TRIAL DESIGN AND TREATMENTS

Patients were randomly assigned in a 2:1 ratio to receive intravenous nivolumab (Opdivo, Bristol-Myers Squibb) or a standard, single-agent therapy of the investigator’s choice, with stratification according to receipt of previous cetuximab therapy (yes or no). Nivolumab was administered at a dose of 3 mg per kilogram of body weight every 2 weeks. Standard therapy consisted of weekly intravenous administration of methotrexate at a dose of 40 to 60 mg per square meter of body-surface area, docetaxel at a dose of 30 to 40 mg per square meter, or cetuximab at a dose of 250 mg per square meter after a loading dose of 400 mg per square meter.

END POINTS AND ASSESSMENTS

The primary end point was overall survival, which was defined as the time from randomization to the date of death from any cause. Secondary end points were progression-free survival (time from randomization to the date of disease progression or death) and the rate of objective response according to RECIST, version 1.1. Additional pre-specified end points included the time to response; associations between PD-L1 level and human papillomavirus (HPV) status and overall survival, progression-free survival, and response rate; safety; and quality-of-life assessments.

Tumor response was assessed by investigators according to RECIST, version 1.1, every 6 weeks beginning at week 9. Patients were treated until an unacceptable level of drug-related toxic effects occurred or until disease progression. However, nivolumab treatment could be continued beyond disease progression, as assessed clinically or radiographically, if the investigator assessed that it was providing clinical benefit. Patients were followed for overall survival every 3 months until death, loss to follow-up, or withdrawal of consent.

At each treatment visit and for 100 days after receipt of the last dose, acute toxic effects were evaluated according to the Common Terminology Criteria for Adverse Events, version 4.0. Adverse events with potential immunologic causes were classified as select adverse events. The criteria for a dose delay or the discontinuation of nivolumab or standard therapy because of treatment-related adverse events were specified in the protocol, available with the full text of this article at NEJM.org. Dose modifications were not permitted for nivolumab but were specified for methotrexate, docetaxel, and cetuximab on the basis of the type and grade of the toxic effect.

Patient-reported outcomes, including symptoms and health-related quality of life, were exploratory end points and were evaluated with the use of the European Organization for Research and Treatment of Cancer (EORTC) Quality of Life Questionnaire–Core 30 module (QLQ-C30) and the head-and-neck–specific module (QLQ-H&N35). Scores for these modules range from 0 to 100, with higher scores indicating better functioning or well-being or higher symptom burden, although scales measuring symptom burden were reverse-scored to facilitate presentation. The proportion of patients reporting health problems was assessed with the use of the three-level version of the European Quality of Life–5 Dimensions (EQ-5D-3L) questionnaire. Patients also completed the EQ-5D-3L visual-analogue scale, for which scores range from 0 to 100 and higher scores indicate better perceived health status.

BIOMARKER ANALYSIS

Fresh or archived pretreatment tumor specimens were obtained after the last therapy and before trial entry from 90.6% of the patients. For patients with oropharyngeal cancer, tumor HPV status, assessed by means of p16 immunohistochemical testing, was required to be documented by local or central analysis and was defined as positive if diffuse staining was present in at least 70% of the tumor cells.¹⁵ Immunochemical testing for p16 was not performed for nonoropharyngeal cancers because of the low prevalence of HPV-positive tumors and poor specificity for HPV status at these anatomical sites.¹⁶ Tumor PD-L1 membrane expression was evaluated centrally by means of immunohistochemical testing (Dako North America) with the use of a rabbit antihuman PD-L1 antibody (clone 28–8, Epitomics) and was scored at prespecified expression levels, including levels of 1% or more, 5% or more, and 10% or more in a minimum of 100 tumor cells that could be evaluated.¹⁷

TRIAL OVERSIGHT

This trial was registered with the National Cancer Institute and was approved by the institutional review board at each participating institution. Written informed consent was obtained from all the patients before enrollment. The trial was designed by the academic authors in collaboration with the sponsor (Bristol-Myers Squibb). The first and last authors attest to the accuracy and completeness of the data and analyses and vouch for adherence of the trial to the protocol. Medical-writing support, funded by the sponsor, was provided by inScience Communications and Chrysalis Medical Communications.

EFFICACY

Among 361 patients who underwent randomization, 133 deaths (55.4% of patients) occurred in the nivolumab group and 85 deaths (70.2% of patients) occurred in the standard-therapy group. The median duration of follow-up for overall survival was 5.1 months (range, 0 to 16.8).

The median overall survival was 7.5 months (95% confidence interval [CI], 5.5 to 9.1) in the nivolumab group versus 5.1 months (95% CI, 4.0 to 6.0) in the standard-therapy group. Overall survival was significantly longer with nivolumab than with standard therapy, and nivolumab-treated patients had a risk of death that was 30% lower than the risk among patients assigned to standard therapy (hazard ratio, 0.70; 97.73% CI, 0.51 to 0.96; P = 0.01) (Fig. 1A). The delayed separation of the Kaplan–Meier curves for overall survival is indicative of nonproportionality, and the hazard ratio should be thought of as an average over time.²³ The estimated rate of overall survival at 1 year among patients treated with nivolumab (36.0%; 95% CI, 28.5 to 43.4) was more than double the rate with standard therapy (16.6%; 95% CI, 8.6 to 26.8).

Open in a separate window

Figure 1

Overall Survival, Progression-free Survival, and Treatment Effect on Overall Survival According to Subgroup

Panel A shows the Kaplan–Meier curves for overall survival among all the patients who underwent randomization and were assigned to receive either nivolumab or standard therapy. In the planned interim analysis, the boundary for statistical significance for overall survival required the P value to be less than 0.0227. Panel B shows the Kaplan–Meier curves for progression-free survival among all the patients who underwent randomization. Symbols indicate censored observations. Hazard ratios (and confidence intervals) were computed with the use of a stratified Cox proportional-hazards model, and the P values were from a stratified log-rank test. Panel C shows a forest plot of unstratified hazard ratios for death in the analysis of the treatment effect according to demographic and clinical subgroups at baseline. Hazard ratios were not calculated for subgroups that included fewer than 20 patients across the two groups. Platinum-refractory disease in the context of primary therapy refers to cancer progression within 6 months after platinum therapy administered in the context of primary or adjuvant therapy (a post hoc derived analysis).

Nivolumab was associated with longer median overall survival than all the options for standard therapy: methotrexate (median, 4.6 months; hazard ratio for death, 0.64; 95% CI, 0.43 to 0.96), docetaxel (median, 5.8 months; hazard ratio, 0.82; 95% CI, 0.53 to 1.28), and cetuximab (median, 4.1 months; hazard ratio, 0.47; 95% CI, 0.22 to 1.01). Across prespecified demographic and clinical subgroups, the estimate of the hazard ratio for death in the analysis of overall survival with nivolumab versus standard therapy was less than 1 (Fig. 1C, and Fig. S2 in the Supplementary Appendix).

No significant difference between groups was observed with regard to the rate of progression-free survival (hazard ratio for disease progression or death, 0.89; 95% CI, 0.70 to 1.13; P = 0.32). The crossing of the Kaplan–Meier curves is indicative of nonproportionality. The median progression-free survival was 2.0 months (95% CI, 1.9 to 2.1) in the nivolumab group versus 2.3 months (95% CI, 1.9 to 3.1) in the standard-therapy group (Fig. 1B). However, a late separation in the Kaplan–Meier curves was observed, and the estimated rates of progression-free survival at 6 months were 19.7% (95% CI, 14.6 to 25.4) in the nivolumab group and 9.9% (95% CI, 5.0 to 16.9) in the standard-therapy group.

The response rate among nivolumab-treated patients was 13.3% (95% CI, 9.3 to 18.3), including 6 complete responses and 26 partial responses. In the standard-therapy group, the response rate was 5.8% (95% CI, 2.4 to 11.6), including 1 complete response and 6 partial responses. The median time to response was 2.1 months with nivolumab versus 2.0 months with standard therapy. Tumor reductions were more durable with nivolumab, as indicated by the tumor-burden plots over time for patients who had either a partial response or a complete response (Fig. S3 in the Supplementary Appendix).

PD-L1 EXPRESSION AND P16 STATUS

A prespecified, exploratory analysis was performed to evaluate the consistency of the treatment effect in subgroups defined according to tumor PD-L1 expression level (≥1% vs. <1%) (Table 2). Tumor PD-L1 expression status could be evaluated in 260 of 361 patients (72.0%) (Table S2 in the Supplementary Appendix). Among the patients who could be evaluated, 57.3% had a PD-L1 expression level of 1% or more.

Table 2

Exploratory Analysis of Overall Survival According to Tumor PD-L1 Expression and p16 Status Subgroups.^*

Variable	Nivolumab (N = 240)		Standard Therapy (N = 121)		Hazard Ratio for Death (95% CI)
	Patients	Median Survival	Patients	Median Survival
	no. (%)	mo	no. (%)	mo
All patients	240 (100.0)	7.5	121 (100.0)	5.1	0.69 (0.53–0.91)
PD-L1 expression level
≥1%	88 (36.7)	8.7	61 (50.4)	4.6	0.55 (0.36–0.83)
≥5%	54 (22.5)	8.8	43 (35.5)	4.6	0.50 (0.30–0.83)
≥10%	43 (17.9)	8.7	34 (28.1)	5.2	0.56 (0.31–1.01)
<1%	73 (30.4)	5.7	38 (31.4)	5.8	0.89 (0.54–1.45)
<5%	107 (44.6)	7.0	56 (46.3)	5.1	0.81 (0.55–1.21)
<10%	118 (49.2)	7.2	65 (53.7)	4.6	0.73 (0.50–1.06)
Not quantifiable	79 (32.9)	7.8	22 (18.2)	5.8	0.79 (0.44–1.44)
p16 status
Positive	63 (26.2)	9.1	29 (24.0)	4.4	0.56 (0.32–0.99)
Negative	50 (20.8)	7.5	36 (29.8)	5.8	0.73 (0.42–1.25)
Combined subgroup
PD-L1 ≥1% and p16-positive	23 (9.6)	8.8	14 (11.6)	3.9	0.50 (0.21–1.19)
PD-L1 ≥1% and p16-negative	17 (7.1)	8.8	16 (13.2)	5.6	0.44 (0.18–1.10)
PD-L1 <1% and p16-positive	24 (10.0)	10.0	10 (8.3)	6.4	0.55 (0.22–1.39)
PD-L1 <1% and p16-negative	14 (5.8)	7.1	12 (9.9)	7.4	0.82 (0.31–2.19)

Open in a separate window

^*Expression of the programmed death 1 ligand 1 (PD-L1) was measured in 260 patients (161 patients in the nivolumab group and 99 in the standard-therapy group), and the p16 level in 178 patients (113 in the nivolumab group and 65 in the standard-therapy group). Hazard ratios are from unstratified Cox proportional-hazards models.

In the analysis of overall survival in the subgroup of patients with a PD-L1 expression level of 1% or more, the hazard ratio for death among patients treated with nivolumab versus standard therapy was 0.55 (95% CI, 0.36 to 0.83) (Fig. 2A), whereas in the subgroup of patients with a PD-L1 expression level of less than 1%, the hazard ratio was 0.89 (95% CI, 0.54 to 1.45; P = 0.17 for interaction) (Fig. 2B). The estimates of the hazard ratio for death in the analysis of overall survival in the subgroups of patients with PD-L1 expression levels of 5% or more and of 10% or more were similar to those among patients with PD-L1 expression levels of 1% or more (Table 2).

Open in a separate window

Figure 2

Overall Survival According to Baseline PD-L1 Status and Quality of Life and Symptom Burden

Kaplan–Meier curves for overall survival according to tumor programmed death 1 ligand 1 (PD-L1) expression of 1% or higher and of less than 1% are shown in Panels A and B, respectively. Symbols indicate censored observations. Hazard ratios (and 95% confidence intervals) were computed with the use of a Cox proportional-hazards model. Panel C shows the results of multivariable analyses of adjusted mean changes from baseline in patient-reported outcomes at weeks 9 and 15, stratified according to treatment group, for 129 patients with questionnaire responses. Least-squares mean estimates were based on analyses of covariance of changes in scores from baseline with adjustment for treatment group, visit, status with respect to previous cetuximab use, and baseline score. Physical, role, and social functioning were assessed by means of the European Organization for Research and Treatment of Cancer (EORTC) Quality of Life Questionnaire–Core 30 module (QLQ-C30), and pain, sensory problems, and social-contact problems were assessed by means of the EORTC head-and-neck–specific module (QLQ-H&N35). All scales range from 0 to 100 and were scored such that higher values indicated better functioning or lower symptom burden. A clinically meaningful score change was regarded as one of 10 points (dashed lines) or more.^21,22 I bars indicate 95% confidence intervals.

In our post hoc exploratory analysis involving the 178 patients for whom tumor p16 status was reported, the median overall survival appeared to be longer with nivolumab than with standard therapy regardless of p16 status (Table 2). Among patients with p16-positive tumors, the median overall survival was 9.1 months in the nivolumab group versus 4.4 months in the standard-therapy group (hazard ratio for death, 0.56; 95% CI, 0.32 to 0.99); among patients with p16-negative tumors, the median overall survival was 7.5 versus 5.8 months (hazard ratio, 0.73; 95% CI, 0.42 to 1.25; P = 0.55 for interaction) (Figs. S4 and S5 in the Supplementary Appendix).

We further explored the effect of nivolumab versus standard therapy on overall survival in subgroups defined according to both PD-L1 expression (≥1% vs. <1%) and tumor p16 status (positive vs. negative) (Table 2). The estimated hazard ratios for death in the analysis of overall survival with nivolumab versus standard therapy were less than 1 in all four subgroups. Results of the exploratory analysis of the treatment effect on response rates in the subgroups defined according to tumor PD-L1 level and p16 status are provided in Table S3 in the Supplementary Appendix.

SAFETY

The most common treatment-related adverse events are shown in Table 3 (see also Tables S4, S5, and S6 in the Supplementary Appendix). The rates of treatment-related adverse events of any grade were similar in the two groups, but fewer events of grade 3 or 4 were reported in the nivolumab group than in the standard-therapy group (occurring in 13.1% vs. 35.1% of patients). In the nivolumab group, the most frequent adverse events of any grade were fatigue, nausea, rash, decreased appetite, and pruritus. Among the select adverse events, gastrointestinal events were less common with nivolumab than with standard therapy (occurring in 6.8% vs. 14.4% of the patients; primarily diarrhea), whereas adverse events of the skin were more common with nivolumab (in 15.7% vs. 12.6%; primarily rash and pruritus), as were adverse events of the endocrine system (in 7.6% vs. 0.9%; primarily hypothyroidism). Pneumonitis was observed in 2.1% of the patients treated with nivolumab. Two treatment-related deaths were reported in the nivolumab group (pneumonitis and hypercalcemia in one patient each), and one patient in the standard-therapy group died from a treatment-related lung infection.

Supported by Bristol-Myers Squibb.

We thank the patients and their families; the study teams involved in the trial; the staff of Ono Pharmaceutical, Osaka, Japan; the staff of Dako North America for collaborative development of the automated programmed death 1 ligand immunohistochemical assay; Ludovic Astier for serving as the Check-Mate 141 protocol manager; Naveed Imshad, M.S., and Karthik Darbha, M.S., for statistical programming; Kim Cocks, Ph.D., Fiona Taylor, M.Biochem., and Michael DeRosa, M.A., for inferential patient-reported outcome analyses; and Michelle Daniels, M.D., of inScience Communications, Springer Healthcare, and Daniel Hutta, Ph.D., of Chrysalis Medical Communications, for medical writing and editorial assistance (funded by Bristol-Myers Squibb) with an earlier version of the manuscript.