCLC. In particular, the integration of kinetic parameters provides a deeper understanding of tumor dynamics and its response. Although further validation in larger populations is needed, this approach holds promise for potentially complementing or hopefully surpassing current evaluation methods.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-940/rc

Data Sharing Statement: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-940/dss

Peer Review File: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-940/prf

Funding: This work was supported by a National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (No. NRF-2022R1A2C1003999) and by the Future Medicine 20*30 Project of Samsung Medical Center (#SMO1250061). It was also partly supported by an Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by the Korean government (MSIT) (No. RS-2021-II212068, Artificial Intelligence Innovation Hub).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-940/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This prospective study was approved by institutional review board of Samsung Medical Center (Nos. 2017-09-120 and 2018-05155). Informed consent was taken from all the patients.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Fu L, Zhou X, Zhang X, et al. Circulating tumor DNA in lymphoma: technologies and applications. J Hematol Oncol 2025;18:29. [Crossref] [PubMed]
2. Marinello A, Tagliamento M, Pagliaro A, et al. Circulating tumor DNA to guide diagnosis and treatment of localized and locally advanced non-small cell lung cancer. Cancer Treat Rev 2024;129:102791. [Crossref] [PubMed]
3. Connal S, Cameron JM, Sala A, et al. Liquid biopsies: the future of cancer early detection. J Transl Med 2023;21:118. [Crossref] [PubMed]
4. Guler GD, Ning Y, Coruh C, et al. Plasma cell-free DNA hydroxymethylation profiling reveals anti-PD-1 treatment response and resistance biology in non-small cell lung cancer. J Immunother Cancer 2024;12:e008028. [Crossref] [PubMed]
5. Choi Y, Dharia NV, Jun T, et al. Circulating Tumor DNA Dynamics Reveal KRAS G12C Mutation Heterogeneity and Response to Treatment with the KRAS G12C Inhibitor Divarasib in Solid Tumors. Clin Cancer Res 2024;30:3788-97. [Crossref] [PubMed]
6. Zhang K, Fu R, Liu R, et al. Circulating cell-free DNA-based multi-cancer early detection. Trends Cancer 2024;10:161-74. [Crossref] [PubMed]
7. Borgeaud M, Olivier T, Bar J, et al. Personalized care for patients with EGFR-mutant nonsmall cell lung cancer: Navigating early to advanced disease management. CA Cancer J Clin 2025;75:387-409. [Crossref] [PubMed]
8. Xu J, Wan R, Cai Y, et al. Circulating tumor DNA-based stratification strategy for chemotherapy plus PD-1 inhibitor in advanced non-small-cell lung cancer. Cancer Cell 2024;42:1598-1613.e4. [Crossref] [PubMed]
9. Widman AJ, Shah M, Frydendahl A, et al. Ultrasensitive plasma-based monitoring of tumor burden using machine-learning-guided signal enrichment. Nat Med 2024;30:1655-66. [Crossref] [PubMed]
10. Tan AC, Lai GGY, Saw SPL, et al. Detection of circulating tumor DNA with ultradeep sequencing of plasma cell-free DNA for monitoring minimal residual disease and early detection of recurrence in early-stage lung cancer. Cancer 2024;130:1758-65. [Crossref] [PubMed]
11. Chen K, Yang F, Shen H, et al. Individualized tumor-informed circulating tumor DNA analysis for postoperative monitoring of non-small cell lung cancer. Cancer Cell 2023;41:1749-1762.e6. [Crossref] [PubMed]
12. Chidharla A, Rapoport E, Agarwal K, et al. Circulating Tumor DNA as a Minimal Residual Disease Assessment and Recurrence Risk in Patients Undergoing Curative-Intent Resection with or without Adjuvant Chemotherapy in Colorectal Cancer: A Systematic Review and Meta-Analysis. Int J Mol Sci 2023;24:10230. [Crossref] [PubMed]
13. Normanno N, Morabito A, Rachiglio AM, et al. Circulating tumour DNA in early stage and locally advanced NSCLC: ready for clinical implementation? Nat Rev Clin Oncol 2025;22:215-31. [Crossref] [PubMed]
14. Noh JM, Kim YJ, Lee HY, et al. Targeted Liquid Biopsy Using Irradiation to Facilitate the Release of Cell-Free DNA from a Spatially Aimed Tumor Tissue. Cancer Res Treat 2022;54:40-53. [Crossref] [PubMed]
15. Boniface CT, Spellman PT. Blood, toil, and taxoteres: Biological determinants of treatment-induced ctDNA dynamics for interpreting tumor response. Pathol Oncol Res 2022;28:1610103. [Crossref] [PubMed]
16. MacManus M, Kirby L, Blyth B, et al. Early circulating tumor DNA dynamics at the commencement of curative-intent radiotherapy or chemoradiotherapy for NSCLC. Clin Transl Radiat Oncol 2023;43:100682. [Crossref] [PubMed]
17. Duffy MJ. Circulating tumor DNA (ctDNA) as a biomarker for lung cancer: Early detection, monitoring and therapy prediction. Tumour Biol 2024;46:S283-95. [Crossref] [PubMed]
18. Stadler JC, Belloum Y, Deitert B, et al. Current and Future Clinical Applications of ctDNA in Immuno-Oncology. Cancer Res 2022;82:349-58. [Crossref] [PubMed]
19. Abbosh C, Hodgson D, Doherty GJ, et al. Implementing circulating tumor DNA as a prognostic biomarker in resectable non-small cell lung cancer. Trends Cancer 2024;10:643-54. [Crossref] [PubMed]
20. Chabon JJ, Hamilton EG, Kurtz DM, et al. Integrating genomic features for non-invasive early lung cancer detection. Nature 2020;580:245-51. [Crossref] [PubMed]
21. Avanzini S, Kurtz DM, Chabon JJ, et al. A mathematical model of ctDNA shedding predicts tumor detection size. Sci Adv 2020;6:eabc4308. [Crossref] [PubMed]
22. van Griethuysen JJM, Fedorov A, Parmar C, et al. Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Res 2017;77:e104-7. [Crossref] [PubMed]
23. Bossé Y, Dasgupta A, Abadier M, et al. Prognostic implication of methylation-based circulating tumor DNA detection prior to surgery in stage I non-small cell lung cancer. Cancer Lett 2024;594:216984. [Crossref] [PubMed]
24. Pan Y, Zhang JT, Gao X, et al. Dynamic circulating tumor DNA during chemoradiotherapy predicts clinical outcomes for locally advanced non-small cell lung cancer patients. Cancer Cell 2023;41:1763-1773.e4. [Crossref] [PubMed]
25. Jung HA, Ku BM, Kim YJ, et al. Longitudinal Monitoring of Circulating Tumor DNA From Plasma in Patients With Curative Resected Stages I to IIIA EGFR-Mutant Non-Small Cell Lung Cancer. J Thorac Oncol 2023;18:1199-208. [Crossref] [PubMed]
26. Gray JE, Markovets A, Reungwetwattana T, et al. Longitudinal Analyses of Circulating Tumor DNA for the Detection of EGFR Mutation-Positive Advanced NSCLC Progression During Treatment: Data From FLAURA and AURA3. J Thorac Oncol 2024;19:1525-38. [Crossref] [PubMed]
27. Rostami A, Lambie M, Yu CW, et al. Senescence, Necrosis, and Apoptosis Govern Circulating Cell-free DNA Release Kinetics. Cell Rep 2020;31:107830. [Crossref] [PubMed]
28. Nygård L, Ahlborn LB, Persson GF, et al. Circulating cell free DNA during definitive chemo-radiotherapy in non-small cell lung cancer patients - initial observations. PLoS One 2020;15:e0231884. [Crossref] [PubMed]
29. Dawood ZS, Alaimo L, Lima HA, et al. Circulating Tumor DNA, Imaging, and Carcinoembryonic Antigen: Comparison of Surveillance Strategies Among Patients Who Underwent Resection of Colorectal Cancer-A Systematic Review and Meta-analysis. Ann Surg Oncol 2023;30:259-74. [Crossref] [PubMed]
30. Assaf ZJF, Zou W, Fine AD, et al. A longitudinal circulating tumor DNA-based model associated with survival in metastatic non-small-cell lung cancer. Nat Med 2023;29:859-68. [Crossref] [PubMed]
31. Rajaram R, Huang Q, Li RZ, et al. Recurrence-Free Survival in Patients With Surgically Resected Non-Small Cell Lung Cancer: A Systematic Literature Review and Meta-Analysis. Chest 2024;165:1260-70. [Crossref] [PubMed]
32. de Langen AJ, Johnson ML, Mazieres J, et al. Sotorasib versus docetaxel for previously treated non-small-cell lung cancer with KRAS(G12C) mutation: a randomised, open-label, phase 3 trial. Lancet 2023;401:733-46. [Crossref] [PubMed]
33. Miyashita H, Kato S, Hong DS. KRAS G12C inhibitor combination therapies: current evidence and challenge. Front Oncol 2024;14:1380584. [Crossref] [PubMed]