Event-Free Survival as a Surrogate for Overall Survival in Gastric and Gastroesophageal Junction Adenocarcinoma: A Meta-analysis in the Neoadjuvant ± Adjuvant Setting

Gastric or gastroesophageal junction (GEJ) adenocarcinoma is a heterogenous and deadly malignancy for which surgical resection is the only potentially curative therapeutic option (1). However, the risk of recurrence following curative resection is high with a 1-year recurrence rate of 48% (2). Treatment options for localized gastric/GEJ cancer include surgery alone, perioperative chemotherapy, or chemoradiotherapy (1). Treatment approaches that integrate perioperative therapies have substantially improved clinical outcomes for patients with gastroesophageal cancer, although the 5-year overall survival (OS) rate remains only around 36% to 59% (3–6).

OS is a well-established and commonly used endpoint in clinical trials evaluating the effectiveness of perioperative treatment among patients with gastric or GEJ adenocarcinoma (7), but necessitates an extended follow-up time. In addition, OS may be impacted by subsequent lines of therapy following the perioperative treatment. Therefore, solely relying on OS to evaluate treatment effectiveness may substantially delay decisions regarding the clinical benefits of a new therapeutic regimen as well as patient access to effective therapies (8). To overcome this challenge, the use of surrogate endpoints has been proposed by regulatory agencies to enable early access to new and potentially efficacious treatments (9). In trials of neoadjuvant therapies with/without adjuvant therapies for gastric or GEJ adenocarcinoma, event-free survival (EFS) or disease-free survival (DFS) are commonly used surrogate endpoints for OS (10–13).

Establishing the validity of an intermediate outcome, such as EFS, as a surrogate OS endpoint in the neoadjuvant ± adjuvant setting of gastric and GEJ adenocarcinoma is crucial for accelerating treatment access, yet it requires appropriate validation. This can be achieved by various approaches, but trial-level association (i.e., the association between the treatment effect on a surrogate outcome and the treatment effect on OS) is the highest level of evidence in surrogate endpoint validation (9). To date, there are only two published meta-analyses evaluating the trial-level association between EFS/DFS and OS in the neoadjuvant setting in gastric or gastroesophageal cancer. A meta-analysis conducted by Petrelli and colleagues (2017) included neoadjuvant trials in gastroesophageal cancer but presented a subgroup analysis in gastric cancer only (14). The results showed a strong trial-level surrogacy between DFS and OS in gastric cancer (14). However, those results were based on just 4 randomized controlled trials (RCT) published a decade ago. Another meta-analysis conducted by Ronellenfitsch and colleagues (2019) used individual patient-level data (IPD) from 8 RCTs of neoadjuvant therapy in gastroesophageal adenocarcinoma and also demonstrated a strong trial-level surrogacy between DFS and OS (15). However, similar to the first meta-analysis, that study only included RCTs conducted before 2011. Thus, while the two prior studies have suggested that DFS is a potential surrogate for OS in gastric or gastroesophageal carcinoma, both were based on older trials and had small sample sizes.

The validation of EFS as a surrogate for OS should ideally be informed by the most recent data. Therefore, this study aimed to evaluate the trial-level association between OS and the intermediate outcome of EFS in the neoadjuvant ± adjuvant setting using the most current data from published RCTs.

A systematic literature review (SLR) was performed to identify RCTs that assessed EFS and OS outcomes of neoadjuvant treatments (regardless of the administration of adjuvant treatment) for gastric or GEJ adenocarcinoma. Keywords such as “gastric cancer” or “GEJ adenocarcinoma,” “neoadjuvant treatment” or “perioperative treatment,” and “survival outcomes” were used to identify relevant studies. To be eligible for inclusion in the main analysis, studies were required to be published in English and meet the following criteria: (i) study populations that only included patients with gastric, GEJ, or distal (or lower) esophagus adenocarcinoma, but not proximal/upper esophagus carcinoma; (ii) patients treated with neoadjuvant therapy irrespective of treatment type; (iii) classified as a RCT; (iv) EFS and OS were evaluated from the same starting point; and (v) reported HRs or Kaplan–Meier (KM) curves for EFS and OS. Of note, the study included clinical trials of different phases as long as they met the above criteria.

Published RCTs were identified from MEDLINE, Embase, and Cochrane Central Register of Controlled Trials with publication dates from inception until December 2020. Conference abstracts were obtained from the American Society of Clinical Oncology and European Society for Medical Oncology conference proceedings that were available from January 2018 to December 2020.

This study included EFS as the intermediate outcome and OS as the final long-term outcome. Different terminologies were used to describe the intermediate outcome in clinical trials, including EFS, DFS, and progression-free survival (PFS). Therefore, in the current study, EFS was used as a general term to describe all these intermediate outcomes. Although the definitions used for EFS, DFS, and PFS varied across trials, they had many similarities (Supplementary Table S1). First, regardless of EFS, DFS, or PFS, the starting point for these outcomes was from the initiation of neoadjuvant therapy in the trials that explicitly reported the starting point. Of note, this definition of DFS was different from that used in the adjuvant trials, which typically starts at adjuvant therapy initiation. Similarly, this definition of PFS should be differentiated from that used in the advanced/metastatic settings, which starts at advanced/metastatic treatment initiation. In addition, the reported definitions of the intermediate outcomes all included events of progression/recurrence and death, regardless of the terminology used. However, while some trials explicitly described all the events included in the definition [e.g., disease progression, irresectable disease at surgery, locoregional or peritoneal tumor recurrence, distant metastases, or death from any cause, whichever occurred first (10, 11)], others were vague about the events included or did not report the events at all. Such variations may potentially lead to inconsistency in the intermediate outcome definition. Despite this limitation, the current study included all EFS, DFS, and PFS definitions in the analysis because focusing on a specific definition of an intermediate outcome would have substantially reduced the sample size, making the meta-analysis less reliable. More importantly, given the substantial variations in the EFS definition, arbitrarily imposing a standard definition may exclude important studies and thus lead to selection bias. This is a common practice in oncology studies using surrogate outcome analysis (14, 16). In most trials, OS was defined as the interval from randomization to death.

The treatment effect on EFS and OS was evaluated using the HR. Most HRs were directly extracted from publications that met the inclusion criteria. If HRs were not reported, pseudo-IPD were reconstructed based on the published KM curves of EFS and/or OS using the algorithm described in Guyot and colleagues (17). HRs and the 95% CIs were then estimated from the pseudo-IPD using Cox proportional hazards models. This is a well-established method and commonly used in the health technology assessment of novel treatments (18).

The associations between treatment effects on EFS and OS (i.e., HR of EFS and HR of OS) between two randomized treatment arms in the eligible RCTs were assessed in main analyses and in sensitivity analyses with slightly modified inclusion criteria. Weighted linear regressions were performed with log(HR) of EFS as the independent variable and log(HR) of OS as the dependent variable, with natural logarithm. Weights in the performed analyses were based on the number of patients in each comparison. A positive coefficient for log(HR) of EFS indicated that an increase in the HR of EFS was associated with an increase in the HR of OS. The coefficient of determination (R²), correlation coefficient (R) and their 95% CIs were calculated to assess trial-level associations between the treatment effects on EFS and OS. The 95% CIs for R and R² were estimated using the percentile method with 10,000 bootstrap iterations. The strength of the association between the treatment effect on EFS and the treatment effect on OS was evaluated considering the proposed criteria of the following regulatory agencies and research groups (19). Ciani and colleagues (2017) proposed that the threshold indicative of a good surrogate outcome is an R of 0.8 or R² of 0.65 (9), based on a SLR of surrogate-endpoint validation in solid tumors (20). Lassere and colleagues (2018) suggested that an R² close to 1 (greater than 0.9) is indicative of a strong correlation, a value between 0.9 and 0.75 represents a very good correlation, and a value below 0.25 indicates a weak-poor correlation (21). The Institute for Quality and Efficiency in Health Care (IQWiG), a German health technology assessment agency, has also proposed a framework to evaluate the strength of a correlation between an intermediate outcome and a long-term outcome (22). According to the IQWiG framework, a high correlation is considered when the lower limit of the 95% confidence interval (CI) of R is ≥ 0.85, a low correlation is considered when the upper limit of the 95% CI of R is ≤ 0.7, and a medium correlation is considered otherwise (22).

In addition, the surrogate threshold effect (STE) for HR of EFS was also estimated in the main analysis of this study and was defined as the minimum treatment effect on the surrogate (i.e., EFS) that would be required to predict a statistically significant nonzero effect (i.e., HR < 1) on OS.

A leave-one-out cross-validation was used to confirm the validity of the model in the main analysis. Specifically, the regression model was refit with one observation excluded and a prediction of HR for OS was made based on the new model and the observed HR for EFS in the excluded observation. This process was repeated for every observation. Consistency in the direction of the observed and predicted HR for OS (e.g., both above or below 1.0), and whether the observed HR fell within the 95% CIs of the predicted HRs, was evaluated.

In addition to the main analyses, three sensitivity analyses were conducted to explore the impact of the population, intervention, and definition of outcomes in the RCTs in the trial-level association between EFS and OS. Sensitivity analysis 1 included RCTs with a mixed population of patients with gastric/GEJ and distal or proximal esophagus carcinoma. Sensitivity analysis 2 included RCTs in which at least one treatment arm included chemotherapy without radiation in both the neoadjuvant and adjuvant settings. We did not restrict the treatment for the comparator arms, which could include chemoradiation therapy or a different chemotherapy in the neoadjuvant and adjuvant settings, or surgery only. Sensitivity analysis 3 included RCTs that explicitly defined the initiation of neoadjuvant therapy as the starting time of EFS and OS.

All statistical analyses were conducted using R software (version 3.6.3). A P < 0.05 was used to determine statistical significance.

The data generated in this study are available within the article and its Supplementary Data files. Raw data may be provided upon reasonable request to the corresponding author.

Of 825 publications identified from the literature search, 15 RCTs met the inclusion criteria for the main analyses and 17 RCTs met the expanded inclusion criteria for certain sensitivity analyses (see the preferred reporting items for systematic reviews and meta-analysis (PRISMA) diagram in (ref. 23; Fig. 1). Specifically, the 15 RCTs included in the main analyses enrolled patients with gastric or GEJ adenocarcinoma or adenocarcinoma of the distal (or lower) esophagus only, while an additional 2 RCTs recruited a mixed population including patients with the aforementioned conditions or proximal/upper esophagus carcinoma, which were included in sensitivity analysis 1 (24–27).

Detailed characteristics of the included studies are listed in Supplementary Table S1. The 17 RCTs enrolled a total of 4,935 patients, had publication years ranging from 2006 to 2020, and had median follow-up times ranging from 9.6 to 126.5 months. One included RCT used a 2 × 2 factorial design to concurrently evaluate two investigational agents and thus contributed two comparisons (28). Nine studies from 6 RCTs investigated chemotherapy in neoadjuvant settings only (13, 24–27, 29–32), while 18 studies from 11 RCTs examined chemotherapy in both neoadjuvant and adjuvant settings (4–6, 10, 11, 28, 33–44). Thirteen of the 17 RCTs used the initiation of neoadjuvant therapy as the starting time of EFS and OS, and the other 4 RCTs did not report the starting time (28, 29, 39, 41, 44).

Figure 2 shows the relationship between treatment effect on EFS and treatment effect on OS using 16 observations from the 15 included RCTs involving 4,388 patients with gastric, GEJ, or distal (or lower) esophagus adenocarcinoma only. On the basis of the weighted linear regression model, the log(HR) of EFS was a significant predictor of OS with an estimated coefficient of 0.72 (P < 0.001), and was considered to be a good surrogate outcome for the log(HR) of OS with an R² of 0.75 (95% CI, 0.49–0.95), based on the criteria of Ciani and colleagues [i.e., good: 0.65 for R² or 0.8 for R (9)]. According to Lassere and colleagues [i.e., very good: R² = 0.75 to 0.9 (21)], the association between EFS and OS represented a very good correlation. The estimated R between treatment effects on EFS and OS was 0.87 (95% CI, 0.70–0.98), which qualified as a medium correlation per IQWiG criteria (22).

The results from the cross-validation analyses confirmed the validity of the model (Fig. 3). The observed and predicted HRs for OS were consistent regarding the direction of the treatment effect on OS (i.e., both HRs were either above or below 1.0) for all comparisons. The observed HRs fell within the 95% CIs of the predicted HRs in 68.8% (11 of 16) of the comparisons.

The STE for OS corresponded to an EFS HR of 1.00 (Fig. 2). This indicates that an EFS HR of at most 1.00 would need to be ascertained to predict a nonzero treatment effect on OS in a future trial with a similar treatment type. However, clinical input and other judgements should also be considered in addition to the STE during the decision-making process.

The results of all three sensitivity analyses, with R² ranging from 0.76 to 0.89 and R ranging from 0.87 to 0.95, indicated similar associations between EFS and OS as the main analysis, supporting the robustness of the main analysis findings (Table 1). Sensitivity analysis 1, which included a broader patient population (i.e., with gastric/GEJ, distal and proximal esophagus carcinoma), yielded an R² of 0.78 (95% CI, 0.53–0.96). Sensitivity analysis 2, which included studies with patients treated with chemotherapy in both the neoadjuvant and adjuvant settings, had the lowest R² of 0.76 (95% CI, 0.43–0.98). Sensitivity analysis 3, which included 11 RCTs that reported defining starting time of both EFS and OS as the initiation of neoadjuvant therapy, yielded the strongest association between treatment effects on EFS and OS, with an R² of 0.89 (95% CI, 0.66–0.98).

The use of the surrogate measures for regulatory approval has been a matter of substantial debate in oncology, health policy, and regulatory communities (45). The FDA accepts validated surrogate endpoints only when reliable evidence of surrogacy is provided (46). Our study aimed to examine whether EFS could be validated as a surrogate endpoint for OS in trials among patients with gastric and GEJ adenocarcinoma in the neoadjuvant ± adjuvant settings. Our analyses included 17 RCTs, representing more than 4,900 patients with gastric or GEJ adenocarcinoma. To our knowledge, this is the largest and most comprehensive study to evaluate the association between EFS and OS among this population in this setting.

Using aggregated data from eligible published RCTs on neoadjuvant treatment (with or without adjuvant treatment) for gastric or GEJ adenocarcinoma, this study found a statistically significant association between the treatment effect on EFS and the treatment effect on OS. The R² for EFS and OS was 0.75 in the main analysis, thus, EFS can be considered a good surrogate endpoint for OS in gastric or GEJ adenocarcinoma based on the criteria of Ciani and colleagues (good correlation: R² > 0.65) or a very good surrogate endpoint based on Lassere and colleagues [very good correlation: R² = 0.75 to 0.90 (9, 21)]. When using the German-specific IQWiG framework, the association between EFS and OS (R = 0.87; 95% CI, 0.70–0.98) qualified as a medium correlation (22). However, several studies have noted that most surrogate outcomes do not meet the criteria for a high correlation based on the IQWiG framework (20, 47, 48). In fact, among new cancer therapies approved by the FDA based on surrogate endpoints during 1992 to 2019, only 5% met this threshold for a high correlation (47). Therefore, the practicality of the IQWiG framework in guiding regulatory decisions is unclear.

The results of the sensitivity analyses were consistent with the findings from the main analysis, suggesting that the surrogate relationship between EFS and OS was not strongly affected by study population, treatment, or outcome definitions. The lowest correlation for EFS and OS observed in this study was in sensitivity analysis 2, with an R² of 0.76 when restricting to RCTs with at least one treatment arm including chemotherapy only (without radiation) in both the neoadjuvant and adjuvant settings. However, the results still met the commonly cited criteria for a good or very good correlation (9, 21). Conversely, the strongest correlation for EFS and OS was observed in sensitivity analysis 3 (R² of 0.89), which included studies explicitly defining the starting time of both EFS and OS as the initiation of neoadjuvant therapy.

The findings from the current study are consistent with prior literature demonstrating a strong correlation between EFS and OS in neoadjuvant trials in gastric or gastroesophageal adenocarcinoma. For example, the meta-analysis of neoadjuvant therapy in esophageal, gastric or GEJ cancer by Petrelli and colleagues concluded that DFS was strongly correlated with OS (R² = 0.78) in their subgroup trial-level analysis focusing specifically on gastric cancer, even though it was only based on 4 RCTs (14). Of note, Petrelli and colleagues included studies using DFS or PFS, which was consistently described as DFS in the study and defined as time from randomization to disease progression or death of any cause, unless otherwise specified. The definition of DFS used by Petrelli and colleagues is functionally similar to that of EFS in our study. Interestingly, the main analysis conducted by Petrelli and colleagues did not find a strong association between DFS and OS (R² = 0.27 based on 22 RCTs). This is likely due to the fact that 17 of 22 RCTs were conducted in esophageal cancer only, which likely included squamous cell carcinoma. Although sensitivity analysis 1 in the current study included RCTs with a mixed population of gastric, GEJ and esophageal cancers, RCTs focusing on esophageal cancer only were excluded. This may explain the consistent findings between the sensitivity analysis 1 and the main analysis.

The findings of the current study are also consistent with those of a meta-analysis conducted by Ronellenfitsch and colleagues, which reported a R² of 0.91 between DFS and OS in the neoadjuvant setting of gastroesophageal adenocarcinoma, indicating a very strong trial-level association (15). Ronellenfitsch and colleagues defined DFS as the time from a landmark of 6 months after randomization to recurrence or death. Because they had IPD from the included RCTs, they were able to use a uniform definition of the intermediate outcome across all trials. This might partially explain a stronger association between DFS and OS compared with our study. In sensitivity analysis 3, where we standardized the starting time as neoadjuvant therapy initiation, we also observed a stronger association between EFS and OS. However, in a study by Nakamura and colleagues, three different PFS definitions yielded very similar HRs (49). Thus, the impact of the intermediate outcome definition on the association between EFS and OS needs to be further explored. Another major difference is that Ronellenfitsch and colleagues included only RCTs comparing neoadjuvant therapy + surgery with surgery alone while our study did not put any restriction in the comparator arm. However, it is unclear whether the comparator in the RCTs has a major impact on the association. In addition, Ronellefitsch and colleagues included a broader population – gastroesophageal cancers, similar to Petrelli and colleagues in terms of tumor locations. However, because Ronellefitsch and colleagues focused exclusively on adenocarcinoma, their findings were more similar to ours than to the results of Petrelli and colleagues, which included both adenocarcinoma and squamous cell carcinoma. This indicates that histological type plays an important role in the association between EFS and OS.

There are other studies focusing on the surrogacy assessment in the perioperative setting (i.e., including both neoadjuvant and adjuvant trials) or the adjuvant setting only. For example, Oba and colleagues reported that the trial-level association between DFS and OS in the adjuvant settings was strong, with an R² of 0.96 (7). Ajani and colleagues (2022) conducted a meta-analysis including both neoadjuvant treatments and adjuvant treatments in esophageal and GEJ cancers and demonstrated that DFS/PFS was a valid surrogate for OS in the perioperative settings (R = 0.89) (50). Their results were quite different from those of Petrelli and colleagues, which included esophageal cancers, possibly due to the inclusion of studies in the adjuvant settings. As the adjuvant setting could be different from the neoadjuvant setting, the results from these studies may be less relevant in the context of our study.

In summary, our study used the most current data from the literature to demonstrate a strong trial-level association between EFS and OS. The findings from this study, along with the results of previous studies, provide strong evidence that EFS is a good surrogate outcome for OS in the neoadjuvant ± adjuvant setting in gastric and GEJ adenocarcinoma. The demonstration of a trial-level association is considered the highest level of evidence (level 1) to establish surrogacy and is commonly used to illustrate the validation of proposed surrogate outcomes in the literature (9). Therefore, as the first surrogate outcome analysis between EFS and OS among patients with gastric or GEJ adenocarcinoma specifically, our study provides important and up-to-date evidence regarding the validity of EFS as a surrogate endpoint for OS in the neoadjuvant ± adjuvant setting among this patient population.

The results of this study should be interpreted within the context of specific limitations. First, as discussed previously, the definition of EFS varied across the included studies and could not be standardized in a literature-based meta-analysis. In addition, beyond the definition, it is challenging to standardize the assessment of EFS/DFS/PFS in the perioperative setting due to the variations in the follow-up intervals, type of follow-up imaging, and use of endoscopy etc. Therefore, the intermediate outcome, EFS, used in this analysis is a simplification of a complex assessment. Lack of consistency across EFS assessments may impact the correlation between EFS and OS and thus the results should be interpreted in the context of this limitation. However, as previously discussed, we included all definitions of EFS to preserve the sample size and reduce selection bias. While heterogeneity in EFS definition is a common limitation of surrogate analysis in oncology, it is reassuring that the conclusion of the main analysis is upheld in the sensitivity analyses and is also consistent with the prior meta-analyses (14, 15). IPD can enhance the analysis by estimating the treatment effects using standard approaches across the trials and thus should be considered in future studies, if feasible. Second, although the current study had a much larger sample size compared with the previous meta-analyses in similar indications, the sample size is still relatively small, which limited the precision of the estimates and led to a wide CI in some analyses. In addition, the small sample size limited the feasibility to conduct stratified analyses based on patient-level characteristics (e.g., age, tumor size/grade/location, or type of neoadjuvant regimen). However, the study included a sensitivity analysis restricting to the trials with both neoadjuvant and adjuvant chemotherapies in one of the randomized arms and the results were consistent with the main analysis. Finally, the RCTs included in this study all focused on neoadjuvant chemotherapies or chemoradiotherapies. As surrogacy may vary by treatment class, the association between EFS and OS may need to be reevaluated when data from clinical trials investigating treatments with different mechanisms of action are available.

In conclusion, this study synthesized the evidence from recent clinical trials in gastric or GEJ adenocarcinoma in the neoadjuvant ± adjuvant setting and provides important insight regarding the use of EFS as a surrogate endpoint for OS. The findings suggest that EFS can be considered a good surrogate endpoint for OS in gastric or GEJ adenocarcinoma in the neoadjuvant ± adjuvant settings. As more trials become available, additional research is needed to further evaluate factors that may impact the validity of EFS as a surrogate endpoint for OS in this patient population. Particularly, as innovative treatments like immunotherapy and targeted therapy in the perioperative setting are expected to substantially improve EFS and OS, it is important to evaluate the EFS and OS association specific to these treatments in the future.

Z.A. Wainberg reports personal fees from Merck, Amgen, AstraZeneca, Daiichi, Arcus, BMS, Lilly, Novartis, Ipsen, Roche, and Astellas outside the submitted work. J. Xie was an employee of Analysis Group, Inc. at the time of the study's conduct. A. Valderrama reports other support from Merck during the conduct of the study; other support from Merck outside the submitted work; and employment with Merck. L. Yin reports other support from Merck during the conduct of the study. S. Zhang reports other support from Analysis Group, Inc. during the conduct of the study; other support from Merck outside the submitted work. C.S. Shih reports other support from Merck and Co. during the conduct of the study; other support from Merck and Co. outside the submitted work. P. Bhagia reports personal fees from Merck during the conduct of the study; personal fees from Merck outside the submitted work. Q. Gu reports grants from Merck & Co., Inc. during the conduct of the study. K. Shitara reports grants and personal fees from Astellas Pharma, Ono Pharmaceutical, Daiichi Sankyo, Taiho Pharmaceutical, Merck Pharmaceutical, and Amgen; personal fees from Eli Lilly and Company, Bristol-Myers Squibb, Takeda Pharmaceuticals, Pfizer Inc., Novartis, AbbVie Inc., GlaxoSmithKline, Boehringer Ingelheim, Janssen, and Guardant Health Japan; and grants from Chugai Pharma, Medi Science, and Eisai outside the submitted work. Y.Y. Janjigian reports personal fees from Amerisource Bergen, Arcus Biosciences, AstraZeneca, Basilea Pharmaceutica, Bayer, Bristol-Myers Squibb, Daiichi-Sankyo, Eli Lilly, Geneos Therapeutics, GlaxoSmithKline, Imedex, Imugene, Lynx Health, Merck, Merck Serono, Mersana Therapeutics, Michael J. Hennessy Associates, Paradigm Medical Communications, PeerView Institute, Pfizer, Research to Practice, RGENIX, Seagen, Silverback Therapeutics, and Zymeworks Inc. outside the submitted work. J. Tabernero reports other support from Array Biopharma, AstraZeneca, Bayer, Boehringer Ingelheim, Chugai, Daiichi Sankyo, F. Hoffmann-La Roche Ltd, Genentech Inc, HalioDX SAS, Hutchison MediPharma International, Ikena Oncology, Inspirna Inc, IQVIA, Lilly, Menarini, Merck Serono, Merus, MSD, Mirati, Neophore, Novartis, Ona Therapeutics, Orion Biotechnology, Peptomyc, Pfizer, Pierre Fabre, Samsung Bioepis, Sanofi, Scandion Oncology, Scorpion Therapeutics, Seattle Genetics, Servier, Sotio Biotech, Taiho, Tessa Therapeutics and TheraMyc, Oniria Therapeutics, and Imedex, Medscape Education, MJH Life Sciences, PeerView Institute for Medical Education and Physicians Education Resource (PER) outside the submitted work.

Z.A. Wainberg: Conceptualization, methodology, writing–original draft, writing–review and editing. J. Xie: Conceptualization, formal analysis, validation, methodology, writing–original draft, writing–review and editing. A. Valderrama: Conceptualization, funding acquisition, validation, methodology, writing–original draft, writing–review and editing. L. Yin: Conceptualization, formal analysis, validation, methodology, writing–original draft, writing–review and editing. S. Zhang: Conceptualization, methodology, writing–original draft, writing–review and editing. C.S. Shih: Conceptualization, methodology, writing–original draft, writing–review and editing. P. Bhagia: Conceptualization, methodology, writing–original draft, writing–review and editing. Q. Gu: Conceptualization, formal analysis, validation, methodology, writing–original draft, writing–review and editing. K. Shitara: Conceptualization, methodology, writing–original draft, writing–review and editing. Y.Y. Janjigian: Conceptualization, methodology, writing–original draft, writing–review and editing. J. Tabernero: Conceptualization, methodology, writing–original draft, writing–review and editing.

This study was funded by Merck & Co., Inc.

Statistical analysis was performed by Fang Han and Zhaocheng Yi of Analysis Group, Inc. Medical writing assistance was provided by Gloria DeWalt, Shelley Batts, and Christine Tam, employees of Analysis Group, Inc. and funded by Merck & Co., Inc.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.