Disparity in survival benefits of pembrolizumab between Asian and non‐Asian patients with advanced cancers: A systematic review and meta‐regression analysis

Abstract Background Immune checkpoint inhibitors have revolutionized the treatment of malignancies. However, disproportionate enrollment among races and ethnicities places the generalizability of global trial results in doubt. Methods In this systematic review, phase 3 randomized controlled trials investigating pembrolizumab in advanced cancers and providing subgroup analyses of Asian and non‐Asian participants were included. The primary and secondary effect measures were the mean differences (MDs) in the natural logarithms of the hazard ratios (HRs) for overall survival (OS) and progression‐free survival (PFS) between these two subgroups, respectively. We used random‐effects meta‐analysis to calculate the pooled ratios of HRs (i.e., exp(MD)) and implemented a meta‐regression analysis to identify significant covariates. Results A total of 17 and 11 trials were included in the meta‐analyses of OS and PFS, respectively. These trials included 2732 (25.49%) Asian and 7000 (65.32%) non‐Asian participants in the OS analysis and 1438 (22.5%) Asian and 4129 (64.61%) non‐Asian participants in the PFS analysis. The pooled ratio of HRs for OS was 0.87 (95% CI: 0.76–0.99; p = 0.0391), favoring Asian participants, but no significant difference was found in PFS (pooled ratio of HRs: 0.93; 95% CI: 0.82–1.07; p = 0.2391). Both linear meta‐regression analyses revealed an open‐label design as a crucial covariate, which indicated more benefits for non‐Asian participants. Conclusions Compared with non‐Asian patients, Asian patients with advanced cancers may derive superior OS benefits from pembrolizumab. Although the results warrant further exploration, this meta‐analysis provides insight into clinical research design.


| INTRODUCTION
Recent advances in immunotherapy have revolutionized the treatment landscape for multiple cancer types, and immune checkpoint inhibitors (ICIs) are one of the most promising therapeutics.Clinical trials have repeatedly demonstrated the overall survival (OS) benefits of ICIs in patients with cancer of various types, and health authorities worldwide have granted regular approval for several ICIs, including a cytotoxic T-lymphocyte antigen 4 (CTLA4) inhibitor (ipilimumab); programmed cell death 1 (PD-1) inhibitors (cemiplimab, dostarlimab, nivolumab, and pembrolizumab); and programmed cell death ligand 1 (PD-L1) inhibitors (atezolizumab, avelumab, and durvalumab).Among these, pembrolizumab is the most widely indicated and is anticipated to be the most sold drug in 2023. 1 Patients from different racial and ethnic groups experience different epidemiology, management, and survival outcomes in various cancers. 2How the efficacy of ICIs varies with race and ethnicity has not been elucidated.Unique racial or ethnic features may aid in identifying potential biomarkers and advancing precision medicine.The POPLAR and OAK studies, in which patients with non-small cell lung cancer (NSCLC) were treated with atezolizumab, reported favorable prognoses in Asian subgroups; different frequencies of specific genetic alterations were suggested to be an explanation for these results. 35][6] By contrast, non-Asian patients with gastric cancers have exhibited more immune-supportive antitumor signatures than Asian patients. 7However, the pharmacokinetics of pembrolizumab are similar between Asian and White patients. 8,9The findings mentioned above shed light on the value of racial and ethnic differences in trial design for immuno-oncology.
Racial and ethnic disparities in global trial enrollment could result in disproportionately more research on Western patients than on Eastern patients.Most trial participants are recruited from North America or Western Europe, and few participants are recruited from Asia. 10 Most global trials have reported the results of racial and ethnic subgroup analyses, but no studies have convincingly demonstrated that the efficacy of ICIs for Asian and non-Asian patients differs.Although some trials have indicated superior efficacy in Asian patients, [11][12][13] within-study comparisons have complicated these findings. 14Therefore, we performed a systematic review and meta-analysis to examine whether the efficacy of pembrolizumab for Asian and non-Asian patients differs significantly across various types of advanced metastatic cancers.

| METHODS
This systematic review and meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines. 15,16 2.1 | Literature search Two authors, SH Peng and YS Lu, independently searched the medical literature databases of PubMed, Cochrane Library, Embase, and Web of Science for relevant articles using the terms "pembrolizumab" and "randomized controlled trial." Te details of the search strategies used for each database are provided in Table S1.Studies were included in our meta-analysis if they (1) were phase 3 trials investigating the efficacy of pembrolizumab in a palliative setting and (2) included subgroup analysis by race or ethnicity.Phase 2 studies, trials in adjuvant or neoadjuvant settings, and irrelevant studies were excluded.We (SH Peng and YS Lu) discussed and resolved selection discrepancies with other authors until a consensus was reached.The Cochrane tool for randomized trials was used to assess the risk of bias in the following domains: the randomization process, deviations from the intended intervention, missing outcome data, outcome measurement, and selective reporting (Table S2).17

| Data extraction
SH Peng and YS Lu independently extracted and verified relevant information from the published manuscripts and supplementary appendices of 26 selected studies to obtain the following metadata: trial name, first author, publication year, study design (open-label or double-blind, placebo-controlled), cancer type, setting (first-line treatment or beyond), experimental treatment strategy (monotherapy, add-on, or combination), criteria for PD-L1 expression, primary endpoints, patient numbers, and demographic characteristics (e.g., median age, sex distribution, and the percentage of Eastern Cooperative Oncology Group performance status [ECOG PS] scores of 0).The selected cutoffs for PD-L1 expression agreed with the primary analysis by statistical testing hierarchy, available demographic characteristics, or label indications (Table S3).

| Statistical analysis
We used the metafor package (version 3.8-1) in R 4.2.1 (R Foundation for Statistical Computing) to perform meta-analyses and meta-regression analyses. 18A twosided p ≤ 0.05 was considered statistically significant.The primary and secondary outcomes were OS and PFS, respectively.The mean difference (MD) in the natural logarithms of the HRs for OS or PFS between the Asian and non-Asian subgroups was determined as the effect measure to compare the efficacy of pembrolizumab in the Asian and non-Asian subgroups, as the following equation: where log is the natural logarithm; MD values <0 and >0 imply that pembrolizumab treatment benefits Asian and non-Asian patients more, respectively.
In forest plots, the estimated pooled ratios of HRs (i.e., exp(MD)) for OS and PFS for Asian to non-Asian patients were obtained through random-effects meta-analysis with Knapp-Hartung adjustment.The heterogeneity among the selected studies was evaluated using the chi-square Q test and the I 2 statistic, where a p < 0.15 in the Q test or I 2 of >50% indicated substantial heterogeneity. 19f the statistical tests for heterogeneity and the I 2 or substantive knowledge suggested substantial heterogeneity among the analyzed studies, a fixed-effects linear metaregression analysis of MDs was fitted to the metadata through the weighted least-squares method to identify the relevant covariate that might account for the observed heterogeneity.If the residual heterogeneity remained statistically significant, a mixed-effects linear meta-regression analysis of the MDs was performed with added random effects to account for unknown sources of heterogeneity.
Simple and multiple generalized additive models (GAMs) of the MDs were constructed and used to draw GAM plots for detecting the nonlinear effects of continuous covariates and identifying the appropriate cutoffs for continuous covariates, if necessary, during the stepwise variable selection procedure.We used the vgam() function with the default values of the smoothing parameters (e.g., s(age, df = 4, spar = 0) for cubic smoothing splines) in the VGAM package in R to fit the GAMs for continuous responses and then used the plotvgam() function of the same package to draw GAM plots to visualize the linear or nonlinear effects of continuous covariates. 20ariable selection, goodness-of-fit assessment, and regression diagnostics were used in the meta-regression analysis to ensure the quality of the results.Specifically, with the aid of the likelihood ratio test, a stepwise variable selection procedure (including iterations between the forward and backward steps) was implemented to obtain the candidate variables for the final linear meta-regression model.The coefficient of determination, R 2 , for the final model is the square of the Pearson's correlation coefficient between the observed and predicted response values.Finally, regression diagnostics were performed, including multicollinearity, residual analysis, influential studies identification, and publication bias assessments.

TABLE 1 (Continued)
T     S4). d In KN-048, the enrollment of the combination arm had been temporarily postponed, and the 22 patients allocated to the control arm during this period were excluded.e In the Asian subgroup of KN-119, patient numbers were reported, but the HR for OS was not available.In the non-Asian subgroup of KN-355, the HR for OS was available, but patient numbers were not. f In KN-045, 113 patients were enrolled from East Asia.

T A B L E 2 (Continued)
based on race or ethnicity were available for OS for 17 trials and PFS for 11 trials and comprised 10,717 and 6391 patients, respectively.In the OS analysis, the Asian and non-Asian subgroups had 2732 (25.49%) and 7000 (65.32%) patients, respectively.In the PFS analysis, the Asian and non-Asian subgroups had 1438 (22.50%) and 4129 (64.61%) patients, respectively.Table S4 lists the countries participating in each trial.The top three Asian countries participating in global trials were Japan, South Korea, and Taiwan.The non-Asian countries for global trials were in North America (Canada and the United States), Europe (mainly France, Germany, Spain, and the United Kingdom), Oceania (Australia and New Zealand), Africa (South Africa), the Middle East (Israel and Turkey), and Latin America (mainly Argentina and Chile).
Table 3A summarizes the results of the linear fixedeffects meta-regression analysis of the MDs in the natural logarithms of the HRs for OS between the Asian and non-Asian subgroups.After adjustment for the effects of other covariates, the mean value of the MDs was −0.7424, favoring the Asian subgroups (p < 0.001; estimated ratio of HRs: 0.4760; 95% CI: 0.3261-0.6948).This value was 0.4897 higher in the trials using add-on pembrolizumab (to ChT or MKI) or combination pembrolizumab-MKI with an open-label design (p = 0.0069; estimated ratio of HRs: 1.6318; 95% CI: 1.1442-2.3273),0.4257 higher if the percentage of ECOG PS scores of 0 was ≤50.28% in the experimental arm (p = 0.0085; estimated ratio of HRs: 1.5307; 95% CI: 1.1148-2.1018),and 0.2787 higher if the percentage of men was ≤81.92% in the experimental arm (p = 0.0411; estimated ratio of HRs: 1.3214; 95% CI: 1.0113-1.7265).
F I G U R E 2 Random-effects meta-analysis of overall survival differences with Knapp-Hartung adjustment (m = 19).The pooled ratio of HRs for OS between the Asian and non-Asian subgroups (0.87; 95% CI, 0.76-0.99;p = 0.0391; chi-square Q test χ 2 = 20.84;p = 0.29; I 2 = 23.9%).Each paired arm corresponds to a square centered at the point estimate of the effect measure (i.e., the ratio of the HRs of the Asian to non-Asian subgroup), with the size being proportional to the study weight.The horizontal line extending from both sides of the square represents the 95% CI. "RE" denotes "random-effects." The normal Q-Q plot of the regression residuals indicated that the assumption of normality was not violated (Figure S2A).A funnel plot and Egger's test for publication bias did not demonstrate significant asymmetry (p = 0.3686; Figure S3A).
Table 3B presents the results of the linear fixedeffects meta-regression analysis of the MDs in the natural logarithms of the HRs for PFS between the Asian and non-Asian subgroups.After adjustment for the effects of other covariates, the mean value of the MDs was −0.2247 (p = 0.0402; estimated ratio of HRs: 0.7988; 95% CI: 0.6444-0.9900).This value was 0.2948 higher in openlabel trials (p = 0.0495; estimated ratio of HRs: 1.3429; 95% CI: 1.0007-1.8020).
The normal Q-Q plot of the regression residuals indicated that the assumption of normality was not violated (Figure S2B).A funnel plot and Egger's test did not suggest significant publication bias (p = 0.2217; Figure S3B).

| DISCUSSION
In this systematic review, approximately 25% of the patients in the formal analyses of OS and PFS were Asian (25.49% and 22.50%, respectively).Similar to a previous report, 66 our meta-analysis results suggest that Asian patients may derive a greater OS benefit from pembrolizumab than non-Asian patients (pooled ratio of HRs: 0.87; 95% CI: 0.76-0.99;p = 0.0391).However, the efficacy in PFS did not significantly differ between the Asian and non-Asian subgroups (pooled ratio of HRs: 0.93; 95% CI: 0.82-1.07;p = 0.2391).Generally, OS is a relatively solid endpoint in oncology trials compared to PFS.Consistent with our analysis results (Figure S1A,B), the magnitude of the effects in the PD-(L)1 inhibitors was greater for OS than for PFS in a meta-analysis report. 67The traditional assessment of PFS by the response evaluation criteria in solid tumors (version 1.1) inadequately reflects the advantage of PD-(L)1 inhibitors, particularly in the context of T A B L E 3 Linear fixed-effects meta-regression analysis of mean differences in natural logarithms of hazard ratios for overall survival (A) and progression-free survival (B) between Asian and non-Asian subgroups.atypical response patterns (e.g., pseudoprogression).Accordingly, OS is suggested to be the gold standard endpoint in PD-(L)1 inhibitor trials. 67When exploring the heterogeneity accounting for the observed differences, both metaregression analyses identified an open-label design as a significant contributing factor, with the regression coefficient estimates indicating more benefits for non-Asian patients (estimated ratio of HRs for OS and PFS, 1.6318 and 1.3429; p = 0.0069 and 0.0495).
According to the regression diagnostics of the two models (Figure S4), KEYNOTE (KN)-010, 45 KN-048-C (one of the trial arms using pembrolizumab as an add-on to ChT for comparison with SOC in untreated HNSCC), 26 KN-181 (second-line pembrolizumab versus ChT in esophageal cancer), 11 and KN-062-P (paired arms comparing pembrolizumab monotherapy and SOC in untreated gastric cancer) 13 were influential studies in the OS meta-regression analysis, and KN-590 (pembrolizumab as an add-on to ChT compared with SOC in untreated esophageal cancer) 12 and KN-048-C were influential studies in the PFS meta-regression analysis.In KN-010, 45 patients with NSCLC harboring epidermal growth factor receptor (EGFR) mutations and disease progression after platinum-based ChT and an EGFR tyrosine kinase inhibitor were eligible and constituted 8.8% of the patients in the pembrolizumab monotherapy arm and 7.6% of the patients in the docetaxel arm.Moreover, a metaanalysis demonstrated that pembrolizumab significantly prolonged OS for patients with wild-type EGFR, not mutant EGFR. 68Several studies have demonstrated that the frequency of driver mutations in lung adenocarcinoma varies by race and ethnicity, with EGFR mutations being the most prevalent in Asian patients. 69In KN-181 70 and KN-590, 71,72 the histology of the Asian participants was primarily esophageal squamous cell carcinoma, consistent with the epidemiologic differences between Eastern and Western countries. 73imilar phenomena have been observed in HNSCC, which is characterized by regional heterogeneity in risk factors and anatomical sites.In Southeast Asia, a high prevalence of oral cavity cancer is associated with the consumption of areca nuts, whereas oropharyngeal human papillomavirus infection contributes to greater rates of oropharyngeal cancer in North America and Western Europe. 74Additionally, in KN-062-P, the proportions of Asian participants with ECOG PS scores of 0 and CPSs of 10 or greater were higher than those of the global population. 43Collectively, our findings highlight the unique features of diverse races and ethnicities and how different subpopulations may unequally benefit from pembrolizumab.In RCTs, masking is crucial to ensure an unbiased estimate of the treatment effect; an open-label design can introduce bias and exaggerate the results. 75,76However, masking is only achievable in certain circumstances based on the nature of the experimental treatment.For instance, participants or investigators may be aware of the assigned treatment because some adverse events, such as immune-related adverse events, could be indicative of ICI monotherapy.In this systematic review, 88.24% of the trials with pembrolizumab monotherapy arms (m = 17) used an open-label design.The use of placebos may help reduce bias, but placebo production can be challenging and time-consuming.The accrual rate may be influenced by the placebo design, and accrual projections are crucial, particularly if competitive studies are being conducted simultaneously.Consequently, objective and reliable outcome measurements, independent outcome adjudications (e.g., blinded independent central radiology review for PFS), and sensitivity analyses with censoring rules could be employed to minimize this risk.These practical issues are unavoidable in trial design.Further investigation is warranted to understand the relationship between an open-label design and any observed differences between Asian and non-Asian patients.
The strengths of this meta-analysis include its rigorous data extraction and thorough meta-regression analysis; these provide comprehensive assessments of the relative efficacy of pembrolizumab for Asian participants and insight into global trial design.These encouraging results may help promote the expansion of global trials in Asia.The inaccessibility of individual patient data is a major limitation, rendering comparisons of demographic and disease characteristics between subgroups challenging.Because of restricted data sources, the efficacy for Asian patients was not compared to other subdivided races and ethnicities.In the meta-regression model for OS, the lower proportions of ECOG PS scores of 0 (≤50.28%)and men (≤81.92%) in the experimental arms resulted in superior OS benefits for the non-Asian subgroups (estimated ratio of HRs, 1.5307 and 1.3214; p = 0.0085 and 0.0411, respectively).However, sex was discovered in another meta-analysis not to be significantly associated with the efficacy of ICIs. 77Therefore, the two aforementioned covariates could be ascribed to other unidentified factors affecting the observed differences.Second, post-study treatment, especially ICIs, could be a confounding source, but analyzing disparities in drug availability is beyond the scope of this study. 78he percentages of patients receiving any subsequent anticancer treatment or post-progression PD-(L)1 inhibitor (Table S5) had no significant effect on the results of the meta-regression model for OS (Table S6).Finally, our attrition due to the exclusion of trials without available subgroup analyses may reduce the statistical credibility of our meta-analysis.The inherent nature of study selection in this meta-analytic study could partially explain the vulnerability of the pooled ratio of HRs for OS and the inconsistent results between OS and PFS.

| CONCLUSIONS
This meta-analysis of phase 3 RCTs investigating the efficacy of pembrolizumab in advanced cancers indicates that pembrolizumab may provide a greater OS benefit for Asian than non-Asian patients.

ACKNOWLEDGMENTS
We would like to thank Dr. Rebecca Cheng for the idea generation and data analysis consultation.This manuscript was edited by Wallace Academic Editing.

FUNDING INFORMATION
This work was supported by the National Science and Technology Council, Taiwan (MOST 110-2314-B-002-218).

CONFLICT OF INTEREST STATEMENT
Ching-Hung Lin reports honoraria from AstraZeneca, Eli Lilly, Novartis, and Pfizer, consulting for AstraZeneca, Eli Lilly, Novartis, and Pfizer.Tom Wei-Wu Chen reports honoraria from AstraZeneca, Daiichi Sankyo, Eisai, Eli Lilly, Novartis, and Roche, consulting for

A B L E 2
Patient numbers in total populations, Asian subgroups, and non-Asian subgroups.First author (year)

F I G U R E 3
Random-effects meta-analysis of progression-free survival differences with Knapp-Hartung adjustment (m = 12).The pooled ratio of HRs for PFS between the Asian and non-Asian subgroups (0.93; 95% CI: 0.82-1.07;p = 0.2391; chi-square Q test χ 2 = 7.37; p = 0.77; I 2 ≈ 0%).Each paired arm corresponds to a square centered at the point estimate of the effect measure (i.e., the ratio of the HRs of the Asian to non-Asian subgroup), with the size being proportional to the study weight.The horizontal line extending from both sides of the square represents the 95% CI. "RE" denotes "random-effects."| 20047ENG et al.
Trial and demographic characteristics of selected studies.
Abbreviations: 1/2/3L, first-/second-/third-line; ABC, advanced breast cancer; ADC, adenocarcinoma; BSC, best supportive care; C, combination arm; CC, cervical cancer; ccRCC, clear cell renal cell carcinoma; ChT, chemotherapy; CPS, combined positive score; CRC, colorectal cancer; EC, esophageal cancer; ECOG PS, Eastern Cooperative Oncology Group performance status; EMC, endometrial cancer; ETOP, European Thoracic Oncology Platform; GC, gastric cancer; HCC, hepatocellular carcinoma; HNSCC, head and neck squamous cell carcinoma; KN, KEYNOTE; MKI, multikinase inhibitor; MSI-H, microsatellite instability-high; NPC, nasopharyngeal cancer; NR, not reported; NSCLC, non-small cell lung cancer; OS, overall survival; P, pembrolizumab monotherapy arm; PD-L1, programmed death ligand-1; PFS, progression-free survival; PM, pleural mesothelioma; SQ, squamous cell carcinoma; TPS, tumor proportion score; UC, urothelial carcinoma.a Double-blind, placebo-controlled design (n = 9); KN-062 was a partially blinded trial.b The percentage of patients with squamous cell carcinoma was 64.9% in KN-181 and 73.2% in KN-590.The remaining approximately one-third of patients had ADC.c stratified by race (155 patients from East Asia and 22 patients from non-Asian countries), all selected studies were stratified by geographic region.The reported Asian subgroups contained only Japanese patients (KN-024, -045 and -189), East Asian patients without Japanese patients (KN-240), and East Asian patients without Thai or Malaysian patients (KN-590).Otherwise, the Asian subgroup was defined as those patients enrolled from countries in East Asia (Table Abbreviations: C, combination arm; Ctrl, control arm; ETOP, European Thoracic Oncology Platform; Exp, experimental arm; KN, KEYNOTE; N/A, not available; OS, overall survival; P, pembrolizumab monotherapy arm; PFS, progression-free survival.aThecomposition of each non-Asian subgroup is specified in TableS4.b Total patient numbers of Asian subsets divided by entire trial populations, unless otherwise specified.c Except for KN-775