Modality-specific survival outcomes of biomarker-matched therapy in unresected non-small cell lung cancer: a clinicogenomic cohort study Authors: Jiantong Sun, Zhenyu Yang, Yin Ku, Mao Liu, Dongsheng Wu, Shiyou Wei, Lunxu Liu Published: 2026 ABSTRACT AbstractBackgroundBiomarker-matched therapy (BMT), including targeted therapy (TT) and immune checkpoint blockade (ICB), has improved outcomes in selected patients with non-small cell lung cancer (NSCLC). However, real-world evidence separating the survival associations of TT and ICB within BMT remains limited. We evaluated BMT and modality-specific survival associations in patients with unresected NSCLC using the Memorial Sloan Kettering Clinicogenomic Harmonized Oncologic Real-World Dataset (MSK-CHORD).MethodsIn total, 1,992 patients with unresected NSCLC were analyzed. BMT was defined as receipt of either matched TT or ICB according to the recorded tumor biomarker profile. Overall survival (OS) was compared using propensity-score overlap-weighted Cox models. Modality-specific analyses separately assessed TT in the TT-actionable group and ICB in the ICB-actionable group. Specimen-source, race-exclusion, landmark, and time-dependent Cox analyses were performed as sensitivity analyses.ResultsAmong 1,992 patients with unresected NSCLC, 894 received BMT. After overlap weighting, BMT was associated with longer OS than no BMT [adjusted hazard ratio (aHR): 0.68; 95% confidence interval (CI): 0.59–0.79], with a stronger association in stage IV disease (aHR: 0.58; 95% CI: 0.49–0.69). In modality-specific analyses, TT was associated with longer OS in the TT-actionable group (aHR: 0.66; 95% CI: 0.53–0.82), with the clearest association in stage IV disease (aHR: 0.51; 95% CI: 0.40–0.65). ICB was not associated with longer OS in the ICB-actionable group (aHR: 1.09; 95% CI: 0.89–1.33). Findings were consistent across specimen-source and race sensitivity analyses, whereas landmark and time-dependent analyses attenuated the BMT and overall TT estimates.ConclusionsIn unresected NSCLC, BMT was associated with longer OS in sequencing-anchored real-world analyses, particularly in stage IV disease. Separate modality-specific analyses showed that the clearest survival association was observed for TT, whereas ICB did not show a comparable association in the ICB-actionable group. These findings support reporting BMT, TT, and ICB separately when interpreting real-world outcomes of biomarker-matched systemic therapy.Keywords: Non-small cell lung cancer (NSCLC), biomarker-matched therapy (BMT), targeted therapy (TT), immune checkpoint blockade (ICB), overlap weighting Highlight box Key findings • In unresected non-small cell lung cancer (NSCLC), biomarker-matched therapy (BMT) was associated with longer overall survival (OS) in sequencing-anchored models. Separate analyses showed the clearest survival association for targeted therapy (TT) in stage IV TT-actionable disease, whereas immune checkpoint blockade (ICB) was not associated with longer OS in the ICB-actionable group. What is known and what is new? • BMT has improved outcomes in selected patients with NSCLC, but TT and ICB differ in biomarker requirements, mechanisms, and clinical indications. • This study separates BMT from modality-specific TT and ICB analyses in an unresected real-world NSCLC cohort. Timing-sensitive analyses further showed that sequencing-anchored survival estimates may be influenced by treatment initiation timing. What is the implication, and what should change now? • Real-world precision-oncology studies should report BMT, TT, and ICB separately when evaluating survival outcomes in unresected NSCLC, while using sensitivity analyses to assess potential immortal time bias. Introduction Lung cancer is the leading cause of cancer mortality worldwide. Non-small cell lung cancer (NSCLC) accounts for most cases, and the 5-year survival rate for NSCLC remains poor at roughly 15–25% (1,2). Biomarker-matched therapy (BMT), including targeted therapy (TT) and immune checkpoint blockade (ICB), has transformed the clinical management of NSCLC by linking systemic treatment to tumor genomic and immunologic profiles (3). Precision oncology approaches have significantly extended survival for patients whose tumors harbor specific actionable genomic alterations or immunologic markers (4). However, phase 3 trials often exclude patients with poor performance status, significant comorbidities, or complex treatment histories. This exclusion creates a knowledge gap regarding precision-oncology outcomes in heterogeneous real-world populations (5). Although real-world evidence supports the clinical value of companion diagnostics and biomarker-driven treatment in molecularly selected NSCLC (6), a major limitation of current observational research is its tendency to evaluate biomarker matching as a single aggregate intervention. Aggregating distinct therapeutic modalities obscures their different biological mechanisms, clinical indications, and prognostic trajectories. For instance, while ICB provides substantial benefit in appropriately selected patients, its efficacy is poor in tumors driven by select genomic alterations, such as EGFR mutations, which require specific TT (7). To accurately evaluate the real-world survival associations of BMT, systemic disease management must be isolated, and therapeutic modalities separated into strictly defined sub-populations (8). Therefore, we used the Memorial Sloan Kettering Clinicogenomic Harmonized Oncologic Real-World Dataset (MSK-CHORD) to evaluate precision-oncology survival associations within an unresected NSCLC cohort. Focusing on unresected patients captured a clinical context in which systemic therapy is central to disease management while reducing heterogeneity from therapeutic resection, perioperative treatment, and the overlap between adjuvant and palliative intent. We decomposed BMT into separate analyses of TT and ICB within their actionable populations. We applied overlap weighting, a propensity-score weighting approach, to balance clinical stage, histology, and molecular features (9). Through this modality-specific approach, together with an exploratory treatment-pattern analysis in EGFR-mutated disease, we aimed to provide a more precise assessment of BMT in patients with unresected NSCLC. We present this article in accordance with the STROBE reporting checklist (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0476/rc). Methods Study design and data source We conducted a retrospective cohort study using the MSK-CHORD, a de-identified clinicogenomic dataset from Memorial Sloan Kettering Cancer Center that links tumor sequencing with longitudinal clinical data from routine care (10). Genomic alterations in MSK-CHORD were derived from tumor biopsy sequencing using Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT), a Food and Drug Administration-authorized targeted sequencing assay with matched blood sequencing to filter germline and clonal hematopoiesis variants. We included patients diagnosed with NSCLC from the MSK-CHORD data release based on the 9 September 2023 data snapshot. To mitigate unmeasured confounding related to perioperative treatment dynamics, surgical eligibility, and the overlap between adjuvant and palliative intent, we strictly excluded patients who underwent any therapeutic surgical procedure. Surgical status was ascertained from procedural records; diagnostic sample acquisitions, such as biopsies, were not classified as therapeutic surgeries. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Exposures and outcomes The primary exposure was BMT, defined as receipt of either matched TT or ICB according to the patient’s recorded tumor biomarker profile. Patients who received either or both modalities were counted as BMT-exposed in the BMT analysis. TT and ICB were then evaluated separately in their respective actionable groups. The operational matching rules are provided in Table S1. For modality-specific analyses, we used predefined actionable groups. The TT-actionable group was defined using the analytic TT eligibility field aligned with the reviewed biomarker-treatment matching framework. The ICB-actionable group was defined as programmed death-ligand 1 (PD-L1) positivity or high tumor mutational burden (TMB). PD-L1 positivity was defined as ≥1%, and high TMB was defined as ≥10 mutations/Mb. Patients with unknown PD-L1 status entered the ICB-actionable group if high TMB was present. Within these groups, TT was compared with no matched TT, and ICB was compared with no ICB. TT was defined as receipt of a targeted agent matched to a recorded actionable genomic alteration in the MSK-CHORD treatment and genomic profiles. The reviewed matching framework incorporated established NSCLC driver alterations and tumor-agnostic actionable targets, and required the concurrent recording of both a relevant genomic target and a corresponding targeted agent to classify TT exposure. The primary outcome was overall survival (OS). In MSK-CHORD, OS was defined from the date of the first sequenced sample to death from any cause or last contact. We right-censored surviving patients at the date of their last available follow-up. Vital status and follow-up time were derived from the MSK-CHORD release. Statistical analysis Baseline characteristics were summarized using frequencies and percentages. We evaluated between-group differences using the Pearson chi-square test or Fisher’s exact test. To address non-randomized treatment assignment and mitigate selection bias, we applied propensity-score-based overlap weighting. We estimated propensity scores for each treatment contrast using unpenalized multivariable logistic regression with prespecified baseline covariates. Overlap weights were calculated as the probability of assignment to the opposite treatment level, thereby emphasizing patients with the greatest covariate overlap between treated and untreated groups. We evaluated post-weighting covariate balance using absolute standardized mean differences (SMDs), with an SMD <0.1 considered adequate balance. We compared survival outcomes using overlap-weighted Cox proportional hazards models. Adjusted hazard ratios (aHRs) and 95% confidence intervals (CIs) were obtained from these covariate-adjusted models. We conducted an exploratory four-group treatment-pattern analysis in the EGFR-mutated subgroup. Patients were categorized into four groups according to the reviewed TT definition and ICB exposure: TT without ICB, no TT or ICB, ICB without TT, and ICB plus TT. Overlap weighting was not applied in this subset due to the limited sample size in the ICB-containing arms. The model was adjusted for age, sex, race, smoking status, histology, clinical stage, and specimen source. Because OS was anchored at the first sequenced sample, treatment could start before, at, or after cohort entry. Classifying patients who started treatment after cohort entry as treated from baseline can misclassify treatment-free person-time and introduce immortal time bias (11). We therefore performed landmark analyses and time-dependent Cox models as sensitivity analyses for potential immortal time bias (12-14). First, we conducted landmark analyses at 3 and 6 months after sequencing. Patients who died or were lost to follow-up before the landmark time were excluded, treatment status was assigned according to whether the relevant therapy had started by the landmark time, and OS follow-up began at the landmark time. Second, we implemented time-dependent Cox proportional hazards models. Treatments initiated before or on the sequencing date were treated as baseline exposure; for treatments initiated after cohort entry, patients contributed unexposed person-time before treatment initiation and exposed person-time thereafter. Because the primary BMT contrast was conducted in the full unresected cohort, we added a sensitivity analysis restricted to the combined actionable group, defined as patients in either the TT-actionable group or the ICB-actionable group. This analysis used the same BMT definition, OS endpoint, overlap-weighted Cox framework, and covariate structure whenever estimable. This analysis assessed whether the BMT association was consistent among patients with biomarker-defined treatment opportunities. Prespecified supplementary analyses included sequencing specimen-source-stratified models and leave-one-out sensitivity models where estimable. We also summarized actionable alteration distributions and PD-L1 unknown status to support interpretation of the biomarker-defined target populations. Results Patient selection and baseline characteristics A total of 1,992 patients with unresected NSCLC met the inclusion criteria, including 894 patients who received BMT by the revised operational definition and 1,098 who did not (Figure 1). Prior to weighting, baseline clinical and molecular characteristics exhibited significant imbalances between the two cohorts (Table 1). Study cohort derivation. BMT, biomarker-matched therapy; ICB, immune checkpoint blockade; MSK-CHORD, Memorial Sloan Kettering Clinicogenomic Harmonized Oncologic Real-World Dataset; NSCLC, non-small cell lung cancer; TT, targeted therapy. Table 1Baseline characteristics of the unresected NSCLC cohort stratified by BMT exposureCharacteristicAll (N=1,992)No BMT (N=1,098)BMT (N=894)P valueAge0.023 ≤70 years1,028 (51.6)541 (49.3)487 (54.5) >70 years964 (48.4)557 (50.7)407 (45.5)Sex<0.001 Male883 (44.3)530 (48.3)353 (39.5) Female1,109 (55.7)568 (51.7)541 (60.5)Race<0.001 White1,511 (75.9)877 (79.9)634 (70.9) Asian Far East/Indian subcontinent233 (11.7)87 (7.9)146 (16.3) Black or African American117 (5.9)66 (6.0)51 (5.7) Other131 (6.6)68 (6.2)63 (7.0)Smoking status<0.001 Former/current1,386 (69.6)834 (76.0)552 (61.7) Never490 (24.6)173 (15.8)317 (35.5) Unknown116 (5.8)91 (8.3)25 (2.8)Histology<0.001 Adenocarcinoma1,482 (74.4)751 (68.4)731 (81.8) Non-adenocarcinoma510 (25.6)347 (31.6)163 (18.2)Clinical stage (AJCC 8th edition)<0.001 Stage I182 (9.1)126 (11.5)56 (6.3) Stage II75 (3.8)61 (5.6)14 (1.6) Stage III307 (15.4)225 (20.5)82 (9.2) Stage IV1,329 (66.7)632 (57.6)697 (78.0) Unknown99 (5.0)54 (4.9)45 (5.0)Specimen source<0.001 Primary tumor971 (48.7)585 (53.3)386 (43.2) Metastasis996 (50.0)503 (45.8)493 (55.1) Local recurrence9 (0.5)4 (0.4)5 (0.6) Unknown16 (0.8)6 (0.5)10 (1.1)PD-L1 status<0.001 Negative/not positive699 (35.1)463 (42.2)236 (26.4) Positive763 (38.3)241 (21.9)522 (58.4) Unknown530 (26.6)394 (35.9)136 (15.2)High TMB0.034 No1,530 (76.8)823 (75.0)707 (79.1) Yes462 (23.2)275 (25.0)187 (20.9)TT-actionable alteration†<0.001 No459 (23.0)378 (34.4)81 (9.1) Yes1,533 (77.0)720 (65.6)813 (90.9)Values are presented as N (%) unless otherwise indicated. P values were calculated using Pearson chi-square test or Fisher’s exact test as appropriate in the unweighted cohort. †, TT-actionable alteration indicates the analytic TT eligibility field aligned with the reviewed biomarker-treatment matching framework. Actionable alteration categories are summarized in Table S2. AJCC, American Joint Committee on Cancer; BMT, biomarker-matched therapy; NSCLC, non-small cell lung cancer; PD-L1, programmed death-ligand 1; TMB, tumor mutational burden; TT, targeted therapy. Baseline characteristics of the unresected NSCLC cohort stratified by BMT exposure Values are presented as N (%) unless otherwise indicated. P values were calculated using Pearson chi-square test or Fisher’s exact test as appropriate in the unweighted cohort. †, TT-actionable alteration indicates the analytic TT eligibility field aligned with the reviewed biomarker-treatment matching framework. Actionable alteration categories are summarized in Table S2. AJCC, American Joint Committee on Cancer; BMT, biomarker-matched therapy; NSCLC, non-small cell lung cancer; PD-L1, programmed death-ligand 1; TMB, tumor mutational burden; TT, targeted therapy. Compared with the no-BMT cohort, patients in the BMT cohort were significantly younger, with higher proportions of female and Asian patients, as well as never-smokers and those with adenocarcinoma (all P<0.05; Table 1). Clinical stage IV disease was more prevalent in the BMT group (78.0% vs. 57.6%; P<0.001), and sequencing was more frequently performed on metastatic specimens (55.1% vs. 45.8%; P<0.001). Regarding molecular profiles, PD-L1 positivity (58.4% vs. 21.9%; P<0.001) and TT-actionable alterations (90.9% vs. 65.6%; P<0.001) were enriched in the BMT group. High TMB was slightly less frequent in the BMT group (20.9% vs. 25.0%; P=0.034). The distribution of actionable alterations is summarized in Table S2, post-weighting balance diagnostics are shown in Table S3, and handling of PD-L1 unknown status is summarized in Table S4. After overlap weighting, all primary treatment contrasts achieved adequate covariate balance, with maximum weighted SMDs below 0.1 across the BMT, TT, and ICB models (range, 0.002 to 0.083) (Figure 2). Covariate balance before and after overlap weighting. Love plot displaying the absolute standardized mean differences of prespecified model covariates before and after overlap weighting for the three primary treatment contrasts. BMT, biomarker-matched therapy; ICB, immune checkpoint blockade; PD-L1, programmed death-ligand 1; TMB, tumor mutational burden; TT, targeted therapy. Survival outcomes by treatment modality The sequencing-anchored, overlap-weighted survival estimates are summarized in Table 2 and Figure 3. In the overall unresected cohort, BMT was associated with longer OS than no BMT (aHR: 0.68; 95% CI: 0.59–0.79; P<0.001; Figure 3A). The association was also observed in the clinical stage IV subgroup (aHR: 0.58; 95% CI: 0.49–0.69; P<0.001; Table 2). In a sensitivity analysis restricted to the combined actionable group, the BMT association was similar in the overall subgroup (aHR: 0.67; 95% CI: 0.58–0.77; P<0.001) and the stage IV subgroup (aHR: 0.57; 95% CI: 0.48–0.67; P<0.001; Table S5). In the TT-actionable group, TT was associated with longer OS (aHR: 0.66; 95% CI: 0.53–0.82; P<0.001; Figure 3B), with a stronger association in clinical stage IV disease (aHR: 0.51; 95% CI: 0.40–0.65; P<0.001; Table 2). ICB was not associated with longer OS in the ICB-actionable group compared with no ICB (aHR: 1.09; 95% CI: 0.89–1.33; P=0.418; Figure 3C) or in the clinical stage IV ICB-actionable group (aHR: 0.91; 95% CI: 0.72–1.16; P=0.456; Table 2). Table 2Decomposition of biomarker-matched therapy into modality-specific survival associations in unresected NSCLCExposure & populationNOverlap-weighted adjusted HR (95% CI)aP valueBMT All NSCLC1,9920.68 (0.59–0.79)<0.001 Stage IV NSCLC1,3290.58 (0.49–0.69)<0.001TT TT-actionable groupb1,5330.66 (0.53–0.82)<0.001 Stage IV TT-actionable group1,0620.51 (0.40–0.65)<0.001ICB ICB-actionable groupc1,0301.09 (0.89–1.33)0.418 Stage IV ICB-actionable group6530.91 (0.72–1.16)0.456aHazard ratios and 95% confidence intervals were estimated using covariate-adjusted, overlap-weighted Cox proportional hazards models. Models were adjusted for age, sex, race, smoking status, histology, clinical stage, sequencing specimen source, driver alteration class, PD-L1 status, and high tumor mutational burden, as applicable to each analysis. Variables that were constant by design within a target cohort or used to define a subgroup or stratum were omitted from the corresponding model. Maximum standardized mean differences after weighting were <0.1 across all analyses, indicating adequate covariate balance. bThe TT-actionable group was defined using the analytic TT eligibility field aligned with the reviewed biomarker-treatment matching framework. cThe ICB-actionable group was restricted to patients with PD-L1 positivity or tumor mutational burden ≥10 mutations/Mb. BMT, biomarker-matched therapy; CI, confidence interval; HR, hazard ratio; ICB, immune checkpoint blockade; NSCLC, non-small cell lung cancer; PD-L1, programmed death-ligand 1; TT, targeted therapy. Decomposition of biomarker-matched therapy into modality-specific survival associations in unresected NSCLC aHazard ratios and 95% confidence intervals were estimated using covariate-adjusted, overlap-weighted Cox proportional hazards models. Models were adjusted for age, sex, race, smoking status, histology, clinical stage, sequencing specimen source, driver alteration class, PD-L1 status, and high tumor mutational burden, as applicable to each analysis. Variables that were constant by design within a target cohort or used to define a subgroup or stratum were omitted from the corresponding model. Maximum standardized mean differences after weighting were <0.1 across all analyses, indicating adequate covariate balance. bThe TT-actionable group was defined using the analytic TT eligibility field aligned with the reviewed biomarker-treatment matching framework. cThe ICB-actionable group was restricted to patients with PD-L1 positivity or tumor mutational burden ≥10 mutations/Mb. BMT, biomarker-matched therapy; CI, confidence interval; HR, hazard ratio; ICB, immune checkpoint blockade; NSCLC, non-small cell lung cancer; PD-L1, programmed death-ligand 1; TT, targeted therapy. Overlap-weighted Kaplan-Meier estimates of overall survival according to treatment modality. (A) Overall survival comparing BMT versus no BMT in the entire unresected NSCLC cohort. (B) Overall survival comparing TT versus no matched TT in the TT-actionable group. (C) Overall survival comparing ICB versus no ICB in the ICB-actionable group. Shaded areas represent 95% CIs of the overlap-weighted survival curves. Numbers at risk represent unweighted patient counts. BMT, biomarker-matched therapy; CI, confidence interval; ICB, immune checkpoint blockade; NSCLC, non-small cell lung cancer; TT, targeted therapy. Subgroup and sensitivity analyses by sequencing specimen source and race Specimen source did not materially change the overall pattern of association. BMT was associated with longer OS in both primary-tumor specimens (aHR: 0.66; 95% CI: 0.53–0.81; P<0.001) and metastatic specimens (aHR: 0.68; 95% CI: 0.55–0.83; P<0.001; Table S6). Race-exclusion sensitivity analyses were then performed by repeating the primary overlap-weighted models after separately excluding Asian, White, or Black patients. The BMT association remained directionally consistent in all three exclusion cohorts, with aHRs of 0.68 after excluding Asian patients, 0.70 after excluding White patients, and 0.68 after excluding Black patients (Tables S7,S8). Similar source-stratified patterns were observed for TT, with significant associations in both primary-tumor specimens (aHR: 0.62; 95% CI: 0.42–0.93; P=0.019) and metastatic specimens (aHR: 0.61; 95% CI: 0.47–0.79; P<0.001; Table S6 and Figure S1). By contrast, ICB was not associated with longer OS in either primary-tumor specimens (aHR: 0.93; 95% CI: 0.67–1.27; P=0.634) or metastatic specimens (aHR: 1.16; 95% CI: 0.89–1.51; P=0.282; Table S6 and Figure S1). Exploratory treatment-pattern analysis in EGFR-mutated NSCLC Among 524 patients with EGFR-mutated unresected NSCLC, 396 received TT without ICB, 82 received neither TT nor ICB, 16 received ICB without TT, and 30 received both modalities using the reviewed TT definition (Table S9). Median OS was 30.5 months for TT without ICB, 17.0 months for no TT or ICB, 11.0 months for ICB without TT, and 17.3 months for ICB plus TT. In an exploratory multivariable Cox model using TT without ICB as the reference group, significantly higher mortality hazards were observed for patients receiving neither TT nor ICB (aHR: 1.95; 95% CI: 1.32–2.88; P<0.001), ICB without TT (aHR: 3.46; 95% CI: 2.28–5.25; P<0.001), and ICB plus TT (aHR: 1.80; 95% CI: 1.22–2.65; P=0.003). Because the ICB-containing groups were small and likely influenced by treatment selection and prior disease trajectory, this analysis should be interpreted as exploratory rather than confirmatory evidence for comparative effectiveness (Table S9). Sensitivity analyses for potential immortal time bias Sensitivity analyses for potential immortal time bias are summarized in Table S3. In the overall unresected cohort, BMT remained statistically significant in the 3-month landmark analysis (aHR: 0.83; 95% CI: 0.71–0.97; P=0.017), but not in the 6-month landmark analysis (aHR: 0.87; 95% CI: 0.73–1.03; P=0.096) or time-dependent Cox model (aHR: 0.95; 95% CI: 0.83–1.08; P=0.415). In the clinical stage IV subgroup, BMT remained associated with lower mortality in the 3-month landmark analysis (aHR: 0.73; 95% CI: 0.61–0.87; P<0.001), 6-month landmark analysis (aHR: 0.76; 95% CI: 0.63–0.92; P=0.004), and time-dependent Cox model (aHR: 0.79; 95% CI: 0.68–0.92; P=0.003). In modality-specific analyses, the TT association in the TT-actionable group was attenuated in these sensitivity analyses: the 3-month landmark estimate (aHR: 0.99; 95% CI: 0.77–1.28; P=0.942), the 6-month landmark estimate (aHR: 1.01; 95% CI: 0.77–1.33; P=0.931), and the time-dependent Cox model (aHR: 1.20; 95% CI: 0.96–1.51; P=0.103) were not statistically significant. In the clinical stage IV TT-actionable group, the landmark estimates were also attenuated and no longer statistically significant (3-month aHR: 0.81; 95% CI: 0.61–1.06; P=0.119; 6-month aHR: 0.83; 95% CI: 0.61–1.13; P=0.225), and the time-dependent Cox model showed no significant association (aHR: 0.95; 95% CI: 0.73–1.23; P=0.687). Within the ICB-actionable group, ICB was not associated with longer OS in landmark analyses, while the time-dependent Cox model yielded higher estimated mortality hazards (overall aHR: 1.53; 95% CI: 1.26–1.86; P<0.001; clinical stage IV aHR: 1.32; 95% CI: 1.04–1.67; P=0.020). Discussion In this clinicogenomic cohort of patients with unresected NSCLC, BMT was associated with longer OS in sequencing-anchored analyses. Separate analyses of TT and ICB showed different survival patterns. The clearest signal was observed for TT, particularly in stage IV TT-actionable disease. By contrast, ICB was not associated with longer OS in the ICB-actionable group. These findings support evaluating BMT, TT, and ICB separately rather than treating biomarker-matched systemic therapy as a single homogeneous category. The favorable association for TT is clinically plausible. Matched targeted agents have established activity in oncogene-driven NSCLC, particularly in metastatic disease. Because the reviewed TT-exposed group was numerically dominated by EGFR-directed therapy, the overall TT estimate should be interpreted primarily as a real-world signal for matched TT in a driver-enriched group rather than as equivalent evidence across all actionable alterations. Our results for stage IV TT-actionable disease are consistent with the established benefit of molecularly matched systemic therapy in patients with targetable driver alterations (15-17). The findings for ICB require careful interpretation and do not imply that ICB is ineffective in NSCLC. Our ICB analysis compared ICB with no ICB within the ICB-actionable group, defined by PD-L1 positivity or high TMB. However, detailed PD-L1 tumor proportion score (TPS), antibody clone, treatment line, chemotherapy combination, disease burden, and performance status were not uniformly available. These factors can influence ICB outcomes in routine practice (18-20). The null ICB association should therefore be interpreted as a real-world finding within the available biomarker data rather than as a definitive estimate of ICB efficacy. This caution is reinforced by the exploratory analysis in EGFR-mutated NSCLC. Current evidence suggests that EGFR-mutant tumors respond poorly to programmed cell death protein 1 (PD-1) pathway blockade (7,21), and recent trials have not clearly established the benefit of adding ICB to chemotherapy after EGFR-TKI progression (22,23). Recent guideline updates also continue to emphasize biomarker-directed systemic therapy for oncogene-addicted NSCLC, including EGFR, ALK, ROS1, BRAF, MET, RET, NTRK, KRAS G12C, and related actionable alterations (16,17). The landmark and time-dependent Cox analyses were included to examine potential immortal time bias after cohort entry. These analyses were not designed to identify the optimal time to start therapy. Rather, they assessed whether the sequencing-anchored associations changed when treatment started after sequencing was handled differently. The attenuation of some BMT and TT estimates suggests that part of the primary association may reflect treatment access after cohort entry. These models should still be interpreted cautiously, because treatment initiation may be influenced by disease progression, treatment line, clinical fitness, and other dynamic factors that were not uniformly available. We therefore present these models as sensitivity analyses, not as replacements for the sequencing-anchored primary analysis. Key strengths of this study are our focus on an unresected cohort to reduce heterogeneity from perioperative care, the separation of TT from ICB within prespecified actionable groups, and the provision of a transparent drug-gene matching table for TT classification. Overlap weighting achieved adequate balance across measured baseline covariates in the primary models. Our study has several limitations. This was an observational study at a single tertiary center, which may limit generalizability. A uniformly defined treatment-line variable was not available in the de-identified public dataset and could not be incorporated into the primary models. We were also unable to adjust for Eastern Cooperative Oncology Group (ECOG) performance status, quantitative disease burden, metastatic sites, symptoms, organ dysfunction, radiographic response, or clinical deterioration at treatment initiation. BMT was intentionally evaluated as an overall cohort-level contrast in the full unresected cohort. A restricted sensitivity analysis in the combined actionable group showed similar results, but BMT estimates should still be interpreted as aggregate real-world associations rather than causal effects among biomarker-eligible patients. PD-L1 antibody clone, scoring system, and percentage, or TPS were not available; PD-L1 unknown cases were therefore handled as a separate category. Rare or pan-cancer operational matches were retained to avoid opaque post hoc exclusion, but they may not represent uniformly established NSCLC indications; therefore, the primary TT findings should not be interpreted as evidence for all listed rare or pan-cancer matches. Finally, reasons for unresected management among stage I–III patients, such as medical inoperability, patient preference, or unresectability, were not uniformly captured. Despite these limitations, the use of modality-specific target populations, overlap weighting, transparent biomarker-treatment matching rules, and sensitivity analyses for potential immortal time bias provides a more cautious and clinically interpretable framework for evaluating real-world precision-oncology associations in unresected NSCLC. Conclusions In unresected NSCLC, BMT was associated with longer OS in sequencing-anchored real-world analyses. Separate modality-specific analyses showed that the clearest association was observed for TT in stage IV TT-actionable disease, whereas ICB was not associated with longer OS in the ICB-actionable group. These findings support reporting BMT, TT, and ICB separately when evaluating real-world precision-oncology outcomes in unresected NSCLC. Supplementary The article’s supplementary files as Acknowledgments We gratefully acknowledge the patients and their families whose data were included in this study. We extend our sincere appreciation to the investigators and staff at Memorial Sloan Kettering Cancer Center (MSKCC) for their substantial effort in establishing, curating, and making the MSK-CHORD clinicogenomic dataset publicly available to the scientific community. We also thank the cBioPortal for Cancer Genomics team for providing the platform to access and visualize these data. 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. The MSK-CHORD dataset is publicly released in a de-identified form. This analysis used only de-identified data and did not require additional institutional review board approval beyond the oversight of the original MSK-CHORD project. Footnotes Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0476/rc Funding: This work was supported by the 1·3·5 project for disciplines of excellence from West China Hospital of Sichuan University (ZYGD23010 to L.L.); the Postdoctoral Fellowship Program of China Postdoctoral Science Foundation (No. GZB20240495 to Z.Y.); and the ‘Qimingxing’ Research Fund for Young Talents (HXQMX0183 to S.W.). Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0476/coif). Z.Y. reports support from the Postdoctoral Fellowship Program of China Postdoctoral Science Foundation. S.W. reports support from the ‘Qimingxing’ Research Fund for Young Talents. L.L. reports support from the 1·3·5 project for disciplines of excellence from West China Hospital of Sichuan University. The other authors have no conflicts of interest to declare. References Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0476/prf Supplementary Materials The article’s supplementary files as