Clinical, Pathologic, and Molecular Prognostic Factors in Patients with Early-Stage EGFR-Mutant NSCLC

This article is featured in Highlights of This Issue, p. 4159

Current guidelines recommend platinum-based chemotherapy after complete surgical resection for patients with stage II and III non–small cell lung cancer [NSCLC; American Joint Committee on Cancer (AJCC) 7th edition] and for a subset of patients with stage IB tumors >4 cm in diameter (1, 2). High-risk pathologic features for relapse in accordance with the National Comprehensive Cancer Network guidelines include visceral pleural invasion, lymphatic invasion, vascular invasion, high-risk histologic findings (high-grade or undifferentiated neuroendocrine tumors), wedge resection, and unknown lymph node status. Despite the use of platinum-based adjuvant chemotherapy, recurrence and cancer mortality rates remain high, so further adjuvant treatment strategies are urgently needed (3, 4).

EGFR mutation–positive (EGFR-M⁺) NSCLC defines a distinct molecular subset that has seen great treatment outcome improvements with the introduction of EGFR-tyrosine kinase inhibitors (TKI) for patients with advanced/metastatic EGFR-M⁺ NSCLC. The role of oncogenic drivers, however, in the prognosis of early-stage NSCLC remains debatable (5). Because the 3-year recurrence-free survival (RFS) rate is only 27% in patients with pathologic stage (pStage) II–IIIA EGFR-M⁺ NSCLC (1, 6), a recent adjuvant trial attempted to improve RFS and overall survival (OS) with EGFR-TKIs. These studies demonstrated an RFS benefit, but not an OS benefit, of first-generation or third-generation EGFR-TKIs in EGFR-M⁺ NSCLC (7–9). In these trials, adjuvant EGFR-TKIs were administered for 2 to 3 years after complete surgical resection without any risk stratification strategy.

In addition, there are unanswered questions regarding adjuvant EGFR-TKIs (10). Do all patients need or benefit from adjuvant TKIs? What is the appropriate treatment duration of adjuvant EGFR-TKIs? Do they prevent or just delay recurrence? What happens after stopping treatment? In the ADJUVANT/CTONG 1104 trial, the majority of patients experienced recurrence after 2 years of gefitinib treatment. The 5-year RFS rate was 22.6%, but further treatment options were limited after recurrence (7). This result suggests that adjuvant EGFR-TKI in unselected patients without risk stratification could be inappropriate. In personalized adjuvant strategies, we must select the group that is most likely to benefit. To address these questions, we need to better understand the natural history and outcomes of EGFR-M⁺ NSCLC after curative surgical resection. If we can identify clinicopathologic and/or molecular risk factors to predict high-risk groups that are likely to benefit from adjuvant EGFR-TKI, we could avoid unnecessary treatment and potential harm due to adverse effects, psychosocial impact, and the financial burden of an expensive 2- or 3-year-long adjuvant treatment. This study aimed to identify clinicopathologic and molecular risk factors for recurrence and survival outcomes and propose personalized adjuvant strategies in resected early-stage EGFR-M⁺ NSCLC.

From January 2008 to August 2020, 2,340 patients with pStage IB–IIIA (AJCC 8th edition) non-squamous NSCLC underwent complete surgical resection at the Samsung Medical Center, Seoul, Korea. Of the total study population, 367 patients did not undergo EGFR mutation testing. A total of 1,181 of 1,973 patients (59.9%) with common EGFR-M⁺ non-squamous NSCLC were included.

Using an in-house algorithm to retrieve a medical big data-based cohort called ROOT (Realtime autOmatically updated data warehOuse in healTh care; ref. 11) and the Registry for Thoracic Cancer Surgery at Samsung Medical Center, detailed clinical data were extracted and analyzed to investigate clinicopathologic risk factors associated with recurrence.

EGFR-M⁺ NSCLC was identified using the PNAclampTM kit or real-time polymerase chain reaction, cobas EGFR Mutation Test v.2, or next-generation sequencing (NGS). Mutations other than a deletion in exon 19 or the L858R point mutation were excluded. To identify molecular risk factors for relapse, comprehensive genomic analysis with whole-exome sequencing (WES) and whole-transcriptome sequencing (WTS) were performed in 56 patients with matched case–controls (pStage and the type of EGFR mutation).

All patients provided written informed consent, and this study was reviewed and approved by the Institutional Review Board of Samsung Medical Center (IRB No. 2021-02-081). The study was conducted in accordance with the principles of the Declaration of Helsinki.

The data cut-off date for the survival analyses was March 31, 2020. RFS was calculated using the Kaplan–Meier estimator and compared using the log-rank test. For multivariate analysis, a Cox regression model was fitted by adjusting the prognostic variables deemed significant (P < 0.05) in the univariate analysis of RFS. RFS was defined as the date of complete surgical resection to the date of recurrence or death due to any cause, whichever came first. Patients with no relapse or death at the time of analysis were censored on the last day of follow-up. OS was defined as the time from complete surgical resection to death due to any cause. Progression-free survival (PFS) was defined as the date of starting EGFR-TKIs treatment after recurrence in a palliative setting to progression or death due to any cause. All P values were two-sided, and P values of <0.05 were considered statistically significant. Data were analyzed using PASW-21 software (IBM SPSS Statistics. Inc., Chicago, IL).

All tumor and matched normal tissue samples were acquired from fresh frozen samples. To purify genomic DNA and RNA, we used the AllPrep DNA/RNA kit (Qiagen), according to the manufacturer's protocol. For WES, we constructed a library using SureSelect XT human All exon V6 (Agilent Technologies) with purified genomic DNA and sequenced on a HiSeq 2500 multiplexing platform (Illumina). The DNA sequencing data were aligned to the human reference genome (hg19) using the MEM algorithms in BWA v.0.7.5, and duplicated reads were removed using Picard v.2.1.5. Aligned DNA sequencing was recalibrated based on public SNP databases or insertion/deletion (indel) databases, according to the recommended GATK protocols. For transcriptome sequencing, we used a TruSeq RNA Access Library Prep kit to build a library and sequenced on a HiSeq 2500 with 100-bp paired-end reads. RNA sequencing reads were aligned to the human reference genome (GRCh37) using STAR v.2.5.1b with default options, and gene-level counts were quantified using RSEM v.1.2.18. The expected count matrix from RSEM was used for further transcriptome analysis.

We called SNPs and small indels using Mutect2 (12) and Manta/Strelka2 (13, 14). We filtered common variants [higher than a minimum variant allele frequency (VAF) of 0.001 in gnomAD v.2.0.2] and low variants allele fraction (with a VAF lower than 0.03). Filtered variants were annotated using Variant Effect Predictor (VEP 56; ref. 15) from the Ensembl database. The tumor mutation burden (TMB) was calculated as the number of non-synonymous variants divided by library size (50.2 Mb). To identify distinct mutation signature in our cohort, we extracted trinucleotide sequence contexts across somatic nucleotide substitutions and configure three different mutation signatures. Each signature was compared with validate signature (COSMIC_2, COSMIC_1 and COSMIC_5) with relatively high cosine similarity (16). Ploidy and purity were estimated by FACETS (v.0.5.14; ref. 17) using tumor and matched normal tissue samples.

WTS subtypes were classified into terminal respiratory unit (TRU) and non-TRU [proximal inflammatory (PI)] group by non-negative matrix factoral algorithms using 3,000 most variable gene sets (18).

According to pathologic classification by cell of origin, we classified into the type II pneumocyte-like tumor cell which had dome-shaped protruding cytoplasm (Type A) and the bronchial surface epithelial cell–like tumor cells which had flat cytoplasmic surface or mucous containing cytoplasm (Type B; refs. 19–21).

The data generated in this study are available within the article and its Supplementary Data file. The additional data are available upon request from the corresponding author.

From January 2008 to August 2020, 1,181 patients of 1,973 (59.9%) with pStage IB–IIIA, EGFR-M⁺, non-squamous NSCLC underwent EGFR testing and curative surgery and were included in the final analysis to evaluate the natural clinical outcomes and clinicopathologic risk factors for RFS and OS (Supplementary Table S1 and Supplementary Fig. S1). At baseline, 48.9%, 28.0%, and 23.1% of patients had pStage IB, II, and IIIA NSCLC, respectively. As expected, females and never-smokers were prevalent. In total, 52.7% of patients exhibited the EGFR deletion 19 mutation and 47.3% exhibited the EGFR L858R mutation. Of the total study population, 43.3% of patients received adjuvant chemotherapy or adjuvant concurrent chemoradiotherapy. In pStage IB, II, and IIIA, 6.4%, 70.4%, and 88.3% of patients received adjuvant treatment, respectively.

During a median follow-up of 38.8 months (range, 0.5–156.2), 472 patients experienced disease recurrence. At the first recurrence, 23.9% of patients had brain metastasis, which increased to 41.5% during follow-up. The median RFS was 49.8 months [95% confidence interval (CI), 45.4–54.2], and the 2-year and 3-year RFS rates were 72.8% and 60.2%, respectively. For pStage IB, II, and IIIA, the median RFS was 73.5 months (95% CI, 62.1–84.9), 48.7 months (95% CI, 41.2–56.3), and 22.7 months (95% CI, 19.4–26.0), respectively (Supplementary Fig. S2A). The median RFS was not significantly different between patients with exon 19 deletion and those with L858R mutation (55.6 vs. 54.6 months; P = 0.93).

The median OS of the total study population was 128.3 months (95% CI, 103.4–153.1). For pStage IB, II, and IIIA, the median OS was 134.6 months (95% CI, 131.2–138.0), 124.3 months (95% CI, 86.8–161.8), and 82.1 months (95% CI, 68.4–95.8), respectively (Supplementary Fig. S2B).

Supplementary Table S2 shows univariate and multivariate analyses of clinicopathologic risk factor for RFS. In the univariate analysis, pStage, grade of differentiation, histologic subtype, lymphatic invasion, vascular invasion, nerve invasion, and pleural invasion showed a statistically significant association with RFS (P < 0.001). According to pathologic classification by cell of origin, the type A was associated with favorable RFS compared with the type B (P = 0.20). In the multivariate analysis, pStage (HR, 1.44, 95% CI, 1.34–1.54; P < 0.001), micropapillary subtype (HR, 1.28; 95% CI, 1.06–1.54; P = 0.01), vascular invasion (HR, 1.23; 95% CI, 1.03–1.47; P = 0.03), and pleural invasion (HR, 1.34; 95% CI, 1.22–1.64; P = 0.04), and pathologic classification (type A vs. type B; HR, 2.30; 95% CI, 1.10–1.50; P < 0.001) were predictive factors for RFS.

Fresh tumor tissue and matched normal tissue samples were available for 98 of 1,181 patients. We performed WES and WTS in 56 matched cases of pStage (pStage II vs. IIIA) and type of EGFR mutation (deletion 19 vs. L858R), and 29,651 somatic variants were detected, which included 6,713 non-synonymous and 22,938 synonymous variants (Fig. 1). Table 1 showed baseline characteristics of 56 patients with matched case–controls.

Among the matched 56 patients, 31 patients (55.4%) had the TRU subtype, which showed downregulation of proliferation-related pathways and low ploidy and TMB, and 25 patients (44.6%) had the non-TRU subtype, which showed several proliferation-related pathways, such as the G2M checkpoint and E2F_targets, and had high ploidy and TMB (Fig. 1).

TP53 was the most frequently co-mutated gene (48%), followed by MUC16 (16%), CTNNB1 (14%), RBM10 (11%), APC (11%), MUC4 (11%), CSMD3 (9%), and PRPRT (9%; Fig. 1). We found three distinct mutation profiles (Signature_1, Signature_2, and Signature_3) that were highly similar to the Apolipoprotein B mRNA Editing Catalytic Polypeptide-like (APOBEC) signature (COSMIC_2), unknown signature (COSMIC_5), and aging signature (COSMIC_1) (Supplementary Fig. S3A and S3B), respectively. Signature_1 had a high frequency of TCW (W = A or T) mutation context and varied within the cohort. Therefore, we calculate APOBEC mutation count and classified into APOBEC-high (≥2) and low (<2) signature groups. Patients with a dominant APOBEC mutation signature (Signature_1, Supplementary Fig. S3A) were enriched more frequently in the non-TRU group (8 of 25, 32%) than in the TRU group (4 of 31, 12.9%; P = 0.16). APOBEC mutation signature or high APOBEC enrichment scores (≥2), however, were not significantly related to RFS and OS.

The non-TRU subtype group showed inferior RFS (HR, 3.49; 95% CI, 1.72–7.09; P < 0.001) compared with the TRU subtype group (Supplementary Fig. S4A). In patients with pStage II NSCLC, the non-TRU subtype group showed inferior RFS (HR, 4.46; 95% CI, 1.53–13.02; P = 0.003) compared with the TRU group (Supplementary Fig. S4B). In patients with pStage IIIA NSCLC, the non-TRU group showed a tendency toward inferior RFS (HR, 2.17; 95% CI, 0.83–5.66; P = 0.1) compared with the TRU group (Supplementary Fig. S4C).

The TP53 mutation group showed inferior RFS (HR, 2.50; 95% CI, 1.24–5.04; P = 0.009; Supplementary Fig. S4D). In pStage II and IIIA, the TP53 mutation group showed a tendency toward inferior RFS (HR, 2.64; 95% CI, 0.90–7.74; P = 0.07; HR, 2.51; 95% CI, 0.91–6.93; P = 0.68, respectively; Supplementary Fig. S4E and S4F).

Figure 2A shows RFS according to the RNA subtype and TP53 mutation (P < 0.01). Three molecular risk groups, group 1 (favorable group; TRU subtype and TP53 wild-type), group 2 (moderate risk group; non-TRU subtype or TP53 mutation), and group 3 (high-risk group; non-TRU subtype and TP53 mutation) for RFS showed statistically significant differences among the 56 patients (Fig. 2). The median RFS was not reached, 25.5 months (95% CI, 1.4–10.4), and 14.2 months (95% CI, 2.9–24.8) for groups 1, 2, and 3, respectively (P < 0.01; Fig. 2A). The 2-year and 3-year RFS rates were 80%, 45%, 20% for groups 1, 2, and 3, respectively. The median RFS were not reached, 36.0 months (95% CI, 0.7–11.9), and 14.7 months (95% CI, 2.1–40.9) for the three groups, respectively, for those with pStage II (P = 0.003; Fig. 2B) and 55.6 months, 25.2 months (95% CI, 0.9–17.6), and 13.7 months (95% CI, 1.2–32.6) for those with pStage IIIA (P = 0.06; Fig. 2C). In multivariate analysis, molecular risk factors (non-TRU subtype and TP53 mutation) were associated with poor RFS independent of pStage and clinicopathologic risk factors (micropapillary subtype, vascular invasion, and pleural invasion; Table 2). Figures 2D and E shows typical cases of type A and type B according to pathologic classification by cell of origin, respectively. The type A of pathologic classification was related with TRU subtype (concordance rate 67.8%), and the type B with non-TRU subtype (concordance rate 80%). Figure 2F shows that RFS was stratified into three risk group by combination of pathologic classification of cell of origin (type A vs. type B) and molecular subtype (TP53 mutation).

Figure 3A shows the APOBEC enrichment score and objective response rate, PFS, and OS (from the date of starting of palliative EGFR-TKIs to death) of 27 patients who experienced recurrence and were treated with EGFR-TKIs. Among 24 patients (except for three who were not evaluated), 5 of 8 (62.5%) patients with APOBEC enrichment scores ≥2 achieved objective response after EGFR-TKI treatment compared with 15/16 (93.7%) patients with a APOBEC enrichment scores <2 (P = 0.18). Patients with high APOBEC enrichment scores (≥2) were associated with inferior PFS of EGFR-TKI (8.6 vs. 28.8 months; HR, 4.16; 95% CI, 1.28–13.46; P = 0.012; Fig. 3B). High APOBEC enrichment score (≥ 2) was also associated with inferior OS compared with lower APOBEC enrichment score (16 months vs. not reached; HR not available; P ≤ 0.001; Fig. 3C).

We investigated the natural history of patients with pStage IB–IIIA EGFR-M⁺ NSCLC after complete surgical resection. Many patients with pStage IB-IIIA EGFR-M⁺ NSCLC, even after complete surgical resection experienced recurrence, especially within 3 years after surgery (6). To improve outcomes of patients with early-stage EGFR-M⁺ NSCLC, efforts with targeted therapy have been attempted (8, 22, 23). In the ADAURA study, 3-year osimertinib treatment improved the RFS with HR of 0.17 in patients with stage II or IIIA disease. Unfortunately, RFS of EGFR-TKI treatment does not always translate into long-term OS benefits. In the ADJUVANT/CTONG1104 study, the survival benefit of adjuvant EGFR-TKI seemed to be outstanding up to 24 months after surgery. However, after discontinuing EGFR-TKIs at 24 months, the Kaplan–Meier curves for RFS began to converge, meeting at 36 months. Therefore, the efficacy of adjuvant EGFR-TKI should not be predicted prematurely before termination of treatment, and an OS benefit should be demonstrated with a sufficient follow-up period.

Circulating tumor DNA (ctDNA) is a noninvasive method for early diagnosis, prognostic stratification, and treatment response monitoring. In the ongoing study (NCT04385369), the efficacy of immunotherapy with standard of chemotherapy was assessed in the minimal residual disease (MRD)-positive patients after surgery (24). Detection of MRD after surgery via ctDNA analysis may be a useful approach to identify patients who need alternative strategies.

Adjuvant EGFR-TKIs may not completely eradicate tumor cells, so there is delayed recurrence, not a cure. Therefore, unlike treatment of advanced-stage disease in which response or PFS is important, adjuvant treatment should be prioritized in high-risk groups for recurrence. Previously, studies have investigated prognostic factors of EGFR mutation status, but there have been no studies identifying factors predicting recurrence in patients with early-stage EGFR-M⁺ NSCLC. Although we included a large sample of over 1,100 patients, almost all significant clinicopathologic factors had HRs <2, so, although statistically significant, their impact was relatively low. At the molecular level, RNA subtype and TP53 mutation status were not only statistically significant but were strong prognostic factors, with HRs of 2.5 or higher regardless of pStage.

By analyzing clinicopathologic and molecular risk factors associated with RFS, we identified groups most benefiting from EGFR-TKIs. Specific gene alterations, such as TP53 mutation, that coexist with EGFR mutations were associated with unfavorable EGFR-TKI efficacy and survival outcomes and are well known in the advanced or metastatic setting (25). Two thirds of patients with EGFR-M⁺ NSCLC have coexisting TP53 mutations, associated with shorter time on EGFR-TKIs and shorter OS in palliative setting (26).

In addition to concurrent genomic alterations, prognosis varies depending on the transcriptional subtype with histopathologic features and the origin of the cells. Transcriptional subtypes with histopathologic features of lung adenocarcinoma were classified as TRU (low proliferation pathway), non-TRU (high proliferation pathway), and proximal-proliferative (formerly magnoid) subtypes in European cohort (27). In the recent East Asia study, the transcriptional subtype of patients with lung adenocarcinoma in East Asia are classified as TRU (low proliferation), TRU-I (low proliferation and high immune infiltration), and non-TRU (PI, high proliferation; ref. 18). This proliferation axis can stratify patients when considering potential confounding factors, but the inflammation axis may predict a patient's response to immunotherapy based on the level of immune infiltration. In a previous study, RNA subtypes, especially for TRU versus non-TRU, were investigated for OS, but not for recurrence in early-stage NSCLC.

Coexisting mutation was different according to EGFR subtype. It has been previously reported that non-TRU is more frequently associated with TP53 (28), while TRU is more often associated with TP 53 wild-type. In the current study, we evaluated the difference of transcriptomic subtype and TP 53 mutation between deletion 19 and L858R, but it was not different statistically (Supplementary Table S3).

Both RNA subtypes and mutations coexisting with EGFR mutations are well-known prognostic factors for PFS and OS in advanced or metastatic EGFR-M⁺ NSCLC but have not been well defined in early-stage EGFR-M⁺ NSCLC after curative resection. Notably, our study showed that patients with the non-TRU subtype or TP53 mutation had shorter RFS than those with TRU subtypes or had the TP53 wild-type independent of pStage. The high-risk group might achieve better outcomes with aggressive strategies, including EGFR-TKIs with or without other treatments, such as antiangiogenic agents, chemotherapy, or others. Concurrent genomic alteration including TP53 mutation might identify different biological subsets of EGFR-M⁺ NSCLC (26). A recent study of apatinib plus gefitinib showed a PFS benefit in patients with TP53 exon 8 mutations (29). In the high-risk group with TP53 mutation, a personalized adjuvant strategy with EGFR-TKIs plus antiangiogenic agents might result in better outcomes than EGFR-TKIs alone. Our study suggests that the low-risk group with TRU subtype and TP53 wild-type without clinicopathologic risk factors might not need adjuvant EGFR-TKI to prevent relapse regardless of pStage II or IIIA.

After recurrence, most patients received EGFR-TKI as palliative treatment with a high response rate and long PFS over 12 months. In our study, the APOBEC mutation signature or high APOBEC enrichment score were related to poor PFS and objective response rate with EGFR-TKIs after recurrence. However, the APOBEC mutation signature or high APOBEC enrichment score were not related to RFS and OS after complete surgical resection. The APOBEC mutation signature is related to the polynucleotide cytosine deaminase protein family (16), known to play key roles in mutagenic progress, contributing to sub-clonal diversification, intra-tumor heterogeneity, and tumor evolution (30). In Asians with lung adenocarcinoma, the APOBEC mutation signature was related to younger age, nonsmokers, and female sex (31). APOBEC-related mutagenesis has been associated with immune cell marker expression (high PD-L1 expression and immune cell infiltration) and immunotherapy response (32, 33). The APOBEC mutation signature or high APOBEC enrichment score are related to EGFR-TKI resistance in advanced-stage NSCLC. Here, we revealed that the APOBEC mutation signature exists in some patients with early-stage EGFR-M⁺ NSCLC, and in these cases, the efficacy of EGFR-TKI was not satisfactory. This supports the idea that a subset of patients with the APOBEC mutation signature may need an alternative treatment strategy such as immunotherapy, antiangiogenic agents, or chemotherapy. For personalized strategy in early-stage NSCLC, we need to consider both recurrence risk and post-recurrence outcome. Clinical risk factor, coexisting mutation (TP53 mutation), and RNA subtype are risk factors for recurrence. In the current study, APOBEC mutation signature seems to define a specific subset of poor response to EGFR-TKI suggesting they might need an alternative adjuvant strategy, e.g., combination with chemotherapy or antiangiogenic agents. A further prospective study with a large number of patients is warranted to confirm the role of the APOBEC mutation signature (34). In the recent study, antiangiogenic agents in combination with EGFR-TKIs improved PFS in patients with EGFR-M⁺ NSCLC who had smoking history in the subgroup analysis (35, 36). Antiangiogenic agents in combination with EGFR-TKIs might be a treatment option in patients with smoking history. EGFR-M⁺ NSCLC showed low expression of PD-L1, the absence of tumor-infiltrating lymphocytes, and increased B7-H4 increased which contributes to low response to immunotherapy (37).

As other ongoing trials for early-stage EGFR-M⁺ NSCLC, LAURA study will assess the efficacy and safety of osimertinib as maintenance therapy in patients with locally advanced, unresectable, EGFR-M⁺, stage III NSCLC without disease progression following definitive chemoradiation therapy (22). ALCHEMIST study investigated efficacy of erlotinib for patients with early-stage EGFR-M⁺ NSCLC who have completed the usual treatment after curative intent surgery (23). The same concerns are raised in studies such as LAURA and ALCHEMIST about selecting a group who are most likely to benefit from adjuvant targeted therapies, so results of these ongoing trials are eagerly awaited.

Our study has several limitations, including its retrospective design and no patients who received EGFR-TKI as adjuvant treatment. We analyzed clinical outcomes and clinicopathologic risk factors for RFS in 1,181 patients with EGFR-M+ NSCLC after complete surgical resection. However, to identify the molecular risk factors for RFS and OS, we could analyze WTS and WES data in only 56 patients. It is a challenge to obtain fresh frozen specimens for WTS. WTS analysis using formalin-fixed paraffin-embedded tissue could be an alternative option. The RNA subtype identified in WTS was a concept originating from the cell of origin, and if TRU and non-TRU classification could be made by pathologic examination, it could be more easily implemented in clinical practice. Adenocarcinoma can originate either from type II pneumocytes or Clara cell being associated with TTF-1 and peripheral tumor, which is type A of our pathologic classification (21). In contrast, others originate from ciliated columnar cell and characterized by mucin MUC5AC or MUC5B expression and centrally located tumor, which is type B of our pathologic classification. It is of interest the pathologic classification of cell of origin was closely correlated in the current study with RNA subtype (TRU and non-TRU) and also predictive of RFS. Pathologic classification by cell of origin might be a potential surrogate marker for RNA subtype, which is under investigation. From a pathologic point of view, we tried to investigate whether these WTS and WES could be replaced by the simpler pathologic method. We have a plan to prospectively validate these pathologic subtypes in a larger number of patients by using a more objective way such as artificial intelligence.

Another possible issue is the ethnicity; this study only included Asians. Given that the OS benefit was not evident in the Asian ethnicity in the FLAURA study and less complex genomic architecture in Asians with marked ancestry differences between Asians and Caucasians (18), it needs to be investigated in the non-Asian ethnic group as well in the future study. This study is a hypothesis-generating study and warrants further prospective investigation in a large number of patients.

In summary, our study highlights that patient with early-stage EGFR-M⁺ NSCLC had diverse clinical outcomes depending on clinicopathologic risk factors, concurrent molecular alterations, and RNA subtype. The low-risk group who had the TRU subtype and TP53 wild-type without clinicopathologic risk factors might not need to receive adjuvant EGFR-TKIs. In the higher risk group with non-TRU subtype and/or TP 53 mutation, or clinicopathologic risk factors, an active adjuvant strategy with EGFR-TKIs with or without other treatments (antiangiogenic agents, chemotherapy, or others) should be considered. In patients with the APOBEC mutation signature, an alternative adjuvant strategy such as immunotherapy, chemotherapy, antiangiogenic agents, or other agents might be needed due to poor response to EGFR-TKIs.

S.H. Lee reports personal fees from AstraZeneca/MedImmune, Rochen, and Pfizer and grants and personal fees from Merck outside the submitted work. J.S. Ahn reports personal fees from Takeda Pharmaceutical, Hanmi, BC World, Novartis Korea, Pfizer, Yuhan, Roche Korea, Amgen Korea, Boehringer Ingelheim, AstraZeneca Korea, Menarini Korea, Bayer Korea, Yooyoung, Pharmbio Korea, and Bixink outside the submitted work. M.J. Ahn reports personal fees from AstraZeneca, Merck, Amgen, MSD, Arcus, Takeda, Alpha Pharmaceutical, and YUHAN outside the submitted work. W.Y. Park reports personal fees from Geninus Inc. outside the submitted work. No disclosures were reported by the other authors.

H.A. Jung: Conceptualization, resources, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. J. Lim: Formal analysis, validation, visualization, methodology, writing–original draft, writing–review and editing. Y.-L. Choi: Formal analysis, supervision, visualization, writing–original draft. S.-H. Lee: Resources, supervision, validation. J.-G. Joung: Conceptualization, supervision, validation, methodology. Y.J. Jeon: Resources, investigation, writing–review and editing. J.W. Choi: Resources, investigation, writing–review and editing. S. Shin: Resources, investigation, writing–review and editing. J.H. Cho: Resources, investigation, writing–review and editing. H.K. Kim: Resources, investigation, writing–original draft. Y.S. Choi: Resources, investigation, writing–review and editing. J.I. Zo: Resources, investigation, writing–review and editing. Y.M. Shim: Resources, investigation, writing–review and editing. S. Park: Resources, writing–review and editing. J.-M. Sun: Resources, investigation, writing–review and editing. J.S. Ahn: Resources, investigation, writing–review and editing. M.-J. Ahn: Resources, investigation, writing–review and editing. J. Han: Formal analysis, writing–review and editing. W.-Y. Park: Conceptualization, supervision, validation, writing–review and editing. J. Kim: Conceptualization, resources, validation, visualization, writing–original draft, project administration, writing–review and editing. K. Park: Conceptualization, resources, supervision, funding acquisition, writing–original draft, writing–review and editing.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.