Healthcare expenditures in the United States now make up almost 18 percent of GDP, with pharmaceutical spending accounting for an increasing share of overall costs (CMS, 2024a). The health economics literature has debated whether technological innovations deliver commensurate value for their costs, particularly as patent-protected medical technologies command monopoly prices upon market entry (Chandra and Skinner, 2012; Cutler and McClellan, 2001). The challenge is especially acute for breakthrough innovations, where clinical trial efficacy may not translate into population-level effectiveness due to patient selection, heterogeneity in provider expertise, and healthcare system constraints (Lakdawalla, 2018). Consequently, the productivity of pharmaceutical innovation has become a critical concern, given its implications for overall health care spending and, ultimately, patient outcomes. This issue is particularly salient in cancer care, where breakthrough treatments offer potentially transformative benefits, but often at a very high cost (Howard et al., 2015; IQVIA, 2023).

Innovation in cancer screening and treatment decreased mortality from invasive cancer by 17 % between 1980 and 2010 (NCI, 2020; SEER, 2022). However, this progress was not uniform across cancer types. The mortality rate for metastatic melanoma actually increased by 15 % over that same period (NCI, 2020; SEER, 2022). Melanoma is the deadliest type of skin cancer, with an estimated one-year mortality rate of 49 % for patients with metastatic disease (Tas, 2012). Melanoma has been largely non-responsive to traditional cancer treatments such as chemotherapy, leaving patients with few treatment options (Davis et al., 2019; Schmitt and Larkin, 2023).

A breakthrough came in 2011 with the FDA approval of the first immune checkpoint inhibitor (ICI), Yervoy (ipilimumab), an immunotherapy cancer drug that harnesses the body’s immune system to recognize and attack cancer cells. In 2014, two additional ICIs were approved, Keytruda (pembrolizumab) and Opdivo (nivolumab), which became the most widely adopted ICIs. Although these ICIs were eventually approved to treat >27 types of cancer, they were first approved to treat metastatic melanoma.1 As shown in Fig. 1, from 2011 to 2022, the mortality rate for melanoma decreased sharply by 26 %, marking the first sustained decline in mortality for this type of cancer in decades. This compares to a 16 % decline during the same period for all other cancer types, which follows a continuation of a pre-existing decreasing mortality trend.

While ICIs are widely recognized as a revolutionary innovation in oncology care (Davis et al., 2019; Fellner, 2012; Hamid et al., 2017; Lamba et al., 2022; Robert et al., 2011), they are also expensive. The cost of a full course of treatment with an ICI can exceed $150,000.2 In 2022, ICIs accounted for 44 % of the $17.5 billion in spending on cancer drugs in Part B of traditional Medicare, with Keytruda, Opdivo, and Yervoy ranked 1st, 2nd, and 11th with respect to total Medicare FFS spending for cancer treatments by the end of our study period (CMS, 2024b; SEER, 2023). Moreover, as shown in Fig. 2, total Medicare spending on cancer drugs has increased by 122 % since their introduction, and 80 % of this growth can be attributed to ICIs. Despite the rapid diffusion of ICIs, there is surprisingly little evidence on the “real-world” (as opposed to clinical trial) impact of this treatment on health outcomes and utilization for patients with cancer or its effects on health care spending.

In this paper, we investigate the impact of the introduction and diffusion of ICIs, focusing on patients with metastatic melanoma. This is the ideal setting to examine the impact of this transformative healthcare technology for multiple reasons. First, melanoma was the first cancer indication for which ICIs were approved to treat, creating clear pre- and post-time periods. Second, FDA labeling indicates that all patients diagnosed with metastatic melanoma are eligible to be treated with ICIs without any additional biomarker testing requirements, meaning that we can readily define the group that is eligible for treatment in claims data. Lastly, prior existing treatments for melanoma had limited efficacy; thus, there was not a compelling outside option to ICIs, again making it reasonably clear which patients were candidates for treatment.

A persistent challenge in studying the impact of new health technologies is that patients who access treatments differ systematically from those who do not. We address this by estimating a difference-in-differences specification that compares patients with metastatic melanoma to patients with metastatic colorectal cancer (CRC) before and after ICIs were introduced. We select patients with CRC as our comparison group because ICI approval for CRC occurred years after melanoma, and then only for patients with microsatellite instability high (MSI-H) CRC, as determined through biomarker diagnostic testing, affecting only about 15 % of patients.3 As a result, we observed almost no use of ICIs for CRC during our study period. The variation in approval timing across cancer types allows us to isolate the effect of ICIs from other changes in cancer treatment.

Our data comes from administrative fee-for-service claims for a 20 % random sample of traditional Medicare patients between 2008 and 2018.4 Medicare beneficiaries are 4.3 times more likely to be diagnosed with melanoma than the rest of the adult population in the U.S., which makes this a particularly important group to assess the use and impact of ICIs (SEER, 2022). We use standard claims-based algorithms to define patients with metastatic melanoma and metastatic colorectal cancer (Barzilai et al., 2004; CCW, 2023).

Our research design provides estimates of the intent-to-treat impact of ICIs. The study treatment period is defined by the three ICIs that were approved in 2011 (Yervoy) and 2014 (Keytruda and Opdivo), which we refer to as the first ICI and the second ICIs going forward.5,6 For melanoma, use of ICIs increased from 5.4 % in the first year of introduction to 33.7 % among patients diagnosed in 2018; use of ICIs in CRC was much lower at essentially zero through 2016 and then just 3.8 % in 2017 and 5.0 % in 2018. The lower take-up among CRC patients reflects the later date of ICI approval for CRC and relatively narrow eligibility criteria once ICIs were approved for CRC in 2017. To address the comparability of patients with melanoma and CRC, we estimate event studies and show consistency in pre-trends between our treatment and comparison groups. We also conduct multiple robustness checks, including entropy weighting to ensure CRC and melanoma patients are balanced on observable characteristics, estimating an alternative specification that compares outcomes for melanoma patients with high versus low propensity for ICIs use, and, finally, comparing outcomes for melanoma patients diagnosed before versus after ICI introduction.

Across all these approaches, we find robust evidence that ICIs transformed melanoma care. Average treatment costs to Medicare increased by $8321 per patient after the first ICI was introduced and an additional $10,840 after the second ICIs were introduced, a total of $19,161 for melanoma patients relative to CRC. The increase was driven by spending in the outpatient setting (where ICIs are administered) with limited offsets from decreased inpatient spending. We also find evidence of substitution away from existing standards of care, with chemotherapy and radiation decreasing by 39.1 % and 7.2 %, respectively, from pre-ICI use. We find a small, but not statistically significant increase, in 1-year mortality from the approval of the first ICI, but a 3.6pp reduction in 1-year mortality after the approval of the second relative to the first ICIs. When scaled by the first-stage estimate of ICI use, this represents a 28.0 % improvement in survival among ICI users. Our estimated treatment effect aligns closely with the efficacy demonstrated in clinical trials, where the second-generation ICIs showed a 30.8 % improvement in 1-year survival over traditional chemotherapy in the pivotal trial for Opdivo. Similarly, a 10-year follow-up study of Keytruda found a 29 % reduction in overall mortality versus Yervoy (see Appendix Table A1 for other key trials).7 The similarity between our real-world estimates and the results from randomized controlled studies is noteworthy because it demonstrates that the survival benefits observed in curated trial participants in highly controlled clinical studies translate to broader patient populations in clinical practice.

Despite clear evidence of substantial health benefits, adoption was uneven across patient groups, with older and dual Medicare-Medicaid eligible patients showing the lowest usage rates. This finding raises concerns about equitable access for many of the most vulnerable patients to innovative treatments and highlights potential barriers to the diffusion of new high-cost medical technologies. Further, ICIs impose high costs on Medicare, raising questions about cost-effectiveness. A back-of-the-envelope cost-benefit analysis suggests that ICIs would need to extend survival by 4 to 6 times the observed first-year gains for benefits to align with costs. This is based on a first-year survival benefit of $13,440 to $20,160 per treated patient when applying standard valuations of $100,000 to $150,000 per life-year gained, compared to an incremental Medicare spending increase of $85,403 per patient using ICIs.8 However, long-term follow-up of clinical trials demonstrates that survival benefits of this magnitude are plausible (Wolchok et al., 2025) and suggest that ICIs may be cost-effective despite high upfront costs.

Our study contributes to several strands of literature. First, it adds to the health economics literature on the productivity of medical innovations. Prior research has shown that even low-cost, evidence-based innovations have variation in productivity due to disparities in adoption and access; yet, these technologies tend to have the largest impact on improving health (Baicker and Chandra, 2004; Chandra and Staiger, 2007; Skinner, 2011; Skinner and Staiger, 2007; 2015). Conversely, high-cost treatment innovations may increase costs without corresponding benefits, and productivity is often related to physician expertise, spillovers in care, and heterogeneous patient treatment effects (Carroll et al., 2023; Chandra et al., 2023; Chandra and Skinner, 2012; Chandra and Staiger, 2020; Cutler et al., 2010; Duggan, 2005; Horn et al., 2022; Hsia et al., 2024; O’Connor et al., 2018). We build on this literature by focusing on a high-cost cancer technology that has diffused particularly rapidly. Given the limited evidence regarding the productivity of this new technology, we produce new evidence to assess its impact.

This study also adds to our understanding of how novel cancer technologies are used in clinical practice and impact patients outside of highly controlled research studies. Innovation in cancer care has accelerated since the advent of immunotherapy, but our understanding of how new cancer technologies diffuse and impact outcomes in the real world remains limited. The results demonstrated in clinical trials may not clearly translate into clinical practice because of the very controlled nature of clinical trials, the distinct features of the United States (US) healthcare system, and individual decision-making. Specifically, differences in provider practice styles, patient adherence to treatment protocols, and the characteristics of patients treated outside of trials can generate differences in outcomes in real-world vs. clinical trial settings (Klonoff, 2020). Furthermore, patients may not be fully aware of or face the full cost of care, nor understand the risks and benefits of treatment due to information asymmetries. This leads some patients who are marginal or inappropriate for treatment to use therapies that may not be efficacious for them (Booth and Detsky, 2019; Fundytus et al., 2021; Tayapongsak Duggan et al., 2017). Relatedly, individuals may be willing to take greater risks and pay more for a mortality reduction toward the end of life, and doctors typically have wide discretion to provide treatments even when empirical evidence to support use is thin (Fischer et al., 2018; Lakdawalla et al., 2012; NCI, 2022). On the other hand, there are multiple barriers to cancer treatment, including patient cost, insurance coverage differences, and provider propensity to take up new innovations, which can make diffusion uneven even when indicated (Dusetzina et al., 2014; Kaisaeng et al., 2014; Streeter et al., 2011).

Lastly, our paper contributes to the literature on the value of cancer innovations. Much of the existing work on the value of new cancer drugs uses estimates from clinical trials and does not account for differences between trial efficacy and the effectiveness of treatments in real-world settings, which, as discussed above, can diverge because of patient adherence and the characteristics of those treated in real-world settings relative to those enrolled in clinical trials. Consequently, prior evidence on whether the benefits of new cancer treatments are aligned with costs may not be fully informative (Chen et al., 2019; Del Paggio et al., 2017; Dusetzina, 2016; Fellner, 2012; Howard et al., 2015; Lakdawalla, 2018; Ridley and Lee, 2020; Tayapongsak Duggan et al., 2017; Vokinger et al., 2020). The findings from this study can inform policymakers about the opportunities and challenges of translating novel medical treatments into improved patient outcomes, which is particularly important given the increasing speed and cost of innovation in cancer care.

The remainder of this paper is organized as follows. Section 2 provides background on melanoma and ICIs and reviews the relevant literature. Section 3 details our research design and empirical strategy. Section 4 details our data sources and presents summary statistics. Section 5 presents our results. Section 6 discusses the implications of our findings and concludes.