Economic Burden of Checkpoint Inhibitor Immunotherapy for the Treatment of Non–Small Cell Lung Cancer in US Clinical Practice
Abstract
Purpose
The efficacy of checkpoint inhibitor (CPI) immunotherapy in patients with non–small cell lung cancer (NSCLC) is limited by a lack of strongly predictive response markers, subjecting patients to potential underutilization of alternative effective treatments, increased risk for futile care, and unnecessary costs. Here, we characterize the extent to which basic molecular tumor-marker testing has been performed for NSCLC therapy selection in the United States, and compare medical resource utilization and costs in CPI-treated patients versus CPI-eligible patients treated with other therapies.
Methods
We identified a cohort of CPI-treated patients with NSCLC and a propensity score–matched cohort of CPI-eligible patients with NSCLC treated with non-CPI therapies (3095 patients in each group), using US administrative claims data covering the pre- and postinitial FDA-approval period for nivolumab, pembrolizumab, and atezolizumab (October 2012 to September 2017). We describe the utilization of recommended baseline molecular testing for CPI selection (pre–index date for CPI or other anticancer therapy), including programmed death ligand 1 (PD-L1) immunohistochemistry, ALK rearrangement and EGFR mutation testing, and pre- and postindex treatment patterns. All-cause medical resource utilization and semiannual total reimbursement (costs) were compared between CPI-treated and non-CPI–treated patients.
Findings
At baseline, in the propensity score–matched CPI- and non–CPI-treated patient cohorts, mean PD-L1 immunohistochemistry test utilization for CPI selection was moderate (0.6 vs 0.7 per patient, respectively). However, we observed much lower mean utilization of testing for EGFR mutations (0.1 vs 0.1 per patient) and ALK rearrangements (0.1 vs 0.2 per patient). Postindex, the use of both chemotherapy and ALK- and EGFR-targeted therapies were decreased in both cohorts. The CPI-treated group had significantly higher mean medical resource utilization in nearly all categories in the postindex period, and total per-patient semiannual costs, than did the CPI-eligible patients who received other therapies (141,537 vs 75,429 US dollars [USD]; P < 0.0001), driven by CPI drug reimbursement. Median (interquartile range) time on CPI was longest with pembrolizumab (113 [106–127] days), followed by nivolumab (105 [97–106] days) and atezolizumab (64 [50–85] days). Despite being associated with the lowest drug cost and the shortest treatment duration, atezolizumab was associated with the highest mean total per-patient semiannual costs (160,540 USD) compared with pembrolizumab (153,003 USD) and nivolumab (138,542 USD).
Implications
The advent of CPI treatment for NSCLC has added substantial care-related costs for patients and payers, concurrent with underutilization of minimum recommended molecular testing for therapy selection. Broad uptake of panel-based comprehensive targeted-therapy and immunotherapy profiling can promote optimal treatment selection and sequencing, reduce the likelihood of futile treatment, and further improve patient outcomes.
Introduction
CPI immunotherapies, including pembrolizumab, nivolumab, and atezolizumab, uniquely offer patients with NSCLC a chance for durable response, as demonstrated by long-term survival in clinical trials.
,
However, up to 60% of patients with NSCLC do not benefit from these drugs.
,
,
,
,
,
,
This lack of benefit is largely due to the limited predictive value of current testing of programmed death ligand 1 (PD-L1) by immunohistochemistry (IHC) for CPI patient selection,
,
,
,
underscoring the critical need for better testing to identify likely responders.
, , Recommendations from major US professional oncology-practice organizations, including the American Society of Clinical Oncology and the National Comprehensive Cancer Network, echo these CPI label indications, stating that patients should receive oral therapies targeted specifically at ALK and EGFR prior to IV CPI.
,
,
Second, to conserve tissue and time, concurrent testing for PD-L1 positivity by IHC is recommended for the identification of potential CPI responders, and specifically as a companion diagnostic assay for the identification of patients eligible for treatment with pembrolizumab and, very recently, combination ipilimumab + nivolumab, among patients who are negative for ALK and EGFR mutations. Patients with alterations in EGFR or ALK have much lower rates of response to, and increased toxicity with, CPI, and the expression of PD-L1 on IHC itself appears to be even less predictive of response to CPI in these patients.
,
Lack of CPI efficacy was markedly demonstrated in a Phase II clinical trial of pembrolizumab, in which treatment-naive patients with NSCLC were both EGFR-mutation positive and PD-L1 high (≥50%); the trial was terminated.
Indeed, the IHC technique for measuring PD-L1 has subjective interpretation and scoring that can cause discord, coupled with tumor heterogeneity, demonstrating the need for more objective and quantitative testing methods.
,
Patients with NTRK fusions, for example, which are currently best identified by next-generation RNA sequencing, are eligible for treatment with larotrectinib, a targeted agent that has demonstrated a 75% objective response rate across multiple tumors, including NSCLC.
Despite these recommendations, however, broad-panel marker testing for targeted therapy options in NSCLC continues to be underperformed.
,
,
Patients who are not comprehensively tested for targeted therapy options, and who instead opt for CPI treatment, with or without PD-L1 testing, may not only receive futile immunotherapy but also miss opportunities to receive targeted agents that are more effective and potentially less toxic and costly.
,
,
,
In a study using a willingness-to-pay threshold of 100,000 USD per quality-adjusted life-year, in which all patients who received pembrolizumab were PD-L1 positive and pembrolizumab was shown to be cost-effective, nivolumab was reported to be cost-effective among only PD-L1–positive patients and not PD-L1–negative patients.
However, a simulation study using data from clinical practice (nontrial) found that, compared to chemotherapy alone, first-line pembrolizumab or pembrolizumab + chemotherapy was not cost-effective in patients with NSCLC and comorbidities that would render them ineligible for trials, even among those who were PD-L1 high.
The investigators concluded that, despite high CPI drug prices, there was an overall reduction in NSCLC treatment costs in the period following FDA approval of CPI, owing to fewer emergency department (ED) visits and hospitalizations. Given these unexpected findings, the present study aimed to describe pretreatment marker testing and treatment patterns, and to compare medical resource utilization (MRU) and costs in 2 mutually exclusive treatment cohorts of patients with NSCLC: CPI versus other, non-CPI therapies.
Participants and Methods
Study Design and Data Source
This retrospective, observational cohort study was conducted using reimbursed administrative claims data from the IBM Watson Health MarketScan Commercial and Medicare Supplemental Claims and Encounters databases. The MarketScan databases include annual US medical and outpatient prescription paid drug claims in over 26.3 million individuals with insurance plans managed by commercial payers, including both employer-based and Medicare Advantage plans. An institutional review board exemption was granted for this study.
NSCLC Cohort Identification
Patients were identified for potential inclusion using the MarketScan claims databases (diagnosed October 1, 2012, to June 30, 2017, with the claims observation period ending September 30, 2017) based on the lung-cancer diagnosis codes defined in the International Classification of Diseases, Ninth Revision–Clinical Modification (ICD-9-CM) and ICD-10.
In the CPI-treated cohort, patients were considered for inclusion if they had at least 1 CPI claim based on the Healthcare Common Procedure Coding System (HCPCS) or National Drug Codes (NDC) for an anti–PD-1 axis CPI indication for advanced or metastatic NSCLC in either the first- or second-line setting at the time, including nivolumab, pembrolizumab, or atezolizumab.
FDA-label NSCLC indications for nivolumab, pembrolizumab, or atezolizumab
during the study period required progression on platinum-based chemotherapy or, if applicable, on EGFR- or ALK-targeted therapy prior to treatment with CPI, if such treatment was undertaken. Appendix 1 (see the online version at doi:https://doi.org/10.1016/j.clinthera.2020.06.018) shows all codes (ICD-9-CM, ICD-10, HCPCS, NDC, and Current Procedural Terminology) used for determining CPI or non-CPI cohort inclusion.
Patients were excluded from the study if they: (1) did not meet the inclusion criteria for the CPI or non-CPI cohort, (2) did not have at least 6 months of continuous medical and pharmacy benefit eligibility in the preindex period, (3) had <1 month of continuous medical and pharmacy benefit eligibility in the postindex period, (4) were younger than 18 years of age, and/or (5) had negative medical expenditures in the postindex period.
Index Dates
Figure 1Index dates for non–small cell lung cancer in cohorts treated with a checkpoint inhibitor (CPI) or with non-CPI treatment. In the CPI cohort, a patient’s index date was the first CPI claim date occurring between March 4, 2015 (initial CPI approval for nivolumab), and June 30, 2017. In the non-CPI cohort, a patient’s index date was identified, based on criteria for concurrent CPI approvals, as: (1) the first date of discontinuation of platinum chemotherapy, EGFR or ALK inhibitor treatment as required by initial nivolumab approval, prior approval of first-line pembrolizumab on October 24, 2016, whichever occurred last; (2) October 24, 2016, in patients who started non-CPI treatment prior to October 24, 2016, and discontinued it after October 24, 2016; or (3) the claim date of first-line non-CPI treatment initiated after October 24, 2016. TKI = tyrosine kinase inhibitor.
The preindex period was defined as the time from October 1, 2012, to each patient’s index date, as previously described. The postindex period was defined as the time from the index date to: (1) 3 months after CPI discontinuation (CPI cohort); (2) loss of eligibility to receive benefits; or (3) the end of the claims-data observation period (September 30, 2017), whichever occurred first. CPI discontinuation was defined as a gap between CPI claims of >3 months or a modification to the index CPI therapy regimen.
Patients in the CPI cohort were propensity score (PS) matched in a 1:1 ratio to patients in the non-CPI cohort using a nearest-neighbor matching algorithm to control for potential imbalances in patient characteristics. PSs were evaluated from a logistics model that included the following prespecified preindex period covariates: the patient’s age and geographic region of residence; insurance type; comorbid conditions including brain metastasis, Crohn disease, rheumatoid arthritis, ulcerative colitis, and/or lupus; calculated Charlson comorbidity index score; total MRU costs; total length of hospital stay; number of primary care physician office visits; and number of prescriptions.
Outcomes Measures
Statistical Analyses
Preindex baseline patient characteristics were compared for balance between the cohorts before and after PS matching using standardized differences. In the CPI cohort, time to CPI nonpersistence was assessed with Kaplan–Meier analysis. Mean medical expenditures and MRU in each cohort were weighted by the number of days in the follow-up period and converted to semiannual estimates by dividing the respective sum by the total number of months of follow-up time and multiplying by 6. t Tests were used for comparing MRU mean counts between the CPI and non-CPI cohorts, assuming a Poisson distribution. We performed nonparametric Wilcoxon signed-rank tests to compare total medical expenditures between the CPI and non-CPI cohorts. All analyses were performed using SAS statistical software version 9.4 (SAS Institute, Cary, North Carolina).
Discussion
To our knowledge, this is the first analysis of medical resource use and costs based on reimbursed claims for anti–PD-1 axis CPI immunotherapy from clinical practice in the treatment in NSCLC. We found significantly higher total and category-specific mean MRU and 6-month reimbursement in CPI-treated patients (141,537 USD) in direct comparison to a PS-matched CPI-eligible patient cohort that received other therapies (74,529 USD). The main drivers of higher postindex-period costs in the CPI cohort were for outpatient services, which included, and were driven by, CPI drug reimbursement (costs 2.5-fold greater than non-CPI treatment), and specialist office visits (costs 3.1-fold greater than non-CPI treatment).
While the study acknowledged that anti–PD-1 axis CPI drugs cost more than other treatments, it reported that total costs of care decreased in all patients with NSCLC following FDA approval of CPI, owing to a reduction in ED visits and hospitalizations in the NSCLC population overall. In our analysis, however, we found post–index-period ED and hospitalization utilization rates and costs to be higher in CPI-treated versus non-CPI–treated patients.
,
,
our data suggest substantial underutilization of basic molecular testing in NSCLC. At baseline, the mean number of reimbursed EGFR- and ALK-targeted therapy tests was far below 1 (0.1–0.2 per patient), and only modest for PD-L1 IHC (0.7 per patient). It is possible that EGFR and ALK testing may have been performed in some patients as a part of a larger comprehensive genomic profiling (CGP) assay. Unlike billing codes for single-gene tests (see Appendix 1 in the online version at doi:https://doi.org/10.1016/j.clinthera.2020.06.018), CGP testing is not necessarily billed with single test codes and is difficult to identify in claims data. It is unlikely, however, that including known CGP testing would materially impact results, given that it has significantly lower documented utilization in NSCLC than do single-gene tests.
The investigators found PD-L1 IHC testing rates to be low, although they had increased over the follow-up period, from 3% in 2015 to 70% in 2017. Similar to our findings on pembrolizumab, Khozin et al
reported a median duration of CPI treatment of 113 days. In a larger follow-up study in 5257 patients who received CPI treatment in clinical practice, progression-free survival was 3.2 months, with longer progression-free survival and overall survival associated with increased PD-L1 staining.
These findings were described as similar to, or slightly worse than, those in patients in the pivotal trials of these therapies.
,
,
,
,
Thus, as has been recently noted by others,
,
reimbursement for molecular testing appears to be disproportionately low in light of its substantial cost-utility value to patients and payers relative to the high cost of CPI treatment, despite its current shortcomings.
In particular, KEYNOTE-024 (pembrolizumab versus chemotherapy for PD-L1–positive non–small cell lung cancer) and KEYNOTE-021 excluded patients with sensitizing mutations in EGFR or ALK.
,
The only CPI trial that included patients with EGFR mutations and PD-L1 expression was stopped prematurely due to a lack of efficacy.
There is also evidence that patients with other oncogenic driver mutations, including ROS1 and MET, who receive CPI prior to targeted therapy for these gene aberrations, experience increased immune-related toxicity once they do receive targeted therapy for these gene aberrations, discontinue treatment, and lose potential benefit for more appropriate treatment.
Thus, NSCLC treatment selection based only on PD-L1 IHC testing may result in costlier CPI treatment, unnecessary exposure to side effects, and worse outcomes in patients with co-expression of PDL-1 and a driver mutation that responds to more effective targeted agents. While we could not call out and specifically quantify adverse event–related costs in our analysis because the claims data were limited to diagnosis and procedural billing code information, adverse event–related costs were included in the total costs of care and may have contributed to the higher costs in the CPI-treated versus the non-CPI–treated patients in our study.
for example, patients with NSCLC who had PD-L1 expression of ≥50% and also had high tumor mutational burden demonstrated an objective response rate of 75% with nivolumab. This rate was >2-fold the response rates in subgroups with high tumor mutational burden and low PD-L1 expression on IHC (≥1%–49% [32%]), or high PD-L1 expression on IHC (≥50%) and low tumor mutational burden (34%).
In a 2017 national-coverage analysis of next-generation sequencing in cancer, the investigators proposed coverage with evidence development as a reimbursement pathway that could have included immunotherapy testing,
but this component was removed in the final 2018 determination.
Similarly, while US commercial payers may have interest in risk-sharing agreements (at least for new drugs), they have been reticent to engage, reportedly due to perceived administrative burden, with a lack of access to clinical outcomes data noted as a key barrier.
Laboratories that partner with treating clinicians for clinical data sharing are, therefore, best positioned to enlist payer support of novel comprehensive genomic and immune profiling tests that can stem growing CPI costs and get patients to optimal alternative treatment strategies.
Limitations
and included only patients with a diagnosis of metastasis based on ICD diagnosis codes. The Charlson comorbidity index scores in both the unmatched (6.4–6.9) and PS-matched (6.9–7.0) cohorts were in the expected range for this population,
and were much higher than those reported in the aforementioned study (2.2–3.0),
providing additional confirmation of accurate ascertainment of patients with advanced NSCLC. We used CPI drug label indications to restrict the inclusion of patients in the non-CPI cohort to those who were eligible for CPI but received other agents; however, the impact of other patient characteristics, such as comorbidities, could not be accounted for. MRU and costs did not include in-hospital oral medications unless they were physician administered and a separate professional claim was filed, as those items are not captured on claims forms or separately reimbursed.
Comments are closed.