Randomized Phase II Trial of Consolidation Pembrolizumab After Definitive Chemoradiotherapy in Locally Advanced Nasopharyngeal Carcinoma: KCSG HN19‐09, CONPELAN Study

ABSTRACT

Background

The benefit of consolidation programmed cell death protein 1 (PD‐1) blockade after chemoradiotherapy in locoregionally advanced nasopharyngeal carcinoma (NPC) remains unclear.

Methods

Patients with stages II–IVB NPC were randomized (2:1) to pembrolizumab 200 mg or placebo every 3 weeks for up to 17 cycles after chemoradiotherapy. The primary endpoint was 3‐year progression‐free survival (PFS).

Results

Fifty‐three patients were randomized (pembrolizumab, n = 34; placebo, n = 19). The 3‐year PFS rates were 56.5% and 57.8% in the pembrolizumab and placebo arms, respectively (one‐sided p = 0.142; stratified HR, 0.408). Grade 3 or higher adverse events occurred only in the pembrolizumab arm (8.8%). In an exploratory analysis (n = 40), high intratumoral TIL density was associated with reduced recurrence in the pembrolizumab arm (OR, 0.04; p = 0.010) but not the placebo arm.

Conclusions

Consolidation pembrolizumab did not improve PFS in unselected locally advanced NPC. Exploratory immune profiling suggested a predictive signal warranting prospective validation.

1. Introduction

Nasopharyngeal carcinoma (NPC) is an Epstein–Barr virus (EBV)‐associated epithelial malignancy endemic to Southern China and Southeast Asia [1]. Concurrent chemoradiotherapy (CCRT) remains the standard of care for patients with locally advanced disease (Stages III–IVA). However, 20%–30% of patients experience relapse within 3 years [2, 3]. Although consolidation chemotherapy after CCRT has been used, its role remains controversial owing to mixed survival benefits and substantial toxicities, such as severe mucositis and nephrotoxicity, which frequently lead to treatment discontinuation [4, 5]. Subsequent studies investigating consolidation chemotherapy after CCRT have yielded mixed results regarding survival benefit [4], and the optimal post‐CCRT strategy remains undefined.

The high expression of programmed death ligand 1 (PD‐L1) in EBV‐associated NPC provides a strong biological rationale for a programmed cell death protein 1 (PD‐1)/PD‐L1 blockade [6]. Pembrolizumab, a monoclonal antibody that targets PD‐1, has demonstrated antitumor activity in recurrent and metastatic NPC [7]. In the Phase Ib KEYNOTE‐028 trial, pembrolizumab monotherapy yielded an objective response rate of 25.9% in patients with PD‐L1‐positive recurrent or metastatic NPC [7]. Furthermore, recent trials, such as RATIONALE‐309 [8], CAPTAIN‐1st [9], and JUPITER‐02 [10], demonstrated that the addition of PD‐1 inhibitors can improve progression‐free survival (PFS) in the recurrent and/or metastatic first‐line setting. However, whether the consolidation of a PD‐1 blockade after definitive CCRT can improve outcomes in an unselected population has not yet been established in a randomized trial. Therefore, the present study (CONPELAN) investigated the efficacy and safety of consolidation pembrolizumab versus a placebo in patients with Stages II–IVB NPC.

2. Materials and Methods

2.1. Study Design and Participants

This multicenter, randomized, double‐blind, placebo‐controlled, investigator‐initiated, Phase II trial was conducted at academic centers affiliated with the Korean Cancer Study Group in South Korea. The study protocol was approved by the Institutional Review Board of each participating center and was conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. The clinical trial was also registered with ClinicalTrials.gov (NCT04227509). Written informed consent was obtained from all patients before enrollment.

Eligible patients were aged 19 years or older with histologically confirmed nonkeratinizing NPC (World Health Organization [WHO] type II, differentiated or undifferentiated), Stages II–IVB according to the American Joint Committee on Cancer eighth edition, an Eastern Cooperative Oncology Group (ECOG) performance status of 0 or 1, adverse events of CCRT of Grade 2 or lower before enrollment, and adequate organ function at enrollment. All patients underwent curative‐intent CCRT with a radiation dose of at least 60 Gy and a cumulative cisplatin dose of at least 180 mg/m². Prior induction chemotherapy was permitted. The key exclusion criteria were confirmed progressive disease following curative‐intent treatment and active autoimmune disease requiring systemic immunosuppression.

2.2. Randomization and Treatment

Patients were randomly assigned in a 2:1 ratio to receive pembrolizumab or a placebo. Randomization was stratified according to age (< 70 vs. ≥ 70 years) and clinical response to CCRT (complete response [CR] vs. non‐CR). Patients in the experimental arm received 200‐mg pembrolizumab intravenously every 3 weeks for up to 17 cycles (maximum duration, 12 months). Treatment was continued until disease progression, unacceptable toxicity, patient withdrawal, or completion of 17 cycles. Patients in the control group received a matching placebo on the same schedule. Per protocol (PP), randomization and the first study drug administration were performed 4–8 weeks after completion of CCRT, conditional on resolution of CCRT‐related toxicities to grade ≤ 2. In practice, all randomized patients initiated study drug between 7 and 8 weeks after completion of CCRT, with a mean interval of 55 days (approximately 7.9 weeks).

2.3. Endpoints

The primary endpoint was the 3‐year PFS rate, defined as the proportion of patients alive and free from disease progression or recurrence at 3 years from the date of the first study drug administration. Disease progression was assessed by the investigators using the Response Evaluation Criteria in Solid Tumors version 1.1, based on imaging findings obtained at protocol‐specified intervals.

The secondary endpoints included the overall response rate (ORR), overall survival (OS), PFS, and safety. OS was defined as the time from randomization to death from any cause. Safety was assessed by monitoring adverse events, graded according to the Common Terminology Criteria for Adverse Events version 5.0.

2.4. Statistical Analysis

This study was designed to detect an improvement in the 3‐year PFS rate from 70% in the control arm to 86% in the experimental arm (hazard ratio [HR] = 0.42), with 80% power at a one‐sided significance level of 0.10. The planned sample size was 112 patients (experimental arm, 75; control arm, 37). Three analysis populations were defined: the intention‐to‐treat (ITT) population (all randomized patients), PP population (ITT patients with treatment compliance of 80% or higher), and safety population (all patients who received at least one dose of the study drug).

The 3‐year PFS rate was estimated using the Kaplan–Meier method. The primary comparison between the treatment arms was performed using a one‐sided log‐rank test at a significance level of 10%. A stratified Cox proportional hazards model adjusted for the CCRT response (CR vs. non‐CR) and baseline variables with clinically meaningful imbalances was used to estimate the HR with 80% confidence intervals (CIs). Secondary endpoints were compared using two‐sided tests at a significance level of 20%. The ORR was compared using the chi‐squared test or Fisher exact test. OS and PFS distributions were compared using the log‐rank test. All statistical analyses were performed using SAS (version 9.4; SAS Institute, Cary, NC, USA), and figures were generated using Python (version 3.9; Python Software Foundation) with the matplotlib library.

2.5. Exploratory Immune Microenvironment Analysis

A prespecified exploratory analysis of the tumor immune microenvironment was performed on available pretreatment biopsy specimens (hematoxylin and eosin slides) using an artificial intelligence (AI)‐powered whole‐slide image analyzer (Lunit SCOPE IO; Lunit Inc., Seoul, South Korea) [11, 12]. A refined version of the model, trained on an expanded dataset, was used in this study. The resulting whole‐slide images were analyzed to quantify cell densities (per mm²) of six cell types: tumor cells, lymphocytes (tumor‐infiltrating lymphocytes [TIL]), fibroblasts, endothelial cells, macrophages, and other cells. The model also identified and quantified mitotic cells within the cancer area. Cell counts were measured separately within the cancer area (intratumoral) and cancer‐associated stroma. Intratumoral TIL (iTIL) density, inflamed score (IS), and three categories of immune phenotypes (inflamed, excluded, and desert) were derived from these measurements. Associations between immune features and recurrence were assessed using odds ratios (ORs) with 95% CIs and the Fisher exact test, stratified by treatment arm.

3. Results

3.1. Baseline Characteristics

Between August 2020 and April 2024, 59 patients with locally advanced NPC were screened. The trial was closed prematurely with 53 randomized patients because of slower‐than‐anticipated accrual, which was attributable to the low incidence of NPC in South Korea and the disruptions caused by the coronavirus disease pandemic. The remaining 53 patients were randomly assigned to receive consolidation pembrolizumab (n = 34) or a placebo (n = 19) (Figure 1). The ITT population comprised 53 randomized patients who received at least one dose of the study drug. Twelve patients (pembrolizumab arm, 5; placebo arm, 7) did not meet the PP criteria because treatment compliance was below 80%, yielding a PP population of 41 patients (29 and 12, respectively). All 53 patients were included in the safety analysis.

FIGURE 1.

Consolidated standards of reporting trials flow diagram. A total of 59 patients were assessed for eligibility, of whom 53 were randomized in a 2:1 ratio to receive pembrolizumab (n = 34) or a placebo (n = 19). All randomized patients were included in the intention‐to‐treat and safety analyses. The exploratory immune microenvironment analysis was performed on 40 patients with available pretreatment biopsy specimens. [Color figure can be viewed at wileyonlinelibrary.com]

Baseline characteristics were balanced between the two arms (Table 1). Mean ages were 49.8 and 51.9 years in the pembrolizumab and placebo arms, respectively (p = 0.618). The ECOG performance status was 0 in 41.2% and 36.8% of patients in the pembrolizumab and placebo arms, respectively (p = 0.757). The WHO histological subtype distribution differed between the arms (p = 0.044): WHO type II (nonkeratinizing carcinoma) was present in 91.2% of patients in the pembrolizumab arm and 84.2% of those in the placebo arm, whereas WHO type III was observed in 8.8% of patients in the pembrolizumab arm and none of those in the placebo arm. Stage IVA disease was the most common disease category in the pembrolizumab and placebo arms (50.0% and 52.6%, respectively), followed by Stage III (29.4% and 31.6%, respectively) and Stage II (20.6% and 15.8%, respectively). Prior induction chemotherapy was administered to 29.4% and 42.1% of the patients, respectively. The rates of CR after chemoradiotherapy were 32.4% (11/34) and 42.1% (8/19) in the pembrolizumab and placebo arms, respectively. The interval from completion of CCRT to the first dose of study drug ranged from 7 to 8 weeks in all randomized patients (mean, 55 days; approximately 7.9 weeks), with no clinically meaningful difference between arms.

TABLE 1.

Baseline characteristics of the intention‐to‐treat population.

Characteristic	Placebo (N = 19)	Pembrolizumab (N = 34)	p
Age, years
Mean (SD)	51.89 (9.60)	49.79 (11.40)	0.618
Median (range)	54 [32–70]	51.5 [20–68]
Sex, n (%)			0.279
Male	15 (78.95)	22 (64.71)
Female	4 (21.05)	12 (35.29)
ECOG PS, n (%)			0.757
0	7 (36.84)	14 (41.18)
1	12 (63.16)	20 (58.82)
WHO histologic type, n (%)			0.044
Type II (nonkeratinizing)	16 (84.21)	31 (91.18)
Type III	0 (0.00)	3 (8.82)
Other	3 (15.79)	0 (0.00)
AJCC 8th stage, n (%)			0.912
II	3 (15.79)	7 (20.59)
III	6 (31.58)	10 (29.41)
IVA	10 (52.63)	17 (50.00)
EBV DNA normal, n (%)	16 (84.21)	29 (85.29)	> 0.999
Prior induction chemotherapy, n (%)	8 (42.11)	10 (29.41)	0.380
CCRT response: CR, n (%)	8 (42.11)	11 (32.35)	0.556
CCRT response: non‐CR, n (%)	11 (57.89)	23 (67.65)

Abbreviations: AJCC, American Joint Committee on Cancer; CCRT, concurrent chemoradiotherapy; CR, complete response; EBV, Epstein–Barr virus; ECOG PS, Eastern Cooperative Oncology Group performance status; WHO, World Health Organization.

3.2. Treatment Compliance

The mean treatment compliance rates were 90.8% (standard deviation [SD], 20.4%) and 76.8% (SD, 32.4%) in the pembrolizumab and placebo arms, respectively (p = 0.131) (Table S1). The median compliance rates were 100% in both arms (pembrolizumab, 11.76%–100.0%; placebo, 5.88%–100.0%). Five patients in the pembrolizumab arm and seven in the placebo arm had compliance below the 80% threshold required for PP inclusion.

3.3. Primary Efficacy

At a median follow‐up of 19.8 months (95% CI, 14.4–28.1) in the pembrolizumab arm and 18.7 months (95% CI, 8.3–25.3) in the placebo arm (reverse Kaplan–Meier method), and at the data cutoff point, 28 of 34 patients (82.4%) in the pembrolizumab arm and 14 of 19 those (73.7%) in the placebo arm did not experience disease progression or death. Consequently, the Kaplan–Meier estimated 3‐year PFS rates were 56.5% (80% CI, 30.86%–75.8%) in the pembrolizumab arm and 57.8% (80% CI, 35.79%–74.7%) in the placebo arm, which were lower than the crude event‐free proportions because of the reduced number of patients at risk at the 3‐year time point (Figure 2). According to the one‐sided log‐rank test, this difference did not reach statistical significance at the prespecified alpha level of 0.10 (p = 0.142).

FIGURE 2.

Progression‐free survival (PFS) in the intention‐to‐treat population. Kaplan–Meier estimates of PFS by treatment arm. The stratified hazard ratio is 0.408 (80% confidence interval, 0.159–1.047), and the one‐sided log‐rank p‐value is 0.142. [Color figure can be viewed at wileyonlinelibrary.com]

Stratified Cox proportional hazards regression yielded a HR of 0.408 (80% CI, 0.159–1.047; one‐sided p = 0.111), favoring the pembrolizumab arm. The upper bound of the 80% CI crossed 1.0, confirming failure to meet the primary endpoint.

Early PFS rates favored pembrolizumab, with 1‐year PFS rates of 90.1% and 81.0% and 2‐year PFS rates of 75.3% and 57.8% in the pembrolizumab and placebo arms, respectively. The overall PFS distribution did not differ between the arms according to the log‐rank test (p = 0.284) (Figure 2). The median PFS was not achieved in either arm.

OS outcomes are summarized in Table S2 and shown in Figure S1.

In an exploratory analysis stratified by prior induction chemotherapy, the treatment HR numerically favored pembrolizumab in both subgroups (induction‐receiving subgroup: HR 0.39, 95% CI 0.04–4.12, n = 18; induction‐naïve subgroup: HR 0.89, 95% CI 0.17–4.63, n = 35), and the formal arm × induction interaction term was not significant (HR 0.33, 95% CI 0.02–5.76, p = 0.45) (Table S3 and Figures S2 and S3). With only 11 PFS events overall, this analysis was severely underpowered and is presented as hypothesis‐generating only.

In the PP population, the 3‐year PFS rates were 60.3% (80% CI, 32.18%–79.8%) and 71.4% (80% CI, 43.33%–87.3%) in the pembrolizumab and placebo arms, respectively (one‐sided log‐rank p = 0.432). The stratified Cox HR in the PP population was 0.312 (80% CI, 0.093–1.051; one‐sided p = 0.109).

3.4. Secondary Efficacy Endpoints

3.4.1. Objective Response Rate

The ORRs in the ITT population were 90.9% and 94.7% in the pembrolizumab and placebo arms, respectively (Fisher's exact test, p > 0.999) (Table 2). CR was achieved in 87.9% (29/33) and 89.5% (17/19) of the evaluable patients in the pembrolizumab and placebo arms, respectively, and partial response was achieved in 3.0% (1/33) and 5.3% (1/19) of patients, respectively. Stable disease was observed in three patients (9.1%) in the pembrolizumab arm, and progressive disease occurred in one patient (5.3%) in the placebo arm. One patient in the pembrolizumab arm could not be evaluated for tumor response. Stratified logistic regression analysis did not identify a treatment difference in the ORR (p = 0.612). In the PP population, ORRs were 93.1% and 100.0% in the pembrolizumab and placebo arms, respectively (p > 0.999).

TABLE 2.

Objective response rate in the intention‐to‐treat population.

Response	Placebo (N = 19)	Pembrolizumab (N = 33 a )	p
CR, n (%)	17 (89.47)	29 (87.88)
PR, n (%)	1 (5.26)	1 (3.03)
SD, n (%)	0 (0.00)	3 (9.09)
PD, n (%)	1 (5.26)	0 (0.00)
ORR, % (80% CI)	94.74 [81.02–99.45]	90.91 [80.86–96.61]	> 0.999

Abbreviations: CI, confidence interval; CR, complete response; ORR, overall response rate; PD, progressive disease; PR, partial response; SD, stable disease.

^a One patient in the pembrolizumab arm was not evaluable for tumor response.

3.4.2. OS

At the time of analysis, one death occurred in the pembrolizumab arm and no deaths were recorded in the placebo arm (Table S2 and Figure S1). The 3‐year OS rates were 100% in both arms. The 4‐year OS rate was 50.0% (80% CI, 7.71%–82.9%) in the pembrolizumab arm; the placebo arm did not reach the 4‐year landmark. Median OS was not reached in either arm, owing to the premature closure of the trial and subsequent data immaturity.

3.5. Safety

All 53 patients were included in the safety analysis (Table 3). Treatment‐emergent adverse event (TEAEs) of any grade occurred in 28 patients (82.3%) and 15 patients (79.0%) in the pembrolizumab and placebo arms, respectively, with most events being Grade 1 or 2. Grade 3 or higher TEAEs occurred in three patients (8.8%) in the pembrolizumab arm and none in the placebo arm. The three events—hematemesis (53‐year‐old male, attributed to a preexisting peptic ulcer; recovered), traumatic left foot fracture (64‐year‐old male; recovered), and decreased neutrophil count (resolved without intervention)—were all assessed as unrelated to the study drug. The only grade ≥ 2 treatment‐related serious adverse event was a single episode of Grade 2 hyperthyroidism, managed with temporary treatment interruption and resolved. No Grade 4 or 5 events, treatment‐related deaths, or permanent treatment discontinuation due to adverse events were observed.

TABLE 3.

Treatment‐emergent adverse events.

	Placebo (N = 19)		Pembrolizumab (N = 34)
	n (%)	No. of events	n (%)	No. of events
Any TEAE	15 (78.95)	41	28 (82.35)	117
Grade 1		30		72
Grade 2		11		42
Grade 3	0 (0.00)	0	3 (8.82)	3
SAE	0 (0.00)	0	3 (8.82)	3
Drug‐related AE		17		45
Most common TEAE
Hypothyroidism	5 (26.32)	6	18 (52.94)	19
Maculopapular rash	1 (5.26)	1	6 (17.65)	6
Peripheral neuropathy	5 (26.32)	6	4 (11.76)	4
Presence of sputum	1 (5.26)	1	4 (11.76)	4
Headache	0 (0.00)	0	3 (8.82)	3
COVID‐19 infection	2 (10.53)	2	1 (2.94)	1

Abbreviations: AE, adverse event; COVID‐19, coronavirus disease; no., number; SAE, serious adverse event; TEAE, treatment‐emergent adverse event.

The most common TEAE was hypothyroidism, affecting 18 patients (52.9%) in the pembrolizumab arm and 5 patients (26.3%) in the placebo arm. The drug‐related adverse events comprised 45 of 117 cases (38.5%) and 17 of 41 (41.5%) in the pembrolizumab and placebo arms, respectively. By the data cutoff point, over 60% of the events in both groups were resolved.

3.6. Exploratory Immune Microenvironment Analysis

Whole‐slide images were available for 40 of the 53 patients (75.5%; pembrolizumab arm, 25; placebo arm, 15). The cell densities (per mm²) of lymphocytes (TILs), fibroblasts, endothelial cells, and macrophages were quantified separately in the cancer area and cancer‐associated stroma: mitotic cells, identified exclusively within the cancer area, were analyzed only in that region.

3.6.1. TIL Density and Recurrence

The recurrence rate did not differ between the pembrolizumab and placebo arms in this subset (5/25 [20.0%] and 4/15 [26.7%], respectively, p = 0.920). Within the pembrolizumab arm, patients with a higher iTIL density had a lower probability of recurrence across several cutoff thresholds. At the upper 25% cutoff point, the OR for recurrence in the high iTIL group relative to that in the low iTIL group was 0.04 (95% CI, 0–0.41; p = 0.010). At the upper 33% cutoff point, the OR was 0.07 (95% CI, 0.01–0.68; p = 0.030). This association was not observed in the placebo arm (upper 25% cutoff point: OR = 2.50, 95% CI, 0.23–59.89, p = 0.480; upper 33% cutoff point: OR = 1.20, 95% CI, 0.11–13.31, p = 0.880) (Table 4).

TABLE 4.

Association between intratumoral TIL density, inflamed score, and recurrence by treatment arm.

Biomarker cutoff value	Arm	Subgroup	N a	Recurrence, n (%)	OR (95% CI)	p
iTIL upper 25%	Pembrolizumab	Upper	21	2 (9.5%)	0.04 (0–0.41)	0.01
		Lower	4	3 (75%)	Ref
	Placebo	Upper	9	3 (33.3%)	2.50 (0.23–59.89)	0.48
		Lower	6	1 (16.7%)	Ref
iTIL upper 33%	Pembrolizumab	Upper	20	2 (10%)	0.07 (0.01–0.68)	0.03
		Lower	5	3 (60%)	Ref
IS upper 25%	Pembrolizumab	Upper	20	2 (10%)	0.04 (0–0.41)	0.02
		Lower	4	3 (75%)	Ref
3‐IP vs. recurrence	Pembrolizumab	—	25	—	Fisher's exact test	0.022
	Placebo	—	15	—	Fisher's exact test	0.454

Abbreviations: 3‐IP, three‐category immune phenotype; CI, confidence interval; IS, inflamed score; iTIL, intratumoral tumor‐infiltrating lymphocyte density; OR, odds ratio; Ref, reference group; vs., versus.

^a Analysis performed on 40 patients (pembrolizumab, n = 25; placebo, n = 15) with available specimens. One specimen in the pembrolizumab arm was excluded from immune phenotype analysis due to quality control failure.

3.6.2. IS and Immune Phenotype

Similarly, a higher IS in the pembrolizumab arm was associated with reduced recurrence. At the top 25% IS cutoff point, the OR was 0.04 (95% CI, 0–0.41; p = 0.020), showing no association in the placebo group (Table 4). The three‐category immune phenotype classification showed a statistically significant association with recurrence in the pembrolizumab arm (Fisher's exact test, p = 0.022), but not in the placebo arm (p = 0.454) (Figure 3). The recurrence rates were 11.8% (2/17) among patients classified with an inflamed phenotype in the pembrolizumab arm and 42.9% (3/7) among those classified with a noninflamed phenotype (excluded or desert). In the placebo arm, recurrence rates were 30.0% (3/10) among patients with an inflamed phenotype and 20.0% (1/5) among those with a noninflamed phenotype; this difference was not significant (p = 0.454).

FIGURE 3.

Exploratory immune microenvironment analysis (digital pathology, N = 40; Lunit Scope). (A) Odds ratios for recurrence by intratumoral tumor‐infiltrating lymphocyte (iTIL) density cutoff values, stratified by treatment arm. (B) Odds ratios for recurrence by inflamed score (IS) cutoff values, stratified by treatment arm. (C) Recurrence rates by three‐category immune phenotypes (inflamed vs. noninflamed), stratified by treatment arm. Asterisks indicate statistically significant associations. [Color figure can be viewed at wileyonlinelibrary.com]

3.6.3. Macrophage Density

In the placebo arm, a difference in combined intratumoral and stromal macrophage density was observed between patients with and without recurrence (mean 4.8 vs. 18.9 per mm²; p = 0.030), with lower macrophage density in patients who experienced recurrence. However, this association was not observed in the pembrolizumab arm (p = 0.160).

3.6.4. Other Cell Types

No statistically significant associations were found between recurrence and the density of fibroblasts, endothelial cells, or mitotic cells in either treatment arm (all p > 0.05).

4. Discussion

This multicenter, randomized Phase II trial evaluated consolidation pembrolizumab following definitive chemoradiotherapy in patients with locoregionally advanced NPC. The 3‐year PFS rate did not differ between the pembrolizumab and placebo arms (56.49% vs. 57.82%; one‐sided log‐rank p = 0.142). The stratified Cox regression yielded a HR of 0.408 (80% CI, 0.159–1.047), suggesting a directional trend that did not reach the prespecified significance threshold. Pembrolizumab was well tolerated, with Grade 3 or higher adverse events occurring in only 8.82% of patients and no treatment‐related deaths. An exploratory digital pathology analysis suggested a possible association between high intratumoral TIL density and reduced recurrence in the pembrolizumab arm, though this finding requires independent validation.

The CONPELAN trial was designed to enroll 112 patients (75 pembrolizumab, 37 placebo) to detect an improvement in 3‐year PFS from 70% to 86% (HR = 0.42), with 80% power at a one‐sided significance level of 0.10. The assumed treatment effect of 16 percentage points was larger than the 10‐percentage‐point difference subsequently observed in the DIPPER trial (77.3%–86.9%) [2], which enrolled a substantially larger cohort in an endemic region. Only 53 patients (47% of target) were randomized, primarily because NPC is uncommon in South Korea, and the coronavirus disease (COVID)‐19 pandemic and 2024 physician workforce disruption further limited enrollment. With eight recurrence events and one death, the study was substantially underpowered. This trial should therefore be interpreted as hypothesis‐generating rather than definitive, and we focused on effect size estimation and CIs rather than relying solely on p‐value thresholds.

Several recent trials have demonstrated that PD‐1 blockade can improve outcomes in locoregionally advanced NPC when combined with appropriate patient selection or treatment sequencing. The DIPPER trial (NCT03427827), a Phase 3 study enrolling 450 patients with stage T4N1 or T1‐4N2‐3 disease in China, showed that adjuvant camrelizumab after induction‐CCRT improved 3‐year event‐free survival from 77.3% to 86.9% (HR, 0.56; 95% CI, 0.36–0.89; p = 0.01) [2]. Liu et al. reported that neoadjuvant and adjuvant toripalimab in high‐risk patients (pretreatment plasma EBV DNA ≥ 1500 copies/mL) improved 2‐year PFS from 74.0% to 92.0% (HR, 0.40; 95% CI, 0.18–0.89; p = 0.019) [13]. The CONTINUUM trial added sintilimab throughout induction, concurrent, and adjuvant phases in high‐risk patients (Stage III–IVa, excluding T3‐4N0 and T3N1), and reported a 3‐year event‐free survival of 86% versus 76% (HR, 0.59; p = 0.019) [3]. These trials with positive results shared two common features that were absent from CONPELAN: risk‐based patient enrichment and larger sample sizes within endemic populations.

The discordant outcomes between CONPELAN and these trials may be explained by two key factors. First, CONPELAN enrolled an unselected population across all Stage II–IVB patients, whereas DIPPER restricted enrollment to T4N1 or N2–3 disease and the toripalimab trial required elevated baseline plasma EBV DNA (≥ 1500 copies/mL). Most patients in our cohort had normal baseline EBV DNA, indicating a lower‐risk population. Plasma EBV DNA has been recognized as a biomarker for risk stratification in NPC, and its kinetics after treatment may identify patients who benefit most from adjuvant intensification [14, 15]. Second, the optimal timing of immunotherapy remains uncertain. CONPELAN administered pembrolizumab only in the consolidation phase, after chemoradiotherapy had already reduced tumor burden, whereas the toripalimab and CONTINUUM trials incorporated PD‐1 blockade during the neoadjuvant phase. Initiating PD‐1 blockade before or concurrent with chemoradiotherapy may be advantageous because the intact tumor provides a larger antigen reservoir for T‐cell priming, and radiation‐induced immunogenic cell death can synergize with checkpoint inhibition to amplify antitumor immunity. In contrast, consolidation‐only immunotherapy acts on a minimal residual disease state with a diminished neoantigen load, which may limit the magnitude of immune‐mediated tumor control.

The exploratory immune microenvironment analysis using Lunit digital pathology provided hypothesis‐generating observations. The upper quartile of intratumoral TIL density was associated with reduced recurrence in the pembrolizumab arm (OR, 0.04; 95% CI, 0–0.41; p = 0.01), and higher IS values showed a similar pattern (OR, 0.04; p = 0.02). The three‐category immune phenotype classification also correlated with recurrence in the pembrolizumab arm (p = 0.022). None of these associations were present in the placebo arm (upper 25% iTIL cutoff: OR, 2.50; p = 0.48), which raises the hypothesis that TIL density may carry predictive value for immunotherapy benefit. AI‐based digital quantification of TIL has shown prognostic and predictive value across multiple cancer types [11, 12, 16]. However, only 40 of 53 patients had evaluable specimens, and only five recurrence events occurred in the pembrolizumab arm. Whether digital pathology‐derived immune features can identify candidates for consolidation immunotherapy remains to be determined in larger cohorts.

The interpretation of our findings is constrained by several design and operational factors. First, the trial enrolled less than half of the planned sample (53 of 112), leaving it substantially underpowered. The sample size calculation assumed a 16‐percentage‐point improvement in 3‐year PFS, which was more optimistic than the 10‐percentage‐point difference observed in the DIPPER trial [2]; a more conservative effect size assumption would have required a substantially larger sample. Second, the 3‐year PFS in the control arm (57.82%) was lower than the 70% assumed at design and lower than the 76%–87% range reported in contemporary endemic‐region trials [2, 3, 13]. This discrepancy likely reflects the small sample size and heavy censoring at the 3‐year landmark rather than a true population‐level difference, but it further weakened the statistical comparison. Third, PFS as a composite endpoint does not distinguish between locoregional and distant failures, which may differ in their susceptibility to immunotherapy; separate reporting of locoregional recurrence‐free survival and distant metastasis‐free survival would have been more informative. Fourth, the low event count (eight recurrences, one death) reduced the precision of survival estimates and prevented subgroup analyses. Fifth, follow‐up loss from the COVID‐19 pandemic and the 2024 physician workforce disruption resulted in substantial censoring before the 3‐year landmark. Sixth, the one‐sided α of 0.10, while appropriate for a screening Phase II trial, sets a lower evidentiary bar than Phase III standards. Seventh, enrollment was concentrated at two of four participating institutions, which may limit generalizability. Eighth, induction chemotherapy was permitted but not used as a stratification factor at randomization, leading to a numerical imbalance between arms (29.4% vs. 42.1%); the formal arm × induction interaction test did not detect a differential treatment effect by induction status (p = 0.45), although power was limited. Finally, the Lunit analysis was restricted to 40 of 53 patients with available specimens, introducing potential selection bias.

CONPELAN adds a randomized, placebo‐controlled design to a literature on adjuvant and consolidation immunotherapy in NPC that has been dominated by single‐arm studies from endemic regions. As a trial conducted in a nonendemic population, it offers a perspective distinct from the predominantly Chinese studies that have generated most of the current evidence. The favorable safety profile, with only one drug‐related SAE (Grade 2 hyperthyroidism) among 34 treated patients, supports the tolerability of pembrolizumab consolidation after chemoradiotherapy. The exploratory finding that high intratumoral TIL density was associated with reduced recurrence only in the pembrolizumab arm, if replicated, could inform biomarker‐driven trial designs. Future immunotherapy trials in locoregionally advanced NPC should adopt more conservative effect size assumptions informed by recent data [2, 3] and consider biomarker‐enriched enrollment strategies—whether based on EBV DNA levels, tumor immune phenotype, or other molecular features—to improve the probability of detecting clinically meaningful treatment effects.

Author Contributions

Conceived and designed the analysis: Sang Hoon Chun and Bhumsuk Keam. Collected the data: All authors. Contributed data or analysis tools: All authors. Performed the analysis: Sang Hoon Chun and Bhumsuk Keam. Wrote the paper: Sang Hoon Chun and Bhumsuk Keam.

Funding

This study was supported by the National R&D Program for Cancer Control through the National Cancer Center, funded by the Ministry of Health and Welfare, Republic of Korea (grant number: RS‐2022‐CC125163).

Conflicts of Interest

B.K. received research funding from MSD, AstraZeneca, Bayer, and Ono Pharmaceutical Co. Ltd. and has served as an advisor for Handok, Yuhan, Trialinformatics, TiumBio, and BeOne outside of the current work. The other authors declare no conflicts of interest.

Supporting information

Figure S1: Overall survival in the intention‐to‐treat population.

Figure S2: Forest plot of treatment effect on progression‐free survival by prior induction chemotherapy subgroup.

Figure S3: Kaplan–Meier estimates of progression‐free survival by treatment arm, stratified by receipt of prior induction chemotherapy.

Table S1: Treatment compliance.

Table S2: Overall survival in the intention‐to‐treat population.

Table S3: Treatment effect on progression‐free survival by prior induction chemotherapy receipt (intention‐to‐treat population).

Chun S. H., Seo S., Cho J. H., et al., “Randomized Phase II Trial of Consolidation Pembrolizumab After Definitive Chemoradiotherapy in Locally Advanced Nasopharyngeal Carcinoma: KCSG HN19‐09, CONPELAN Study,” Head & Neck 48, no. 11 (2026): 3103–3112, 10.1002/hed.70329.

Data Availability Statement

The datasets generated in this study are not publicly available owing to patient privacy restrictions and institutional policy.