Development and validation of prognostic nomograms based on peripheral blood T-cell counts and their dynamic changes for patients with nasopharyngeal carcinoma receiving radiotherapy

Highlights

• •Nomograms successfully predict NPC survival using T-cell dynamics post-radiotherapy.
• •Baseline and Δ T-cell counts are key prognostic factors for OS and PFS.
• •The proposed nomograms outperform traditional TNM staging and EBV-DNA-based models.
• •The nomograms' generalizability has been validated in external dataset.
• •High-risk patients can benefit from tailored, aggressive treatment strategies.

Abstract

Aim

To develop and validate prognostic nomograms incorporating peripheral blood lymphocyte counts and their dynamic changes in patients with nasopharyngeal carcinoma (NPC) undergoing radiotherapy.

Methods

In this retrospective cohort study, consecutive patients with NPC who received radiotherapy at Chongqing University Cancer Hospital were included as the internal cohort (randomly divided 70 %/30 % for training and validation), while patients treated at Guangxi Medical University Cancer Hospital served as the external validation cohort. Prognostic nomograms for progression-free survival (PFS) and overall survival (OS) were constructed using multivariable Cox regression analyses.

Results

The internal and external cohorts comprised 746 and 138 patients, respectively. Age, gross tumor volume dose, neoadjuvant chemotherapy, clinical stage, plasma EBV-DNA level, baseline total T-cell count, and its post-treatment change (ΔT cells) were identified as independent prognostic factors. The nomograms demonstrated strong predictive performance, with concordance indices of 0.701, 0.716, and 0.714 for PFS, and 0.759, 0.735, and 0.734 for OS in the training, internal validation, and external datasets, respectively. The areas under the receiver operating characteristic curves for 3-year and 5-year PFS and OS exceeded 0.7 across all cohorts. Calibration plots indicated good agreement between predicted and observed outcomes, and decision curve analysis confirmed greater net clinical benefit compared with TNM staging and EBV-DNA-based models.

Conclusion

The proposed T-cell-based nomograms reliably predict PFS and OS in patients with NPC undergoing radiotherapy and were validated in an external cohort. These models provide improved prognostic discrimination beyond conventional staging systems and may assist in individualized risk stratification and management planning.

Introduction

Nasopharyngeal carcinoma (NPC) arises from the epithelial lining of the nasopharynx, most commonly in the fossa of Rosenmüller [[1], [2], [3]]. Epstein-Barr virus (EBV) infection is the principal etiological factor, particularly for the undifferentiated non-keratinizing subtype prevalent in endemic regions [1,2,4,5]. In 2022, an estimated 120,416 new cases and 73,476 deaths from NPC occurred worldwide, accounting for 0.6 % of all cancer diagnoses and 0.8 % of cancer mortality [6]. Although NPC is highly radiosensitive and early-stage disease is often curable with modern multimodal therapy [5,7], marked interpatient variability in clinical outcomes remains, emphasizing the need for more refined prognostic tools to support individualized management and improve survival.

Lymphocytes have long been recognized as active participants in the development and progression of solid tumors and as important prognostic indicators, including in NPC [8,9]. Regulatory T cells (Tregs) exert immunosuppressive effects that are essential for maintaining immune homeostasis but can be co-opted by tumors to evade immune surveillance. Elevated circulating Treg levels have been reported in patients with untreated or relapsed NPC [10]. Previous studies have also highlighted the prognostic relevance of specific lymphocyte subsets: a higher circulating CD4/CD8 ratio was associated with improved distant metastasis-free survival (DMFS) [11]; low baseline CD4⁺ T-cell counts predicted poorer outcomes [12]; and in patients with EBV-DNA levels exceeding 1500 copies/mL, higher proportions of CD3⁺CD8⁺ T cells correlated with better overall survival (OS) during concurrent chemoradiotherapy (CCRT) [13]. Despite these insights, existing prognostic models incorporating lymphocyte parameters remain suboptimal and lack comprehensive validation.

Furthermore, lymphocyte counts are dynamic biomarkers that fluctuate in response to disease progression, therapeutic interventions, comorbid conditions, and overall patient status [14,15]. Previous studies have shown that CCRT induces a decline in peripheral lymphocyte counts; however, patients who maintained higher levels of circulating CD3⁺CD8⁺ T cells achieved superior OS [13]. In patients with head and neck cancers, including NPC, a greater reduction in the lymphocyte-to-monocyte ratio following radiotherapy has been linked to poorer OS [16]. Weng et al. further reported that low baseline CD8⁺ T-cell levels predicted unfavorable outcomes, whereas a greater post-radiotherapy decline in CD4⁺ T cells was associated with improved prognosis [17]. Although available evidence remains limited, these findings collectively suggest that dynamic alterations in lymphocyte profiles may serve as critical prognostic indicators in patients with NPC undergoing radiotherapy.

Several recent studies have proposed prognostic models that incorporate lymphocyte-based parameters in NPC. For example, Yan et al. developed a nomogram integrating CD8⁺ T-cell counts and the platelet-to-lymphocyte ratio, which demonstrated good predictive accuracy for OS (C-index, 0.724) [18]. Another model combined apparent diffusion coefficient values with extranodal extension and the lymphocyte-to-monocyte ratio, achieving a C-index of 0.795 [19]. Although these models exhibit promising discriminatory performance, they predominantly rely on static clinicopathologic variables and baseline hematologic indices. A major limitation of existing approaches is the absence of robust external validation and, more critically, the omission of dynamic immune markers. Therefore, the present study sought to develop and validate prognostic nomograms incorporating temporal changes in peripheral lymphocyte profiles among patients with NPC undergoing radiotherapy. These models aim to enhance prognostic precision, facilitate individualized risk stratification, and inform clinical decision-making in radiotherapy management.

Methods

Study design and patients

This retrospective cohort study was conducted at two medical centers. The internal cohort comprised consecutive patients with NPC who received radiotherapy at Chongqing University Cancer Hospital between May 1, 2012, and December 31, 2020, and was used for model development and internal validation. External validation was performed using an independent cohort of consecutive patients with NPC treated with radiotherapy at Guangxi Medical University Cancer Hospital between January 1, 2016, and December 31, 2020. Details of treatment protocols, including radiotherapy, chemotherapy, and adjunctive therapies, as well as methods for lymphocyte quantification, have been described previously by the authors [17]. Both centers used BD flow cytometry platforms (FACSCalibur and FACSCanto) following identical quality control procedures and employing the BD Multitest 6-Color TBNK Reagent (Becton, Dickinson and Company), with standardized fluorescent labeling and clone configurations.

Eligible patients were required to have histologically confirmed NPC; be between 18 and 85 years of age; have an Eastern Cooperative Oncology Group (ECOG) performance status score of 0–2; and have received definitive radiotherapy encompassing the nasopharynx and cervical lymphatic drainage regions. Availability of complete data on peripheral blood cell counts, lymphocyte subset profiles, and plasma EBV-DNA levels within 30 days before treatment initiation and 21–30 days after completion of radiotherapy was also required for inclusion.

Patients were excluded if they had severe underlying cardiac or pulmonary disease; experienced acute infections during treatment; had concurrent malignant tumors, hematopoietic disorders, splenomegaly with hypersplenism, acquired immunodeficiency syndrome, active autoimmune disease, or bone metastases; or had received prior radiotherapy to the head and neck, thymic, splenic, or iliac regions. Additional exclusion criteria included oral or intravenous glucocorticoid use for ≥2 weeks during treatment (equivalent to a prednisone dose of ≥10 mg per day); receipt of more than three cycles of induction or adjuvant chemotherapy; administration of more than three cycles of triweekly cisplatin or seven doses of weekly cisplatin during CCRT; and use of granulocyte colony-stimulating factors or thymosin within 14 days prior to peripheral blood sampling.

This study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Ethics Committees of Chongqing University Cancer Hospital (approval no. CZLS2023323-A) and Guangxi Medical University Cancer Hospital (approval no. KYB2024014). Given the retrospective design and the anonymized nature of the data, the requirement for informed consent was waived by both committees.

Follow-up and outcomes

In accordance with institutional practice during the study period, patients were followed up every 1–3 months during the first year after treatment, every 3–6 months in the second year, every 6–12 months from the third to fifth years, and annually thereafter. Progression-free survival (PFS) was defined as the interval from diagnosis to the first documented disease progression, death from any cause, or last follow-up. OS was defined as the interval from diagnosis to death from any cause or last follow-up.

Construction and validation of the prognostic models

Prognostic models were developed and validated for both PFS and OS. The internal cohort was randomly divided into a training set (70 %) and an internal validation set (30 %). Candidate prognostic variables were first evaluated using univariable Cox proportional hazards regression analyses. Variables with a P value <0.05, along with established clinical prognostic factors (including sex, age, tumor stage, EBV-DNA copy number, gross tumor volume (GTV) dose, and neoadjuvant chemotherapy) were subsequently entered into multivariable Cox regression analyses. The final models were selected based on the lowest Akaike Information Criterion (AIC) value to achieve optimal fit. The proportional hazards assumption was rigorously assessed globally and for each covariate in the final multivariable model using Schoenfeld residuals (via the cox.zph function in survival package).

The C-index was calculated to assess model discrimination. Receiver operating characteristic (ROC) curves for 1-, 3-, and 5-year PFS and OS were generated, and the corresponding areas under the curve (AUCs) were used to quantify predictive accuracy. Calibration curves were constructed to evaluate the agreement between predicted and observed outcomes. Patients were stratified into high- and low-risk groups according to their nomogram-derived risk scores, and survival differences between groups were analyzed using the Kaplan-Meier method. Decision curve analysis (DCA) was performed to evaluate the net clinical benefit of the models. For external validation, risk scores were computed for each patient in the external dataset based on the established nomograms, and the same analytic procedures were applied to assess model performance.

Sample size

To ensure the robustness of the final prognostic models, the study required a minimum of 20 outcome events per variable included in the model. With an anticipated seven independent predictors in the nomograms, at least 140 events were necessary. Based on prior literature and an average follow-up duration of approximately 5 years, the expected event rates were about 20 % for OS [20,21] and 26 % for PFS [22,23], corresponding to required sample sizes of at least 700 and 528 patients, respectively. In addition, sample size estimation using the method proposed by Riley et al. (with an allowable outcome event proportion error ≤5 %, expected shrinkage <10 %, and target R² = 0.1) indicated that at least 565 and 595 patients were needed for OS and PFS modeling, respectively [24]. Integrating both approaches, the minimum sample size for this study was conservatively set at 700 patients.

Statistical analysis

Missing data were imputed using mean substitution. The Kolmogorov-Smirnov test was applied to assess the normality of continuous variables. Variables following a normal distribution were analyzed using analysis of variance, whereas non-normally distributed variables were compared using the Kruskal-Wallis test. Categorical variables were analyzed using the chi-square test. Survival curves were generated using the Kaplan-Meier method, and univariable and multivariable Cox proportional hazards models were fitted with the “survival” package in R. Nomograms were constructed using the “rms” package, and graphical visualization was performed with “ggplot2.” Model internal validation was performed with bootstrap resampling (1000 iterations) and ten-fold cross validation. The “rmsr” and “rmda” packages were used to calculate and plot the C-index, ROC curves, calibration plots, and DCA. All statistical tests were two-sided, and P values <0.05 were considered statistically significant. Analyses were conducted using R software, version 4.3.2 (R Foundation for Statistical Computing, Vienna, Austria; https://cran.r-project.org/bin/windows/base/).

Results

Patient characteristics

A total of 874 patients with NPC were screened for inclusion in the internal cohort, of whom 746 met the eligibility criteria. These patients were randomly assigned to the training set (n = 523) and the internal validation set (n = 223). The external validation cohort comprised 138 eligible patients, selected from 677 screened cases. In total, 884 patients were included in the final analysis.

The median age of the study population was 49 years (range, 18–82), and 71.5 % were male. Most patients had an ECOG performance status score of 0 or 1 (94.7 %), and the predominant histologic subtype was non-keratinizing squamous cell carcinoma (77.6 %). Baseline characteristics were comparable across the training, internal validation, and external validation cohorts (all P > 0.05), except for sex distribution (Table 1). Across all three datasets, total T-cell and CD4⁺ T-cell counts declined after treatment. The CD8⁺ T-cell count decreased in the training and external validation cohorts but increased in the internal validation cohort (Table 1).

Table 1.

Characteristics of the patients in the training, internal validation, and external validation datasets.

Factor	Training dataset (n = 523)	Internal validation dataset (n = 223)	External validation dataset (n = 138)	P
Age (years)	49 (16)	50 (13)	45 (12)	0.874
Sex				0.019
Female	145 (28.0)	81 (36.0)	26 (19.0)
Male	378 (72.0)	142 (64.0)	112 (81.0)
ECOG score				0.757
0–1	489 (93.5)	210 (94.2)	138 (100.0)
2	34 (6.5)	13 (5.8)	0 (0.0)
Pathological type				0.789
Non-keratinizing carcinoma	129 (24.7)	55 (24.7)	21 (15.2)
Keratinizing squamous cell carcinoma	382 (73.0)	167 (74.9)	116 (84.1)
Basaloid squamous cell carcinoma	12 (2.3)	1 (0.4)	1 (0.7)
Clinical stage				0.597
I-II	125 (24.0)	54 (24.0)	24 (17.0)
III-IV	398 (76.0)	169 (76.0)	114 (83.0)
Neoadjuvant chemotherapy				0.861
Yes	259 (50.0)	112 (50.0)	25 (18.1)
None	264 (50.0)	111 (50.0)	113 (81.9)
GTV dose (Gy)	70.4 (0.0)	70.4 (0.0)	72.32 (2.0)	0.697
EBV-DNA (copies/ml)	500 (1830)	500 (2088)	722 (625)	0.806
Baseline total T cell count	1046.0 (384.3)	640.4 (119.2)	1049.5 (435.1)
Post-treatment total T cell count	472.7 (206.6)	628.8 (122.6)	497.5 (249.0)
Change (Δ) in total T cell count	573.3 (377.2)	11.6 (128.9)	552.0 (436.4)
Baseline CD4+ T cell count	585.9 (243.6)	336.3 (94.1)	580.6 (256.7)
Post-treatment CD4+ T cell count	180.9 (84.8)	233.4 (108.7)	186.5 (106.9)
Change (Δ) in CD4+ T cell count	405.1 (223.9)	102.9 (114.6)	394.2 (254.5)
Baseline CD8+ T cell count	395.1 (178.5)	234.7 (73.1)	408.3 (214.5)
Post-treatment CD8+ T cell count	253.0 (138.3)	307.0 (113.3)	276.1 (163.8)
Change (Δ) in CD8+ T cell count	142.1 (186.9)	−72.2 (104.6)	132.1 (212.7)

ECOG: Eastern Cooperative Oncology Group; GTV: gross tumor volume; EBV: Epstein-Barr virus.

Survival

The mean follow-up duration was 55.6 months (range, 4.5–142.9). Among the 884 patients, 245 deaths occurred, yielding a median OS of 101.1 months (95 % confidence interval [CI], 95.9–106.0). A total of 352 PFS events were recorded, with a median PFS of 93.0 months (95 % CI, 86.2–99.8). In the internal cohort, the median OS and PFS were not reached; the 3- and 5-year OS rates were 93.7 % and 84.4 %, respectively, and the corresponding PFS rates were 76.5 % and 68.8 %. In the external validation cohort, the median OS and PFS were 81.0 months (95 % CI, 76.1–85.9) and 80.0 months (95 % CI, 76.1–83.8), respectively. The 3- and 5-year OS rates were 91.3 % and 74.4 %, and the 3- and 5-year PFS rates were 88.4 % and 79.7 %, respectively.

Exploration of the prognostic factors

Univariable Cox regression analyses in the training cohort demonstrated that baseline CD4⁺, CD8⁺, and total T-cell counts, their changes from baseline to post-treatment (Δ), and post-treatment CD8⁺ T-cell counts were significantly associated with survival outcomes. These variables, along with established clinical prognostic factors (including sex, age, tumor stage, plasma EBV-DNA level, GTV dose, and neoadjuvant chemotherapy) were incorporated into multivariable Cox regression models.

As shown in Table 2, both baseline total T-cell count (PFS: hazard ratio [HR], 0.600; 95 % CI, 0.456–0.789; P < 0.001; OS: HR, 0.639; 95 % CI, 0.425–0.959; P = 0.031) and Δ total T-cell count (PFS: HR, 0.650; 95 % CI, 0.494–0.855; P = 0.002; OS: HR, 0.646; 95 % CI, 0.431–0.969; P = 0.035) were identified as independent protective factors. For PFS, Δ CD4⁺ T-cell count, baseline CD8⁺ T-cell count, and Δ CD8⁺ T-cell count were also protective, whereas post-treatment CD8⁺ T-cell count was an independent risk factor. For OS, both baseline and Δ CD4⁺ T-cell counts were independent protective predictors, while CD8⁺ T-cell parameters showed no significant association. The proportional hazards assumption was verified for all models using Schoenfeld residuals, with no significant violations detected globally or for any covariate (all P > 0.05; Supplementary Figure S1).

Table 2.

Multivariable analysis results for PFS and OS.

	PFS		OS
	HR (95 %CI)	P value	HR (95 %CI)	P value
Age	1.025 (1.007–1.042)	0.005	1.025 (1.008–1.042)	0.004
Sex	1.410 (1.021–1.948)	0.037	1.565 (0.962–2.548)	0.071
ECOG score	1.477 (0.983–2.218)	0.060	1.646 (0.934–2.901)	0.085
Clinical stage	1.949 (1.635–2.323)	<0.001	2.334 (1.823–2.989)	<0.001
Neoadjuvant chemotherapy	0.580 (0.391–0.865)	0.009	0.543 (0.364–0.809)	0.003
GTV dose	0.966 (0.953–0.980)	<0.001	0.979 (0.961–0.997)	0.020
EBV-DNA copies	1.385 (1.047–1.834)	0.023	1.624 (1.084–2.433)	0.019
Baseline total T cell count	0.600 (0.456–0.789)	<0.001	0.639 (0.425–0.959)	0.031
Post-treatment total T cell count	1.186 (0.907–1.551)	0.214	1.661 (0.799–3.453)	0.175
Change in total T cell count (Δ)	0.650 (0.494–0.855)	0.002	0.646 (0.431–0.969)	0.035
Baseline CD4+ T cell count	0.811 (0.618–1.065)	0.132	0.708 (0.470–0.966)	0.038
Post-treatment CD4+ T cell count	1.046 (0.800–1.367)	0.744	0.780 (0.441–1.380)	0.394
Change in CD4+ T cell count (Δ)	0.642 (0.487–0.848)	0.002	0.574 (0.378–0.872)	0.009
Baseline CD8+ T cell count	0.651 (0.495–0.857)	0.002	0.738 (0.493–1.104)	0.139
Post-treatment CD8+ T cell count	1.350 (1.031–1.768)	0.029	0.819 (0.427–1.572)	0.549
Change in CD8+ T cell count (Δ)	0.701 (0.533–0.920)	0.011	0.733 (0.493–1.090)	0.129

PFS: progression-free survival; OS: overall survival; HR: hazards ratio; CI: confidence interval; ECOG: Eastern Cooperative Oncology Group; GTV: gross tumor volume; EBV: Epstein-Barr virus.

Construction and validation of the nomograms

Nomograms predicting PFS and OS were constructed based on the multivariable Cox regression models (Fig. 1). The full multivariable Cox regression coefficients and nomogram scoring formulas for both the PFS and OS models in Supplementary Table S1. The C-indexes for the PFS nomogram were 0.701, 0.716, and 0.714 in the training, internal validation, and external validation cohorts, respectively. Corresponding C-indexes for the OS nomogram were 0.759, 0.735, and 0.734 (Supplementary Table S2).

Fig. 1.

Prognostic Nomograms. (A) Nomogram for progression-free survival (PFS). (B) Nomogram for overall survival (OS). Use of the nomogram: according to the specific values of each risk factor, the corresponding points in located on the nomogram and the total score is calculated by adding up the points for all risk factors. A vertical line is drawn between the probability scales of the time points corresponding to the total score, and the intersection of this vertical line with the risk probability axis represents the predicted survival probability for the patient.

As shown in Fig. 2, in the training cohort, the AUCs for 1-, 3-, and 5-year PFS were 0.728, 0.732, and 0.750, respectively (Fig. 2A), and those for OS were 0.836, 0.800, and 0.790, respectively (Fig. 2B). In the internal validation cohort, AUCs for both PFS (Fig. 2C) and OS (Fig. 2D) approximated 0.75, while in the external validation cohort, AUCs for both endpoints were approximately 0.7 (Fig. 2E and 2F).

Fig. 2.

Receiver operating characteristics (ROC) curves of the progression-free survival (PFS) and overall survival (OS) nomograms. (A) ROC curve and area under the curve (AUC) of the PFS nomogram in the training dataset. (B) ROC curve and AUC of the OS nomogram in the training dataset. (C) ROC curve and AUC of the PFS nomogram in the internal validation dataset. (D) ROC curve and AUC of the OS nomogram in the internal validation dataset. (E) ROC curve and AUC of the PFS nomogram in the external validation dataset. (F) ROC curve and AUC of the OS nomogram in the external validation dataset.

Calibration curves for the 1-, 3-, and 5-year PFS and OS models in all three cohorts closely followed the 45° reference line, confirming excellent agreement between predicted and observed outcomes (Supplementary Figure S2). DCA further demonstrated that, across a broad range of threshold probabilities, the nomograms provided greater net clinical benefit than the “treat all” or “treat none” strategies (Supplementary Figure S3).

Risk scores were computed for each patient based on the nomogram formulas, and individuals were stratified into high- and low-risk groups according to the median risk score within each cohort. Kaplan-Meier analyses showed that both PFS and OS were significantly longer in the low-risk groups compared with the high-risk groups (Fig. 3).

Fig. 3.

Survival curves for low-risk and high-risk groups identified by the nomograms. (A) Progression-free survival (PFS) survival curves for the low- and high-risk groups in the training dataset. (B) PFS survival curves for the low- and high-risk groups in the internal validation dataset. (C) PFS survival curves for the low- and high-risk groups in the external validation dataset. (D) Overall survival (OS) survival curves for the low- and high-risk groups in the training dataset. (E) OS survival curves for the low- and high-risk groups in the internal validation dataset. (F) OS survival curves for the low- and high-risk groups in the external validation dataset.

Comparison of the nomograms with models containing canonical prognostic factors

TNM staging remains the standard prognostic framework for NPC, and plasma EBV-DNA copy number has recently emerged as an additional prognostic biomarker [5]. To evaluate the incremental value of the present nomograms, their performance was compared with models based on TNM staging alone and on TNM staging combined with EBV-DNA level. As shown in Supplementary Table S2, the C-indexes for predicting PFS and OS were 0.669 and 0.689, respectively, for the TNM model, and 0.686 and 0.726, respectively, for the TNM plus EBV-DNA model, both lower than those achieved by the proposed nomograms. Similarly, the AUCs for the two conventional models (Supplementary Figures S4 and S5) were inferior to those of the T-cell-based nomograms. DCA further demonstrated that, across a range of threshold probabilities, the nomograms consistently yielded greater net clinical benefit than the TNM- or EBV-DNA-based models (Supplementary Figure S6). These findings indicate that incorporating dynamic T-cell metrics provides superior prognostic discrimination and clinical utility.

Discussion

This retrospective cohort study developed and externally validated prognostic nomograms that integrate dynamic changes in peripheral blood lymphocyte counts to predict clinical outcomes in patients with NPC undergoing radiotherapy. Both baseline T-cell levels and post-treatment variations were independently associated with PFS and OS. The resulting nomograms demonstrated strong predictive performance and effectively stratified patients according to risk, offering a practical tool to support individualized management and treatment optimization.

Although CD8⁺ T cells are principal effector lymphocytes in antitumor immunity, our findings do not establish a definitive association between peripheral blood CD8⁺ T-cell counts measured at a single post-treatment time point and OS. The interpretation of post-radiotherapy CD8⁺ T-cell levels is confounded by several biological processes, including chemotherapy-induced lymphodepletion, subsequent lymphocyte proliferation, and the dynamic redistribution of lymphocytes between the circulation and the tumor microenvironment (TME). Peripheral blood counts primarily reflect systemic depletion and recovery, whereas tumor immune activity is determined largely by the extent and activation state of intratumoral infiltration. Lymphocyte trafficking to peritumoral lymph nodes, tertiary lymphoid structures, or tumor margins may further alter circulating levels without accurately reflecting local immune engagement [25]. In head and neck squamous cell carcinoma, parallel activation and proliferation of CD8⁺ T cells have been observed in both the tumor bed and peripheral blood during immunotherapy [26], yet this concordance remains observational and mechanistically unproven. The CD8⁺ T-cell counts obtained approximately one month after treatment in the present study represent a single, indirect snapshot of a highly dynamic process. To clarify the prognostic relevance of these findings, future research should integrate temporal mapping of lymphocyte migration, spatial-omics characterization of immune localization, and direct functional validation of cytotoxic activity in relation to long-term clinical outcomes [27].

Several previous studies have underscored the prognostic importance of T cells [[10], [11], [12], [13],17], B cells [28,29], and their subsets in NPC. Tregs, which play a pivotal role in maintaining immune tolerance within the TME, have been associated with poorer clinical outcomes in NPC [10]. Tao et al. reported that a low CD4/CD8 ratio, reflecting either a reduced CD4⁺ T-cell count or an elevated CD8⁺ T-cell count, was linked to adverse prognosis, consistent with the findings of the present study regarding CD4⁺ T-cell levels [11]. Similarly, Shen et al. demonstrated that a low baseline CD4⁺ T-cell count predicted unfavorable outcomes [12]. In contrast, Zhu et al. observed that among patients with high EBV-DNA levels (>1500 copies/mL), elevated proportions of CD3⁺CD8⁺ T cells correlated with improved OS [13], a pattern that paralleled our results for PFS but not OS. Weng et al. further showed that higher pretreatment CD8⁺ T-cell levels and greater post-radiotherapy reductions in CD4⁺ T cells were both associated with better prognosis [17], aligning with our observations. Collectively, these studies highlight the prognostic relevance of T-cell dynamics in NPC. However, most prior analyses did not integrate key clinical parameters alongside immune variables. In contrast, our nomograms incorporated seven clinically and biologically significant factors, thereby enhancing predictive accuracy and minimizing the risk of overfitting.

Our nomograms demonstrated substantially better performance than previously reported models based on TNM staging and TNM combined with EBV-DNA levels, which represent canonical prognostic frameworks for NPC [5]. In developing the models, we carefully balanced the number of included variables with the available sample size to ensure statistical robustness and avoid overfitting. Variables were rigorously selected through univariable and multivariable Cox analyses, resulting in parsimonious models with strong predictive performance. Unlike several prior studies, some of which used large databases but did not explicitly account for sample size considerations; our approach prioritized model reliability and interpretability. Expanding the sample size in future research may allow inclusion of additional variables and further improve prognostic precision. Importantly, external validation using an independent patient cohort yielded consistent and satisfactory results. As many previously published models lacked external validation [18], this study addresses a critical gap by confirming the reproducibility and generalizability of the nomograms across institutions. Furthermore, both the higher concordance indices and the superior DCA profiles of our models underscore their enhanced predictive accuracy and potential clinical benefit compared with conventional TNM- and EBV-DNA-based systems.

Several previously published nomograms based solely on clinicopathologic variables have reported C-index values ranging from 0.712 to 0.772 for predicting outcomes in NPC [[30], [31], [32], [33]]; however, none incorporated T-cell parameters. A few models have integrated immune-related variables: for example, a nomogram for predicting distant metastasis that included CD4⁺ T-cell counts and lactate dehydrogenase (C-index, 0.763) [34], and another for OS based on CD8⁺ T-cell counts and the platelet-to-lymphocyte ratio (C-index, 0.724) [18]. Although these models successfully discriminated between risk groups, they did not specifically address patients receiving radiotherapy. The present study uniquely focused on OS and PFS in patients treated with radiotherapy, a population rarely targeted in prior modeling efforts. Mi et al. reported a nomogram for DMFS based on age, chemotherapy, N stage, and residual tumor (C-index, 0.807) [35], whereas another model was developed to predict radiation-induced mucositis [36], an important quality-of-life outcome but not a survival endpoint. Nomograms offer practical clinical utility by translating multivariable risk estimates into individualized predictions readily accessible through digital tools. In our study, patients identified as high-risk by the nomograms exhibited significantly poorer prognoses, suggesting that such patients may benefit from intensified chemotherapy, escalated radiation doses, or closer post-treatment surveillance [5]. Although the present nomograms require post-treatment data, specifically, the change in total T-cell count, and therefore cannot be applied before therapy, dynamic T-cell alterations after treatment have been consistently shown to carry prognostic significance in NPC [14,16]. These findings underscore the potential of our models to advance precision risk stratification and inform personalized treatment strategies in this disease.

The TME is a complex and dynamic ecosystem that profoundly influences tumor behavior and treatment response. Tumor-infiltrating lymphocytes (TILs) exert potent antitumor effects in solid malignancies, including NPC, and enhancing TIL activity through immunotherapeutic approaches has become an established therapeutic strategy [37]. A recent meta-analysis demonstrated that higher densities of TILs, particularly CD3⁺, CD4⁺, and CD8⁺ T cells, are associated with improved OS [8]. Conversely, low infiltration of CD3⁺ or CD8⁺ T cells in locally advanced NPC has been linked to shorter OS and disease-free survival (DFS) following platinum-based CCRT [9]. Radiotherapy can promote the release of tumor-associated antigens, thereby priming the immune system and enhancing TIL-mediated cytotoxicity [38,39]. The persistence and activation of TILs within the TME are therefore critical to sustaining the late therapeutic effects of radiotherapy. Immunotherapy has recently emerged as a transformative treatment paradigm in oncology, including NPC [40]. TILs are central to both the antitumor efficacy of immune checkpoint inhibitors and the prediction of therapeutic response [41,42]. In this study, we focused on peripheral blood lymphocytes, which reflect the systemic immune milieu and are closely correlated with intratumoral immune activity. Peripheral lymphocyte profiling offers a minimally invasive, reproducible, and cost-effective alternative to tissue-based immune assessment [43]. Future investigations should evaluate the applicability and prognostic performance of the present nomograms in NPC cohorts receiving immunotherapy.

This study has several limitations. First, as a retrospective analysis, it was constrained by the completeness and accuracy of data extracted from hospital records, and potential selection bias cannot be entirely excluded. Second, treatment protocols evolved during the study period, and prior to 2020, established clinical guidelines did not recommend immunotherapy as first-line therapy for NPC. Accordingly, none of the patients in this cohort received immunotherapy as part of their initial treatment. With accumulating evidence supporting the efficacy of immune checkpoint inhibitors in NPC and their growing incorporation into contemporary clinical practice, future work will aim to expand the dataset to include patients treated with immunotherapy and radiotherapy, allowing recalibration and external validation of the proposed nomograms in this emerging therapeutic context. Finally, given current trends toward individualized radiotherapy strategies, further prospective studies are warranted to confirm the clinical utility and generalizability of these models in diverse patient populations.

Conclusion

In summary, we developed and rigorously validated nomograms for predicting OS and PFS in patients with NPC treated with radiotherapy. These models incorporate both baseline total T-cell counts and their post-treatment changes, reflecting the dynamic nature of host immune responses. The nomograms demonstrated strong predictive performance and can effectively identify patients at high risk of adverse outcomes, thereby supporting more individualized treatment planning and follow-up strategies.

Funding

This work was supported by grants from the Key Research and Development Program Project of Guangxi Zhuang Autonomous Region (No. Guike AB23026020 to Xiaodong Zhu); Chongqing Science and Health Joint Medical Research Project (No. 2023MSXM129 to Qianqian Lei); Natural Science Foundation of Chongqing City (No. CSTB2024NSCQ-MSX0629 to Qianqian Lei); Science and Technology Research Program of Chongqing Municipal Education Commission (No. KJQN202500134 to Qianqian Lei); and the Key Research and Development Project of Sichuan Province (No. 2022YFSY0012 to Qianqian Lei).

Data Availability Statement

In response to reasonable requests, the corresponding author will provide access to the data generated and/or analyzed during the current study.

Ethics statement

This study was approved by the Ethics Committee of Chongqing University Cancer Hospital (approval #CZLS2023323-A) and Guangxi Medical University Cancer Hospital (approval #KYB2024014). The study was conducted according to the principles of the Helsinki Declaration.

CRediT authorship contribution statement

Kegui Weng: Writing – review & editing, Writing – original draft, Validation, Software, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Qianqian Lei: Writing – review & editing, Writing – original draft, Supervision, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Ye Hong: Writing – review & editing, Writing – original draft, Supervision, Software, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Kaihua Chen: Writing – review & editing, Writing – original draft, Supervision, Software, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Yongchu Sun: Writing – review & editing, Writing – original draft, Supervision, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Ying Wang: Writing – review & editing, Writing – original draft, Visualization, Validation, Software, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Xiaodong Zhu: Writing – review & editing, Writing – original draft, Supervision, Resources, Project administration, Methodology, Formal analysis, Data curation, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.