Re-examining Post-operative Chemoradiotherapy in Head and Neck Cancer: An Updated Long-Term Combined Analysis of RTOG 9501/EORTC 22931

Abstract

Background:

Post-operative chemoradiation (CRT) is generally recommended for patients with extranodal extension (ENE) and/or positive margins, but not for patients without these features, based on a post-hoc analysis of RTOG 9501 and EORTC 22931. However, this analysis lacked tests of interaction necessary to identify a predictive biomarker. In addition, updated data is now available.

Patients:

This study assessed 744 patients enrolled on RTOG 9501 and EORTC 22931, randomized trials comparing CRT to RT following surgery. Overall survival (OS) was analyzed with Cox regression. Cancer-specific mortality (CSM), other-cause mortality (OCM), and recurrence outcomes were analyzed with competing risk methodology. Tests of interaction assessed for differential benefits of CRT in various subgroups.

Results:

Median follow-up was 6.9 years. Among all patients, CRT improved OS (HR=0.81, 95% CI: 0.68–0.97, P=0.026). Although CRT improved OS in the subgroup with ENE and/or positive margins (HR=0.71, 95% CI: 0.57–0.89, P=0.003) and not in those without these features (HR=0.94, 95% CI: 0.68–1.30, P=0.7), tests of interaction showed no evidence of a differential effect of CRT in these subgroups (P-interaction=0.17). There was also no evidence of interaction when analyzing other outcomes, or when assessing ENE and margin status individually. While CRT significantly reduced CSM (HR=0.68, 95% CI: 0.55–0.83, P<0.001), it also significantly increased OCM (HR=1.51, 95% CI: 1.07–2.12, P =0.018). PO-CRT improved locoregional recurrence (HR=0.64, 95% CI: 0.48–0.85, P=0.002), but not distant metastasis (HR=0.83, 95% CI=0.64–1.08, P=0.17).

Conclusions:

Concurrent chemotherapy improved OS in HNC patients undergoing post-operative radiotherapy in the combined populations of EORTC 22931 and RTOG 9501. ENE and/or positive margins are not predictive biomarkers, and patients without these features may still benefit from CRT. CRT improved CSM, but this was partly offset by higher OCM. Refining the population most likely to benefit from post-operative CRT, taking into consideration both oncologic and patient-related factors, needs further exploration.

INTRODUCTION:

Two landmark randomized controlled trials, RTOG 9501 and EORTC 22931, both showed significantly improved locoregional control with post-operative chemoradiation (CRT) versus radiation (RT) alone following surgery for patients with head and neck cancer.^1–3 Although both trials met their primary endpoint (progression-free survival for EORTC 22931, locoregional control for RTOG 9501), there were discrepant results with respect to overall survival, which was superior with CRT in EORTC 22931,¹ but not in RTOG 9501.^2–3 To reconcile this, an unplanned combined analysis was performed.⁴ This found that the subgroup of patients with positive margins and/or extranodal extension (ENE), which were shared eligibility criteria across both trials, had improved survival with CRT. By contrast, there was no survival benefit for patients without these “high-risk” features. This finding profoundly changed the standard of care internationally. Post-operative CRT for patients with either margin-positivity or ENE is considered a level 1 recommendation by National Comprehensive Cancer Network guidelines, and these risk factors have guided both routine clinical practice and clinical trial design over the past two decades.

However, the identification of a factor that stratifies patients into subgroups with and without a statistically significant benefit from an intervention is not sufficient for defining a predictive biomarker.^5–7 There can still be substantial overlap of the 95% confidence intervals in this situation, especially if the subgroups have different power to detect effects due to disparate sample sizes. To define a predictive biomarker that guides treatment decisions, the relative treatment effect observed with an intervention must be statistically different in the two subgroups as assessed via tests of interaction.^5–7 Examples of validated predictive biomarkers in cancer include estrogen receptor (ER) positivity for ER-antagonists in breast cancer,⁸ epidermal growth factor receptor (EGFR) mutations for EGFR inhibitors in non-small cell lung cancer,⁹ and low-volume metastatic disease burden for local radiotherapy in metastatic prostate cancer.¹⁰ In addition, other plausible prognostic factors, such as stage, perineural invasion, and lymphovascular invasion, were not analyzed and compared to ENE and margin status in the original combined analysis.

Given the critical importance of these trials, and that longer follow-up data is now available for RTOG 9501, it is of interest to revisit the combined analysis with this context. We analyzed the impact of post-operative CRT on survival in these trials, and assessed whether positive margins, ENE, or any other factors can be qualified as predictive biomarkers. We also analyzed the effects of post-operative CRT on patterns of recurrence and both cancer-specific and other-cause mortality using competing risks methodology.

METHODS:

Data Acquisition

This study is a combined analysis of data originally collected in NRG Oncology’s RTOG 9501 (https://clinicaltrials.gov/study/NCT00002670) and EORTC 22931 (https://clinicaltrials.gov/study/NCT00002555). Individual patient data for these trials were obtained through standard data sharing agreements with NRG Oncology and the EORTC.

Included Trials

Study designs for RTOG 9501 and EORTC 22931 are described in detail elsewhere.^1,2 Briefly, both trials randomized patients with oral cavity, larynx, hypopharynx, and oropharynx cancers to post-operative radiotherapy with or without concomitant cisplatin (100mg/m² for 3 cycles) following surgical resection. RTOG 9501 enrolled patients with ENE, positive margins, or ≥2 positive lymph nodes with a primary endpoint of locoregional control, whereas EORTC 22931 enrolled patients with ENE, positive margins, perineural involvement, lymphovascular invasion (LVI), pN2-3, oral cavity and oropharyngeal tumors with nodal level IV/V involvement, or pT3-4 (except pT3N0 larynx) and had a primary endpoint of disease-free survival.

Statistical Analysis

Baseline patient characteristics were compared using the Wilcoxon rank-sum test for continuous variables and Pearson’s Chi-squared test for categorical variables. Overall survival (OS) was estimated using the Kaplan-Meier method, and OS curves were compared using the log-rank test. The cumulative incidence function accounting for competing risks was used to estimate the cumulative incidence curves for cancer-specific mortality (CSM), other-cause mortality (OCM), locoregional recurrence (LRR), and distant metastasis (DM), with comparisons made using Gray’s k-sample test.¹¹ Univariate and multivariable analyses were conducted using Cox proportional hazards models for OS. The proportional hazards assumption for the Cox models was assessed visually using the scaled Schoenfeld residuals and confirmed with the goodness-of-fit test.¹² The Fine-Gray subdistribution hazard model was employed to estimate factors associated with CSM, OCM, LRR, and DM accounting for competing risks (OCM for CSM, CSM for OCM, and any mortality for LRR and DM). Patients with unknown covariates were excluded from analyses requiring those covariates. Sensitivity analyses were performed where models were stratified by trial of enrollment (RTOG 9501 versus EORTC 22931) and where all patients were censored after 5 years of follow-up.

We conducted tests of interaction between treatment arm (RT vs. CRT) and covariates of interest to explore potential effect modifications. Interaction terms were created by multiplying the treatment variable with each covariate. These interaction terms were then included in the Cox proportional hazards models for OS, as well as Fine-Gray models for CSM, OCM, LRR, and DM. To determine whether the effect of treatment varied significantly across these covariates, we evaluated the statistical significance of the interaction terms using the likelihood ratio test comparing models with the interaction terms to those without. Additionally, we estimated the subgroup effects of RT versus CRT across specified covariates. These results were graphically presented in a Forrest plot to illustrate the magnitude and direction of the treatment effects within each subgroup.

All statistical analyses were performed using R, version 4.4.0. Two-sided tests were conducted, and a p-value of less than 0.05 was considered statistically significant for most analyses. To account for multiple hypothesis testing among the various interactions investigated, a Bonferroni correction was applied for these interaction analyses (p < 0.006 = 0.05/9). All data was analyzed at Cedars-Sinai Medical Center and NRG Oncology.

RESULTS:

744 patients were analyzed, including 334 patients from EORTC 22931 and 410 patients from RTOG 9501. Of note, 680 (91%) patients met eligibility criteria for both trials, including all 410 patients from RTOG 9501 and 270/334 (81%) of patients from EORTC 22931. There were no significant differences in baseline characteristics of patients randomized to post-operative CRT versus RT in terms of demographic or clinical characteristics (Supplementary Table 1). Median age was 55, with only 26 patients (3.5%) older than 70. Primary subsites were oropharynx in 37%, oral cavity in 27%, larynx in 22%, and hypopharynx in 15%. 55% of patients had extranodal extension, and 18% had positive margins. RTOG 9501 had slightly older patients (P=0.005), more oropharynx cancers (P<0.001), more patients with N2 disease (P<0.001), and more perineural invasion (P<0.001), whereas EORTC 22931 had higher rates of margin-positivity (P<0.001) (Supplementary Table 2).

Median follow-up was 6.9 years (EORTC=5.0 years, RTOG=10.0 years). There were 174 mortality (52.1% of patients), 116 CSM (34.7%), 58 OCM (17.4%), 83 LRR (24.8%), and 77 DM (23.1%) events in EORTC-22931 during follow-up, compared to 289 (70.4%), 221 (53.4%), 68 (16.6%), 103 (25.1%), and 132 (32.2%) events in these respective categories in RTOG 9501. In the combined analysis across all patients (Table 1), CRT improved OS (hazard ratio (HR): 0.81, 95% confidence interval (CI): 0.68–0.97, P = 0.026), with an estimated 7-year OS of 41% vs. 33% for patients undergoing CRT vs. RT, respectively (Figure 1A). There was no evidence of non-proportional hazards over time. OS remained significantly better in patients treated with post-operative CRT versus RT alone when adjusting for a variety of prognostic factors in multivariable analysis (Table 1; HR=0.77, 95% CI = 0.63–0.93, P=0.008).

Table 1.

Univariate and multivariable Cox regression for overall survival among patients enrolled on RTOG 9501 and EORTC 22931. CI, confidence interval; ENE, extranodal extension; HR, hazard ratio; LVI, lymphovascular invasion; PNI, perineural invasion; ref, reference.

Characteristic	Univariate		Multivariable
	HR (95% CI)	p-value	HR	p-value
Treatment
Radiotherapy	1.00 (ref)	—	1.00 (ref)	—
Chemoradiation	0.81 (0.68–0.97)	0.026	0.77 (0.63–0.93)	0.008
Age	1.01 (0.99–1.02)	0.31	1.00 (0.99–1.01)	0.795
Sex
Male	1 (ref)	—	1.00 (ref)	—
Female	0.90 (0.66–1.22)	0.49	1.01 (0.72–1.40)	0.966
Anatomic Site
Oral Cavity	1.00 (ref)	—	1.00 (ref)	—
Larynx	0.78 (0.60–1.02)	0.07	0.72 (0.55–0.96)	0.025
Hypopharynx	1.11 (0.84–1.47)	0.47	0.88 (0.64–1.21)	0.434
Oropharynx	0.64 (0.51–0.81)	<0.001	0.64 (0.50–0.83)	<0.001
T-classification
T1	1.00 (ref)	—	1.00 (ref)	—
T2	1.82 (1.23–2.68)	0.003	1.98 (1.30–3.01)	0.002
T3	2.11 (1.44–3.10)	<0.001	2.35 (1.54–3.59)	<0.001
T4a	2.31 (1.59–3.35)	<0.001	2.48 (1.64–3.75)	<0.001
N-classification
N0	1.00 (ref)	—	1.00 (ref)	—
N1	1.02 (0.66–1.55)	0.94	0.84 (0.54–1.31)	0.437
N2	1.29 (0.94–1.76)	0.11	1.13 (0.78–1.63)	0.509
N3	1.59 (0.96–2.66)	0.07	1.18 (0.66–2.12)	0.569
ENE
Negative	1.00 (ref)	—	1.00 (ref)	—
Positive	1.53 (1.27–1.85)	<0.001	1.66 (1.35–2.06)	<0.001
Margins
Negative	1.00 (ref)	—	1.00 (ref)	—
Positive	1.07 (0.84–1.35)	0.58	1.29 (1.00–1.67)	0.054
Grade
Low	1.00 (ref)	—	1.00 (ref)	—
Intermediate	1.06 (0.84–1.34)	0.63	0.90 (0.68–1.18)	0.434
High	0.91 (0.70–1.19)	0.48	0.79 (0.58–1.08)	0.146
LVI
Negative	1.00 (ref)	—	1.00 (ref)	—
Positive	1.28 (1.20–1.61)	0.03	0.96 (0.75–1.23)	0.749
PNI
Negative	1.00 (ref)	—	1.00 (ref)	—
Positive	1.54 (1.23–1.91)	<0.001	1.27 (1.00–1.62)	0.051
Study
EORTC 22931	1.00 (ref)	—	1.00 (ref)	—
RTOG 9501	1.16 (0.96–1.41)	0.13	1.31 (1.02–1.68)	0.038

Figure 1.

Overall survival with post-operative chemoradiation versus radiation alone (RT) in EORTC 22931 and RTOG 9501 among A) all patients enrolled, B) patients with positive margins and/or extranodal extension, and C) patients with negative margins and no extranodal extension.

There was improved overall survival with CRT in the subgroup of 475 patients with positive margins and/or ENE (HR = 0.71, 95% CI 0.57–0.89, P=0.003), but not in the 267 patients without these factors (HR = 0.94, 95% CI: 0.68–1.30, P=0.7) (Figure 1B–C). However, there was significant overlap in the 95% confidence intervals for these subgroups (Figure 2), and the presence of margin positivity and/or ENE did not predict for greater relative OS benefit from CRT using tests of interaction (P-interaction=0.17). Additionally, neither margin-positivity (P-interaction=0.25) nor ENE (P-interaction = 0.46) predicted for greater relative benefit from CRT vs RT alone when analyzed as individual factors. We also found no significant interaction when comparing the impact of CRT versus RT on any other oncologic outcome (CSM, LRR, DM) based on ENE or margin-positivity.

Figure 2.

Forrest plot of overall survival for patients randomized to chemoradiation versus radiation alone in various subgroups. P-values given the p-interaction from tests of interaction. With adjustment for multiple hypothesis testing, P-interaction < 0.006 is considered significant. Dots represent the hazard ratio (HR) for chemoradiation versus radiation in a given subgroup, and the bracketed line represents the 95% confidence interval. ENE, extranodal extension; LVI, lymphovascular invasion. PNI, perineural invasion.

Several sensitivity analyses were performed. Given that the follow-up for RTOG 9501 was double that of EORTC 22931, an analysis was performed censoring all patients beyond 5-years of follow-up (Supplementary Figure 1A, Supplementary Table 3). Additionally, because survival differed based on which trial patients enrolled on based in multivariable analysis, we also analyzed multivariable models stratified by trial (Supplementary Figure 1B, Supplementary Table 3). Both of these sensitivity analyses showed results consistent with the main analysis.

Among the tested factors, the only variable associated with increased relative survival benefit from CRT after correcting for multiple hypothesis testing was lack of LVI (P-interaction=0.002). For patients that were LVI-negative, CRT was associated with improved survival (HR=0.56, 95% CI: 0.56–0.85, P<0.001), whereas LVI-positive patients had a trend towards worse survival (HR=1.45, 95% CI: 0.96–2.20, P=0.08). The effect was more pronounced in the EORTC trial versus the RTOG trial (Supplementary Figure 2).

There were opposite associations of CRT with CSM versus OCM (Supplementary Tables 4–5). Patients receiving CRT had markedly lower CSM (HR: 0.68, 95% CI: 0.55–0.83, P<0.001), but they had significantly higher OCM (HR: 1.51, 95% CI: 1.07–2.12, P=0.018). Using competing risk methodology, the 7-year cumulative incidence of CSM was 37% vs 53% for patients receiving post-operative CRT vs RT alone, respectively, compared to 7-year OCM of 21% vs. 14% (Figure 3). These results were unchanged when adjusting for known prognostic factors in multivariable analysis (Supplementary Tables 4–5). In subgroup analysis (Supplementary Figures 3–4), there were trends towards CRT being associated with a greater improvement in CSM in margin positive patients (P-interaction = 0.04) and higher OCM in patients aged 60 or older (P-interaction = 0.02), but these were not significant after adjusting for multiple hypothesis testing.

Figure 3.

Cumulative incidence of cancer-specific mortality (CSM) and other-cause mortality (OCM) with post-operative chemoradiation versus radiation alone among all patients enrolled on RTOG 9501 and EORTC 22931 using competing risks analysis.

With respect to patterns of recurrence, CRT decreased LRR (HR: 0.64, 95% CI: 0.48–0.85, P=0.002), but not DM (HR=0.83, 95% CI: 0.64–1.08, P=0.17) versus RT alone. 7-year cumulative incidence of LRR was 20% vs. 31% (P=0.003), and 7-year cumulative incidence of DM was 26% vs. 31% (P=0.2) for patients treated with CRT vs. RT, respectively (Supplementary Figure 5). In multivariable analysis (Supplementary Table 6), the strongest pathologic factors predicting increased risk of LRR were T-classification (T2: HR 2.18, 95% CI: 1.07–4.46, P=0.03; T3: HR 2.13, 95% CI: 1.03–4.41, P=0.04; T4a: HR 2.41, 95% CI 1.17–4.95, P=0.02), ENE (HR 1.46, 95% CI: 1.05–2.01, P=0.02), and PNI (HR 1.42, 95% CI 1.00–2.01, P=0.05). Margin positivity was not associated with increased LRR (HR=1.23, 95% CI 0.85–1.79, P=0.27)

When comparing survival and recurrence outcomes across the two trials (Figure 4), the impact of CRT was consistent, with no significant differences observed in the effect of CRT on OS, CSM, OCM, LRR, and DM (P-interaction > 0.1 for all).

Figure 4.

Comparisons of outcomes with post-operative chemoradiation versus radiation alone in RTOG 9501 and EORTC 22931. P-values represent tests of interaction of the association of chemoradiation with outcome across both trials. CSM, cancer-specific mortality; DM, distant metastasis; LRR, locoregional recurrence; HR, hazard ratio; OCM, other-cause mortality; OS, overall survival.

DISCUSSION:

In this updated combined analysis of RTOG 9501 and EORTC 22931, we found that in the total population of the two trials, post-operative chemoradiation improved overall survival in comparison to radiation alone. The 19% relative decrease in overall mortality observed was similar to the 17% decrease in mortality observed with concomitant chemotherapy in the updated MACH-NC meta-analysis among 19,805 patients across 107 randomized trials, mostly in the non-operative setting.¹³ This corresponded to an 8% absolute difference in OS at 7 years in our study. The survival benefit from chemoradiation was attributable to improved LRR, without an observed impact on DM. The influence of CRT on all outcomes was fairly similar in both EORTC 22931 and RTOG 9501, with no heterogeneity observed across the trials. This, along with the fact that both trials administered nearly identical treatments to similar populations, with 91% of patients meeting eligibility criteria for both trials, adds strength to the validity of the combined results.

We found no statistical evidence that margin-negativity, lack of ENE, or both factors combined were sufficiently robust predictive biomarkers to identify patients undergoing post-operative radiation who would not benefit from concomitant chemotherapy. Although it is true that patients with positive margins or extranodal extension had statistically improved survival with CRT, while those without these features had no significant benefit, this is not sufficient to identify a predictive biomarker.^5–7 True predictive biomarkers require statistical tests of interaction that demonstrate a differential relative benefit of a treatment in two subgroups.^5–7 In the case of patients in RTOG 9501/EORTC 22931 with and without ENE or margin-positivity, there was substantial overlap in the 95% confidence intervals of CRT’s effect on survival, leading to negative tests of interaction. Additionally, there was no statistical evidence of a differential effect of CRT in these two subgroups for any other outcome. In fact, we identified several other pathologic factors outside of margins and ENE that had similar or stronger differences in survival among subgroups undergoing CRT versus RT, but none were sufficiently robust for clinical use when accounting for multiple hypothesis testing and bioplausibility. As a result, we believe the most parsimonious and accurate interpretation of the result of these trials in the total combined population is that concomitant chemotherapy improved locoregional control and survival irrespective of ENE or margin status. This is consistent with how subgroup analyses have been interpreted in more modern NRG/RTOG head and neck cancer trials. For example, a subgroup analysis of RTOG 1016 showed a large difference in the relative effect of cisplatin versus cetuximab-based radiation on survival for patients with Zubrod performance score 0 (HR=1.08, one-sided 95% upper CI 1.55) versus score 1 (HR=2.66, one-sided 95% upper CI 4.32).¹⁴ However, a test of interaction was non-significant after adjusting for multiple hypothesis testing, and the trial results have been universally interpreted as demonstrating superior efficacy of cisplatin independent of Zubrod score.

Applying these results from trials initiated approximately three decades ago to modern practice is challenging. Given advances in treatment, imaging, and pathologic assessment, stage-for-stage outcomes may be improved in modern practice. However, we note that patients without ENE or positive margins treated with radiation alone in RTOG 9501/EORTC 22931 had nearly identical 5-year locoregional recurrence rates to patients undergoing post-operative RT in RTOG 0920,¹⁵ a more modern trial excluding patients with positive margins and ENE (23% versus 25%, respectively). RTOG 0920 also showed a disease-free survival benefit in patients without ENE or positive margins with the addition of cetuximab.¹⁵ A testable hypothesis is whether the RTOG 0920 population might benefit even more from cisplatin, a more effective agent.

We further note that most validated true predictive biomarkers in oncology have a biologic underpinning that allows them to dichotomize patients into groups that do or do not benefit from treatment (e.g. ER-positivity for ER antagonists in breast cancer, EGFR-mutations for EGFR-inhibitors in non-small cell lung cancer). It is more unlikely that general pathologic features, such as ENE or margin positivity, would be capable of predicting whether a cytotoxic chemotherapy would or would not provide a relative decrease in LRR via radiosensitization. While it is true that prognostic pathologic factors that increase the absolute risk of LRR would be expected to increase the absolute benefit of systemic therapy, ENE and margin positivity do not represent the only pathologic factors that predict for increased recurrence. In fact, T-classification and PNI had similar or stronger impact on LRR to ENE and margin-status in this study. Thus, there are almost certainly patients without ENE and positive margins undergoing post-operative radiation that would benefit from concomitant chemotherapy. It is reasonable to consider concomitant post-operative chemoradiation in healthy patients with other established adverse pathology features putting them at high risk of LRR beyond ENE and positive margins (large volume disease, high number of involved lymph nodes, extensive PNI, very close or unclear margins, etc.). Ultimately, new tools for personalization of treatment intensity are needed. Novel molecular biomarkers, such as the radiosensitivity index,^16–19 post-operative circulating tumor DNA,²⁰ or artificial intelligence-guided predictive models,²¹ are unvalidated but potentially promising tools that may complement traditional pathologic factors to better identify which patients benefit from increased adjuvant treatment intensity. The many advances in personalized biomarker development could provide a basis for design of new trials to optimize post-operative treatment in head and neck cancer.

An unexpected finding of our analysis was that lack of LVI was correlated with increased benefit from chemoradiation (P-interaction=0.002), even after correcting for multiple hypothesis testing. Although this could be explored in future studies, we think this is unlikely to represent a reproducible biomarker for multiple reasons. First, there is no biologically plausible reason why lack of LVI should increase benefit from CRT. If anything, it would seem more likely that patients with LVI, as a potential marker of more aggressive biology, would benefit from more aggressive treatment, not less. Further, the difference in CRT’s effect on survival between LVI-positive and LVI-negative patients was essentially entirely driven by EORTC 22931 and was not clearly recapitulated in RTOG 9501. Last, the disparate outcomes with CRT versus RT in LVI-positive and LVI-negative patients were driven in large part by a more than tripling in OCM observed in LVI-positive patients undergoing CRT versus RT (Supplementary Figure 4). While it is at least conceivable that LVI could impact the oncologic impact of CRT, it would not influence OCM from this treatment. Thus, we believe that the interaction between LVI and CRT is most likely due to random chance, and represents a further example of the caution needed when interpreting subgroup analyses.

Another novel result of this analysis is that although CRT dramatically reduced mortality from head and neck cancer (absolute reduction of 16% at 7-years), approximately half of the absolute survival benefit was lost due to increased non-cancer mortality (absolute increase of 7% at 7-years). This finding should be interpreted with a degree of caution. At least some of the difference in other-cause mortality could be a result of patients treated with CRT avoiding head and neck cancer death, making them more likely to succumb to other causes of death over time. Nevertheless, it should be noted that these trials enrolled a relatively young, healthy cohort, with median age of 55 and only 3.5% of patients older than 70. Further, separation of the other-cause mortality curves for CRT and RT emerged within about 3 years. The reason for a possible increase in non-cancer mortality with CRT is unclear. Non-cancer mortality has been infrequently reported in head and neck cancer clinical trials and is poorly characterized following post-operative treatment. However, the results of this analysis are consistent with RTOG 91-11, where patients with locally advanced larynx cancer randomized to concurrent cisplatin and radiation had substantially higher non-cancer mortality than patients receiving radiation without concomitant cisplatin despite no obvious increase in recorded late toxicity.²² It is possible that aspiration or other unrecorded medical conditions are worse for patients undergoing CRT during long-term follow-up, but this is speculative. Whatever the cause, it highlights the fact that beyond cancer-associated factors, patient-related health factors are critical to optimize the benefit of post-operative CRT. For example, CRT appeared to potentially have a trend towards greater impact on other-cause mortality for patients aged 60 and older, nearly tripling other-cause mortality in these patients, versus younger patients that had no significant change in this outcome (P-interaction=0.02). This is consistent with a large-scale analysis of NRG/RTOG definitive chemoradiation trials that showed decreased cause-specific survival benefits for patients of age 70 years or more.²³ It has been proposed that the ratio of the likelihood of cancer-specific mortality to other-cause mortality, known as the ω-ratio, could be a tool for selection of treatment intensification.^24,25 This concept has been validated within the clinical trials comprising the MACH-NC/MARCH datasets.²⁶ On the other hand, we also note that it is possible that CRT’s increase in non-cancer mortality could be mitigated in modern practice through advances in radiation technology, improved supportive medications during chemotherapy, and increased recognition of the importance of ancillary services like dietetic and swallowing therapy for head and neck cancer survivors. Identifying patients at higher risk of increased other-cause mortality from post-operative chemoradiation is critical to optimize the risks and benefits of this treatment.

This study has some limitations. First, we note that this is an unplanned combined analysis of these trials. There were differences in median follow-up time (10.0 years for RTOG 9501 versus 5.0 years for EORTC 22931), so RTOG 9501 was the dominant contributor to results at later time points. We addressed this by performing two sensitivity analyses. The first included Cox models stratified by trial of enrollment, whereas the second censored all patients at 5-years of follow-up. Both sensitivity analyses were nearly identical to the results of the entire cohort. Additionally, the sample size needed to identify a differential effect of a treatment in two subgroups is substantially higher that the sample size needed to identify an effect in the overall population.²⁷ Thus, despite pooling data from two separate trials, this study could be underpowered for assessing predictive biomarkers that determine benefit from CRT. However, this lack of power is a central point of this analysis. These trials do not have sufficient power to support guidelines limiting chemoradiation only to those with positive margins or ENE. While it is true this study cannot definitively prove that there is no differential effect of chemoradiation in patients with or without certain risk factors following surgery, there is no evidence that lack of ENE and margin-positivity are more promising biomarker candidates than other clinicopathologic factors. We also note certain variables that could influence outcomes were not available. For instance, HPV information was lacking. Of note, the prevalence of HPV-associated oropharyngeal cancer was much lower during accrual to RTOG 9501 and EORTC 22931 versus today. RTOG 9003 and EORTC 24971 (TAX 323), which enrolled patients during a similar time period, had 40% and 16% rates of p16-positivity in oropharyngeal cancer patients, respectively.^28,29 Given the randomized nature of treatment assignment, the relatively low-prevalence of HPV-positivity during this era, and the fact that 63% of the cohort had non-oropharyngeal cancers, it is unlikely that differences in HPV-positivity substantially impacted the conclusions of this study. Another limitation is the lack of data regarding comorbidity, causes of non-cancer mortality, body mass index, and smoking status, limiting our ability to further analyze the non-cancer mortality associated with CRT. However, given that this is an individual patient data-level analysis of two trials randomizing patients to the intervention of interest in nearly identical settings, we do not believe any of these factors undermine the fundamental conclusions of this study.

In summary, concomitant chemotherapy improved overall survival for head and neck cancer patients undergoing post-operative radiation in the combined total population of EORTC 22931 and RTOG 9501 through reduced locoregional recurrence. Lack of ENE and margin negativity did not clearly identify patients undergoing post-operative radiotherapy who would not benefit from concomitant chemotherapy. Concomitant chemotherapy can also potentially increase non-cancer mortality, and this effect may be more pronounced in older and frailer patients. Thus, perhaps the best interpretation of this updated analysis is that post-operative CRT should be considered for healthy, younger patients undergoing post-operative radiation therapy with pathologic features that put them at high risk for locoregional recurrence, including, but importantly not limited to, patients with ENE and positive margins. Careful consideration should be given to the overall health and performance status of the patient given the increase in other-cause mortality with CRT that blunts its benefit even in this relatively young, healthy population. Equally important, improved algorithms for optimization and personalization of post-operative treatment intensity are needed. Additionally, research is needed to optimize treatment strategies in older or more frail populations, where the benefit of treatment intensification could be less or non-existent. This study illustrates how, based on the foundation of RTOG 9501 and EORTC 22931, there is continuing opportunity to gain greater understanding of how to improve outcomes of post-operative head and neck cancer treatment.

HIGHLIGHTS.

• In a combined updated analysis of RTOG 9501 and EORTC 22931, post-operative chemoradiation (CRT) improved overall survival.

• Patients with extranodal extension and/or positive margins had similar benefit from CRT to those without these features.

• CRT improved cancer-specific survival, but also significantly increased non-cancer mortality.

• Post-operative CRT should not be limited exclusively to those with extranodal extension or positive margins.