Abstract
This comparative effectiveness research study examines outcomes associated with pembrolizumab and nivolumab among US veterans with advanced, unresectable, or metastatic melanoma.
Introduction
Monoclonal antibodies targeting programmed cell death protein 1 (PD-1) are mainstays of modern cancer therapy. Pembrolizumab and nivolumab are the most used members of this class, but their comparative effectiveness and safety have not been established in a prospective randomized clinical trial. Comparative effectiveness and safety analyses in the few indications where both are used in the same manner may help support claims of the agents’ interchangeability.1 Initial retrospective analyses of clinical data with limited follow-up suggest the 2 agents’ comparative effectiveness and safety but have not been evaluated with extended follow-up in a national US health system.2,3 We therefore conducted a propensity score-adjusted comparison of the 2 agents’ effectiveness and safety outcomes in advanced, unresectable, or metastatic melanoma using data from the national Veterans Affairs (VA) health system, the largest integrated provider of cancer care in the US.
Methods
For this comparative effectiveness research study, we extracted all pembrolizumab and nivolumab administrations for advanced, unresectable, or metastatic melanoma (using International Classification of Diseases, Ninth Revision [ICD-9] or International Statistical Classification of Diseases and Related Health Problems, Tenth Revision [ICD-10] codes and natural language processing [NLP]) from January 1, 2015, to July 1, 2025, from the VA’s Corporate Data Warehouse (eMethods in Supplement 1). We excluded patients with known BRAF variations at immunotherapy initiation and those who received ipilimumab or nivolumab. Baseline covariates for patient, temporal, and cancer characteristics thought to influence anti–PD-1 prescribing were balanced using propensity score adjustment, accounting for missing data via multiple imputation (eMethods, eFigures 1 and 2 in Supplement 1). The primary outcome was time-to-next treatment (TTNT); secondary outcomes were overall survival (OS) and prolonged, high-dose corticosteroids (a proxy for clinically significant all-cause immune-related adverse events).1 Propensity-weighted Cox proportional hazard models and Kaplan-Meier estimates were pooled across imputations using Rubin rules.
The VA Ann Arbor institutional review board deemed this study exempt and granted a waiver of informed consent for this analysis of existing administrative data. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines. Analysis was performed with R version 4.4.1 (R Project for Statistical Computing).
Results
The cohort included 742 patients with advanced or metastatic melanoma who received anti–PD-1 therapy with either pembrolizumab (413 [55.7%]) or nivolumab (329 [44.3%]); the median (IQR) age was 73 (68-80) years, and 723 (97.4%) were male (Table). Baseline characteristics were similar between treatment groups. Pembrolizumab recipients were older and were more likely to have initiated therapy in recent years. Median follow-up was 5.6 years (95% CI, 4.27-6.61 years) for pembrolizumab and 5.5 years (95% CI, 4.97-6.40 years) for nivolumab. Propensity score adjustment achieved excellent covariate balance on all variables (with a standardized difference below 0.05). Median TTNT was 14.6 months (95% CI, 11.8-18.6 months) for pembrolizumab and 13.2 months (95% CI, 9.95-16.1 months) for nivolumab (adjusted hazard ratio [aHR] for pembrolizumab vs nivolumab, 0.92; 95% CI, 0.78-1.10; P = .38) (Figure, panel A). Median OS was 27.9 months (95% CI, 21.3-32.4 months) for pembrolizumab and 25.3 months (95% CI, 19.6-34.3 months) for nivolumab (aHR, 0.96; 95% CI, 0.79-1.16; P = .66) (Figure, panel B). Two-year cumulative event rates of prolonged, high-dose corticosteroid use were 22.1% (95% CI, 17.3%-26.9%) for pembrolizumab and 23.5% (95% CI, 17.7%-29.2%) for nivolumab (aHR, 1.00; 95% CI, 0.70-1.44; P = .99) (Figure, panel C).
Table. Baseline Characteristics of the Study Population.
| Characteristic | Patients, No. (%) | |
|---|---|---|
| Nivolumab (n = 329) | Pembrolizumab (n = 413) | |
| Sex | ||
| Female | 9 (2.7) | 10 (2.4) |
| Male | 320 (97.3) | 403 (97.6) |
| Race and ethnicity | ||
| Hispanic | 6 (1.9) | 11 (2.8) |
| Non-Hispanic Black | 7 (2.3) | 7 (1.8) |
| Non-Hispanic White | 288 (92.9) | 380 (95.0) |
| Othera | 9 (2.9) | 2 (0.5) |
| Missing | 19 (5.8) | 13 (3.1) |
| Age, y | ||
| ≤59 | 29 (8.8) | 26 (6.3) |
| 60 to 69 | 87 (26.4) | 79 (19.1) |
| 70 to 79 | 134 (40.7) | 195 (47.2) |
| ≥80 | 79 (24.0) | 113 (27.4) |
| Geographic region | ||
| Continental | 45 (13.7) | 43 (10.4) |
| Midwest | 60 (18.2) | 97 (23.5) |
| North Atlantic | 66 (20.1) | 77 (18.6) |
| Pacific | 75 (22.8) | 107 (25.9) |
| Southeast | 83 (25.2) | 89 (21.5) |
| Year of treatment | ||
| 2015-2017 | 103 (31.3) | 127 (30.8) |
| 2018-2020 | 140 (42.6) | 114 (27.6) |
| 2021-2023 | 77 (23.4) | 136 (32.9) |
| 2024-2025 | 9 (2.7) | 36 (8.7) |
| Body mass indexb | ||
| Healthy weight | 79 (24.3) | 118 (28.6) |
| Obese | 113 (34.8) | 145 (35.2) |
| Overweight | 128 (39.4) | 142 (34.5) |
| Underweight | 5 (1.5) | 7 (1.7) |
| Missing | 4 (1.2) | 1 (0.2) |
| Smoking status | ||
| Never | 105 (37.2) | 136 (38.0) |
| Prior | 128 (45.4) | 162 (45.3) |
| Current | 49 (17.4) | 60 (16.8) |
| Missing | 47 (14.3) | 55 (13.3) |
| Charlson Comorbidity Index (excluding malignant neoplasms) | ||
| 0 | 108 (32.8) | 113 (27.4) |
| 1 | 89 (27.1) | 130 (31.5) |
| 2-3 | 75 (22.8) | 98 (23.7) |
| 4-6 | 46 (14.0) | 53 (12.8) |
| ≥7 | 11 (3.3) | 19 (4.6) |
| Distance to primary VA facility, miles | ||
| 0-25 | 172 (52.3) | 207 (50.2) |
| 26-50 | 76 (23.1) | 112 (27.2) |
| 51-75 | 40 (12.2) | 53 (12.9) |
| >75 | 41 (12.5) | 40 (9.7) |
| Missing | 0 | 1 (0.2) |
| Area Deprivation Index | ||
| 1-20 | 48 (14.8) | 76 (18.8) |
| 21-40 | 54 (16.7) | 72 (17.8) |
| 41-60 | 84 (25.9) | 94 (23.2) |
| 61-80 | 78 (24.1) | 86 (21.2) |
| 81-100 | 60 (18.5) | 77 (19.0) |
| Missing | 5 (1.5) | 8 (1.9) |
| Rural-urban classification group | ||
| Urban | 268 (81.5) | 327 (79.2) |
| Large rural city or town | 28 (8.5) | 48 (11.6) |
| Small rural town | 18 (5.5) | 27 (6.5) |
| Isolated small rural town | 15 (4.6) | 11 (2.7) |
| ECOG performance status | ||
| 0 | 83 (33.2) | 112 (37.8) |
| 1 | 113 (45.2) | 122 (41.2) |
| 2 | 41 (16.4) | 50 (16.9) |
| 3 | 13 (5.2) | 10 (3.4) |
| 4 | 0 | 2 (0.7) |
| Missing | 79 (24.0) | 117 (28.3) |
| Hospitalization in prior year | 127 (38.6) | 171 (41.4) |
| Original stage (at diagnosis) | ||
| I-III | 138 (61.6) | 171 (63.6) |
| IV | 86 (38.4) | 98 (36.4) |
| Missing | 105 (31.9) | 144 (34.9) |
| Brain metastasis | 47 (20.8) | 56 (18.0) |
| Missing | 103 (31.3) | 102 (24.7) |
Abbreviations: ECOG, Eastern Cooperative Oncology Group; VA, Veterans Affairs.
Other race and ethnicity includes American Indian or Alaska Native, Asian, and Native Hawaiian or other Pacific Islander.
Thresholds for body mass index (calculated as weight in kilograms divided by height in meters squared) categories were: underweight, below 18.5; healthy weight, 18.5 to below 25; overweight, 25 to below 30; and obese, 30 and higher.
Figure. Kaplan-Meier Graph of Comparative Clinical Effectiveness and Safety of Pembrolizumab and Nivolumab in Advanced or Metastatic Melanoma.
OS indicates overall survival.
Discussion
In this US national cohort with more than 5 years of follow-up and extensive control for confounding, pembrolizumab and nivolumab showed similar effectiveness and safety as first-line therapies for advanced, unresectable, or metastatic melanoma, consistent with anti–PD-1 class effect.1 Given accumulating clinical data from different health systems in multiple countries with total sample size exceeding 2000, these agents may have similar effectiveness and safety across indications.2,3 The decision of which to use should be a joint decision of relevant stakeholders—patients, payers, and health systems—guided by financial, operational, and logistical factors.1 Limitations of our study include reliance on corticosteroid use as a proxy for immune-related adverse events, potential unmeasured confounding, and a limited sample size that cannot not rule out modest differences in safety or effectiveness, particularly in genomically defined subsets.4
eMethods.
eFigure 1. Love plot for covariate balance across imputations
eFigure 2. Propensity score overlap between treatment groups
eReferences.
Data sharing statement
References
- 1.Goldstein DA, Saltz LB, Pond GR, Tannock IF. Pharmacological class effects of anticancer drugs: opportunities for decreasing healthcare spending. BMJ Oncol. 2024;3(1):e000287. doi: 10.1136/bmjonc-2023-000287 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Cybulska-Stopa B, Piejko K, Ostaszewski K, et al. Long-term clinical evidence of comparable efficacy and toxicity of nivolumab and pembrolizumab in advanced melanoma treatment. Melanoma Res. 2023;33(3):208-217. doi: 10.1097/CMR.0000000000000885 [DOI] [PubMed] [Google Scholar]
- 3.Moser JC, Wei G, Colonna SV, Grossmann KF, Patel S, Hyngstrom JR. Comparative-effectiveness of pembrolizumab vs. nivolumab for patients with metastatic melanoma. Acta Oncol. 2020;59(4):434-437. doi: 10.1080/0284186X.2020.1712473 [DOI] [PubMed] [Google Scholar]
- 4.Monson KR, Ferguson R, Handzlik JE, et al. ; IO-GEM Consortium . Inherited mitochondrial genetics as a predictor of immune checkpoint inhibition efficacy in melanoma. Nat Med. 2025;31(7):2385-2396. doi: 10.1038/s41591-025-03699-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
eMethods.
eFigure 1. Love plot for covariate balance across imputations
eFigure 2. Propensity score overlap between treatment groups
eReferences.
Data sharing statement
