With nearly 750,000 new cases and 365,000 deaths each year, head and neck cancer is the eighth most common cancer worldwide (1). The locally advanced (LA) form of head and neck squamous cell carcinoma (HNSCC), defined as stages III-IVb based on the seventh edition of the American Joint Committee on Cancer (AJCC) staging system (2), is generally managed through various combinations of surgery, radiotherapy (RT), and systemic therapy (3). Specifically, chemoradiation therapy (CRT) that is concurrently administered, was shown to have a greater benefit than RT sequentially administered before or following chemotherapy in the meta-analysis of chemotherapy in head and neck cancer (MACH-NC) study (4–7). This, along with de-escalation RT strategies to lower toxicity rates, has led to improved survival in the subgroup of patients with human papillomavirus (HPV)-positive oropharyngeal squamous cell carcinoma (OPSCC); however, 5-year survival rates with CRT approaches still do not exceed 60% in the overall LA-HNSCC population, highlighting the unmet need to identify new treatment strategies that can increase the efficacy of CRT (8–10).

Overall survival (OS) is considered the most reliable outcome in oncology clinical trials and is one of the most valuable patient-centered outcomes that oncology treatments aim to improve (11, 12). The disadvantage of the OS outcome (with death as its only endpoint) in clinical trials is that it requires a large number of patients and an extended follow-up period to detect statistically significant differences, especially in earlier stages of cancer (13). In contrast, early time-to-event outcomes (e.g. event-free survival [EFS], progression-free survival [PFS], disease-free survival [DFS], and recurrence free survival [RFS]) additionally include disease progression in their definition, which occurs more frequently than death, especially in earlier cancer stages. As a result, these early time-to-event outcomes mature faster than OS, allowing for smaller sample sizes, shorter follow-up durations and earlier readouts for trials evaluating novel cancer therapies. However, for adoption of these alternate endpoints as primary trial endpoint, strong evidence and careful examination of the relationship between these early time-to-event outcomes and OS must be undertaken to establish a true surrogate relationship. If these endpoints can be shown to be surrogates for OS, more timely evaluations of the benefit-to-risk profile of experimental treatments and consequently earlier patient access to novel therapies can be achieved (14, 15).

A comprehensive examination of EFS and duration of locoregional control as potential surrogates for OS in patients with LA-HNSCC has previously been conducted in a 2009 study by Michiels et al. (16) In that study, individual patient-level data from 104 trials from four meta-analyses on hyperfractionated or accelerated RT and concomitant, induction, or adjuvant chemotherapy were analyzed at an individual level as well as at the trial level. The study concluded that OS was strongly correlated with EFS both at the individual level (range of correlation coefficient [R]: 0.82-0.90) and at the trial level (R=0.98 for RT, and range of 0.79-0.93 for CRT), and that at both levels, EFS was a stronger surrogate for OS compared to duration of locoregional control for patients receiving CRT. These results have broadly established the relevance and suitability of EFS as a surrogate for OS to assess the treatment effect of CRT in randomized controlled trials conducted in patients with LA-HNSCC who received locoregional treatments, including surgical interventions (16).

Targeted immunotherapies, such as inhibitors of programmed cell death protein 1, are being increasingly investigated in LA-HNSCC. Trials of these novel interventions, many of them still ongoing (17–21), are heterogeneous in terms of study design, population characteristics and primary endpoints. These trials were not captured in Michiels et al. (16), as the studies included in that analysis were published in or before 2000 (16). Furthermore, trial populations included in Michiels et al. (16) were a mix of surgery-eligible (e.g., receiving curative surgical intervention as part of their locoregional treatment) and surgery-ineligible patients. Given many novel treatments are evaluated in trials exclusively conducted in surgery-ineligible patients including time-to-event primary endpoints other than OS, it is of interest to further explore the association between EFS and OS in this population.

The current study aimed to assess the correlation between trial-reported OS and EFS using published clinical trial data among newly diagnosed patients with LA-HNSCC, with a special focus on trials evaluating definitive CRT without surgical intervention. Although patients with resectable tumors may still receive definitive CRT instead of surgery if they are suitable for organ preservation, many patients who receive definitive CRT are indeed ineligible for curative surgical interventions. For the purpose of completeness, CRT was defined as the combination of RT and any class of systemic therapies, including chemotherapy, targeted therapy or a combination of both. To minimize risk of bias, trials were identified through a comprehensive systematic literature review (SLR) using pre-defined criteria.

Curative therapeutic options are limited for surgery-ineligible patients with LA-HNSCC, highlighting an unmet need for novel interventions for the treatment of this population. Identifying early clinical outcomes that can be surrogates for OS is key to lowering the cost and duration of trials and, thereby, facilitating patient access to those novel treatments and improving clinical outcomes at a faster pace. The advantage of early time-to-event outcomes such as EFS or PFS in randomized controlled trials is that they take less time to mature relative to OS, thereby providing an earlier indication of likely treatment benefit(68–70). Additionally, they are not as affected by post-progression treatments, which can affect estimates of treatment effects on OS and confound the results (70, 71). For example, salvage surgery for persistent/recurrent disease (e.g. laryngectomy) can affect post-progression survival, possibly diluting the treatment effect on OS, with high complications rates and morbidity (10, 72, 73). These post-progression treatments are not currently included in the definition of time-to-event outcomes such as EFS or DFS in oncology trials (12), including the studies identified in our evidence base. Based on all the above, many phase III oncology studies have therefore adopted these alternate endpoints as the primary endpoint of the trial (11). However, it remains an ongoing debate as to whether there is strong evidence to suggest that these early time-to-event outcomes can be treated as surrogates for OS. Careful examination of the relationship between the early and final endpoints must be undertaken to establish a true surrogate relationship.

Results from the Michiels et al. study have broadly established the relevance and suitability of EFS as a surrogate for OS to assess the treatment effect of RT and chemotherapy in randomized controlled trials conducted in patients with LA-HNSCC (16). Our analysis followed the approach taken by that study, while complementing it by including additional trials and focusing on the subgroup patients receiving definitive CRT. The current study incorporated a comprehensive SLR conducted as early as May 2021. As such, compared to Michiels et al. (16), our evidence base additionally included trials published after 2000, which evaluated novel treatments such as immunotherapies and other targeted therapies used as part of definitive CRT in LA-HNSCC patients who do not receive surgery. In anticipation that the included clinical trials generally did not start after 2017, we used the disease staging criteria presented in the 7^th edition of the AJCC manual (as opposed to the 8^th edition, which was implemented in January 2018) to define the LA disease. Therefore, none of the studies were, for example, excluded based on the new staging system presented in the 8^th edition for p16-positive OPSCC. Once the SLR was completed, all of the included studies were found to have a start date prior to 2018, as initially expected.

EFS in our study additionally included other endpoints such as PFS, DFS, and RFS, because these early time-to-event outcomes have been the primary endpoints of many trials conducted in patients who receive definitive CRT, including studies of immunotherapy and other novel treatments (17–21). While these varied based on event definition (e.g. death, disease progression, second primary cancer, and surgery), all of them included death and disease progression as common events, per the study eligibility criteria of the analysis. Due to a higher percentage of disease progression events in LA-HNSCC compared to surgical events or second primary cancer events, inclusion of other time-to-event outcomes in the EFS definition may be less impactful in determining correlation between EFS and OS.

Results of the current study, while limited to trial-level analyses and being conducted in a slightly different population, support the Michiels et al. (16) results. In the main analysis inclusive of all CRT trials, the correlation coefficient (95% CI) was 0.85 (0.72-0.93), suggesting a strong correlation between EFS and OS. Due to differences in study characteristics, the surrogacy relationship was further evaluated in subsets of trials with similar length of follow up, CRT timing, outcome definition, intervention class, and RT modality. The correlation coefficient in these scenarios always remained above 0.82, suggesting that a similar relationship existed in all subsets. Specifically, CRT timing did not influence the strength of the correlation between EFS and OS. As the number of studies decreased within the analyzed subsets of trials, the 95% CIs became much wider, indicating greater uncertainty. In addition to the consistently strong correlations (estimated R ranging between 0.81 and 0.94), the expected trial-level surrogacy relationships (an intercept sufficiently close to zero and a slope significantly different from zero) were met in all scenarios, with the exception of the model for trials that compared sequential CRT versus concurrent CRT. Lastly, a leave-one-out cross-validation strategy was taken, similar to Michiels et al. (16), in order to ensure the robustness and accuracy of the predictive models. In the leave-one-out analyses, predictions generally fell within the reported CIs.

While the OS-EFS correlation was analyzed in a number of trial subsets based on major study, intervention and outcome characteristics, further scenarios could be explored in the future. For example, it may be of interest to evaluate the OS-EFS correlation in patients with p16-positive OPSCC given new staging and treatment algorithms have been developed for this particular population over the last two decades (3, 33). Among the trials of CRT included in the current analysis, only five reported outcomes of interest in patients with p16-positive OPSCC, precluding conclusive results for the exploratory analysis in that subpopulation. Future research can potentially focus on evidence bases other than the one synthesized in the current study; for example, the OS-EFS correlation could be further explored in trials evaluating RT alone in p16-positive patients, particularly those using the de-escalation treatment strategy with lower toxicity rates (74). Such trials, however, were not directly related to the research questions of the current analysis and were not included in our SLR.

This analysis was subject to some limitations. Our study lacked individual patient-level data. A proper assessment of surrogacy requires that there is a relationship between the two outcomes at the patient-level as well as a relationship between treatment effects at the trial-level (75). Furthermore, patient-level data allows to account for differences in baseline patient characteristics between studies (e.g. age, gender, smoking status, HPV status, ECOG scores) that could influence the EFS and OS relationship. In the absence of individual patient-level data, we were only able to assess the trial-level surrogacy relationships using aggregate data. Furthermore, EFS and OS correlation may vary by the type of event. For instance, distant metastatic events may be more strongly correlated to OS compared to a locoregional event given disease prognosis in LA-HNSCC. This was shown in the Michiels et al. study, where EFS (inclusive of locoregional and distant metastatic events) was a stronger surrogate for OS compared to duration of locoregional control for patients receiving CRT. Due to lack of patient-level data we could not assess this aspect in our study. Finally, the majority of the included CRT studies were completed prior to approval of immunotherapies in recurrent and/or metastatic HNSCC (R/M HNSCC) (76–80). The improvement in OS from use of immunotherapies in R/M HNSCC, might influence the correlation between EFS and OS. Thus, it is important that future correlation research focusses on more recent LA-HNSCC trials (i.e. initiated or completed after approval of immunotherapies in R/M HNSCC) in which patients are eligible to receive immunotherapies after failure of CRT.

Despite some limitations, the trial-level analyses were conducted on a dataset that was obtained from a rigorous SLR, and the models employed were assessed for appropriateness. The cross-validation procedure showed that the correlation models were generally able to predict OS HRs with acceptable accuracy. In conclusion, EFS was strongly correlated with OS in this trial-level analysis. Future research using individual patient-level data can further investigate if EFS could be considered a suitable early clinical endpoint for evaluation of CRT regimens in LA-HNSCC patients receiving definitive CRT.

The original contributions presented in the study are included in the article/ Supplementary Material . Further inquiries can be directed to the corresponding author.

Conception/design was performed by SK, DA, AM, CMB, DC, KR, and NU. Collection and/or assembly of data was performed by SK, DA, and AM. Data analysis and interpretation were performed by SK, DA, DM, AM, CMB, DC, KR, and NU. Manuscript writing was performed by AM and KR. Final approval of manuscript was provided by SK, DA, AM, DM, CMB, DC, KR, and NU. All authors contributed to the article and approved the submitted version.

Funding for the study was provided by Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc., Kenilworth, NJ, USA.

KR is an employee of Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc., Kenilworth, NJ, USA. NU was employed by Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc., Kenilworth, NJ, USA, at the time the research was conducted. CB and DC are employees of Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc., Kenilworth, NJ, USA and stockholders of Merck & Co., Inc., Kenilworth, NJ, USA. SK, DA, AM, and DM are employees of PRECISIONheor, which received funding for this study.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.