Efficacy and Safety of PD‐1/PD‐L1 Inhibitors Combined With Anti‐Angiogenic Tyrosine Kinase Inhibitors for Recurrent or Metastatic Nasopharyngeal Carcinoma: A Systematic Review and Single‐Arm Meta‐Analysis

ABSTRACT

Objective

To evaluate the efficacy and safety of PD‐1/PD‐L1 inhibitors plus small‐molecule anti‐angiogenic drugs (apatinib, anlotinib and famitinib) for recurrent/metastatic nasopharyngeal carcinoma (R/M NPC).

Methods

Systematic searches of major databases up to January 7, 2026 identified prospective single‐arm Phase II trials and retrospective cohort/real‐world studies of PD‐1/PD‐L1 inhibitors combined with apatinib, anlotinib, or famitinib in R/M NPC. Pooled response and safety rates were estimated using random‐effects models, with study quality assessed by appropriate tools.

Results

Nine studies comprising 10 independent cohorts (357 patients) were included. The pooled objective response rate (ORR) was 48% (95% CI: 39%–57%), the disease control rate (DCR) 83% (95% CI: 77%–88%), and the 1‐year overall survival rate 79% (95% CI: 68%–86%). The incidence of grade ≥ 3 treatment‐related adverse events (TRAEs) was 47% (95% CI: 37%–58%). The most frequent serious TRAEs were hypertension (9%), hand‐foot syndrome (10%), and nasopharyngeal necrosis (14%), with the latter requiring close monitoring due to its potential for severe bleeding complications. No treatment‐related deaths were reported across all included studies. Subgroup analysis showed no significant difference in ORR between camrelizumab and toripalimab; apatinib yielded higher ORR than anlotinib; immunotherapy‐naive patients had significantly higher ORR than those previously treated with immune checkpoint inhibitors (ICIs) (p = 0.0028).

Conclusion

The combination shows promising antitumor activity and a generally acceptable safety profile in R/M NPC, particularly for immunotherapy‐naive patients. These findings support further investigation in well‐designed randomized controlled trials, but confirmation of its role relative to standard therapies requires larger comparative studies.

This single‐arm meta‐analysis (9 studies, 357 patients) shows that PD‐1/PD‐L1 inhibitors plus anti‐angiogenic TKIs yield an ORR of 48% and DCR of 83% in recurrent/metastatic nasopharyngeal carcinoma. Median PFS is 9.95 months. ICI‐naive patients benefit more (ORR 60.9% vs. 36.1%, p = 0.0028). Grade ≥ 3 adverse events (47%) include 14% nasopharyngeal necrosis. Efficacy is promising with manageable safety.

1. Introduction

Nasopharyngeal carcinoma (NPC) is a malignant tumor that originates from the epithelial cells of the nasopharyngeal mucosa. It has a significant geographical clustering pattern, with a high incidence in southern China, Southeast Asia, and North Africa, and the incidence is increasing year by year [1]. With the improvement of current treatment regimens, the 5‐year survival rate of locally advanced NPC has increased to over 80%, but still about 20%–30% of patients experience local recurrence or distant metastasis [2]. Patients with recurrent/metastatic nasopharyngeal carcinoma (R/M NPC) have a poor prognosis, with a median overall survival (OS) of typically less than 20 months [3]. Currently, it is widely believed that platinum‐based combination chemotherapy is the standard first‐line treatment for R/M NPC [4].

In recent years, several Phase III randomized controlled trials have confirmed that combining PD‐1 inhibitors with chemotherapy can further significantly prolong the progression‐free survival (PFS), which has become a new first‐line standard treatment [5, 6, 7]. However, after the failure of first‐line treatment, the treatment of R/M NPC patients has fallen into a dilemma. Studies have shown that in second‐line and subsequent‐line treatments, the objective response rate (ORR) of single‐agent chemotherapy, PD‐1 inhibitors alone, and anti‐angiogenic drugs alone is only 20%–30%, and the median PFS is less than 5 months [8, 9]. This indicates that more effective treatment strategies are urgently needed.

Anti‐angiogenic small molecule tyrosine kinase inhibitors (TKIs) inhibit the vascular endothelial growth factor receptor (VEGFR) target to cause tumor vessel regression, and can temporarily repair and normalize deformed vessels [10]. This can alleviate hypoxia and reduce interstitial pressure. Immune checkpoint inhibitors block PD‐1/PD‐L1 to disrupt the immune escape of tumor cells and restore the ability of T cells to kill tumor cells. The theoretical basis for the better effect of combining these two is that the normalization of blood vessels opens a channel for efficient infiltration of activated T cells into the tumor, thereby exerting a better anti‐tumor effect, and inhibiting the VEGF signal itself can directly weaken the negative regulation on immune cells, allowing the immune attack to persist. This synergistic mechanism has been verified in multiple solid tumors, such as liver cancer, kidney cancer, and non‐small cell lung cancer [11, 12, 13].

Several single‐arm Phase II clinical studies have explored the efficacy and safety of PD‐1/PD‐L1 inhibitors combined with small‐molecule TKIs (such as apatinib, anlotinib, and famitinib) in R/M NPC. The results showed that the ORR was between 30% and 70%, demonstrating good anti‐tumor activity [14, 15, 16, 17, 18, 19, 20, 21, 22]. However, these studies had small sample sizes, with differences in results, and some studies included large molecule anti‐angiogenic drugs such as bevacizumab, which had different mechanisms and toxicity profiles from oral small molecule TKIs. To date, there is no systematic review and meta‐analysis specifically for the treatment of R/M NPC with “PD‐1/PD‐L1 inhibitors combined with small molecule anti‐angiogenic drugs,” such as apatinib, anlotinib, and famitinib. Therefore, this study aims to systematically evaluate the efficacy and safety of PD‐1/PD‐L1 inhibitors combined with small‐molecule anti‐angiogenic drugs, especially apatinib, anlotinib, and famitinib, in the treatment of R/M NPC, and through subgroup analysis to explore potential factors affecting the efficacy (PD‐1 drug type, TKI type, previous immunotherapy status), in order to provide higher level evidence‐based medical evidence for clinical decision‐making.

2. Methods

2.1. Search Strategy

This study was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) guidelines and was registered in PROSPERO, with the registration number CRD420261279882. Systematic searches were conducted in the following databases: PubMed, Embase, Web of Science, Cochrane Library, China National Knowledge Infrastructure (CNKI), Chinese Biomedical Literature Database (CBM), and Wanfang Data. The search period covered the period from database inception to January 7, 2026. The search terms included “nasopharyngeal carcinoma”, “recurrent”, “metastatic”, “PD‐1”, “PD‐L1”, “immune checkpoint inhibitor”, “apatinib”, “anlotinib”, “famitinib”, “anti‐angiogenic”, and their corresponding Chinese terms. As an example, the search formula was: (nasopharynx* carcinoma* OR NPC) AND (recur* OR metast*) AND (PD‐1 OR PD‐L1 OR camrelizumab OR toripalimab) AND (apatinib OR anlotinib OR famitinib). Additionally, manual searches were conducted for the reference lists of included studies and related conference abstracts (ASCO, ESMO, and CSCO). Details of the search strategy can be found in Appendix S1.

2.2. Inclusion and Exclusion Criteria

2.2.1. Inclusion Criteria

Patients with recurrent or metastatic nasopharyngeal carcinoma, diagnosed by pathology or cytology; intervention measures were PD‐1/PD‐L1 inhibitors combined with small molecule anti‐angiogenic TKIs (apatinib, anlotinib, and famitinib); eligible study designs included prospective single‐arm Phase II trials and retrospective studies (including retrospective cohort and real‐world studies) that evaluated the predefined combination regimen, with a sample size of ≥ 10 cases (a minimum of 10 patients was required to ensure stable estimation of event rates and to minimize the influence of small‐study effects on pooled estimates); at least one core outcome measure such as ORR, DCR, and adverse events could be extracted.

2.2.2. Exclusion Criteria

Reviews, case reports, meta‐analyses, etc., non‐original studies; combination of chemotherapy, large molecule anti‐angiogenic drugs, or other non‐targeted drugs; duplicate data, inability to extract full text, or incomplete data; non‐single‐arm design or sample size < 10 cases.

2.3. Study Selection and Data Extraction

Two researchers independently screened the literature and extracted the data. In case of disagreement, discussions or other determinations were made. The extracted information included the first author, publication year, sample size, baseline patient characteristics, drug regimen, efficacy, and safety data, etc.

2.4. Quality Assessment

This study employed a multi‐tool combination for quality assessment, aiming to comprehensively evaluate the risk of bias for different study types. For the 7 prospective single‐arm studies (8 cohorts: [14, 17, 18, 19, 21, 23], Xi 2023, and [7] (C1) and [20] (C2)), the quality assessment tool developed by the National Institutes of Health (NIH) for “before‐after non‐controlled group studies” was used. This tool consists of 11 items and is applicable for the internal validity assessment of single‐arm intervention studies. Each item is rated as “yes”, “no”, or “unclear”, and the overall quality rating was determined based on the number of “no” responses: 0–2 “no” = low risk of bias/good, 3–8 “no” = moderate risk/fair, and 9–11 “no” = high risk/poor. For the 2 retrospective studies ([15, 16, 23], where [16] is a real‐world retrospective study that only includes data from its NPC subgroup), due to potential selection bias and confounding bias, the ROBINS‐I (Risk Of Bias In Non‐randomized Studies—of Interventions) tool was supplemented for bias risk assessment. The overall risk of bias for the two retrospective studies [15, 16, 23] using ROBINS‐I was rated as serious, primarily due to insufficient control of confounding factors.

All quality assessments were independently completed by two researchers. Disagreements were resolved through discussion or consultation with a third researcher. The NIH scoring results were as follows: the scores of the 7 prospective studies (8 cohorts) ranged from 8 to 10 (out of 11 points), with a mean score of 9.3. The overall quality grade was “moderate” to “good,” with the main deduction items being the lack of sample size calculation in some studies [14, 22] and the absence of blind outcome assessment in all studies (an inherent limitation of single‐arm studies). The ROBINS‐I evaluation results showed that [15, 16, 23] were at “serious risk” in the “confounding bias” and “moderate risk” in the “selection bias” domains. This was mainly due to the insufficient control of prognostic factors (such as the number of treatment lines, liver metastasis rate, EBV DNA level, etc.) and selection bias in the retrospective design. However, both retrospective studies performed well in outcome measurement and data integrity. The detailed quality assessment results can be found in Appendix S2.

2.5. Statistical Analysis

The analysis was conducted using R 4.5.2 software and the meta package. The rates of single‐arm studies were logit‐transformed to stabilize the variance. The generalized linear mixed model (GLMM) was used for pooling (metaprop function, method = “GLMM”), and the τ ² estimate was calculated using the restricted maximum likelihood (REML) method. The pooled effect size and 95% confidence interval (CI) were calculated using the random effects model. Heterogeneity between studies was evaluated using the Cochran Q test and the I ² statistic: I ² < 25% was considered low heterogeneity, 25%–75% was moderate, and > 75% was high heterogeneity. Predefined subgroups included PD‐1 inhibitor types, TKI types, and previous immunotherapy status. Sensitivity analysis included the leave‐one‐out method, conversion between fixed and random effects models, and exclusion of small sample or retrospective studies.

Publication bias was evaluated using funnel plots and Egger's linear regression test. It should be noted that the Egger test has limitations for single‐arm rate Meta‐analysis: the variance of single‐arm rates is related to the rate value, and the asymmetry of the funnel plot may be due to heterogeneity rather than publication bias. Moreover, the test power is low when the number of studies is < 10. Therefore, this study mainly relied on funnel plot assessment, and the Egger test result was only provided as an auxiliary explanation. Zero‐event studies were processed by GLMM based on the binomial distribution log‐likelihood automatically, without using continuity correction. The prediction interval was calculated using the formula of the random effects model and transformed back to the original scale. The Hartung–Knapp adjustment was not used because the metaprop function does not directly support GLMM mode, and the traditional normal approximation is still acceptable when the number of studies is small.

For the pooled analysis of median PFS, the following approach was used. The reported medians and their corresponding 95% confidence intervals (CIs) were extracted from each study. A natural logarithmic transformation was applied to the median PFS values to approximate a normal distribution. The standard error (SE) of the log‐transformed median was derived from the 95% CI assuming symmetry on the log scale, using the formula: SE = [ln(UCI) − ln(LCI)]/(2 × 1.96). Studies that reported only the median without any measure of variability were excluded from this analysis. The log‐transformed medians were pooled using a random‐effects model with restricted maximum likelihood (REML) estimation. The pooled estimate was then back‐transformed to the original scale using exponentiation. This analysis is considered exploratory rather than confirmatory, given the methodological limitations of pooling medians from survival data. A two‐sided p < 0.05 indicates a statistically significant difference.

3. Results

3.1. Study Selection and Study Characteristics

The initial search yielded 96 documents; after removing duplicates, 62 remained. By reading the titles and abstracts, 35 were initially excluded, and 27 full texts were obtained for a re‐screening. Eventually, 9 studies comprising 10 independent cohorts (357 patients in total) were included. The research screening process is shown in Figure 1. It is particularly important to note that in this study, the two cohorts of [20], Cohort 1 and Cohort 2, were treated as independent studies. The main basis was that the patient populations of the two cohorts were completely mutually exclusive. Cohort 1 was platinum‐resistant and immunotherapy‐naive; Cohort 2 was PD‐1 inhibitor‐resistant. Their inclusion criteria, treatment regimens, and outcome indicators were independently reported, and the median follow‐up time was different (23.3 vs. 18.5 months). Although safety data were combined and reported, the efficacy data were completely independent. This handling may introduce a slight correlation, but due to the absence of overlapping patients and independent effect sizes, the actual impact is extremely small. Sensitivity analysis (re‐analyzing after merging the two cohorts into one study) showed that the main conclusion did not change, indicating that the results were robust.

FIGURE 1.

Flow chart of literature selection and inclusion. *The 9 included studies comprised 10 independent cohorts; Yuan 2023 reported two distinct cohorts (Cohort 1 and Cohort 2).

The basic characteristics included in the study are shown in Table 1. Among them, there are 5 items of the combination regimen of camrelizumab and 3 items of the combination regimen of toripalimab; there are 2 items of other PD‐1/PD‐L1 inhibitors; the distribution of TKI types: 5 items of apatinib, 3 items of anlotinib, 1 item of famitinib, and 1 item of mixed TKI [15, 23]. All the studies are prospective or retrospective single‐arm designs, and the quality evaluation results are presented in Appendix S2.

TABLE 1.

Basic characteristics included in the studies.

Study	Study design	Treatment regimen	Median follow‐up (mo)	N	ORR (%)	DCR (%)	1‐year OS rate (%)	Median PFS (mo)	Grade ≥ 3 TRAEs (%)	Hypertension (≥ 3, %)	Hand‐foot syndrome (≥ 3, %)	Nasopharyngeal necrosis (≥ 3, %)
Cai 2023 [14]	Single‐arm Phase II (interim)	Toripalimab + Anlotinib	10.3	30	36.7	90.0	NR	NR	NR	NR	23.3	NR
Gao 2025 [15]	Retrospective cohort	PD‐1 inhibitor + Apatinib/Anlotinib	28.7	65	47.7	90.8	NR	20.8	38.5	6.2	6.2	NR
Jiang 2024 [16] a	Real‐world retrospective	Prior PD‐1 inhibitor + Anlotinib	17.1	11	45.5	100.0	NR	14.3	NR	NR	NR	NR
Ding 2023 [17]	Single‐arm Phase II	Camrelizumab + Apatinib	12.4	58	65.5	86.2	NR	10.4	58.6	19.0	3.4	15.5
Xi 2023	Single‐arm Phase II	Camrelizumab + Famitinib	16.7	18	33.3	77.8	89	7.2	44.4	5.6	5.6	22.2
Zhang 2024 [19]	Single‐arm Phase II	Toripalimab + Anlotinib	17.4	40	37.5	85.0	73.3	9.5	57.5	7.5	22.5	NR
Yuan 2023 (C1) [20] b	Single‐arm Phase II	Camrelizumab + Apatinib	23.3	40	65.0	80.0	82.5	12.6	65.3	27.8	12.5	12.5
Yuan 2023 (C2) [20] b	Single‐arm Phase II	Camrelizumab + Apatinib	18.5	32	34.4	68.8	68.8	4.5	65.3	27.8	12.5	12.5
Lu 2024 [22]	Single‐arm Phase II	Toripalimab + Apatinib	20.5	37	67.6	86.5	82.9	11.6	32.4	8.1	5.4	8.1
Liao 2024 [21]	Single‐arm Phase II	Camrelizumab + Apatinib	16	26	38.5	61.5	50.0	6.0	26.9	0.0	7.7	7.7

Abbreviation: NR, not reported.

^a The median follow‐up time in [16] is reported for the overall population (21 cases); the NPC subgroup was not reported separately. Safety data were reported only for the overall population (21 cases) and were not provided separately for the NPC subgroup; therefore, they are not listed here.

^b The safety data for both cohorts in Yuan 2023 were reported collectively (72 cases). Given the intention to perform a pooled analysis, the incidence rates of grade ≥ 3 TRAEs and specific adverse events were presented as identical for both cohorts.

3.2. Efficacy Analysis

3.2.1. Objective Response Rate (ORR)

All 10 studies reported ORR. The results from the random effects model combination showed that the ORR of PD‐1/PD‐L1 inhibitor combined with small molecule TKI treatment was 48% (95% CI: 39%–57%). There was moderate heterogeneity (I ² = 63.0%, τ ² = 0.1919, p = 0.0039). The prediction interval was [23%–74%]. The forest plot is shown in Figure 2A.

FIGURE 2.

(A) Forest plot (ORR); (B) Forest plot (DCR); (C) Forest plot (1‐year OS rate); (D) Forest plot (median PFS).

3.2.2. Disease Control Rate (DCR)

All 10 studies reported DCR. The combined DCR was 83% (95% CI: 77%–88%). The heterogeneity I ² = 45.5% (τ ² = 0.1938, p = 0.0571), suggesting moderate heterogeneity. The prediction interval was [63%–94%]. The forest plot is shown in Figure 2B.

3.2.3. One‐Year Overall Survival Rate (1‐Year OS Rate)

Six studies reported the 1‐year OS rate, involving a total of 251 patients. The combined analysis used the random effects model. The 1‐year OS rate was 79% (95% CI: 68%–86%). The heterogeneity I ² = 68.5% (τ ² = 0.2816, p = 0.0073), and the prediction interval was [44%–95%]. The forest plot is shown in Figure 2C.

3.2.4. Median Progression‐Free Survival (Median PFS)

Seven studies reported median PFS data, and the combined analysis used the random effects model. The median PFS was 9.95 months (95% CI: 8.32–11.59 months). The heterogeneity among the studies was low (I ² = 29.2%, p = 0.216). The forest plot is shown in Figure 2D.

3.3. Safety Analysis

3.3.1. ≥ 3 Grade Treatment‐Related Adverse Events (TRAEs)

Seven studies reported ≥ 3 Grade TRAEs. The combined incidence rate was 47% (95% CI: 37%–58%). The heterogeneity I ² = 72.9% (τ ² = 0.2241, p = 0.0011), and the prediction interval was [20%–76%]. The forest plot is shown in Figure 3A.

FIGURE 3.

(A) Forest plot (TRAEs); (B) Forest plot (hypertension incidence); (C) Forest plot (hand‐foot syndrome incidence); (D) Forest plot (nasopharyngeal necrosis incidence).

3.3.2. Common Specific ≥ 3 Grade Adverse Events

3.3.2.1. Hypertension

Ten studies ([15, 17, 18, 23], Xi 2023, [7, 19, 24] combined, [21, 22]) reported that the combined incidence rate was 9% (95% CI: 5%–18%), I ² = 63.8%, prediction interval [2%–37%]. The forest plot is shown in Figure 3B. Hand‐foot syndrome: 7 studies reported that the combined incidence rate was 10% (95% CI: 6%–17%), I ² = 59.7%, prediction interval [2%–37%]. The forest plot is shown in Figure 3C. Nasopharyngeal necrosis: 4 studies [17, 18, 20, 22] reported that the combined incidence rate was 14% (95% CI: 8%–18%), I ² = 0% (p = 0.94). The forest plot is shown in Figure 3D. No treatment‐related deaths were reported in any of the included studies. In Xi 2023, two patients developed Grade 3–4 major epistaxis in the setting of grade ≥ 3 nasopharyngeal necrosis; both were successfully managed with nasal packing and vascular embolization, with no fatal outcomes.

3.4. Subgroup Analysis

3.4.1. Grouped by PD‐1 Drug Type

The DCR in the camrelizumab group (from 5 studies) was 76.5% (95% CI: 67.4%–83.7%), and the ORR was 48.9% (95% CI: 35.3%–62.6%). The DCR in the toripalimab group (from 3 studies) was 86.9% (95% CI: 79.1%–92.1%), and the ORR was 47.3% (95% CI: 31.2%–64.0%). There was no statistically significant difference in ORR between the two groups (p > 0.05), while the DCR showed that the toripalimab group was slightly higher. The forest plots are shown in Figure 4A,B.

FIGURE 4.

(A) Forest plot (DCR in camrelizumab and toripalimab); (B) Forest plot (ORR in the camrelizumab and toripalimab); (C) Forest plots (ORR and DCR in apatinib group); (D) Forest plots (ORR and DCR in anlotinib). (E) Forest plots (ORR in ICI naive and experienced group); (F) Forest plots (DCR in ICI naive and experienced group).

3.4.2. Grouped by TKI Type

The ORR in the apatinib group (from 5 studies) was 55.3% (95% CI: 42.3%–67.6%), and the DCR was 78.4% (95% CI: 69.1%–85.5%). The ORR in the anlotinib group (from 3 studies) was 37.1% (95% CI: 26.7%–49.0%), and the DCR was 87.1% (95% CI: 77.1%–93.2%). The ORR in the famitinib group (from 1 study) was 33.3% (95% CI: 14.4%–58.8%), and the DCR was 77.8% (95% CI: 51.9%–92.6%). The ORR of the apatinib group was higher than that of the anlotinib group (p < 0.05), and the DCR was higher in the anlotinib group. Forest plots are shown in Figure 4C,D.

3.4.3. Based on the Previous Immunotherapy Status

The ORR of the ICI‐naive group (5 studies: [17, 18, 20] (C1), [21, 22]) was 60.9% (95% CI: 50.7%–70.2%), and the DCR was 80.4% (95% CI: 70.5%–87.6%). The ORR of the ICI‐treated group (3 studies: [16], Xi 2023, [20] (C2)) was 36.1% (95% CI: 25.1%–48.8%), and the DCR was 79.5% (95% CI: 56.4%–92.1%). The difference in ORR between the two groups was statistically significant (p = 0.0028), suggesting a potentially greater benefit in ICI‐naive patients. However, this finding is based on subgroup analyses of single‐arm studies and should be interpreted as hypothesis‐generating rather than confirmatory. The difference in DCR was not statistically significant (p = 0.92). The forest plots are shown in Figure 4E,F. The summary forest plots of all subgroups are presented in Figure 5.

FIGURE 5.

(A) Forest plot (subgroup analysis of ORR); (B) Forest plot (subgroup analysis of DCR).

3.5. Sensitivity Analysis

3.5.1. Leave‐One‐Out Method

After eliminating each study one by one, the combined ORR range was between 45% and 51%, and the DCR was between 82% and 84%. There was no significant change. Fixed Effects Model: The ORR was 45% (95% CI: 41%–49%), and the DCR was 82% (95% CI: 79%–85%). This was similar to the results of the random effects model. Elimination of the Study with the Smallest Sample: After eliminating [16] (n = 11), the combined ORR was 48% (95% CI: 39%–57%), which was consistent with the main analysis. Sensitivity Analysis Excluding the Retrospective Study with Mixed TKI Use: After eliminating [15, 23] (the only study that included both apatinib and anlotinib as anti‐angiogenic agents), the ORR was 48% (95% CI: 38%–58%), and the results remained stable. The other retrospective study [16] was not excluded separately due to its small sample size (n = 11), which would have limited the interpretability of the sensitivity analysis; however, it was included in the leave‐one‐out sensitivity analysis, which confirmed the robustness of the findings.

3.6. Publication Bias

The funnel plot (Figure 6) shows that the studies are generally symmetrical. The distribution of research points for the ORR indicator is symmetrical, with no significant publication bias; the funnel plot corresponding to the DCR indicator shows certain asymmetry, suggesting the possibility of publication bias. When interpreting the results, caution is required. The Egger linear regression test did not indicate significant publication bias (ORR: t = −1.17, df = 8, p = 0.275; DCR: t = 1.28, df = 8, p = 0.236). The imputation method analysis did not add missing studies, and the combined effect size remained unchanged. However, given the limited number of included cohorts (n = 10) and the consequently low power of funnel plot assessment and Egger's test in this setting, these results should be interpreted cautiously and considered as descriptive rather than definitive evidence for the absence of publication bias.

FIGURE 6.

ORR and DCR funnel chart.

3.7. Quality Evaluation

Quality assessment was performed using different tools according to study design. For the 7 prospective single‐arm studies (8 cohorts), the NIH tool (11 items) yielded scores ranging from 8 to 10 (out of 11), with a mean score of 9.3, indicating overall good quality. The most frequent limitations were lack of sample size justification [14, 22] and absence of blinded outcome assessment, which is inherent to open‐label single‐arm designs. For the 2 retrospective studies [15, 16, 23], the ROBINS‐I tool was used. Both studies were rated as having a serious overall risk of bias, primarily due to inadequate control for confounding and selection bias. Nevertheless, both studies performed well in terms of outcome measurement and data completeness. The separate assessment results for the two study types are presented in detail in Appendix S2. This distinction ensures that the numerical NIH scores are not conflated with the ROBINS‐I risk categories.

4. Discussion

This study is the first single‐arm meta‐analysis specifically targeting the combination of PD‐1/PD‐L1 inhibitors with small‐molecule anti‐angiogenic drugs (TKIs) for the treatment of recurrent/metastatic nasopharyngeal carcinoma. A total of 9 studies comprising 10 independent cohorts (357 patients) were included. Our main findings are that the objective response rate (ORR) of the combination therapy is 48%, the disease control rate (DCR) is 83%, the median progression‐free survival (PFS) is 9.95 months, and the 1‐year overall survival (OS) rate is 79%; the incidence of Grade 3 or higher treatment‐related adverse events (TRAEs) is 47%, and the common serious adverse events include hypertension (9%), hand‐foot syndrome (10%), and nasopharyngeal necrosis (14%); subgroup analysis shows that the ORR of the apatinib combination regimen is higher than that of anlotinib (55.3% vs. 37.1%, p < 0.05), while the DCR of the toripalimab combination regimen is superior to that of camrelizumab (86.9% vs. 76.5%); patients who had not received immunotherapy (ICI first‐line treatment) have a higher ORR than those who had received ICI treatment (60.9% vs. 36.1%, p = 0.0028). Overall, the combination therapy shows promising anti‐tumor activity in the studied populations. However, given the non‐comparative, single‐arm design of the included studies, these findings should be considered exploratory and hypothesis‐generating rather than confirmatory. Well‐designed randomized controlled trials are needed to validate the efficacy and safety of this strategy relative to standard regimens.

This study is largely consistent with the results of previously published similar meta‐analyses. Gao et al. [23] reported an ORR of 49%, DCR of 80%, and 1‐year OS of 76%, validating the overall efficacy of the combination therapy. However, it is worth noting that this study strictly limits the anti‐angiogenic drugs in the combination to small‐molecule tyrosine kinase inhibitors (TKIs), excluding large‐molecule antibodies (such as bevacizumab). This choice is mainly based on methodological considerations: small‐molecule TKIs and large‐molecule anti‐VEGF antibodies have significant differences in mechanism of action, administration route, pharmacokinetic characteristics, and toxicity profile [25, 26], and conflating the two may introduce unignorable clinical and methodological heterogeneity. Additionally, in real clinical practice, oral small‐molecule TKIs are more widely used in maintenance treatment and subsequent lines of treatment for advanced tumors due to their convenience and compliance advantages, and independent evaluation of their efficacy has greater practical clinical significance [27, 28, 29]. Small‐molecule TKIs can simultaneously inhibit multiple tyrosine kinase receptors (such as VEGFR, PDGFR, FGFR, etc.), theoretically having broader anti‐tumor and immunomodulatory effects, but this assumption still requires more research to verify [30]. Therefore, this study specifically focuses on small‐molecule TKIs, aiming to provide more precise evidence‐based evidence for the clinical application of these drugs.

Our results show that the ORR of the combination therapy (48%) appears higher than historical data for PD‐1 inhibitors alone (20%–34%) or anti‐angiogenic drugs alone (17%–36%), suggesting a potential synergistic anti‐tumor effect. However, this comparison is indirect and based on historical benchmarks rather than head‐to‐head evidence, and should therefore be interpreted with caution. From a theoretical perspective, this synergistic effect can be explained from multiple perspectives. The PD‐1/PD‐L1 axis plays a central role in tumor immune evasion in NPC. PD‐1 is an inhibitory receptor expressed on activated T cells; its engagement with PD‐L1 on tumor cells delivers a co‐inhibitory signal that suppresses T‐cell proliferation, cytokine production, and cytotoxic function. In NPC, PD‐L1 is frequently overexpressed, rendering this pathway a critical immune escape mechanism and a rational therapeutic target for combination strategies. As mentioned earlier, tumor microenvironment remodeling is the most direct mechanism, which is related to the mechanism of the drugs. Anti‐angiogenic drugs can promote the normalization of tumor vessels, improve hypoxia, reverse immunosuppression, increase CD8+ T cell infiltration, while immune checkpoint inhibitors disrupt the immune escape of tumor cells to restore the killing of tumor cells by T cells [31]. Another theory—the tumor ecological evolution theory—provides a more macroscopic perspective. Tumors are not isolated cell clusters, but a dynamically evolving ecosystem where cancer cells, blood vessels, immune cells, and stromal components compete and co‐evolve. In this system, anti‐angiogenic therapy can alter the evolutionary pressure of tumors, inhibit the expansion of angiogenesis‐dependent clones, while immunotherapy can eliminate less immunogenic subclones. The combination of the two may more comprehensively constrain the adaptive evolution of tumors and delay the emergence of drug resistance [32]. Additionally, the metabolic reprogramming aspect can also be discussed. Anti‐angiogenic therapy can improve tumor hypoxia and reduce lactate accumulation, thereby alleviating the metabolic inhibition on T‐cell function and enhancing the efficacy of immunotherapy [24]. These theoretical perspectives provide possible explanations for the superior efficacy of combined therapy in patients who have received initial treatment with ICI (ORR 60.9% vs. 36.1%, p = 0.0028), but further basic and clinical studies are still needed to confirm them.

The ORR of the apatinib combination regimen was higher than that of anlotinib (55.3% vs. 37.1%, p < 0.05) in this pooled analysis. However, this finding is derived from subgroup analyses of single‐arm studies and may be influenced by confounding factors such as differences in baseline characteristics, treatment lines, and study populations across the included studies [33, 34, 35]. It should therefore be considered hypothesis‐generating rather than demonstrating superiority of one TKI over another. Similarly, the higher ORR observed in ICI‐naive patients compared with ICI‐experienced patients (60.9% vs. 36.1%, p = 0.0028) suggests that the combined treatment may more effectively activate the anti‐tumor immune response in the immunotherapy‐naive population, but this observation requires confirmation in prospective comparative studies.

This meta‐analysis included original studies that provided valuable clues in the exploration of biomarkers. From the perspective of the included studies, the original research included in this meta‐analysis provided empirical support for the above theory in terms of biomarker exploration. Ding, Zhang, et al. [17] as the first global prospective evaluation of the combination of camrelizumab and apatinib achieved an ORR of 65.5% and a DCR of 86.2%, making it the most outstanding study in terms of efficacy in this analysis, laying the clinical value foundation for this combination regimen. Yuan et al. [20] discovered through RNA sequencing and multiple immunofluorescence techniques that in platinum‐resistant (ICI initial treatment) patients, B‐cell molecular markers and tertiary lymphoid structures (TLS) were the most distinctive features for differentiating responders from non‐responders, and the presence of TLS was significantly correlated with ORR, while in patients resistant to PD‐1 inhibitors, the expression characteristics of angiogenesis genes were strongly related to efficacy. This discovery precisely revealed the immune ecological differences of different treatment contexts in the population, forming a perfect mutual confirmation with our subgroup results—the ORR of ICI initial treatment patients was significantly higher than that of treated patients. Zhang et al. further expanded the dynamic monitoring perspective, through continuous monitoring of plasma circulating tumor DNA (ctDNA), found that the blood tumor mutation burden in the early stage of treatment was significantly correlated with treatment response and PFS, and the “significant increase” in ctDNA level was 2.3 months earlier than the median time of imaging progression, indicating disease progression. These exploratory analyses provided molecular‐level evidence for the mechanism of combined treatment and offered a reference for the design of biomarker‐driven clinical trials in the future.

In terms of safety, the incidence of ≥ 3 grade TRAEs in the combined treatment group was 47%, higher than that in the single‐agent treatment group, but most adverse events could be managed through dose adjustment or symptomatic treatment. Nasopharyngeal necrosis was a serious adverse event worthy of attention in this combined treatment regimen. In this analysis, nasopharyngeal necrosis was defined as a grade ≥ 3 adverse event, and the original studies all used the CTCAE standard for grading, with basically consistent definitions. The combined incidence rate was 14%, with no heterogeneity among studies (I ² = 0%), suggesting a relatively stable risk. Risk factors included local recurrence, re‐radiation, tumor invasion of major blood vessels, etc. In Xi 2023, two patients developed Grade 3–4 major epistaxis in the setting of nasopharyngeal necrosis; both were successfully managed with nasal packing and vascular embolization, and no fatal outcomes were reported across all included studies [18]. Therefore, although the overall safety profile is generally acceptable, it is necessary to emphasize that high‐risk patients—particularly those with prior radiotherapy or local recurrence—should be screened before treatment through nasal endoscopy and imaging, and close monitoring should be carried out during treatment to enable timely intervention for potential bleeding complications.

It is undeniable that this study has the following limitations: all included studies were single‐arm and without control, and the evidence level was lower than that of randomized controlled trials; although the total sample size reached 357 cases, the sample size of each study was small (11–72 cases), and the results of some subgroups were unstable; some outcome indicators had high heterogeneity, possibly due to differences in drugs, doses, populations, and follow‐up times; although the Egger test was not significant (ORR p = 0.275, DCR p = 0.236), the number of studies was only 10, and the interpretation of the funnel plot should be cautious; the median PFS was based on the log‐normal assumption, and the methodological approach was non‐standard, and the conclusion should be regarded as exploratory; potential prognostic factors such as age, gender, EBV DNA level, and tumor location were not adequately controlled. Therefore, the conclusion of this study should be regarded as an exploratory hypothesis. Future III‐phase randomized controlled trials are urgently needed to verify it, and predictive biomarkers should be explored, and large‐sample real‐world studies should be conducted to further evaluate long‐term survival benefits and rare adverse events.

5. Conclusion

In conclusion, the combination of PD‐1/PD‐L1 inhibitors with small‐molecule anti‐angiogenic drugs for recurrent/metastatic nasopharyngeal carcinoma shows promising efficacy and a generally acceptable safety profile in the studied populations, especially for patients who have not previously received immunotherapy. However, the notable incidence of nasopharyngeal necrosis (14%) and its potential for severe bleeding complications warrant careful patient selection and close monitoring. These findings support further investigation in well‐designed randomized controlled trials. Although the current evidence mainly comes from single‐arm studies, this research provides the highest level of evidence‐based medical evidence for clinical decision‐making to date. Further verification is needed through large‐scale randomized controlled trials in the future.

Author Contributions

Rongyi Hu: writing – original draft, project administration, conceptualization, methodology, writing – review and editing. Tiebin Li: writing – original draft, writing – review and editing, visualization, formal analysis. Ziman Luo: investigation, validation, software, data curation. Huining Chen: project administration, resources, supervision, funding acquisition. Song Qu: writing – review and editing, project administration, resources, supervision, funding acquisition.

Funding

This study is supported by grants from The Key Laboratory of Early Prevention and Treatment for Regional High Frequency Tumor (Guangxi Medical University), Ministry of Education (GKE‐ZZ202506) the Key Research and Development Program Project of Guangxi Zhuang Autonomous Region (grant number Guike AB25069065); the National Natural Science Foundation of China (82460536); and the General Program of Guangxi Natural Science Foundation (2024GXNSFAA010398).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Appendix S1: Search strategy.

Appendix S2. Study quality assessment results.

Data Availability Statement

Data sharing was not applicable to this article as no datasets were generated or analyzed during this study.