Background
Lung cancer is the leading cause of cancer death, representing approximately 25% of all cancer deaths.1 Non–small cell lung cancer (NSCLC) comprises the majority of patients with lung cancer1 and has seen significant improvements in survival over the past decade.2 Advances in personalized therapies for advanced NSCLC (aNSCLC), which allow treatment based on the genetic makeup of a tumor, have significantly contributed to the improved outcomes of patients.2
Despite the therapeutic advancements, the realized benefits of personalized medicine may be hindered by the disparities that exist in the care of patients with NSCLC. Furthermore, insurance has been previously identified as an important factor in dictating the cancer care that patients receive.3–5 Specifically, prior work4,6 suggests that patients with Medicaid experience lower-quality cancer care and worse survival outcomes.6,7 A study examining cancer survival found that those with public insurance (non-Medicare) had a higher risk of mortality across different cancers, ranging from 11% to 25% (increased risk).8
Limited data indicate whether these disparities persist in the era of personalized medicine. An analysis examining ALK biomarker testing in patients with NSCLC found that Medicaid beneficiaries were 40% less likely to be tested compared with commercially insured patients.9 Another study found that Medicaid expansion was associated with an improvement in time to treatment for Black or African American patients.10 Another study found an association with Medicaid expansion and improved survival in states with expanded Medicaid, underscoring the relationship between access to care and cancer outcomes.11 However, it remains unknown whether access to biomarker testing and personalized treatments is driving the differences in outcomes. We examined the associations between Medicaid (vs commercial) insurance and the receipt of biomarker testing and targeted therapy, and explored the association with overall survival (OS) in a contemporary cohort of patients with aNSCLC.
Methods
Study Design and Data Source
We conducted a retrospective analysis using the Flatiron Health database, which is derived from nationwide longitudinal electronic health records (EHRs). During the study period, these data originated from approximately 280 cancer clinics representing >2.2 million US patients with cancer. The deidentified patient-level data include structured and unstructured data, curated via technology-enabled abstraction.12,13 Institutional Review Board approval of the study protocol was obtained before study conduct and included a waiver of informed consent. The EHR data are linked with external mortality data, and the quality of the mortality data has been previously evaluated and validated.14,15 Commercially or Medicaid-insured patients with NSCLC and an advanced diagnosis from January 1, 2011, through September 30, 2019, were included in the study. Additional inclusion criteria included requirement of a clinic visit within 120 days of advanced diagnosis and being aged 18 to 64 years at diagnosis of advanced disease. Patients were excluded if they received clinical study drugs or care at academic sites, or if they had biomarker testing before their advanced diagnosis date. To mitigate the risk of missing biomarker or treatment records because of patients seeking care outside the Flatiron Health network, we excluded patients who had their first biomarker test ≥3 months or their first evidence of treatment ≥4 months from their advanced diagnosis date. Patient attrition is shown in supplemental eTable 1 (available with this article at JNCCN.org).
Construction of Variables
Clinical characteristics included demographics and clinical factors (Table 1). Patients were assigned to Medicaid or commercially insured cohorts based on the record of insurance before or within 90 days of the advanced diagnosis date. Patients lacking information on start or end dates for their coverage were assigned to the Medicaid or commercial insurance cohorts as long as their insurance type was exclusively one or the other. Patients with overlapping insurance types were excluded.
Baseline Patient Characteristics by Insurance Type
Biomarker testing was evaluated during the first 3 months after the advanced cancer diagnosis. Biomarkers examined included ALK, EGFR, ROS1, BRAF, and PD-L1. In addition to individual biomarker testing rates, we also assessed receipt of all recommended biomarker tests, which was defined as the time period after a targeted therapy became available (approved by the FDA). Dates when targeted therapies were first approved by the FDA for each biomarker were as follows: ALK: August 26, 2011 (crizotinib); EGFR: May 14, 2013 (erlotinib); ROS1: March 11, 2016 (crizotinib); PD-L1: October 2, 2015 (pembrolizumab); and BRAF: June 27, 2017 (dabrafenib + trametinib).
We identified first-line (1L) therapy occurring within 4 months after advanced cancer diagnosis. Treatment-related outcomes included the proportion of patients receiving 1L therapy and the frequency of receiving 1L therapy. Treatment classifications included tyrosine kinase inhibitor (TKI)–based (afatinib, alectinib, crizotinib, ceritinib, dabrafenib, dacomitinib, erlotinib, gefitinib, lorlatinib, osimertinib, trametinib), cancer immunotherapy–based (atezolizumab, durvalumab, nivolumab, pembrolizumab), and other chemotherapy/biologic-based (not including those mentioned).
Statistical Analysis
Biomarker testing–related outcomes included testing rates, the proportion of patients receiving recommended biomarker tests (defined earlier, based on the date of availability of a therapy indicated for a given biomarker), and the likelihood of receiving biomarker tests by insurance type. Biomarker testing was assessed via (1) biomarker testing rate by person-time and (2) an overall proportion. The biomarker testing rate by person-time was estimated by dividing the number of patients with a given biomarker test before age 65 years divided by the amount of person-time “at risk,” which included the time from diagnosis until receipt of the test or age 65 years, whichever came first (after age 65 years it was assumed that the patient would be eligible for Medicare and no longer at risk for having Medicaid as the primary insurance). The overall proportion was estimated by dividing the number of patients who received a biomarker test within 3 months of diagnosis by the total number of eligible patients. For biomarker temporal trends analyses, we classified patients by year and quarter (eg, Q2 2012) or semester (eg, S1 2012) of their advanced diagnosis.
Descriptive statistics, including means and percentages for continuous variables and counts and percentages for categorical variables, were used along with the Student t test or chi-square test statistics, respectively, to compare differences in baseline patient characteristics by insurance type. We used a logistic regression model with covariates for insurance (Medicaid vs commercial) and time to assess differences in the receipt of the recommended biomarker tests between insurance cohorts over time by each semester (eg, half/year). Because guideline recommendations for biomarker testing are primarily focused on nonsquamous cell carcinoma histology, we conducted a subgroup analysis assessing receipt of recommended biomarker tests by insurance type among patients with nonsquamous cell carcinoma histology. An interaction term between time and insurance was included to assess whether any differences changed over time. We then used a series of sequential models to conduct an exploratory analysis of the relation between insurance type and OS. Patients were followed up from advanced diagnosis to receipt of testing, treatment, death, or last known visit date. We used Cox regression models to estimate the relative hazard of biomarker testing, treatment, and mortality according to insurance type.
Final hazard ratios (HRs) and 95% confidence intervals for all outcomes were adjusted for baseline characteristics, including gender, age, race/ethnicity, region, smoking status, histology, year of advanced diagnosis, liver metastases, brain and central nervous system metastases, and bone metastases. Models for mortality were sequentially adjusted for testing, then treatment (any 1L treatment, then type of 1L treatment) as time-varying covariates to account for potential immortal time bias resulting from potential differences in timing/receipt of testing and treatment between insurance cohorts. Finally, we constructed separate models for patients with each insurance type to evaluate the association of testing and treatment on mortality separately in patients with Medicaid and commercially insured patients. A sensitivity analysis examining the interaction between histology and insurance was conducted to assess for a differential effect by histology. The Kaplan-Meier method was used to calculate OS distributions, with the log-rank test used to test differences by insurance type. The statistical significance threshold was set a priori using 2-sided tests at P <.05. All analyses were performed using R version 3.5.3 (R Foundation for Statistical Computing).
Results
Baseline Patient Characteristics
A total of 7,010 patients were included in the analysis, of whom 6,145 were in the commercially insured cohort and 865 in the Medicaid cohort. Although most of the demographic and clinical characteristics were similar between the 2 cohorts, race/ethnicity, region, and smoking status were significantly different (P <.001), with the Medicaid cohort having a greater proportion of Black or African American patients (20% vs 9.3%), more patients from the South (55% vs 42%), and more patients likely to be smokers (93% vs 85%; Table 1).
Biomarker Testing Patterns
Compared with commercially insured patients, a smaller proportion of Medicaid beneficiaries were observed to be tested for each individual biomarker in terms of the proportion of patients receiving tests (Figure 1) and the frequency of testing (supplemental eFigure 1). For instance, differences in biomarker testing for potential targeted agents (EGFR, ALK, ROS1, and BRAF) ranged from 12% to 15% (P <.001), whereas differences in PD-L1 testing to identify candidates for cancer immunotherapy was 9% (P <.001). Similar differences were observed when examining the rate of testing, which accounted for variable times of follow-up, although the difference in PD-L1 testing was not statistically significant (P =.15).
In addition to the fewer tests observed for Medicaid beneficiaries in individual biomarkers, a smaller proportion of Medicaid beneficiaries was observed to have been tested for all recommended biomarkers. This difference was consistent regardless of the number of biomarker tests recommended during the time period (P <.001 for insurance and time), and the magnitude of the difference did not change over time (P =.583 for interaction between time and insurance; Figure 2). Lower proportions of testing in patients with Medicaid were generally observed over time for EGFR, ALK, ROS1, and BRAF. However, PD-L1 testing seemed to be more similar between the 2 cohorts (supplemental eFigure 2). After controlling for baseline characteristics, including histology, Medicaid beneficiaries were 19% less likely to be tested compared with commercially insured patients (HR, 0.81; 95% CI, 0.74–0.89; P <.001; Table 2). A similar pattern was observed when biomarker testing rates were stratified by histology. Overall, descriptive analyses showed a higher rate of testing in patients with nonsquamous cell carcinoma histology compared with those with squamous cell carcinoma histology. In addition, testing rates were numerically higher for those with commercial insurance compared with those with Medicaid in both nonsquamous and squamous cell carcinoma histology, although these differences were not tested statistically (supplemental eTable 2).
Biomarker testing patterns by insurance type. All differences between Medicaid and commercial insurance are statistically significant (P<.001).
Citation: Journal of the National Comprehensive Cancer Network 20, 5; 10.6004/jnccn.2021.7083
Proportion of patients receiving all recommended biomarker tests over time. Arrows indicate the half-year period when each biomarker was first recommended (as determined by FDA approval of targeted therapy indicated for that biomarker).
Abbreviation: S, semester.
*P<.001 for difference in receipt of testing by type of insurance (Medicaid vs commercial) in logistic regression model.
Citation: Journal of the National Comprehensive Cancer Network 20, 5; 10.6004/jnccn.2021.7083
Risk of Testing and Treatment
Treatment Patterns
A lower proportion of Medicaid beneficiaries compared with commercially insured beneficiaries was observed to have any 1L treatment (67% vs 79%; P <.001) and any biomarker-driven therapy (TKI- or cancer immunotherapy–based regimens; 14% vs 25%; P <.001). After adjustment for baseline characteristics and tumor histology, Medicaid beneficiaries were 28% less likely to receive any 1L treatment (HR, 0.72; 95% CI, 0.66–0.79; P <.001) and 30% less likely to be treated using a biomarker-driven therapy (HR, 0.70; 95% CI, 0.58–0.85; P <.001; Table 2). Among those tested, Medicaid beneficiaries were 29% less likely to be treated using a biomarker-driven therapy (HR, 0.71; 95% CI, 0.58–0.88; P <.001); yet among the subgroup of patients with a positive driver mutation, there was no difference in receipt of targeted therapy (HR, 0.90; 95% CI, 0.60–1.34; P =.60; Table 2).
Overall Survival
Median OS in the Medicaid cohort was 7.8 months (95% CI, 7.0–8.7 months) compared with 10.3 months (95% CI, 10.0–10.7 months) in the commercial cohort (log-rank P <.01; Figure 3A). After adjustment for baseline clinical factors, Medicaid beneficiaries had a 23% higher risk of mortality (HR, 1.23; 95% CI, 1.13–1.35; Table 3). Adding testing to the model resulted in minimal change to the HR (HR, 1.22; 95% CI, 1.12–1.33); however, after adding testing and type of treatment to the model, the HR approached 1 (HR, 1.12; 95% CI, 1.03–1.22; P =.010) but the result remained statistically significant, indicating that other unmeasured factors may be contributing to the difference in OS between Medicaid beneficiaries and commercially insured patients.
Overall survival by (A) insurance type and (B) receipt of testing among patients with Medicaid.
Citation: Journal of the National Comprehensive Cancer Network 20, 5; 10.6004/jnccn.2021.7083
Risk of Mortality
Among Medicaid beneficiaries, those who had at least 1 biomarker test had a longer median OS (8.7 months; 95% CI, 7.6–10.3 months) compared with those who had no tests (OS, 6.9 months; 95% CI, 6.2–7.9 months) (log-rank P <.01) (Figure 3B). After controlling for clinical factors and histologic type, we found that Medicaid beneficiaries who were not tested had a 27% higher risk of mortality than those who had at least 1 biomarker test (HR, 1.27; 95% CI, 1.06–1.52; P =.013; Table 3). After adjusting for type of treatment, the effect was attenuated minimally (HR, 1.23; 95% CI, 1.02–1.48; P =.026).
Discussion
This analysis of a large national oncology EHR database found that Medicaid (vs commercial insurance) beneficiaries with aNSCLC were less likely to receive recommended biomarker testing or targeted therapy. Notably, Medicaid insurance may also be associated with inferior OS, partially mediated by treatment differences between patients with Medicaid and commercially insured patients. Although previous studies have shown poorer outcomes with Medicaid beneficiaries, these studies have commonly used older data.6,7 Because precision medicine has brought substantial changes to clinical practice and the promise of longer OS for patients with aNSCLC, our findings suggest that disparities in both molecular testing and receipt of targeted therapy may be perpetuating socioeconomic disparities.
Our findings are consistent with earlier studies that found molecular testing to be less likely in Medicaid or Medicare–Medicaid dual-eligible populations9,16; however, neither study examined the magnitude and potential impact of these testing disparities on OS in Medicaid beneficiaries versus commercially insured patients. Although determining why testing disparities exist is beyond the scope of this study, there are a number of potential explanations. In our study, a larger proportion of patients with Medicaid had squamous cell carcinoma histology, for which driver mutations may be less prevalent and less likely to be tested. However, our findings persisted after adjusting for histology. In addition, sensitivity analyses found that a smaller proportion of Medicaid beneficiaries was tested for all recommended biomarkers in the nonsquamous cell carcinoma histology subgroup (P <.001; consistent with the overall cohort); a lack of a differential effect on receipt of biomarker-driven therapy between insurance by histology (interaction P >.6) indicates that these results seem to be consistent across histologies.
Inconsistent Medicaid policies may be another potential explanation for testing variation. A recent report found that only approximately 40% of state Medicaid programs provided coverage for comprehensive biomarker testing.17 Lower access to testing may lead to fewer opportunities for biomarker-driven therapies, and among those tested there was a disparity in the receipt of biomarker-driven therapy, indicating other potential access barriers to treatments. Although the drivers behind the lack of treatment utilization among Medicaid beneficiaries are unclear, similar to testing, a lack of timely updated coverage policies may play a role. Because the evidence base underlying different genetic tests and associated targeted therapies is evolving rapidly, it is critical to develop an approach for payors to promptly and comprehensively evaluate evidence and incorporate it into coverage decisions, because the disparity in the receipt of biomarker-driven therapies may influence outcomes. In this study, Medicaid beneficiaries without biomarker testing had an increased risk of mortality of 27%, indicating the potential impact that updating coverage policies to improve access to biomarker testing, and subsequent treatment, may have in this population.
Similar to other studies that have assessed cancer survival,5,8,18 OS may be shorter in the Medicaid population compared with a commercially insured population in the era of precision medicine. Our sequential modeling approach provides some mechanistic insights regarding this relationship in the context of a modern aNSCLC practice. Although the increased risk of mortality associated with Medicaid was not ameliorated when molecular testing was added to the model, there was a decrease in the Medicaid-associated hazard when receipt of any treatment was added. Furthermore, the risk of death was further attenuated when accounting for the type of treatment. This result suggests that disparities in biomarker testing itself may not drive OS differences between patients with Medicaid and commercially insured patients but that access to treatment, in particular biomarker-driven therapies, may be of importance in narrowing the survival gap. Nonetheless, these factors can only partially explain the survival differences observed, and thus other factors, such as social determinants of health, may play a role. In light of these findings, programs/policies that address not only testing and treatment access barriers but also other issues specific to Medicaid beneficiaries are needed to improve survival outcomes. A cohort analysis of patients in cancer clinical trials with positive findings provided further evidence supporting this need to address social determinants of health. Unger et al19 found a much smaller added survival benefit of an experimental drug compared with the standard of care in patients with Medicaid or no insurance versus those with commercial insurance. The authors speculated that this finding reflected the impacts of socioeconomic status and disparities in access to healthcare services, which may result in differential access to support services and postprotocol therapy.
There are a number of areas of future research that may further inform healthcare policies and quality-of-care improvements. First, additional research aimed at identifying the specific drivers of the observed disparities in OS would be helpful to inform the design of programs that specifically address Medicaid beneficiary needs. Our findings strongly suggest that access to biomarker testing and targeted therapy may be necessary but not sufficient as strategies to address disparities. Future work could focus on patient access to oncology care, the impact of noncancer illnesses on outcomes, Medicaid reimbursement policies, and whether financial toxicity associated with treatment or other health-related social needs is rendering targeted therapies unaffordable for some Medicaid beneficiaries. In addition, further understanding regarding testing, treatment, and outcomes of patients with Medicare compared with commercial and Medicaid beneficiaries may be warranted due to differing coverage policies, particularly in light of the Medicare national coverage determination for next-generation sequencing.
This analysis has a number of limitations to consider. First, as with all studies based on real-world data, it is subject to misclassification and missingness with the potential for patients to receive care outside of the Flatiron Health network. We therefore sought to mitigate this risk through our inclusion criteria requiring patients to have an office visit after their advanced diagnosis and excluding those who had their first treatment >120 days after advanced diagnosis and their first test >90 days after advanced diagnosis. In addition, insurance status may change over time and may not be completely captured in the EHRs; however, this risk was mitigated by excluding patients who had evidence of both commercial insurance and Medicaid. Although this analysis focused on community-based care, academic sites also care for Medicaid beneficiaries and may have different testing rates, so our results may not be generalizable across all sites of care. Last, we had limited information on performance status; however, the missingness and distribution of ECOG performance status scores were balanced between the cohorts.
Conclusions
Medicaid beneficiaries with aNSCLC had a 19% and 30% higher risk of not receiving biomarker testing and biomarker-driven therapy, respectively. Furthermore, Medicaid beneficiaries had an observed 23% higher risk of mortality than commercially insured patients, which may be partially mediated by disparities in receipt of treatment in general and targeted therapies in particular. Improved access to evidence-based molecular testing and targeted therapies may improve outcomes for Medicaid beneficiaries with lung cancer.
References
- 1.↑
American Cancer Society. Key statistics for lung cancer. Accessed November 19, 2020. Available at: https://www. cancer.org/cancer/lung-cancer/about/key-statistics.html
- 2.↑
Lou Y, Dholaria B, Soyano A, et al. Survival trends among non-small-cell lung cancer patients over a decade: impact of initial therapy at academic centers. Cancer Med 2018;7:4932–4942.
- 3.↑
Abdelsattar ZM, Hendren S, Wong SL. The impact of health insurance on cancer care in disadvantaged communities. Cancer 2017;123:1219–1227.
- 4.↑
Parikh-Patel A, Morris CR, Kizer KW. Disparities in quality of cancer care: the role of health insurance and population demographics. Medicine (Baltimore) 2017;96:e9125.
- 5.↑
Parikh-Patel A, Morris CR, Martinsen R, et al. Disparities in stage at diagnosis, survival, and quality of cancer care in California by source of health insurance. UC Davis. Accessed July 22, 2021. Available at: https://escholarship.org/uc/item/8xc078vj
- 6.↑
Pezzi TA, Schwartz DL, Pisters KMW, et al. Association of Medicaid insurance with survival among patients with small cell lung cancer. JAMA Netw Open 2020;3:e203277.
- 7.↑
Parsons HM, Maguire FB, Morris CR, et al. Impact of insurance type and timing of Medicaid enrollment on survival among adolescents and young adults with cancer. Pediatr Blood Cancer 2020;67:e28498.
- 8.↑
Ellis L, Canchola AJ, Spiegel D, et al. Trends in cancer survival by health insurance status in California from 1997 to 2014. JAMA Oncol 2018;4:317–323.
- 9.↑
Illei PB, Wong W, Wu N, et al. ALK testing trends and patterns among community practices in the United States. JCO Precis Oncol 2018;2:1–11.
- 10.↑
Adamson BJS, Cohen AB, Estevez M, et al. Affordable Care Act (ACA) Medicaid expansion impact on racial disparities in time to cancer treatment [abstract]. J Clin Oncol 2019;37(Suppl):Abstract LBA1.
- 11.↑
Lee A, Shah K, Chino JP, et al. Changes in cancer mortality rates after the adoption of the Affordable Care Act [abstract]. J Clin Oncol 2020;38 (Suppl):Abstract 2003.
- 12.↑
Birnbaum BE, Nussbaum N, Seidl-Rathkopf KN, et al. Model-assisted cohort selection with bias analysis for generating large-scale cohorts from the EHR for oncology research. ArXiv. Preprint posted online January 3013, 2020. Available at https://arxiv.org/abs/2001.09765, doi: 2001.09765
- 13.↑
Ma X, Long L, Moon S, et al. Comparison of population characteristics in real-world clinical oncology databases in the US: Flatiron Health, SEER, and NPCR. medRxiv. Preprint posted online March 18, 2020. doi: 10.1101/2020.03.16.20037143v2
- 14.↑
Curtis MD, Griffith SD, Tucker M, et al. Development and validation of a high-quality composite real-world mortality endpoint. Health Serv Res 2018;53:4460–4476.
- 15.↑
Carrigan G, Whipple S, Taylor MD, et al. An evaluation of the impact of missing deaths on overall survival analyses of advanced non-small cell lung cancer patients conducted in an electronic health records database. Pharmacoepidemiol Drug Saf 2019;28:572–581.
- 16.↑
Kehl KL, Lathan CS, Johnson BE, et al. Race, poverty, and initial implementation of precision medicine for lung cancer. J Natl Cancer Inst 2019;111:431–434.
- 17.↑
LUNGevity Foundation. Lung cancer scorecard. Accessed July 22, 2021. Available at: https://lungevity.org/public-policy/lung-cancer-scorecard
- 18.↑
Niu X, Roche LM, Pawlish KS, et al. Cancer survival disparities by health insurance status. Cancer Med 2013;2:403–411.
- 19.↑
Unger JM, Blanke CD, LeBlanc M, et al. Association of patient demographic characteristics and insurance status with survival in cancer randomized clinical trials with positive findings. JAMA Netw Open 2020;3:e203842.
