The present study was conducted at Chi Mei Medical Center (Tainan, Taiwan). Data was retrospectively collected from the Center's cancer registry database covering the period from January 2012 to December 2022. The treatment protocol followed a prospective regimen of IC followed by CCRT. Patients aged 25-84 years were included and disease staging was based on the American Joint Committee on Cancer TNM classification, 8th edition (16). The present study involved patients who met the following inclusion criteria: i) Histopathologically confirmed NPC; and ii) complete blood sample data available, including samples collected within 3 days before the start of IC (pre-IC), within 3 days before the start of CCRT (pre-CCRT) and within 1 week after the completion of radiotherapy (post-RT). The exclusion criteria were as follows: i) Patients who had not received induction chemotherapy (IC) and lacked pre-IC baseline hematological data (n=26; Fig. 1); and ii) patients with a history of other malignancies, hematologic disorders, active infections or febrile illness at baseline.

All patients with NPC received IC followed by definitive CCRT. Radiotherapy was administered using the IMRT technique. The IC and CCRT protocols were implemented according to established clinical guidelines in Taiwan and China (17-19). The IC regimen followed the TPF protocol, comprising docetaxel (60-75 mg/m²), cisplatin (50-75 mg/m²) and 5-fluorouracil (600-1,000 mg/m²/day), administered every 3 weeks for two or three cycles. Following IC, the patients underwent definitive radiotherapy concurrently with cisplatin (100 mg/m²), administered every 3 weeks for two cycles.

All patients underwent either IMRT or tomotherapy at Chi Mei Medical Center. For radiotherapy, the radiation dose for the gross tumor volume in the nasopharynx and involved cervical lymph nodes was 70-72 Gy. The first clinical target volume was treated with a radiation dose of ~60 Gy, and the second clinical target volume was treated with 45-54 Gy. Conventional fractionation was performed with daily doses of 1.80 and 2.12 Gy. Concurrent cisplatin chemotherapy was administered at 100 mg/m² every 3 weeks for a total of two cycles, as previously described (20).

Information regarding patient characteristics, including age, sex, histological type, disease stage, weight changes before and after treatment and treatment modalities (including radiotherapy prescription and chemotherapy regimen), was collected from the registry database. Complete blood count data were obtained at three specific time points: i) Pre-IC; ii) pre-CCRT and; iii) post-RT. NLR, PLR and LMR were calculated at each time point.

Blood specimens (3 ml per collection) were collected in plastic vacuum tubes and analyzed using a Sysmex XR-9000 fully automated hematology analyzer (Sysmex Corporation) to obtain information on routine hematological parameters, including absolute neutrophil, lymphocyte, monocyte and platelet counts. Based on these values, the following hemogram ratios were calculated: i) NLR=neutrophil count/lymphocyte count; ii) PLR=platelet count/lymphocyte count and; iii) LMR=lymphocyte count/monocyte count.

Receiver operating characteristic (ROC) curve analysis was conducted to examine the prognostic utility of the nine inflammatory markers (including NLR, PLR and LMR pre-IC, pre-CCRT and post-RT) and determine the optimal cut-off values for predicting survival outcomes.

Descriptive statistics are presented as frequencies and percentages for categorical variables and as the mean ± SD for continuous variables. A two-sided significance (α) level of 0.05 was predefined, and exact P-values are reported for all statistical tests. Comparisons of hematological biomarkers, including NLR, PLR and LMR, across the three time points (pre-IC, pre-CCRT and post-RT), were conducted using one-way repeated-measures ANOVA since these measurements were obtained from the same patients. When the overall repeated-measures ANOVA exhibited a statistically significant difference, post hoc pairwise comparisons were performed using the Bonferroni correction. Independent samples t-tests were used to compare NLR, PLR and LMR between the survivor and non-survivor groups at the three time points. Variables with significant differences (P<0.05) were analyzed using binary logistic regression to investigate their association with mortality. Additional covariates included in the regression model were patient characteristics, such as sex, age, cancer stage, weight loss percentage and pretreatment EBV status (classified as positive, negative or missing).

The χ² test was used to compare survival status (alive vs. deceased) across categorical variables. Optimal cutoff values for NLR, PLR and LMR at each treatment phase were determined using ROC curve analysis based on the maximum Youden index, with diagnostic metrics detailed in Table SI. Kaplan-Meier analysis was used to estimate survival probabilities, and the log-rank test was used to evaluate differences between survival curves. OS was defined as the period from confirmed diagnosis to mortality or the last follow-up. Cox proportional hazards modeling was performed to explore survival determinants. Owing to the limited number of mortality events, the multivariable model was restricted to variables that were statistically significant in the univariate analysis (P<0.05) to reduce the risk of overfitting and unstable estimates. To address small-sample bias, Firth's penalized likelihood correction was applied in both univariate and multivariable Cox regression analyses. In addition, to evaluate the robustness of the present findings and address potential information loss from dichotomization, sensitivity analyses were performed by treating inflammatory biomarkers as continuous variables in the Cox regression models (Table SII). All analyses were performed using SPSS (version 25.0; IBM Corp.) P<0.05 was considered to indicate a statistically significant difference.

A total of 125 patients with histologically determined NPC were initially screened for eligibility. Of these, 26 patients were excluded as they did not receive IC and lacked the required pre-IC baseline hematologic data. Ultimately, a final cohort of 99 patients met all inclusion criteria and were enrolled in the present retrospective analysis (Fig. 1). Complete data regarding histology, clinical characteristics and follow-up were available for all patients. The demographic and clinical details of the patients are summarized in Table I. The present study comprised 76 male (76.8%) and 23 female patients (23.2%). The average patient age was 50 (range: 25-84) years. Among the patients, 1 (1.0%) had stage I, 12 (12.1%) had stage II, 38 (38.4%) had stage III and 48 (48.5%) had stage IV disease. With regard to pre-treatment viral load, EBV status was detected in 52 patients (52.5%), undetected in 20 (20.2%) and remained unknown in 27 (27.3%). All patients received at least two cycles of IC, followed by definitive CCRT. The median follow-up period was 60 (range: 11-209) months. During the follow-up period, 11 patients experienced local recurrence, 10 had distant metastases and 15 were deceased. The overall 5-year OS rate for the entire cohort was 84.8%.

Optimal pre-IC cut-off values used to stratify patients with NPC into high- and low-level groups were as follows: i) NLR=4.02; ii) PLR=248.67 and; iii) LMR=3.17. These thresholds effectively differentiated patients with poorer vs. better outcomes. Subsequently, the cut-off values were adjusted at the pre-CCRT phase to; i) NLR=1.91; ii) PLR=212.16 and; iii) LMR=2.80, distinguishing the high- and low-risk cohorts. Finally, post-RT, the cut-off values were; i) NLR=9.51; ii) PLR=296.36 and; iii) LMR=1.06, providing a final stratification point. These time-specific cut-off values (Table SI) were determined through ROC curve analyses performed at each treatment phase (pre-IC, pre-CCRT and post-RT) to examine the prognostic performance of NLR, PLR and LMR, and identify the most informative thresholds for predicting survival outcomes.

Table II outlines hematological biomarker values at the three investigated time points: Pre-IC, pre-CCRT and post-RT. Repeated-measures ANOVA revealed significant longitudinal variations throughout the therapeutic course for NLR and PLR (both P<0.001), while a marginal trend was observed for LMR (P=0.06). Specifically, higher mean NLR and PLR values were observed at later time points, whereas LMR values were lower post-RT. Notably, at the post-RT evaluation, non-survivors exhibited higher NLR (P=0.05) and PLR (P=0.02) levels, alongside lower LMR levels (P=0.05), compared with survivors.

The logistic regression analysis showed that being ≥60 years old was significantly associated with a reduced likelihood of elevated PLR post-RT [odds ratio (OR)=0.27; 95% CI: 0.09-0.76; P=0.01]. In addition, patients who experienced weight loss of ≥7.5% body weight exhibited significantly higher odds of elevated NLR post-RT compared with those who lost <7.5% (OR=2.44; 95% CI: 1.02-5.86; P=0.05). No other variables, including sex, tumor stage, nodal stage, pretreatment EBV status and LMR, showed significant associations with inflammatory markers after radiotherapy (Table III).

Patients with PLR pre-IC ≥248.67 exhibited significantly poorer 5-year OS compared with those with PLR pre-IC <248.67 (66.7 vs. 87.4%; P=0.03; Fig. 2D). Similarly, patients with PLR post-RT ≥296.3 exhibited poorer 5-year OS compared with those with PLR post-RT <296.3 (79.7 vs. 96.7%; P=0.02; Fig. 2F).

Table IV presents the results of the univariate and multivariate Cox regression analyses with Firth's correction to evaluate the risk factors associated with mortality in patients with NPC. Advanced nodal stage (N3) and high PLR pre-IC were significantly associated with a poor OS (both P<0.05) in the univariate analysis. In the multivariate analysis, high PLR pre-IC [hazard ratio (HR): 3.81; 95% CI: 1.21-12.00; P=0.02] and advanced nodal stage (N3; HR: 2.91; 95% CI: 1.01-8.38; P=0.05) were found to be independent predictors of mortality.

Furthermore, sensitivity analyses, with NLR, PLR and LMR evaluated as continuous variables using Cox regression models with Firth's correction, yielded results that were not entirely identical to those obtained by the primary ROC-dichotomized analyses. In the univariate analysis, an advanced nodal stage (N3; P=0.04) and an elevated NLR post-RT (P=0.02) were significantly associated with poor OS. However, in the present multivariable continuous model, both advanced nodal stage (N3; HR=2.75; 95% CI: 0.91-8.29; P=0.07) and NLR post-RT (HR=1.02; 95% CI: 0.99-1.04; P=0.12) were not statistically significant (Table SII).

To evaluate the prognostic stability of the hematologic biomarkers, a sex-stratified subgroup analysis was conducted (Table SIII). In male patients (n=76), an elevated PLR pre-IC was notably associated with poorer survival in both univariate (HR, 4.31; 95% CI, 1.22-15.23; P=0.02) and multivariable analyses (HR, 4.21; 95% CI, 1.17-15.13; P=0.03), demonstrating the role of PLR pre-IC as a robust prognostic indicator within this sub-cohort. Although a decreased LMR pre-IC was notably associated with mortality in the univariate analysis (HR, 5.97; 95% CI, 1.02-34.83; P=0.05) for males, this effect was attenuated and lost statistical significance after multivariable adjustment (HR, 5.61; 95% CI, 0.92-34.41; P=0.06). Conversely, no marked prognostic factors were identified among female patients (n=23), including PLR pre-IC in univariate analysis (HR, 14.48; 95% CI, 0.15-1363.45; P=0.25). Notably, the statistical estimates within the female subgroup exhibited wide CIs, reflecting limited statistical precision driven by the smaller sample size and fewer clinical events in the present cohort.