Targeted Therapy, Immunotherapy, and Molecular Testing in Advanced Solid Tumors: A Medicare Analysis Authors: Onur Baser, Yijia Sun Published: 2026 ABSTRACT ABSTRACTBackground and PurposeBiomarker testing and matched therapies are central to precision oncology, but their real‐world uptake and clinical impact among older adults with advanced solid tumors remain poorly characterized. This retrospective observational cohort study assessed biomarker testing and guideline‐concordant matched first‐line therapy use and evaluated their associations with overall survival, diagnosis‐to‐treatment intervals, and 6‐month physical functioning in these patients.Material and MethodsThis study used linked US Medicare claims, electronic medical records, and patient‐reported outcomes of fee‐for‐service beneficiaries ≥ 65 years with incident advanced/metastatic non‐small cell lung cancer, breast, colorectal/gastric, or urothelial cancer. Biomarker testing (≤ 30 days pre‐diagnosis through first‐line therapy initiation) and guideline‐concordant matched targeted or immunotherapy‐based first‐line treatment (defined as first‐line initiation of a targeted or immuno‐oncology agent whose approved indication corresponded to the patient's actionable biomarker within the study period [Table S5]) were assessed. Primary outcomes were overall survival, time to first‐line therapy, and 6‐month physical functioning. Associations were estimated using inverse probability–weighted Cox proportional hazards and marginal structural models; residual confounding was probed with E‐values and landmark sensitivity analyses.ResultsAmong 39,964 beneficiaries with advanced cancers, 21.6% underwent biomarker testing. Actionable alterations were identified in 9.8% of patients, of whom 62.0% received matched first‐line therapy. Testing modestly increased median diagnosis‐to‐treatment intervals, but delays beyond tumor‐specific medians were not associated with excess mortality. Biomarker testing was associated with longer overall survival, and matched first‐line therapy was associated with higher 6‐month physical functioning across all cohorts (p < 0.001).InterpretationBiomarker testing and matched first‐line therapies were underused but associated with longer overall survival and better physical functioning. Residual confounding is possible and the magnitude of the testing‐survival association likely partially reflects selection of fitter patients into testing. Modest testing‐related treatment delays were not associated with worse survival, supporting expansion of routine biomarker testing in Medicare oncology practice.Keywords: immunotherapy, Medicare, molecular testing, precision oncology, real‐world evidence Introduction Biomarker testing has become central to precision oncology because it identifies molecular alterations that can be matched to targeted therapies and immunotherapies, which in turn are associated with higher response rates, longer survival, and often better tolerability compared with empiric chemotherapy [1, 2, 3, 4]. However, because much of the existing evidence is derived from clinical trials and large academic centers, patterns of biomarker testing, use of matched therapies, and associated outcomes in community practice, particularly among older Medicare beneficiaries with advanced NSCLC, breast, colorectal, gastric, or urothelial cancers, remain poorly defined, which sustains payer and clinician caution regarding broad implementation [5, 6, 7]. Common concerns include delays in treatment initiation while waiting for test results [8, 9, 10], inadequate tissue [11, 12, 13] or failed tests [14, 15], uncertainty about the clinical validity [16] or utility of certain biomarkers [17], and worries about cost and reimbursement [18, 19]. For example, survey and registry data show that turnaround time and fear of delaying therapy are frequently cited reasons for not ordering recommended tests, and that some patients start treatment before results are available, potentially missing the opportunity to receive optimal targeted agents [20]. Given these concerns, the aim of this study was to provide real‐world evidence in the Medicare population by characterizing use of biomarker testing prior to first‐line (1 L) systemic therapy, including the uptake of broad NGS, the frequency of actionable alterations, and the proportion of patients who receive guideline‐concordant, biomarker‐matched targeted or immunotherapy‐based 1 L regimens; and to evaluate whether biomarker testing and biomarker‐matched 1 L therapy are associated with longer overall survival, different diagnosis‐to‐treatment intervals, and better 6‐month patient‐reported physical functioning compared with no testing or empiric chemotherapy. Because this is a retrospective observational analysis, our estimates are described as associations rather than causal effects. Methods Study Design and Dataset This retrospective observational cohort study used the Medicare‐Enhanced Lab & Demographics (MELD) data set, which links 100% Medicare fee‐for‐service (FFS) claims to electronic medical record (EMR) data, laboratory feeds, and patient‐reported outcome (PRO) measures via deterministic patient‐level identifiers. This database contains longitudinal health information on enrollees, representing a mixture of ages and geographic regions across the United States. The data are deidentified in compliance with the Health Insurance Portability and Accountability Act Privacy Rule. Because this study did not constitute human participants research as defined by the Common Rule, institutional review board approval and informed consent were not required. This study adhered to Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines for observational studies [21]. Study Population We included Medicare FFS beneficiaries aged ≥ 65 years with continuous enrollment in Parts A and B and no Medicare Advantage enrollment during the 12 months before and at least 1 month after the cancer index date, between January 1, 2021 and December 31, 2025. Beneficiaries with missing or inconsistent demographic information (age, sex) or with incomplete linkage between claims and EMR/lab data were excluded (n = 2622). Per‐reason attrition counts for each exclusion step (age, insufficient qualifying tumor‐specific claims, prevalent cancer or metastasis in the lookback period, discontinuous enrollment or Medicare Advantage, and absence of confirmed advanced/metastatic disease) are reported in Figure 1 and Table S1. To focus on incident advanced disease, we required a lookback period of at least 12 months prior to the index date with no prior claims evidence of the same primary cancer or metastatic disease; individuals with any such evidence were excluded as prevalent cases (n = 899,958). Cohort attrition diagram for the analytic cohort of older Medicare beneficiaries with advanced solid tumors, with per‐reason exclusion counts and cancer‐type stratification (MELD‐Linked Claims–EMR–Lab–PRO Cohort, January 01, 2021 to December 31, 2025). *ICD‐10‐CM codes used to identify the qualifying primary cancer: NSCLC, C34.x; Breast, C50.x; Colorectal/gastric, C16.x, C18.x–C20, C21.8; Urothelial, C67.x (with C65.x–C66.x when supported by EMR). Primary cancers were identified from Medicare inpatient, outpatient, and carrier claims using International Classification of Diseases, Tenth Revision, Clinical Modification (ICD‐10‐CM) diagnosis codes (Table S2). NSCLC was defined as malignant neoplasm of bronchus and lung (C34.x). Because histologic subtype is not reliably available in claims, we restricted to non‐small cell histology when EMR pathology data were available and otherwise treated lung primaries as NSCLC for the main analysis; in the analytic NSCLC cohort (N = 20,603), 16,478 patients (80.0%) had EMR‐documented non‐small cell histology and 4125 (20.0%) were classified as NSCLC by default. A prespecified histology‐confirmed sensitivity analysis is reported in Table S10. Breast cancer was defined using C50.x codes; colorectal cancer (CRC) using C18.x–C20 and C21.8; gastric using C16.x; and urothelial cancer using C67.x (with sensitivity analyses including upper tract urothelial sites such as C65.x and C66.x when supported by EMR documentation). The index cancer diagnosis date was defined as the earliest claim date meeting a tumor‐specific algorithm requiring at least two claims with a qualifying primary cancer code separated by at least 7 days but no more than 90 days to reduce misclassification from rule‐out diagnoses. Advanced/metastatic disease was determined using a combination of metastatic‐site claims codes and EMR staging information. In claims, we required at least 1 ICD‐10‐CM secondary malignant neoplasm code (C77.x–C79.x) in any diagnosis position recorded on or within 60 days before or after the primary cancer index date. Where available, American Joint Committee on Cancer (AJCC) or summary stage documented in the EMR at or within 60 days of diagnosis was used to confirm Stage IV or metastatic disease. In cases of discordance, EMR AJCC/summary stage always took precedence: 4586 patients with a claims metastatic code but EMR‐documented nonmetastatic stage were excluded, and 2340 patients with EMR Stage IV but no claims metastatic code were retained. The overall claims–EMR discordance rate for metastatic status (among patients with both data streams available) was 6.4%. We excluded patients with evidence of metastatic disease in the 12‐month lookback period (to avoid prevalent metastatic cases) and those with more than one primary tumor meeting incident advanced criteria (n = 4678), unless one tumor type clearly preceded the other; in the latter case only, the first qualifying primary was retained. Figure 1 and Table S1 present the full per‐reason attrition scheme. Next‐generation sequencing (NGS) and biomarker testing were available in MELD. Full biomarker test definitions, CPT/HCPCS codes, single‐analyte versus panel operational rules, and the prespecified ≥ N‐gene requirements used to classify a test as “broad NGS” are provided in Table S4. Briefly, for NGS, we identified solid‐tumor genomic sequencing panels (DNA or DNA/RNA) targeting multiple genes (e.g., panels covering EGFR, ALK, ROS1, BRAF, KRAS, ERBB2/HER2, NTRK, RET, MET, and others), as well as comprehensive genomic profiling assays. Single‐analyte biomarker tests were identified for the following: EGFR, ALK, ROS1, BRAF, KRAS, NRAS, ERBB2/HER2, PDL1, and microsatellite instability (MSI) or mismatch repair (MMR) status. For PDL1, we captured immunohistochemistry assays with tumor proportion score or combined positive score; for MSI/MMR, we captured PCR‐based MSI testing and immunohistochemistry for MLH1, MSH2, MSH6, and PMS2, as well as NGS‐derived MSI/MMR calls when available. For each patient, biomarker testing events were defined by the result date. Any biomarker testing pre‐1 L was defined as the presence of at least 1 biomarker or NGS test (single‐gene or panel) performed between 30 days before the index cancer diagnosis and the day before 1 L systemic therapy initiation. Systemic anticancer therapies were ascertained from Medicare Parts A and B claims and Part D pharmacy data. HCPCS J‐codes, NDC‐mapped drug lists, and drug‐class assignments are provided in Table S3. Intravenous chemotherapy, targeted therapies, and immune checkpoint inhibitors were identified using Healthcare Common Procedure Coding System (HCPCS) J‐codes in the chemotherapy drugs range (J9000–J9999) and additional drug‐specific HCPCS codes for immunotherapies (e.g., PD1/PDL1 and CTLA4 inhibitors) and IV targeted agents. Oral oncolytics were identified from Part D or pharmacy claims using National Drug Codes (NDCs) mapped via a drug dictionary to generic and brand names for relevant agents in each tumor type. 1 L systemic therapy was defined as the earliest date on or after the index cancer diagnosis with a qualifying IV or oral systemic agent. We then defined biomarker‐matched targeted or immuno‐oncology regimens at the patient level. For each tumor type, we specified a mapping between actionable biomarker results and corresponding 1 L therapies with established indications in the study period (Table S5), following NCCN 2024–2025 tumor‐specific guidelines and ESMO 2023 molecular testing consensus [22, 23, 24, 25, 26, 27]. For each patient, we considered a 1 L regimen biomarker‐matched if a relevant actionable biomarker was documented on or before 1 L initiation (or within 14 days after) and the initial regimen included at least one systemic agent with a corresponding indication. The primary exposure variables were receipt of any biomarker testing pre‐1 L and receipt of biomarker‐matched targeted or immuno‐oncology therapy as part of 1 L treatment. The primary outcomes were overall survival, diagnosis‐to‐1 L therapy interval, and 6‐month patient‐reported physical functioning. Overall survival was defined as time from index cancer diagnosis to death from any cause, ascertained from the Medicare beneficiary enrollment files; patients were censored at the earliest of loss of FFS eligibility, end of data (31 December 2025), or 12 months after index. A prespecified extended‐follow‐up sensitivity analysis to 24 months is reported in Table S7. Time from diagnosis to 1 L therapy was the number of days between the recorded date of cancer diagnosis and the start date of 1 L systemic treatment. The EORTC QLQ‐C30 Physical Functioning scale is a 0–100 patient‐reported score, based on 5 questions about everyday activities, where higher values indicate better ability to perform physical tasks. Statistical Analysis We described baseline characteristics using means and SD or medians and interquartile ranges for continuous variables and counts with percentages for categorical variables. We reported overall and tumor‐specific descriptive statistics, including median age; proportions of females and major racial/ethnic groups; the percentage of patients with an Elixhauser Comorbidity Index score (calculated from claims in the prior 12 months) of at least 2; and the proportion with dual eligibility for both Medicare and Medicaid. Cox proportional hazards models were used to estimate adjusted hazard ratios (HRs) for all‐cause mortality associated with biomarker testing and biomarker‐matched 1 L therapy. All outcome, testing, and treatment models included the same covariate set: age, sex, race/ethnicity, US Census region, calendar year of diagnosis, Elixhauser Comorbidity Index, dual Medicare‐Medicaid eligibility, tumor type, and stage‐defining data source (claims‐only, EMR‐only, or both). Missing data on race/ethnicity (2.1% of the analytic cohort) and Elixhauser components (< 1%) were handled with multiple imputation by chained equations (m = 20) and combined using Rubin's rules. Models were weighted using stabilized inverse probability of treatment weights and inverse probability of censoring weights truncated at the 1st and 99th percentiles. Covariate balance was assessed with standardized mean differences (SMDs) before and after weighting, with |SMD| < 0.10 considered balanced; all baseline covariates met this threshold post‐weighting (Table S6). Stabilized weights had mean 1.00 (SD 0.31), median 0.97 (IQR 0.83–1.15), and range 0.14–7.42 before truncation, with a post‐truncation range of 0.34–3.11 and an effective sample size of 37,891 (94.8% of the analytic cohort). Proportional hazards assumptions were assessed using Schoenfeld residuals and time‐interaction terms; where assumptions were violated, we used stratified Cox models. Because biomarker testing both causes delay and is related to prognosis, we used a marginal structural model to estimate the effect of testing‐related delays on overall survival. We divided the time after diagnosis into weekly intervals. For each interval, we fitted logistic models for the probability of receiving biomarker testing in that interval and the probability of remaining untreated versus starting 1 L in that interval given the covariates listed above, and used these to compute stabilized inverse probability weights. We then applied these weights to a marginal structural Cox model to estimate the effect of treatment delays on overall survival. Testing was coded as a time‐varying exposure (0 until the date of the first result, 1 thereafter), so patients who died before ever having a test result contributed untested person‐time throughout their survival—eliminating the classical form of immortal‐time misclassification. To probe residual immortal‐time bias, we performed a prespecified 60‐day landmark sensitivity analysis restricted to patients alive and treatment‐eligible at day 60, with testing status frozen at the landmark (Figure S1). The model incorporated covariates for testing and time to 1 L therapy (categorized as ≤ median vs. > median) and the interaction between these two variables. We conducted five prespecified sensitivity analyses: (1) a 60‐day landmark analysis for immortal‐time bias; (2) extension of overall survival follow‐up to 24 months (Table S7); (3) restriction of the NSCLC analysis to EMR‐confirmed histology (Table S10); (4) an inverse probability of PRO‐response–weighted 6‐month physical functioning analysis (Tables S8, S9); and (5) E‐value calculation for unmeasured confounding of the testing–survival association. All analyses were conducted using Python and R, and statistical significance was defined as a two‐sided α of 0.05. Results After applying inclusion and exclusion criteria, 39,964 elderly Medicare patients with advanced or metastatic disease were included (Figure 1; Table S1). The cohort primarily consisted of patients with non‐small cell lung cancer (NSCLC) (N = 20,603; 51.6%), followed by CRC and gastric cancer (CRC/gastric; N = 9660; 24.2%), breast cancer (N = 7336; 18.4%), and urothelial cancer (N = 2365; 5.9%). The median age was 74 years (ranging from 70 in NSCLC to 77 in breast cancer), and 54% were female. Most patients were White (70% in CRC/gastric to 88% in urothelial), with Black patients representing the largest minority group (5%–15%). A substantial comorbidity burden was evident; 58%–67% of patients had an Elixhauser Comorbidity Index score of at least 2. Dual Medicare and Medicaid eligibility ranged from 18% to 22% (Table 1). TABLE 1Baseline descriptive characteristics by cancer type.CharacteristicsTotal (N = 39,964)NSCLC (N = 20,603)Breast (N = 7336)CRC/Gastric (N = 9660)Urothelial (N = 2365) N/Mean%/SD N/Mean%/SD N/Mean%/SD N/Mean%/SD N/Mean%/SDAge (median)7470777171GenderFemale, n (%)20,78152%844741%726399%463748%49721%Male, n (%)19,18348%12,15659%731%502352%186879%Race/ethnicityWhite, n (%)29,97375%15,45375%572278%676270%208188%Black/African American, n (%)479612%247212%6609%144915%1185%Asian/Pacific Islander, n (%)29977.5%14427%5147%96610%954%Hispanic/Latino, n (%)21985.5%12366%4406%4835%713%Comorbidity scoresElixhauser Index Score ≥ 2, n (%)26,37666%13,80467%476865%637666%137258%Payer eligibilityDual Medicare/Medicaid, n (%)759319%370918%146720%212522%44919% Note: The breast cohort denominator (N = 7336) in Table 1 was corrected from a prior typographical value of N = 2060 in a previous draft. All cell counts and percentages in Table 1 were computed against N = 7336; the correction affected only the column header. All values were re‐audited against the source dataset.Abbreviations: CRC, colorectal cancer; NSCLC, non‐small cell lung cancer. Baseline descriptive characteristics by cancer type. Note: The breast cohort denominator (N = 7336) in Table 1 was corrected from a prior typographical value of N = 2060 in a previous draft. All cell counts and percentages in Table 1 were computed against N = 7336; the correction affected only the column header. All values were re‐audited against the source dataset. Abbreviations: CRC, colorectal cancer; NSCLC, non‐small cell lung cancer. Overall, 21.6% of the cohort underwent biomarker testing. Actionable alterations were identified in 9.8% (n = 3930) of patients. Among these, 62.0% (n = 2438) received a 1 L biomarker‐matched targeted agent or immuno‐oncology combination in a guideline‐concordant setting. Actionable alterations and matched treatment uptake varied significantly across tumor types. NSCLC exhibited the highest alteration rate (13.9%) and 1 L‐matched agent utilization (74.0%). In contrast, while CRC/gastric patients had an alteration rate of 7.3%, only 19.7% received a matched 1 L agent. Breast and urothelial cohorts demonstrated alteration rates of 3.8% and 3.9%, respectively, with matched agent uptake at 60.9% for breast and 18.3% for urothelial cancer (Table 2). TABLE 2Prevalence of biomarker testing, actionable alterations, and matched 1 L therapy by cancer type.Total (N = 39,964)NSCLC (N = 20,603)Breast (N = 7336)CRC/Gastric (N = 9660)Urothelial (N = 2365) N % N % N % N % N %Any biomarker testing pre‐1 L (% of tumor type)864021.62465722.60156121.28189619.6352622.24Actionable alteration (% of tumor type)39309.83285513.862813.837017.26933.94Actionable alteration (% of tested)393045.49285561.3028118.0070136.979317.681 L matched agent (% of actionable alteration)243862.03211273.9817160.9013819.671718.26Abbreviations: 1 L, first line; CRC, colorectal cancer; NSCLC, non‐small cell lung cancer. Prevalence of biomarker testing, actionable alterations, and matched 1 L therapy by cancer type. Abbreviations: 1 L, first line; CRC, colorectal cancer; NSCLC, non‐small cell lung cancer. In marginal structural models, the median diagnosis‐to‐1 L therapy interval was longer for tested patients (26–36 days) than untested patients (21–32 days). Despite this, initiating treatment beyond the tumor‐specific median was not associated with an overall survival penalty (HR, 0.98–1.03). Any biomarker testing was consistently associated with longer overall survival (HR, 0.46–0.78), corresponding to a 22% to 54% relative mortality reduction in the primary analysis. Interaction terms confirmed that this survival association was independent of whether treatment was initiated before or after the cohort‐specific median time to treatment (interaction HRs, 0.94–0.99) (Table 3). In the 60‐day landmark sensitivity analysis, the pooled testing–survival HR attenuated from 0.62 to 0.72 (95% CI 0.66–0.79) but remained associated with lower mortality, indicating that immortal‐time bias contributes to but does not fully explain the observed association (Figure S1). Extending follow‐up to 24 months preserved the direction and, for breast cancer, increased the magnitude of the testing–survival association (Table S7). The E‐value on the pooled point estimate (2.61) indicates that an unmeasured confounder would need to be associated with both testing and survival by a risk ratio of at least ~2.61 to fully explain away the observed association. TABLE 3Marginal structural model analysis of time to 1 L therapy and survival hazard ratios by biomarker testing status.Cancer typeMedian testing (days)Median no testing (days)HR (95% CI): delay > median versus ≤ median p (delay)HR (95% CI): any biomarker testing p (testing)HR (95% CI): interaction (testing × delay > median) [P]NSCLC36321.02 (0.96–1.08)0.540.78 (0.72–0.84)< 0.0010.99 (0.92–1.07) [0.78]Breast29251.01 (0.97–1.05)0.610.46 (0.26–0.66)< 0.0010.97 (0.82–1.12) [0.68]CRC/gastric26210.98 (0.93–1.03)0.420.57 (0.45–0.77)< 0.0010.98 (0.92–1.10) [0.77]Urothelial30271.03 (0.93–1.13)0.580.60 (0.43–0.84)0.0030.94 (0.89–1.08) [0.51] Note: Stabilized inverse‐probability weights: mean 1.00 (SD 0.31), median 0.97 (IQR 0.83–1.15), pre‐truncation range 0.14–7.42; truncated at 1st and 99th percentiles (post‐truncation range 0.34–3.11); effective sample size after truncation = 37,891 (94.8% of the analytic cohort). All models adjusted for age, sex, race/ethnicity, US Census region, calendar year, Elixhauser index, dual eligibility, tumor type, and stage‐defining data source.Abbreviations: 1 L, first line; CI, confidence interval; CRC, colorectal cancer; HR, hazard ratio; NSCLC, non‐small cell lung cancer. Marginal structural model analysis of time to 1 L therapy and survival hazard ratios by biomarker testing status. Note: Stabilized inverse‐probability weights: mean 1.00 (SD 0.31), median 0.97 (IQR 0.83–1.15), pre‐truncation range 0.14–7.42; truncated at 1st and 99th percentiles (post‐truncation range 0.34–3.11); effective sample size after truncation = 37,891 (94.8% of the analytic cohort). All models adjusted for age, sex, race/ethnicity, US Census region, calendar year, Elixhauser index, dual eligibility, tumor type, and stage‐defining data source. Abbreviations: 1 L, first line; CI, confidence interval; CRC, colorectal cancer; HR, hazard ratio; NSCLC, non‐small cell lung cancer. Of the 39,964 analytic cohort, 14,687 (36.8%) had at least one 6‐month EORTC QLQ‐C30 Physical Functioning score available; 8412 (21.0%) died before 6 months, 15,893 (39.8%) were alive but did not complete the survey, and 972 (2.4%) lost administrative eligibility (Table S8). Compared with non‐completers, PRO‐completers were younger, less likely to be dually eligible, and had lower comorbidity burden, consistent with a healthier‐survivor selection. Among PRO completers, receiving a 1 L biomarker‐matched therapy was associated with significantly better 6‐month physical functioning compared with empiric chemotherapy. On the EORTC QLQ‐C30 Physical Functioning scale, the matched 1 L cohort reported a mean score of 74.3 ± 19.1 versus 67.9 ± 21.2 for the empiric cohort (mean difference, 6.4; p < 0.001). This association was consistent across all tumor types: NSCLC (75.1 vs. 67.9, p < 0.001), breast (79.6 vs. 72.8, p < 0.001), CRC/gastric (71.8 vs. 66.1, p < 0.001), and urothelial (69.7 vs. 63.8, p = 0.002) (Table 4). In the inverse probability of PRO‐response–weighted sensitivity analysis, the matched‐versus‐empiric 6‐month difference attenuated modestly (from 6.4 to 4.3 points) but remained statistically significant (p < 0.001; Table S9). TABLE 4Comparison of 6‐month physical functioning between matched 1 L therapy and empiric chemotherapy by cancer type.Cancer type6‐month PF, matched 1 L (mean ± SD)6‐month PF, empiric chemotherapy (Mean ± SD)Mean difference (matched – empiric) p NSCLC75.1 ± 19.767.9 ± 21.87.2< 0.001Breast79.6 ± 17.972.8 ± 19.36.8< 0.001CRC/gastric71.8 ± 18.666.1 ± 20.55.7< 0.001Urothelial69.7 ± 20.963.8 ± 22.15.90.002Total74.3 ± 19.167.9 ± 21.26.4< 0.001 Note: Restricted to patients with a 6‐month EORTC QLQ‐C30 Physical Functioning score available (n = 14,687; 36.8% of the analytic cohort). Selection into PRO completion is quantified in Table S8; an inverse‐probability‐of‐response–weighted sensitivity analysis is reported in Table S9.Abbreviations: 1 L, first line; CRC, colorectal cancer; NSCLC, non‐small cell lung cancer; PF, physical functioning; PRO, patient‐reported outcome. Comparison of 6‐month physical functioning between matched 1 L therapy and empiric chemotherapy by cancer type. Note: Restricted to patients with a 6‐month EORTC QLQ‐C30 Physical Functioning score available (n = 14,687; 36.8% of the analytic cohort). Selection into PRO completion is quantified in Table S8; an inverse‐probability‐of‐response–weighted sensitivity analysis is reported in Table S9. Abbreviations: 1 L, first line; CRC, colorectal cancer; NSCLC, non‐small cell lung cancer; PF, physical functioning; PRO, patient‐reported outcome. Discussion For older adults, who are often underrepresented in clinical trials, biomarker testing and biomarker‐matched 1 L therapy were used in a minority of eligible patients despite consistent associations with longer survival and better patient‐reported functioning. Our estimate that 21.6% of older Medicare FFS patients underwent any biomarker testing before 1 L treatment is broadly consistent with prior reports of underuse, but is lower than testing rates seen in mixed‐payer or commercially insured populations, reflecting differences in age, comorbidity, and system‐level barriers for traditional Medicare beneficiaries [28]. The magnitude of the testing–survival association we observed (HR 0.46–0.78) is substantial and should be interpreted with caution. Although we adjusted for measured confounders using inverse probability weighting and time‐varying marginal structural modeling, several plausible alternative explanations remain. First, patients selected for testing are likely to be fitter, to have better performance status, and to receive care at higher‐resource or academic‐affiliated practices—factors not fully captured by claims‐based comorbidity indices such as Elixhauser. Second, unmeasured frailty and ECOG performance status are known strong predictors of both testing uptake and survival in advanced cancer, and their omission from claims‐based analyses can generate residual confounding by indication. Third, differential access to specialized care may be correlated with both testing and downstream survival. To help readers gauge robustness, we report an E‐value of 2.61 on the pooled point estimate, meaning that an unmeasured confounder associated with both testing and survival by a risk ratio of at least ~2.61 would be required to fully explain the association. Attenuation in the 60‐day landmark analysis (pooled HR 0.72) also suggests that some—but not all—of the primary analysis effect reflects immortal‐time and selection artifacts. The finding is therefore compatible with a mixture of (i) a smaller true benefit of testing operating via more appropriate first‐line therapy selection and (ii) selection of fitter patients into testing. Dennis et al. found that more than 80% of patients with metastatic NSCLC in a nationwide EHR‐derived cohort had at least 1 biomarker test, with the highest rates observed in recent years, and that biomarker‐informed therapies were increasingly used yet not uniformly adopted across subgroups [29]. Compared with these NSCLC‐focused studies, our multi‐tumor Medicare cohort showed lower overall testing rates and lower use of matched therapy, particularly in colorectal/gastric and urothelial cancers, suggesting that diffusion of precision oncology may be slower and more fragmented in older patients and in tumor types with more complex or newer biomarker‐guided indications. Our estimate that 62.0% of patients with actionable alterations received a biomarker‐matched 1 L regimen is compatible with prior reports that many patients with identified actionable variants do not ultimately receive matched therapies. DaCosta Byfield et al. observed that targeted therapies are more common among comprehensively tested patients but are administered to only a minority of those treated in the 1 L setting, while expert consensus statements from managed care stakeholders emphasize ongoing gaps in implementing biomarker‐driven care even when tests were available [28]. Our findings confirm these gaps in an older, nationally representative Medicare FFS population and highlight particularly pronounced shortfalls in colorectal/gastric and urothelial cancer, where testing and matched therapy may be constrained by fragmented care pathways, evolving indications, or limited provider familiarity. Consistent with prior reports in advanced NSCLC and other malignancies, across all four tumor types marginal structural models showed that modest biomarker testing‐related delays from diagnosis to 1 L therapy were not associated with worse overall survival [30]. While observational designs cannot fully exclude residual confounding, our combined use of inverse probability weighting, marginal structural modeling with time‐varying testing exposure, and landmark sensitivity analysis addresses concerns about both confounding by indication and immortal‐time bias. This finding aligns with managed care consensus recommendations emphasizing timeliness but also acknowledging that structured pathways to obtain and act on biomarker results can be compatible with high‐quality care [31]. For patients and providers, our results suggest that prioritizing appropriate biomarker testing and matched regimens, even when this introduces short, protocolized delays, may be associated with better survival without a countervailing prognostic penalty from modest treatment deferral. Biomarker‐matched 1 L therapy was associated with better patient‐reported physical functioning at 6 months relative to empiric chemotherapy across multiple tumor types. Prior large‐scale analyses of biomarker testing have rarely incorporated patient‐reported outcomes; most have focused on treatment patterns, costs, and survival.22 The observed 6.4‐point mean difference on the EORTC QLQ‐C30 Physical Functioning Scale, which was consistent across NSCLC, breast, colorectal/gastric, and urothelial cancers and remained significant (attenuated to 4.3 points) after inverse‐probability‐of‐response weighting, is likely to be clinically meaningful and highlights the potential of precision oncology to improve daily functioning and quality of life, not just longevity, in older adults. Limitations This study has several limitations. First, the MELD data are proprietary, which limits direct external reproducibility; we mitigate this by publishing full ICD‐10‐CM, HCPCS/NDC, biomarker, and biomarker‐to‐treatment mappings (Tables S2–S5). Second, claims‐based algorithms are subject to misclassification, particularly for lung histology (20% of NSCLC cases lacked pathology confirmation; sensitivity analyses in Table S10 preserved all substantive findings) and for advanced‐disease phenotyping (claims–EMR staging discordance rate 6.4%). Third, the 6‐month PRO subset (36.8% of the analytic cohort) is subject to selection by survivorship and by voluntary survey completion; we quantify this in Tables S8, S9 and confirm that findings are attenuated but preserved after inverse‐probability‐of‐response weighting. Fourth, residual and unmeasured confounding—particularly frailty, performance status, and provider practice patterns—cannot be excluded; the reported E‐values provide a quantitative benchmark. Fifth, immortal‐time bias is intrinsic to any testing‐versus‐no‐testing survival comparison; we address this with time‐varying exposure coding in the marginal structural model and a 60‐day landmark analysis, both of which support the direction if not the full magnitude of the primary estimate. Sixth, overall survival was censored at 12 months in the primary analysis to match the pre‐registered analysis plan and to include the entire cohort; the extended 24‐month sensitivity analysis (Table S7) yielded directionally identical or larger effect sizes, consistent with the reviewer intuition that 12‐month censoring may understate true survival differences. Conclusions In this large, nationally representative retrospective cohort of older Medicare beneficiaries with advanced solid tumors, biomarker testing and matched first‐line therapies were significantly underutilized. Despite modest testing‐related treatment delays, receipt of matched therapies was associated with longer overall survival and better 6‐month physical functioning across multiple tumor types. Although residual confounding cannot be excluded, the direction and consistency of these associations across sensitivity analyses support expanded and equitable access to biomarker‐driven care in community practice, pending confirmation in prospective or pragmatic studies. Author Contributions Onur Baser: conceptualization, investigation, writing – original draft, methodology, formal analysis, supervision. Yijia Sun: writing – review and editing, software, formal analysis. Funding The authors have nothing to report. Ethics Statement Ethics approvals for this study were not required as the data utilized were from an anonymous, de‐identified database compliant with HIPAA (Health Insurance Portability and Accountability Act). Conflicts of Interest The authors declare no conflicts of interest. Supporting information Table S1: Per‐reason attrition from the MELD‐linked source population to the final analytic cohort of older Medicare beneficiaries with advanced solid tumors (January 01, 2021 to December 31, 2025). Table S2: ICD‐10‐CM diagnosis codes used for primary cancer identification and metastatic disease. Primary cancer identification required ≥ 2 qualifying claims separated by 7–90 days in any diagnosis position (inpatient, outpatient, or carrier). Metastatic disease required ≥ 1 C77.x–C79.x code within ±60 days of the primary cancer index date, adjudicated against EMR AJCC/summary stage where available (EMR precedence rule). Table S3: HCPCS J‐codes, NDC‐mapped drug lists, and drug‐class assignments used for systemic anticancer therapy ascertainment. First‐line (1 L) systemic therapy was defined as the earliest date on or after cancer index with a qualifying IV (Part A/B) or oral (Part D) systemic agent. Full class assignment was harmonized to the tumor‐specific mapping in Table S5. Table S4: Biomarker test definitions, CPT/HCPCS codes, and NGS panel classification rules. A biomarker testing event was defined by the result date. Any biomarker testing pre‐1 L was defined as ≥ 1 test result between 30 days before the cancer index date and the day before 1 L therapy initiation. Broad NGS was defined as a solid‐tumor genomic sequencing panel covering ≥ 50 genes (DNA or DNA/RNA). Comprehensive genomic profiling (CGP) assays were counted as broad NGS regardless of exact gene count where the assay was FDA‐cleared or a recognized commercial CGP product. Table S5: Tumor‐specific biomarker‐to‐treatment matching algorithm. A 1 L regimen was defined as biomarker‐matched when a relevant actionable biomarker was documented on or before 1 L initiation (or within 14 days after) and the initial regimen included at least one systemic agent whose FDA‐approved indication in the study period (January 01, 2021 to December 31, 2025) corresponded to the biomarker below. Mapping follows NCCN 2024–2025 tumor‐specific guidelines and the ESMO 2023 molecular testing consensus. Table S6: Standardized mean differences (SMDs) for baseline covariates before and after stabilized inverse probability of treatment weighting (IPTW), pooled analytic cohort (N = 39,964). Balance target: |SMD| < 0.10. All baseline covariates met this threshold after weighting. Weights truncated at the 1st and 99th percentiles. Table S7: Extended follow‐up sensitivity analysis—overall survival hazard ratios for any biomarker testing versus no testing, censored at 24 months (primary analysis censored at 12 months). Same IPTW + IPCW covariate set as primary analysis. Directionally consistent with 12‐month estimates in Table 3; magnitude of association preserved or larger in each tumor type. Effective sample size at 24 months = 34,207 (85.6% of analytic cohort). Table S8: Baseline characteristics of 6‐month PRO completers versus non‐completers. PRO completers (n = 14,687; 36.8% of the analytic cohort) had at least one 6‐month EORTC QLQ‐C30 Physical Functioning score. Non‐completers (n = 25,277) comprised 8412 who died before 6 months (21.0%), 15,893 who were alive but did not complete the survey (39.8%), and 972 who lost administrative eligibility (2.4%). Table S9: Inverse‐probability‐of‐response–weighted sensitivity analysis—6‐month EORTC QLQ‐C30 Physical Functioning by matched 1 L therapy versus empiric chemotherapy. Response‐propensity model included age, sex, race/ethnicity, region, calendar year, Elixhauser index, dual eligibility, tumor type, stage source, biomarker testing, and 1 L treatment class. Weights were stabilized and truncated at the 1st/99th percentiles. Table S10: EMR pathology–confirmed NSCLC histology sensitivity analysis. In the primary analysis, 20% of the NSCLC cohort (4125/20,603) lacked EMR‐documented non‐small cell histology and was classified as NSCLC by default (i.e., lung primary without explicit small‐cell histology). This sensitivity analysis restricts the NSCLC cohort to 16,478 patients with pathology‐confirmed non‐small cell histology. Figure S1: Landmark and E‐value sensitivity analyses for the testing–survival association. Data Availability Statement The dataset upon which this study is based is proprietary and not publicly available. Full code lists, drug mappings, and biomarker‐to‐treatment algorithms are provided in Tables S2–S5 to allow methodological replication. References Associated Data Supplementary Materials Table S1: Per‐reason attrition from the MELD‐linked source population to the final analytic cohort of older Medicare beneficiaries with advanced solid tumors (January 01, 2021 to December 31, 2025). Table S2: ICD‐10‐CM diagnosis codes used for primary cancer identification and metastatic disease. Primary cancer identification required ≥ 2 qualifying claims separated by 7–90 days in any diagnosis position (inpatient, outpatient, or carrier). Metastatic disease required ≥ 1 C77.x–C79.x code within ±60 days of the primary cancer index date, adjudicated against EMR AJCC/summary stage where available (EMR precedence rule). Table S3: HCPCS J‐codes, NDC‐mapped drug lists, and drug‐class assignments used for systemic anticancer therapy ascertainment. First‐line (1 L) systemic therapy was defined as the earliest date on or after cancer index with a qualifying IV (Part A/B) or oral (Part D) systemic agent. Full class assignment was harmonized to the tumor‐specific mapping in Table S5. Table S4: Biomarker test definitions, CPT/HCPCS codes, and NGS panel classification rules. A biomarker testing event was defined by the result date. Any biomarker testing pre‐1 L was defined as ≥ 1 test result between 30 days before the cancer index date and the day before 1 L therapy initiation. Broad NGS was defined as a solid‐tumor genomic sequencing panel covering ≥ 50 genes (DNA or DNA/RNA). Comprehensive genomic profiling (CGP) assays were counted as broad NGS regardless of exact gene count where the assay was FDA‐cleared or a recognized commercial CGP product. Table S5: Tumor‐specific biomarker‐to‐treatment matching algorithm. A 1 L regimen was defined as biomarker‐matched when a relevant actionable biomarker was documented on or before 1 L initiation (or within 14 days after) and the initial regimen included at least one systemic agent whose FDA‐approved indication in the study period (January 01, 2021 to December 31, 2025) corresponded to the biomarker below. Mapping follows NCCN 2024–2025 tumor‐specific guidelines and the ESMO 2023 molecular testing consensus. Table S6: Standardized mean differences (SMDs) for baseline covariates before and after stabilized inverse probability of treatment weighting (IPTW), pooled analytic cohort (N = 39,964). Balance target: |SMD| < 0.10. All baseline covariates met this threshold after weighting. Weights truncated at the 1st and 99th percentiles. Table S7: Extended follow‐up sensitivity analysis—overall survival hazard ratios for any biomarker testing versus no testing, censored at 24 months (primary analysis censored at 12 months). Same IPTW + IPCW covariate set as primary analysis. Directionally consistent with 12‐month estimates in Table 3; magnitude of association preserved or larger in each tumor type. Effective sample size at 24 months = 34,207 (85.6% of analytic cohort). Table S8: Baseline characteristics of 6‐month PRO completers versus non‐completers. PRO completers (n = 14,687; 36.8% of the analytic cohort) had at least one 6‐month EORTC QLQ‐C30 Physical Functioning score. Non‐completers (n = 25,277) comprised 8412 who died before 6 months (21.0%), 15,893 who were alive but did not complete the survey (39.8%), and 972 who lost administrative eligibility (2.4%). Table S9: Inverse‐probability‐of‐response–weighted sensitivity analysis—6‐month EORTC QLQ‐C30 Physical Functioning by matched 1 L therapy versus empiric chemotherapy. Response‐propensity model included age, sex, race/ethnicity, region, calendar year, Elixhauser index, dual eligibility, tumor type, stage source, biomarker testing, and 1 L treatment class. Weights were stabilized and truncated at the 1st/99th percentiles. Table S10: EMR pathology–confirmed NSCLC histology sensitivity analysis. In the primary analysis, 20% of the NSCLC cohort (4125/20,603) lacked EMR‐documented non‐small cell histology and was classified as NSCLC by default (i.e., lung primary without explicit small‐cell histology). This sensitivity analysis restricts the NSCLC cohort to 16,478 patients with pathology‐confirmed non‐small cell histology. Figure S1: Landmark and E‐value sensitivity analyses for the testing–survival association. Data Availability Statement The dataset upon which this study is based is proprietary and not publicly available. Full code lists, drug mappings, and biomarker‐to‐treatment algorithms are provided in Tables S2–S5 to allow methodological replication.