Prognostic value of ROS1 rearrangement in lung adenocarcinoma stratified by clinicopathologic and radiological features: a retrospective cohort study Authors: Yi Liu, Zelin Ma, Shuxiao Ma, Abuduwaili Yasheng, Zelin Deng, Chenglin Guo, Lunxu Liu Published: 2026 ABSTRACT BackgroundThe prognostic significance of ROS1 rearrangement in lung adenocarcinoma (LUAD) remains a subject of debate. This study systematically evaluated the clinicopathologic characteristics and prognostic impact of ROS1 rearrangement in early- and advanced-stage LUAD, stratified by distinct clinicopathologic and radiological features.MethodsWe retrospectively analyzed comprehensive clinical data and follow-up records of patients with resected invasive LUAD who underwent ROS1 testing between 2018 and 2020. Patients were stratified by tumor stage (I vs. II–III), radiographic pattern (part-solid vs. solid), histologic differentiation (moderate vs. poor), and pathologic aggressiveness (passive vs. aggressive, defined by the presence of lymphovascular invasion, visceral pleural invasion, or spread through air spaces). The prognostic significance of ROS1 rearrangement was assessed within each subgroup.ResultsAmong 3,162 enrolled patients, 145 (4.5%) were ROS1 rearrangement-positive. ROS1 rearrangement was associated with non-smoking status, pure solid nodules, poor differentiation, and lymphovascular invasion. ROS1 rearrangement was associated with significantly worse recurrence-free survival (RFS) and overall survival (OS) in patients with stage II–III disease, solid radiographic pattern, poor differentiation, and aggressive pathologic features. In contrast, RFS and OS were comparable between ROS1-positive and -negative patients in the stage I, part-solid, moderately differentiated, and passive pathologic subgroups. Multivariable Cox regression further confirmed these subgroup-specific findings. The rate of bone metastasis was significantly higher in ROS1 rearrangement-positive than in ROS1 rearrangement-negative patients.ConclusionsROS1 rearrangement plays a significant prognostic role in advanced progression of LUAD; however, the generally favorable survival outcomes of early-stage LUAD may obscure its prognostic value. Highlight box Key findings • In lung adenocarcinoma (LUAD), ROS1 rearrangement was correlated with significantly worse recurrence-free survival and overall survival in patients with stage II–III disease, solid radiographic pattern, poor differentiation, and aggressive pathologic features. • ROS1-positive tumors show an increased propensity for bone metastasis. What is known and what is new? • ROS1 rearrangement defines a targetable molecular subtype of LUAD, but its independent prognostic value remains controversial. • This study shows that the prognostic value of ROS1 rearrangement depends on underlying tumor biology and clinicopathologic stratification. What is the implication, and what should change now? • ROS1 status should be integrated with clinicopathologic and radiological factors for refined postoperative risk stratification. • ROS1-rearranged patients with high-risk features may benefit from intensified bone surveillance and risk-adapted adjuvant management strategies. Introduction Lung cancer remains the leading cause of cancer death worldwide, with an estimated over 2.2 million new cases and 1.8 million deaths each year (1). Over 80% of lung cancers are classified as non-small cell lung cancer (NSCLC) and lung adenocarcinoma (LUAD) represents the most prevalent subtype (2). ROS proto-oncogene 1 (ROS1) was initially identified in the 1980s as a viral oncogene and was found to encode a transmembrane receptor tyrosine kinase structurally related to the insulin receptor family (3,4). The oncogenic role of ROS1 was first recognized in glioblastoma, where chromosomal rearrangements were shown to drive aberrant kinase activation and tumorigenesis (5). Subsequent studies confirmed that ROS1 rearrangement promotes constitutive activation of downstream signaling pathways, promoting enhanced proliferation, survival, and metastatic potential (6-8). In 2007, ROS1 rearrangement was first discovered in NSCLC (9) and has been reported to occur in approximately 1% to 3% of LUADs, with a higher prevalence in younger patients and never-smokers (10-12). The therapeutic landscape for ROS1-rearranged NSCLC has evolved rapidly with the introduction of targeted agents, which have demonstrated notable antitumor activity and transformed ROS1-rearranged disease into a clinically actionable entity (13,14). Crizotinib, originally developed as a mesenchymal-epithelial transition factor inhibitor (15) and later approved for anaplastic lymphoma kinase fusion NSCLC (16), was found to potently inhibit ROS1 and to produce high response rates in early clinical studies (17-19), leading to its approval as the first ROS1-targeted tyrosine kinase inhibitor (TKI) (20). The marginal blood–brain barrier penetration of crizotinib results in limited central nervous system activity (21) and acquired resistance also commonly emerges (22). To overcome these limitations, next-generation agents such as entrectinib, lorlatinib, and repotrectinib were developed and subsequently approved for the treatment of ROS1-positive NSCLC (14). Although the predictive value of ROS1 rearrangement for targeted therapy responsiveness is well recognized, its intrinsic prognostic significance independent of ROS1-TKIs in LUAD remains poorly defined, and the limited available studies have reported conflicting results (23-27). Stratified analysis based on tumor–node–metastasis (TNM) stage, radiologic appearance and histologic pattern has proven valuable for delineating the prognostic impact of oncogenic driver mutations within specific clinical settings in LUAD (28,29). Moreover, lymphovascular invasion (LVI), visceral pleural invasion (VPI) and spread through air spaces (STAS) are recognized as additional pathologic descriptors linked to aggressiveness and unfavorable outcomes in LUAD (30). Therefore, this study aims to comprehensively investigate the prognostic value of ROS1 rearrangement in resected invasive LUAD stratified by TNM stage, radiologic, histologic and pathologic aggressive characteristics. We present this article in accordance with the REMARK reporting checklist (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0292/rc) (31). Methods The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Ethical approval was granted by the Institutional Ethics Committee for Clinical Research of West China Hospital, Sichuan University (No. 2025-2518). Written informed consent was waived for this study due to its retrospective design. Cohort and baseline characteristics This retrospective cohort study was conducted by enrolling consecutive patients who underwent complete resection for invasive LUAD and were tested for ROS1 status in the Department of Thoracic Surgery at West China Hospital, Sichuan University, between 2018 and 2020. The study excluded patients who had received neoadjuvant therapy or presented with stage IV disease at pathology. Entire clinicopathologic characteristics were systematically collected for each case, capturing demographic variables (age at diagnosis, sex, smoking history), surgical approach, adjuvant treatment, and other histopathologic and radiologic characterization, including computed tomography (CT) imaging patterns, predominant histological subtype, status of LVI, VPI, STAS, and pathologic TNM stage according to the 9th edition of the TNM classification. Histopathologic and radiologic characterization All tumor specimens underwent centralized pathologic review. Histological subtyping adhered to the 2015 WHO Classification of Thoracic Tumors (32) and was subsequently condensed into a three-tier grading system: well-differentiated (lepidic predominant), moderately differentiated (papillary, acinar, or invasive mucinous predominant), and poorly differentiated (micropapillary or solid predominant). The presence of LVI, VPI, and STAS was determined based on established morphologic criteria (30). Patients were then stratified into two pathologic behavioral categories: an “aggressive” subgroup (featuring at least one of LVI, VPI, or STAS) and a “passive” subgroup (devoid of all three features). The consolidation tumor ratio (CTR) was calculated as the proportion of the maximum diameter of the consolidation component relative to the maximum tumor diameter on the final preoperative CT scan. Accordingly, nodules were classified as pure ground-glass opacity (CTR =0), part-solid (0< CTR <1), or solid (CTR =1) nodules. Detection of driver gene alterations ROS1 and ALK rearrangements were assessed using fluorescence in situ hybridization (FISH). A specimen was considered FISH-positive if split signals were observed in more than 15% of tumor cells. EGFR mutation analysis was performed using PCR-based methods or next-generation sequencing. All FISH-positive results underwent subsequent confirmation via immunohistochemistry (33). Survival endpoints The study defined two primary survival endpoints: overall survival (OS) and recurrence-free survival (RFS). OS represented the time from surgery to death from any cause, while RFS represented the time from surgery to the first documented recurrence (any site). In RFS analysis, non-cancer-related deaths were censored. All documented recurrences were mapped to specific anatomic regions: thorax, abdomen, neck, brain, and bone. Follow-up Postoperative monitoring followed a structured protocol. Clinical and radiological evaluations were scheduled every 3–6 months for the initial two years, every 6 months for the subsequent three years, and annually thereafter. The standard imaging workup constituted chest CT, brain magnetic resonance imaging (MRI) or CT, and upper abdominal surveillance. Any clinical suspicion of recurrence prompted further diagnostic procedures, including bone scan, positron emission tomography-computed tomography (PET-CT), or biopsy. Follow-up data were supplemented with telephone interviews until June 2025. Statistical analysis Data analysis was performed in R (version 4.2.3). Categorical variable associations were tested using Pearson’s chi-square or Fisher’s exact test. Survival probabilities for RFS and OS were estimated with the Kaplan-Meier method, and group comparisons were made with the log-rank test. Variables demonstrating a univariate association with outcomes (P<0.1) were entered into a multivariable Cox proportional hazards model to identify independent prognostic factors. A two-sided P value <0.05 indicated statistical significance. Results Clinicopathologic features of ROS1-rearranged LUADs Among 3,162 patients with invasive LUAD, ROS1 rearrangements were identified in 145 (4.5%). The distribution of major driver alterations in the cohort, including EGFR mutations and ALK rearrangements, is summarized in Table S1. ROS1-positivity was significantly more prevalent among non-smokers (P=0.049). Radiologically, ROS1-rearranged tumors were more frequently observed as solid nodules than ROS1-negative tumors (56.6% vs. 47.5%, P=0.001). Histologically, ROS1-positive patients were more frequently diagnosed with the solid-predominant subtype compared with ROS1-negative patients (12.4% vs. 7.7%, P=0.003). Moreover, among pathologic features, ROS1-positive tumors were strongly associated with LVI (P=0.01) (Table 1). Table 1Clinicopathologic features of patients with resected invasive lung adenocarcinomaCharacteristicsROS1− (n=3,017)ROS1+ (n=145)P valueAge, median (IQR), years60 (52, 67)58 (51, 67)0.18Age, n (%)0.25 ≥60 years1,563 (51.8)68 (46.9) <60 years1,454 (48.2)77 (53.1)Sex, n (%)0.12 Female1,634 (54.2)88 (60.7) Male1,383 (45.8)57 (39.3)Smoking history, n (%)0.049* Never2,189 (72.6)116 (80) Ever828 (27.4)29 (20)Pathologic TNM stage, n (%)0.11 Stage I2,524 (83.7)115 (79.3) Stage II213 (7.1)9 (6.2) Stage III280 (9.3)21 (14.5)Image pattern, n (%)0.001* Pure GGO313 (10.4)2 (1.4) Part-solid1,272 (42.2)61 (42.1) Solid1,432 (47.5)82 (56.6)Histologic subtype, n (%)0.003* Lepidic306 (10.1)2 (1.4) Acinar1,770 (58.7)94 (64.8) Papillary544 (18)26 (17.9) Invasive mucinous109 (3.6)2 (1.4) Micropapillary55 (1.8)3 (2.1) Solid233 (7.7)18 (12.4)LVI, n (%)0.01* Absent2,808 (93.1)127 (87.6) Present209 (6.9)18 (12.4)VPI, n (%)0.19 Absent2,593 (85.9)119 (82.1) Present424 (14.1)26 (17.9)STAS, n (%)0.69 Absent2,393 (79.3)113 (77.9) Present624 (20.7)32 (22.1)Surgery type, n (%)0.48 Lobectomy1,931 (64)97 (66.9) Sublobar resection1,086 (36)48 (33.1)Adjuvant chemotherapy, n (%)0.55 No2,633 (87.3)129 (89) Yes384 (12.7)16 (11)Adjuvant TKIs, n (%)0.005* No2,579 (85.5)136 (93.8) Yes438 (14.5)9 (6.2)*, statistically significant. GGO, ground-glass opacity; IQR, interquartile range; LVI, lymphovascular invasion; STAS, spread through air spaces; TKIs, tyrosine kinase inhibitors; TNM, tumor-node-metastasis; VPI, visceral pleural invasion. Clinicopathologic features of patients with resected invasive lung adenocarcinoma *, statistically significant. GGO, ground-glass opacity; IQR, interquartile range; LVI, lymphovascular invasion; STAS, spread through air spaces; TKIs, tyrosine kinase inhibitors; TNM, tumor-node-metastasis; VPI, visceral pleural invasion. Stratified survival outcomes of ROS1-rearranged LUADs To evaluate the prognostic impact of ROS1 rearrangement, patients with pure ground-glass opacities or lepidic-predominant adenocarcinoma were excluded from survival analyses due to excellent long-term outcomes. Additionally, patients who received adjuvant TKIs were removed to avoid confounding by targeted treatment effects. Then RFS and OS were analyzed in resected LUAD stratified by stage, radiologic, histologic, and pathologic aggressive characteristics. The final analytical cohort consisted of 2,300 patients, with a median follow-up duration of 66.8 months (Figure 1). Study flow diagram and stratified analysis framework. GGO, ground-glass opacity; OS, overall survival; RFS, recurrence-free survival; TKI, tyrosine kinase inhibitor. In the overall cohort, no statistically significant differences in RFS (P=0.37) and OS (P=0.13) were observed between ROS1-positive and ROS1-negative patients (Figure 2). Among patients with stage II–III disease, ROS1 rearrangement was significantly associated with worse RFS (P=0.03) and OS (P=0.02). This association was further substantiated in aggressive subgroup, where ROS1 rearrangement significantly predicted inferior RFS (P=0.02) and OS (P=0.02). Similarly, within the subgroup of solid nodules, ROS1-positivity conferred significantly poorer OS (P=0.02), with a non-significant trend toward poorer RFS (P=0.09). Within the subset of poorly differentiated tumors, ROS1-positive status was significantly linked to reduced OS (P=0.03) and demonstrated a trend for reduced RFS (P=0.08). In contrast, both RFS and OS were comparable between ROS1-positive and ROS1-negative patients with stage I, part-solid, moderately differentiated, or passive subgroups (Figure 3). Kaplan-Meier curves comparing overall survival (A) and recurrence-free survival (B) between ROS1-positive and ROS1-negative patients after complete resection of invasive lung adenocarcinoma. Shaded areas represent 95 percent confidence intervals. Subgroup Kaplan-Meier analyses of overall survival (A) and recurrence-free survival (B) by ROS1 status across four clinicopathological stratification domains: (I) tumor pathological stage (Stage I vs. Stage II–III), (II) radiographic pattern (part-solid vs. solid), (III) histologic differentiation (moderately vs. poorly differentiated), and (IV) pathologic aggressiveness (passive vs. aggressive). To further evaluate the prognostic impact of ROS1 rearrangement, we performed Cox multivariable analysis for RFS and OS in each subgroup. ROS1 rearrangement persisted as an independent factor of worse RFS and OS, respectively, in patients with stage II–III, solid, poorly differentiated, and aggressive subgroups. Conversely, it failed to demonstrate independent prognostic value in the stage I, part-solid, moderately differentiated and passive subgroups (Tables 2,3). Table 2Multivariable analyses on recurrence-free survival in each subgroup of lung adenocarcinoma patientsVariablesStage I (n=1,915)Stage II-III (n=385)Part-solid (n=986)Solid (n=1,314)Moderately (n=2,058)Poorly (n=242)Passive (n=1,531)Aggressive (n=769)Hazard ratio (95% CI)P valueHazard ratio (95% CI)P valueHazard ratio (95% CI)P valueHazard ratio (95% CI)P valueHazard ratio (95% CI)P valueHazard ratio (95% CI)P valueHazard ratio (95% CI)P valueHazard ratio (95% CI)P valueROS1 rearrangement Positive/negative1.463 (0.767–2.789)0.252.461 (1.370–4.419)0.003*0.532 (0.129–2.198)0.382.259 (1.437–3.552)<0.001*1.499 (0.853–2.636)0.163.077 (1.585–5.988)<0.0011.650 (0.794–3.431)0.181.905 (1.120–3.239)0.02*Age1.011 (0.996–1.027)0.141.013 (0.997–1.029)0.11––1.012 (1.000–1.024)0.0501.011 (0.998–1.025)0.10––––1.014 (1.001–1.027)0.04*Sex Male/female1.082 (0.718–1.630)0.711.064 (0.703–1.608)0.77––0.980 (0.716–1.340)0.900.952 (0.664–1.366)0.79––––1.179 (0.846–1.642)0.33Smoking history Ever/never0.975 (0.623–1.525)0.911.121 (0.742–1.695)0.591.705 (0.936–3.107)0.081.068 (0.774–1.475)0.691.365 (0.929–2.006)0.110.760 (0.519–1.112)0.161.143 (0.702–1.861)0.590.951 (0.680–1.333)0.77Stage II–III/INANANANA4.220 (2.111–8.435)<0.001*2.880 (2.198–3.774)<0.001*3.677 (2.693–5.022)<0.001*1.524 (1.030–2.254)0.04*4.081 (2.336–7.131)<0.001*2.523 (1.910–3.333)<0.001*Radiologic pattern Solid/part-solid2.023 (1.395–2.936)<0.001*2.525 (1.326–4.808)0.005*NANANANA1.990 (1.437–2.756)<0.001*––1.233 (0.762–1.994)0.393.279 (2.075–5.155)<0.001*Histologic differentiation Poorly/moderately2.989 (2.030–4.401)<0.001*1.223 (0.881–1.698)0.23––1.787 (1.377–2.318)<0.001*NANANANA1.957 (1.046–3.661)0.04*1.606 (1.213–2.126)<0.001*Pathologic feature Aggressive/passive4.554 (3.249–6.383)<0.001*2.591 (1.770–3.793)<0.001*2.113 (1.189–3.755)0.01*4.464 (3.311–5.988)<0.001*3.805 (2.827–5.122)<0.001*2.880 (1.737–4.775)<0.001*NANANANASurgery type Sublobar/lobectomy1.154 (0.812–1.641)0.42––0.671 (0.370–1.218)0.191.131 (0.803–1.594)0.480.934 (0.650–1.343)0.71––0.665 (0.390–1.134)0.13––Adjuvant chemotherapy Yes/no1.222 (0.809–1.846)0.34––––0.947 (0.720–1.246)0.701.069 (0.767–1.489)0.69––1.322 (0.719–2.430)0.370.805 (0.603–1.075)0.14Variables with P≥0.1 on univariate analysis were excluded from multivariable modeling. *, statistically significant. CI, confidence interval; NA, not applicable as there is only one variable in the category. Multivariable analyses on recurrence-free survival in each subgroup of lung adenocarcinoma patients Variables with P≥0.1 on univariate analysis were excluded from multivariable modeling. *, statistically significant. CI, confidence interval; NA, not applicable as there is only one variable in the category. Table 3Multivariable analyses on overall survival in each subgroup of lung adenocarcinoma patientsVariablesStage I (n=1,915)Stage II-III (n=385)Part-solid (n=986)Solid (n=1,314)Moderately (n=2,058)Poorly (n=242)Passive (n=1,531)Aggressive (n=769)Hazard ratio (95% CI)P valueHazard ratio (95% CI)P valueHazard ratio (95% CI)P valueHazard ratio (95% CI)P valueHazard ratio (95% CI)P valueHazard ratio (95% CI)P valueHazard ratio (95% CI)P valueHazard ratio (95% CI)P valueROS1 rearrangement Positive/negative1.966 (0.990–3.904)0.0543.612 (1.957–6.668)<0.001*0.459 (0.062–3.405)0.453.597 (2.245–5.764)<0.001*1.742 (0.963–3.149)0.075.291 (2.571–10.870)<0.001*2.407 (0.929–6.241)0.072.848 (1.699–4.772)<0.001*Age1.032 (1.013–1.051)0.001*1.023 (1.006–1.041)0.008*––1.028 (1.015–1.042)<0.001*1.027 (1.011–1.043)0.111.028 (1.006–1.049)0.011*––1.034 (1.021–1.049)<0.001*Sex Male/female2.558 (1.629–4.017)<0.001*0.901 (0.572–1.418)0.656.334 (2.607–15.390)<0.001*1.159 (0.819–1.639)0.411.428 (0.961–2.122)0.08––1.929 (0.855–4.348)0.111.375 (0.969–1.957)0.08Smoking history Ever/never0.616 (0.387–0.980)0.04*1.534 (0.984–2.387)0.060.641 (0.275–1.494)0.301.206 (0.857–1.698)0.281.124 (0.749–1.687)0.57––0.767 (0.307–1.915)0.571.082 (0.769–1.522)0.65Stage II-III/INANANANA3.467 (1.389–8.653)0.008*3.382 (2.521–4.536)<0.001*3.979 (2.803–5.649)<0.001*1.932 (1.254–2.979)0.003*7.940 (3.890–16.205)<0.001*2.565 (1.953–3.368)<0.001*Radiologic pattern Solid/part-solid2.026 (1.274–3.222)0.003*2.525 (1.326–4.808)<0.001*NANANANA2.388 (1.582–3.606)<0.001*––1.048 (0.483–2.271)0.913.413 (2.092–5.556)<0.001*Histologic differentiation Poorly/moderately2.375 (1.529–3.691)<0.001*1.159 (0.829–1.620)0.39––1.506 (1.143–1.986)0.004*NANANANA1.565 (0.610–4.014)0.351.475 (1.113–1.955)0.007*Pathologic feature Aggressive/passive12.813 (7.734–21.225)<0.001*4.638 (2.856–7.532)<0.001*5.982 (2.833–12.634)<0.001*10.417 (6.849–15.873)<0.001*8.871 (5.859–13.433)<0.001*7.179 (3.441–14.977)<0.001*NANANANASurgery type Sublobar/lobectomy––––––1.585 (1.102–2.279)0.011.535 (1.020–2.310)0.04*––0.543 (0.211–1.399)0.21––Adjuvant chemotherapy Yes/no0.961 (0.591–1.563)0.87––––0.936 (0.701–1.251)0.661.084 (0.756–1.555)0.66––––––Variables with P≥0.1 on univariate analysis were excluded from multivariable modeling. *, statistically significant. CI, confidence interval; NA, not applicable as there is only one variable in the category. Multivariable analyses on overall survival in each subgroup of lung adenocarcinoma patients Variables with P≥0.1 on univariate analysis were excluded from multivariable modeling. *, statistically significant. CI, confidence interval; NA, not applicable as there is only one variable in the category. Postoperative recurrent patterns of ROS1-rearranged LUADs Within the survival analysis cohort, no significant difference between the overall recurrence rate of ROS1-positive patients and ROS1-negative patients was observed. (P=0.44). ROS1-positive patients experienced metastases to the thorax (n=12), abdomen (n=4), brain (n=6), and bone (n=9). ROS1 rearrangement was associated with a higher incidence of bone recurrence, with a prevalence of 6.7% in ROS1-positive patients versus 3.1% in ROS1-negative patients (P=0.047). Recurrence at other metastatic sites was comparable between the two groups (Table 4). Table 4Recurrence sites of ROS1-positive and ROS1-negative patients with lung adenocarcinomaRecurrenceROS1+ (n=134)ROS1− (n=2,166)P valueOverall, n (%)0.44 No111 (82.8)1,847 (85.3) Yes23 (17.2)319 (14.7)Thorax, n (%)0.70 No122 (91)1,992 (92) Yes12 (9)174 (8)Brain, n (%)0.81 No128 (95.5)2,059 (95.1) Yes6 (4.5)107 (4.9)Bone, n (%)0.047* No125 (93.3)2,098 (96.9) Yes9 (6.7)68 (3.1)Abdomen, n (%)0.86 No130 (97)2,115 (97.6) Yes4 (3)51 (2.4)Neck, n (%)0.48 No134 (100)2,144 (99) Yes0 (0)22 (1)*, statistically significant. Recurrence sites of ROS1-positive and ROS1-negative patients with lung adenocarcinoma *, statistically significant. Discussion Our study delineated the clinicopathologic profile of ROS1-rearranged LUAD, confirming its association with classic demographic and high-risk pathologic features. Consistent with prior literature, ROS1 rearrangement was significantly more prevalent among never-smokers (10-12,23). Radiologically and histologically, ROS1-positive tumors demonstrated a strong predilection for the solid growth pattern, presenting more frequently as pure solid nodules on CT and as solid-predominant subtype upon pathologic examination. Furthermore, the significant association with LVI underscores a potentially more aggressive phenotype inherent to a subset of these tumors. The observed frequency of ROS1 rearrangement in this cohort was notably higher (4.5%); the most critical reason was that our cohort was exclusively composed of patients with resected invasive LUAD and explicitly excluded pre-invasive lesions. In addition, the study population showed a high proportion of never-smokers (72.6%). The intrinsic prognostic significance of ROS1 rearrangement, independent of TKI therapy, has been a subject of ongoing debate. Early foundational studies suggested that ROS1-rearranged LUAD might constitute a tendency toward high histologic grade, but ROS1-positive and ROS1-negative patients showed no difference in survival (23,25,26). Yu et al. (27) reported that ROS1 rearrangement was associated with better prognosis in LUAD, conversely, Cai et al. (24) reported significant association linking ROS1-positivity with unfavorable survival outcomes. These conflicting findings likely stem from methodological variations, including differences in cohort size, histology distribution, and the confounding effects of adjuvant therapy. Our analysis, conducted on a large, surgically resected cohort with adjuvant TKI patients excluded, found no significant prognostic impact in the overall population. However, we further demonstrated that the prognostic effect of ROS1 rearrangement is not universal but is context-dependent, becoming apparent only when specific high-risk features are present. The core finding of our study is that ROS1-positivity emerged as an independent factor of worse RFS and OS specifically in subgroups representing more advanced or aggressive disease, specifically stage II-III, solid radiological appearance, poor histological differentiation, and the presence of aggressive pathologic features (LVI, VPI, or STAS). A plausible explanation for this stratified effect is that tumors at a more advanced stage of progression have accumulated additional genetic and microenvironmental changes. The oncogenic drive from the ROS1 rearrangement may synergize with these coexisting alterations, thereby accelerating disease recurrence and mortality (12,34,35). In contrast, in early-stage or indolent subgroups (stage I, part-solid, moderately differentiated, passive pathologic features), the generally favorable biology and low metastatic potential may overshadow the negative impact of the ROS1 driver, rendering its prognostic value negligible. Furthermore, the tumor microenvironment or intrinsic cellular differentiation state in these less aggressive contexts might suppress the full oncogenic potential of the ROS1 rearrangement. Beyond survival outcomes, our analysis revealed a distinct pattern of recurrence associated with ROS1 rearrangement. While the overall recurrence rate was comparable to ROS1-negative patients, ROS1-positive patients exhibited a significantly higher propensity for bone metastasis. This site-specific tropism suggests a unique biology that may be influenced by the ROS1-driven signaling axis, which may facilitate tumor progression within the bone marrow microenvironment. Wang et al. (36) reported that CD74-ROS1 rearrangement was detected in bone metastasis cells but not in lung metastasis cells from in vivo model of LUAD. Further, they demonstrated that targeting the CD74-ROS1 and the downstream axis with crizotinib in vivo strongly impeded bone metastasis and secondary metastasis. This finding has direct clinical implications, suggesting that ROS1-positive patients, particularly those with other high-risk features, may benefit from more vigilant bone surveillance. Several limitations of our study warrant consideration. First, its retrospective and single-center design, despite the large cohort, may introduce selection bias and limit generalizability. Second, molecular testing was not uniformly performed across all patients, and the study population was limited to those with available ROS1 testing results. Third, while FISH supplemented by immunohistochemistry (IHC) provides a reliable approach for detecting ROS1 rearrangement (33,37), reverse transcription polymerase chain reaction and next-generation sequencing were not routinely implemented during the initial phase of this study. Consequently, the absence of co-mutations data limits our ability to elucidate the mechanistic basis for the pronounced prognostic impact of ROS1 in advanced disease, highlighting an important avenue for future investigation. In addition, incomplete temporal information on TKI therapy precluded its appropriate incorporation into survival analyses, and exclusion of these patients may introduce selection bias. On the other hand, propensity score matching was not utilized as it would have substantially diminished the already limited sample size of ROS1-positive cases within stratified subgroups, thereby critically compromising statistical power. Our primary aim was to evaluate prognostic heterogeneity of ROS1 rearrangement across predefined stratifications rather than estimate causal effects, for which multivariable Cox regression represents an appropriate methodological approach while preserving cohort size for meaningful stratified analyses (38,39). This statistical methodology has been validated and applied in studies investigating the stratified prognostic effects of other driver mutations in LUAD (28,29). Conclusions ROS1 rearrangement plays a significant prognostic role in advanced progression of LUAD; however, the generally favorable survival outcomes of early-stage LUAD may obscure its prognostic value. Supplementary The article’s supplementary files as Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0292/prf