Efficacy and Safety of Celecoxib Combined with Gabapentin for Head and Facial Pain in Nasopharyngeal Carcinoma: A Retrospective Cohort Study Authors: Xiangrong Cao, Linzong Xu, Weiwei Ta Published: 2026 ABSTRACT AbstractBackgroundHead and facial pain caused by local invasion in nasopharyngeal carcinoma (NPC) is notoriously severe and traditionally managed with opioids, which carry substantial dependence and tolerability risks. Because this tumor-related pain intrinsically involves both neuropathic and inflammatory pathways, we retrospectively evaluated a novel opioid-sparing analgesic strategy combining celecoxib and gabapentin.MethodsThis exploratory retrospective cohort analysis included 60 newly diagnosed NPC patients experiencing severe head and facial pain secondary to cranial nerve or skull base invasion. Patients received either opioid monotherapy (extended-release oxycodone, n=30) or a combined regimen of celecoxib and gabapentin (n=30). The primary endpoint was the absolute reduction in Numeric Rating Scale (NRS) scores at 48 hours. Secondary endpoints included sleep quality improvement, assessed via the Insomnia Severity Index (ISI), and adverse events. Longitudinal data were rigorously analyzed using Generalized Estimating Equations (GEE) to account for repeated measures and adjust for baseline clinical covariates.ResultsBaseline clinical characteristics were broadly comparable between the cohorts, with no statistically significant differences in premedication NRS scores (6.5 in both groups) or ISI scores (16.5 vs 19.5, P = 0.118); however, the combined group had a numerically higher baseline ISI score, and all ISI comparisons were therefore based on change-from-baseline scores. The GEE longitudinal analysis revealed a significant treatment-by-time interaction for pain relief. At the 48-hour primary endpoint, the combined regimen demonstrated a significantly greater absolute reduction in NRS scores compared to the opioid group (Estimated Mean Difference [MD], 2.15; 95% CI, 1.45 to 2.85; P < 0.001). Furthermore, the combined therapy yielded a profoundly larger improvement in sleep architecture (ISI reduction MD, 9.85; 95% CI, 7.50 to 12.20; P < 0.001). Safety profiles analyzed via Fisher’s exact test confirmed the combined group had a significantly lower incidence of nausea and vomiting (0.0% vs 20.0%, P = 0.024); no significant between-group differences were observed for other adverse events including constipation, dizziness, and somnolence.ConclusionCombining celecoxib and gabapentin appears to offer an effective, opioid-sparing alternative for managing complex tumor-related pain in NPC, and was associated with greater pain reduction over the 48-hour observation period, accelerated sleep recovery, and significantly better gastrointestinal tolerability compared to conventional opioid therapy in this retrospective cohort. These findings are hypothesis-generating and warrant prospective validation.Keywords: Nasopharyngeal carcinoma, Pain, Celecoxib, Gabapentin, Clinical efficacy Introduction Nasopharyngeal carcinoma (NPC) accounts for approximately 130,000 new cases globally each year1 and represents one of the most prevalent malignant tumors in Asia, particularly in China and Southeast Asia.2 Due to the complex anatomy of the nasopharynx and the typically insidious onset of symptoms, most patients are diagnosed at a locally advanced or advanced stage.3 The clinical manifestations of NPC are often non-specific; however, approximately 16% of patients experience head and facial pain, primarily driven by cranial nerve invasion-related neuralgia and inflammatory pain secondary to skull base erosion.4,5 This tumor-related pain is typically persistent, presenting as a continuous dull ache with intermittent exacerbations. Although NPC is highly sensitive to chemoradiotherapy, pain alleviation generally occurs only about two weeks after the initiation of anti-tumor treatment.6 Consequently, effective pain management during the critical window from initial presentation to the onset of tumor shrinkage is frequently overlooked. The chronic and severe nature of this pain often leads to sleep disturbances, anxiety, and depression, substantially impairing patients’ quality of life. Currently, guidelines from both the National Comprehensive Cancer Network (NCCN) and the European Association for Palliative Care (EAPC) recommend opioids, such as oxycodone and fentanyl, as the first-line treatment for moderate to severe cancer-related pain.7,8 However, the clinical utility of short-acting opioids is often constrained by their limited duration of action9 and the necessity for complex dosage titrations.10 Furthermore, previous studies have demonstrated that high-dose opioid therapy increases the incidence of adverse effects and is associated with reduced overall survival and increased mortality risks.11 Given the dual neuropathic and inflammatory mechanisms underlying NPC-related head and facial pain, developing novel, opioid-sparing analgesic strategies holds significant clinical value for patients with locally advanced disease. Multimodal analgesic strategies, which combine agents with different mechanisms of action, have been increasingly adopted in various cancer pain settings to reduce opioid requirements and improve tolerability.12 However, evidence specifically addressing multimodal analgesia for pain in nasopharyngeal carcinoma remains remarkably limited, with most existing studies focusing on postoperative pain or chronic pain syndromes unrelated to malignancy.13 This evidence gap underscores the need for clinical investigations evaluating combination non-opioid regimens in the NPC population. Celecoxib, a selective cyclooxygenase-2 (COX-2) inhibitor, exerts robust anti-inflammatory and analgesic effects and has demonstrated clinical benefits in managing pain associated with osteoarthritis, rheumatoid arthritis, and other inflammatory conditions.14 Meanwhile, gabapentin is a well-established agent for treating neuropathic pain; however, monotherapy is often insufficient to achieve optimal pain control.15 Conceptually, combining these two agents could theoretically provide complementary analgesic effects by simultaneously targeting the inflammatory and neuropathic pain pathways characteristic of NPC. Nevertheless, clinical evidence supporting this combined approach in NPC patients remains scarce. Therefore, this exploratory retrospective study aims to evaluate the efficacy and safety of a combination regimen comprising celecoxib and gabapentin for the management of tumor-related head and facial pain in NPC patients, hypothesizing that this approach might be associated with greater pain relief and improved sleep quality compared to conventional opioid therapy. Methods Study Design and Participants This study was a retrospective, exploratory cohort analysis based on clinical data from 60 newly diagnosed NPC patients treated at the Lanzhou University Second Hospital between 2019 and 2022. The study was reported in accordance with the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines.16 All included patients had a pathologically confirmed diagnosis and experienced varying degrees of head and facial pain symptoms secondary to cranial nerve invasion and/or skull base bone destruction. General demographic and comprehensive clinical data were extracted from the electronic medical records. To comprehensively adjust for baseline characteristics, extracted covariates included age, sex, disease course, TNM clinical stage (Stage III vs IVA), imaging-confirmed invasion sites (isolated cranial nerve, isolated skull base bone, or mixed involvement), presence of a neuropathic pain component, trigeminal nerve involvement, and time to the initiation of anti-tumor therapy. This study was approved by the Ethics Committee of the Lanzhou University Second Hospital (Ethics approval number: 2024A-959), and the requirement for informed consent was waived due to the retrospective nature of the study. Inclusion and Exclusion Criteria Patients were eligible for inclusion in this study if they met all of the following criteria: 1) pathologically confirmed initial diagnosis of NPC; 2) presence of head and facial pain attributed to local tumor invasion of the cranial nerves and/or skull base bone, confirmed by clinical manifestations and imaging examinations; 3) a baseline Numeric Rating Scale (NRS) score17 of ≥4 at initial presentation; and 4) good general condition at the time of consultation, with no significant dysfunction of major organ systems. Conversely, patients were excluded if they met any of the following criteria: 1) presence of severe systemic diseases, such as concurrent malignancies or coagulation disorders; 2) documented dysfunction of major organs (eg, heart, liver, kidneys, lungs, or central nervous system); 3) poor treatment compliance or interruption of treatment; 4) inability to independently cooperate with clinical assessments; and 5) any other condition that could significantly interfere with data integrity. A total of 86 patients were initially assessed for eligibility. Of these, 26 patients were excluded: 11 did not meet the inclusion criteria (NRS < 4 or pain not attributable to local invasion), and 15 met one or more exclusion criteria (6 with severe systemic diseases, 4 with major organ dysfunction, and 5 with incomplete medical records). The remaining 60 patients were included in the final analysis. Treatment Allocation and Regimens Due to the retrospective and observational nature of this study, treatment allocation was not randomized. The choice of analgesic regimen was determined by the attending physicians’ clinical judgment, based on the patients’ baseline pain characteristics, history of NSAID contraindications, prior opioid tolerance, and patient preference. Institutional clinical practice guidelines at Lanzhou University Second Hospital during the study period recommended the combined regimen (celecoxib plus gabapentin) as the preferred first-line approach for NPC patients presenting with head and facial pain who had: (1) imaging-confirmed cranial nerve or skull base involvement, (2) a documented neuropathic pain component, and (3) no contraindications to NSAIDs or gabapentin. Opioid monotherapy was generally reserved for patients who had relative contraindications to the combined regimen, including a history of NSAID-related gastrointestinal intolerance, renal function concerns, or patient preference for opioid-based analgesia after shared decision-making. Patients with absolute contraindications to NSAIDs or gabapentin, or those with documented prior opioid tolerance or chronic opioid use, were not included in this analysis. These allocation criteria were applied consistently throughout the study period. Patients were categorized into two groups: the opioid group (n = 30) and the combined therapy group (celecoxib in combination with gabapentin, n = 30). Patients in the opioid group received extended-release oxycodone hydrochloride tablets (Mundipharma, China), initiated at 10 mg every 12 hours. Patients in the combined group received oral celecoxib capsules (Pfizer Inc., USA) at 100 mg every 12 hours, combined with oral gabapentin capsules (Jiangsu Hengrui Medicine Co., Ltd., China) at 600 mg every 8 hours. Both groups were provided with a standardized rescue analgesia protocol. If the NRS score remained ≥3, an immediate-release rescue dose (equivalent to 10 mg of oral morphine hydrochloride or 50% of the regular extended-release oxycodone dose) was administered. Dose titration was continued until the NRS score dropped below 3. To compare the opioid-sparing effect, the total consumption of rescue medications over the first 48 hours was converted into cumulative oral morphine equivalents (OME) for both groups, in accordance with established clinical practice guidelines for cancer pain.7 Outcome Measures and Safety Monitoring The primary endpoint of this exploratory analysis was defined as the change in NRS score from baseline at 48 hours post-treatment. The NRS was assessed prior to treatment and at 2, 4, 6, 8, 12, 24, and 48 hours after medication administration. Sleep quality was evaluated as a secondary endpoint using the adapted Insomnia Severity Index (ISI)18 prior to treatment and at 24 and 48 hours post-medication. To accommodate the short-term assessment window, the standard ISI recall period of the past two weeks was modified to the past 24 hours for the post-treatment assessments, consistent with the approach used in previous perioperative and acute pain studies. All assessments were administered by trained research nurses who were blinded to the study hypothesis but not to treatment allocation. Adverse events were retrospectively extracted from electronic medical records, including nursing records, physician progress notes, medication orders, and laboratory results. A standardized data extraction checklist was applied to capture the following predefined events: nausea and vomiting, constipation, dizziness, somnolence, gastrointestinal bleeding, hepatic or renal dysfunction, and major cardiovascular events. Statistical Analysis Data processing and statistical analyses were performed using SPSS version 26.0 and R software (version 4.2.1). The normality of continuous variables was evaluated using the Shapiro–Wilk test. Normally distributed data were expressed as mean ± standard deviation and compared using the independent samples t-test. Non-normally distributed continuous variables (such as baseline NRS, ISI, and disease course) were presented as medians with interquartile ranges (IQR) and analyzed using the Mann–Whitney U-test. Categorical variables were reported as absolute numbers and percentages n (%). To address the repeated-measures nature of the longitudinal NRS and ISI data, Generalized Estimating Equations (GEE) models were fitted with an identity link function, a Gaussian variance function, an unstructured working correlation matrix, and robust (sandwich) standard errors. The GEE models evaluated the main effects of treatment group and time, as well as the group-by-time interaction, adjusting for baseline covariates to mitigate potential selection bias. The following covariates were included in all adjusted GEE models: age (continuous), sex (male/female), TNM clinical stage (Stage III vs IVA), invasion sites (isolated cranial nerve, isolated skull base bone, or mixed involvement), presence of neuropathic pain component (yes/no), trigeminal nerve involvement (yes/no), baseline NRS score, and baseline ISI score. For the NRS model, baseline NRS was included as a covariate; for the ISI model, baseline ISI was included as a covariate. For the comparison of adverse events, due to the low expected frequencies, Fisher’s exact test was strictly applied instead of the chi-square test. Effect sizes, including mean differences (MD) and 95% confidence intervals (CI), were reported where appropriate. A two-sided p < 0.05 was considered statistically significant for all analyses. Model Parsimony and Sensitivity Analysis Given the relatively small sample size (n = 60) and the number of repeated measurements (eight time points for NRS, three for ISI), the use of an unstructured working correlation matrix in the GEE models required estimation of 28 correlation parameters for the NRS model. To assess whether model overparameterization influenced our findings, we performed sensitivity analyses using alternative working correlation structures (exchangeable and AR-1) for both NRS and ISI outcomes. The estimated treatment effects and significance levels were materially unchanged across all correlation structures (data not shown), supporting the robustness of our primary findings. Nonetheless, the complexity of the unstructured model, combined with the use of robust sandwich standard errors in a modest sample, may result in some imprecision; therefore, our results should be interpreted with appropriate caution. Sample Size Considerations Given the retrospective and exploratory nature of this study, no formal a priori sample size calculation was performed. The sample size was determined by the number of eligible patients treated at our institution during the study period who met the predefined inclusion criteria. The achieved sample size of 30 patients per group is comparable to or exceeds that of similar exploratory retrospective studies in cancer pain management. Using G*Power software (version 3.1.9.7), with a two-sided significance level of 0.05 and 30 patients per group, the minimum detectable difference in NRS reduction (assuming 80% power and a pooled standard deviation of 1.85 based on the observed variance) was estimated to be approximately 1.55 points. The minimum detectable difference was estimated using G*Power 3.1.9.7 with the following parameters: two-sided t-test, α = 0.05, power = 0.80, allocation ratio 1:1, and pooled SD derived from the observed variance in the primary outcome. For the primary endpoint, the observed between-group difference in NRS reduction at 48 hours (MD = 2.15, 95% CI: 1.45 to 2.85) yielded a relatively precise estimate, with the lower bound of the 95% confidence interval exceeding the minimal clinically important difference for pain reduction (MCID = 1.0–1.5). However, the study was underpowered to detect smaller effect sizes (eg, <1.55 points), to perform robust subgroup analyses, or to draw definitive conclusions regarding secondary endpoints or infrequent adverse events. The relatively small sample size increases the risk of Type II error for secondary endpoints and rare adverse events, which should be considered when interpreting the null findings for constipation, dizziness, somnolence, and other safety outcomes. These methodological constraints underscore the preliminary and hypothesis-generating nature of our findings, as emphasized in the Discussion and Conclusion. Results Comparison of Baseline Clinical Characteristics Between the Two Groups A total of 60 patients with newly diagnosed NPC experiencing tumor-related head and facial pain were included in this exploratory cohort analysis. Based on the analgesic regimens administered, patients were categorized into the opioid group (n = 30) and the combined therapy group (celecoxib plus gabapentin, n = 30). The cohort comprised 41 males and 19 females, with ages ranging from 18 to 77 years. The median age was 56.0 years in the opioid group and 53.5 years in the combined group. Both groups demonstrated a median disease course of 6.0 months (Figure 1). Flow diagram of the patient selection process. To evaluate potential selection bias, baseline tumor characteristics and specialized pain variables were comprehensively assessed. There were no significant differences between the two cohorts regarding the TNM clinical stage, patterns of local tumor invasion (isolated cranial nerve, isolated skull base bone, or mixed involvement), the presence of a neuropathic pain component, or trigeminal nerve involvement (all P > 0.05). Both groups experienced moderate-to-severe pain at presentation (median premedication NRS score: 6.5 in both groups) and concomitant moderate clinical insomnia (median premedication ISI score: 16.5 vs 19.5, P = 0.118). Additionally, the time from analgesic consultation to the initiation of definitive anti-tumor therapy (radiotherapy and/or chemotherapy) was comparable between the two groups (Table 1). Table 1Baseline Demographic and Clinical Characteristics of the Two Groups of PatientsCharacteristicsOpioid Group (n = 30)Combined Group (n = 30)χ2/Z ValueP ValueSex, n (%)0.8140.367Male22 (73.3)19 (63.3)Female8 (26.7)11 (36.7)Age (years), Median (IQR)56.00 (47.00, 59.00)53.50 (44.00, 64.00)−0.0670.947Course of disease (months), Median (IQR)6.00 (6.00, 8.00)6.00 (5.00, 7.50)−0.7250.468TNM clinical stage, n (%)0.2720.602Stage III12 (40.0)14 (46.7)Stage IVA18 (60.0)16 (53.3)Invasion sites, n (%)0.3340.846Isolated cranial nerve8 (26.7)7 (23.3)Isolated skull base bone10 (33.3)12 (40.0)Mixed involvement12 (40.0)11 (36.7)Neuropathic pain component, n (%)0.1410.707Yes24 (80.0)25 (83.3)No6 (20.0)5 (16.7)Trigeminal nerve involvement, n (%)0.2720.605Yes18 (60.0)16 (53.3)No12 (40.0)14 (46.7)Premedication NRS score, Median (IQR)6.50 (5.00, 8.25)6.50 (5.00, 8.25)−0.1420.887Premedication ISI score, Median (IQR)16.50 (12.00, 22.50)19.50 (14.00, 24.25)−1.5650.118Time to anti-tumor therapy (days), Median (IQR)14.50 (10.00, 18.25)15.00 (11.00, 20.00)−0.4520.651Notes: Continuous variables with non-normal distributions are presented as Median (Interquartile Range, IQR) and were compared using the Mann–Whitney U-test (Z value). Categorical variables are presented as absolute numbers and percentages n (%) and were compared using the Pearson chi-square test (χ2 value) or Fisher’s exact test where appropriate.Abbreviations: NPC, nasopharyngeal carcinoma; TNM, tumor-node-metastasis; NRS, Numeric Rating Scale; ISI, Insomnia Severity Index. Baseline Demographic and Clinical Characteristics of the Two Groups of Patients Notes: Continuous variables with non-normal distributions are presented as Median (Interquartile Range, IQR) and were compared using the Mann–Whitney U-test (Z value). Categorical variables are presented as absolute numbers and percentages n (%) and were compared using the Pearson chi-square test (χ2 value) or Fisher’s exact test where appropriate. Abbreviations: NPC, nasopharyngeal carcinoma; TNM, tumor-node-metastasis; NRS, Numeric Rating Scale; ISI, Insomnia Severity Index. Primary Endpoint: Change in NRS Scores at 48 Hours and Longitudinal Analysis There was no significant difference in premedication NRS scores between the opioid group and the combined group (Median [IQR]: 6.50 [5.00, 8.25] vs 6.50 [5.00, 8.25]; Z = −0.142, P = 0.887; Table 1). However, following the initiation of analgesic treatment, the combined therapy group demonstrated a more rapid, sustained, and pronounced reduction in pain intensity (Table 2 and Figure 2A). For the primary endpoint, the absolute reduction in NRS score from baseline at 48 hours was significantly greater in the combined group compared to the opioid group (Estimated MD in reduction, 2.15; 95% CI, 1.45 to 2.85; P < 0.001) (Table 3). Table 2Comparison of Post-Treatment NRS Scores Between the Two Groups Over TimeTimepointOpioid Group (n = 30)Combined Group (n = 30)2 h6.00 (5.00, 8.00)3.00 (2.00, 4.25)4 h5.00 (3.00, 6.00)2.00 (2.00, 3.00)6 h4.50 (3.00, 5.00)2.00 (1.00, 2.00)8 h4.00 (3.00, 4.25)1.00 (1.00, 2.00)12 h3.00 (3.00, 5.00)1.00 (1.00, 2.00)24 h3.00 (3.00, 4.25)1.00 (1.00, 1.00)48 h3.00 (3.00, 3.00)1.00 (1.00, 1.00)Notes: NRS scores are non-normally distributed continuous variables and are presented as Median (Interquartile Range, IQR).Abbreviation: NRS, Numeric Rating Scale. Comparison of Post-Treatment NRS Scores Between the Two Groups Over Time Notes: NRS scores are non-normally distributed continuous variables and are presented as Median (Interquartile Range, IQR). Abbreviation: NRS, Numeric Rating Scale. Longitudinal trajectory of pain relief and absolute reduction in NRS scores between the two groups. (A) Changes in median Numeric Rating Scale (NRS) scores over the 48-hour treatment period. Data are presented as medians with error bars representing the interquartile range (IQR). (B) Absolute reduction in NRS scores from baseline across post-treatment time points. Bars represent the median reduction, and error bars indicate the IQR. P values were derived from the post-hoc analyses of the Generalized Estimating Equations (GEE) model with Bonferroni correction. *** P < 0.001 compared to the opioid group at the corresponding time point; ns, no significant difference at baseline. Table 3Comparison of the Absolute Reduction in NRS Scores from Baseline and GEE-Estimated Effect Sizes Between the Two GroupsTimepointOpioid Group Reduction (Median [IQR])Combined Group Reduction (Median [IQR])Estimated MD (95% CI)aP Valueb2 h0.50 (0.00, 1.00)3.50 (2.50, 4.00)2.85 (2.20, 3.50)< 0.0014 h1.50 (1.00, 2.00)4.50 (3.50, 5.00)2.70 (2.00, 3.40)< 0.0016 h2.00 (1.50, 3.00)4.50 (4.00, 5.50)2.55 (1.90, 3.20)< 0.0018 h2.50 (2.00, 3.50)5.50 (4.50, 6.50)2.40 (1.75, 3.05)< 0.00112 h3.50 (2.50, 4.50)5.50 (4.50, 6.50)2.00 (1.30, 2.70)< 0.00124 h3.50 (2.50, 4.50)5.50 (5.00, 6.50)2.00 (1.30, 2.70)< 0.00148 hc3.50 (3.00, 4.50)5.50 (5.00, 6.50)2.15 (1.45, 2.85)< 0.001Notes: a Reduction values are presented as medians with interquartile ranges (IQR). The Estimated Mean Difference (MD) and 95% Confidence Interval (CI) represent the adjusted between-group difference in the absolute reduction of NRS scores (Combined group vs Opioid group) derived from the Generalized Estimating Equations (GEE) model, which estimates marginal means rather than medians. A positive MD indicates a greater reduction in pain intensity in the combined group. b P values for between-group differences at each time point were derived from post-hoc comparisons of the GEE model, with Bonferroni correction applied for multiple testing. c The change in NRS score from baseline at 48 hours serves as the primary endpoint of this exploratory analysis.Abbreviations: NRS, Numeric Rating Scale; GEE, Generalized Estimating Equations; MD, Mean Difference; CI, Confidence Interval. Comparison of the Absolute Reduction in NRS Scores from Baseline and GEE-Estimated Effect Sizes Between the Two Groups Notes: a Reduction values are presented as medians with interquartile ranges (IQR). The Estimated Mean Difference (MD) and 95% Confidence Interval (CI) represent the adjusted between-group difference in the absolute reduction of NRS scores (Combined group vs Opioid group) derived from the Generalized Estimating Equations (GEE) model, which estimates marginal means rather than medians. A positive MD indicates a greater reduction in pain intensity in the combined group. b P values for between-group differences at each time point were derived from post-hoc comparisons of the GEE model, with Bonferroni correction applied for multiple testing. c The change in NRS score from baseline at 48 hours serves as the primary endpoint of this exploratory analysis. Abbreviations: NRS, Numeric Rating Scale; GEE, Generalized Estimating Equations; MD, Mean Difference; CI, Confidence Interval. To rigorously account for the within-subject correlation inherent in the repeated-measures design, a GEE model was employed. The longitudinal analysis revealed a significant main effect of the treatment group (Wald χ2 = 38.45, P < 0.001) and a significant main effect of time (Wald χ2 = 145.20, P < 0.001). More importantly, a significant group-by-time interaction was observed (Wald χ2 = 22.15, P < 0.001), indicating that the trajectory of pain relief differed fundamentally between the two regimens. Specifically, at 2 hours post-administration, the median NRS score in the combined group significantly decreased to 3.00, further declining to 1.00 by 8 hours, and stably plateauing at this minimal pain level through 48 hours. In contrast, the opioid group maintained a median NRS score of 6.00 at 2 hours, which slowly declined to 3.00 by 12 hours and remained at that level thereafter. The between-group differences in NRS reduction, estimated by the GEE model, consistently favored the combined group across all evaluated time points with robust effect sizes (Table 3 and Figure 2B). Secondary Endpoint: Improvement in Sleep Quality (ISI Scores) There was no significant difference in baseline ISI scores between the opioid group and the combined group. However, consistent with the trajectory of pain relief, sleep quality improved significantly more in the combined therapy group (Table 4 and Figure 3A). Table 4Comparison of Post-Treatment ISI Scores Between the Two Groups Over TimeTimepointOpioid Group (n = 30)Combined Group (n = 30)Adjusted P Value*24 h12.50 (8.00, 16.50)5.00 (3.75, 6.00)< 0.00148 h12.00 (8.00, 16.00)3.00 (2.75, 4.00)< 0.001Notes: Raw ISI scores are non-normally distributed continuous variables and are presented as Median (Interquartile Range, IQR). *P values for the between-group comparisons at each specific time point were derived from the estimated marginal means of the Generalized Estimating Equations (GEE) model, with Bonferroni correction applied for multiple comparisons.Abbreviation: ISI, Insomnia Severity Index. Comparison of Post-Treatment ISI Scores Between the Two Groups Over Time Notes: Raw ISI scores are non-normally distributed continuous variables and are presented as Median (Interquartile Range, IQR). *P values for the between-group comparisons at each specific time point were derived from the estimated marginal means of the Generalized Estimating Equations (GEE) model, with Bonferroni correction applied for multiple comparisons. Abbreviation: ISI, Insomnia Severity Index. Impact of analgesic regimens on sleep architecture and insomnia severity. (A) Longitudinal changes in the median Insomnia Severity Index (ISI) scores from baseline (Pre) to 48 hours post-treatment. Error bars represent the interquartile range (IQR). (B) Absolute reduction in ISI scores from baseline at 24 and 48 hours. Bars represent the median reduction, and error bars indicate the IQR. P values were derived from post-hoc analyses of the Generalized Estimating Equations (GEE) model with Bonferroni correction for multiple comparisons. *** P < 0.001 compared to the opioid group at the corresponding time point; ns, no significant difference at baseline. To ensure methodological consistency, a GEE model was similarly applied to the longitudinal ISI data. The analysis revealed a significant treatment-by-time interaction (Wald χ2 = 18.32, P < 0.001), indicating that the combined regimen provided a statistically significant and more rapid improvement in sleep architecture compared to opioid monotherapy. At 24 and 48 hours post-treatment, the median ISI scores in the opioid group were 12.50 (subthreshold insomnia) and 12.00, respectively. In stark contrast, the median scores in the combined group decreased to 5.00 and 3.00 (indicating a return to normal clinical sleep). Importantly, all between-group comparisons were based on the absolute reduction from baseline rather than raw post-treatment scores alone, accounting for the numerically higher baseline ISI score in the combined group (Table 5). Table 5Comparison of the Absolute Reduction in ISI Scores from Baseline and GEE-Estimated Effect SizesTimepointOpioid Group Reduction (Median [IQR])Combined Group Reduction (Median [IQR])Estimated MD (95% CI)aP Valueb24 h4.00 (2.50, 6.00)14.50 (10.00, 18.25)9.20 (6.80, 11.60)< 0.00148 h4.50 (2.50, 6.50)16.50 (11.00, 20.25)9.85 (7.50, 12.20)< 0.001Notes: a Estimated Mean Difference (MD) and 95% Confidence Interval (CI) represent the adjusted between-group difference in the absolute reduction of ISI scores (Combined group vs Opioid group) derived from the GEE model, which estimates marginal means rather than medians. A positive MD indicates a greater improvement (reduction in score) in sleep quality in the combined group. b P values for between-group differences at each time point were derived from post-hoc comparisons of the GEE model, with Bonferroni correction applied for multiple testing.Abbreviations: ISI, Insomnia Severity Index; GEE, Generalized Estimating Equations; MD, Mean Difference; CI, Confidence Interval. Comparison of the Absolute Reduction in ISI Scores from Baseline and GEE-Estimated Effect Sizes Notes: a Estimated Mean Difference (MD) and 95% Confidence Interval (CI) represent the adjusted between-group difference in the absolute reduction of ISI scores (Combined group vs Opioid group) derived from the GEE model, which estimates marginal means rather than medians. A positive MD indicates a greater improvement (reduction in score) in sleep quality in the combined group. b P values for between-group differences at each time point were derived from post-hoc comparisons of the GEE model, with Bonferroni correction applied for multiple testing. Abbreviations: ISI, Insomnia Severity Index; GEE, Generalized Estimating Equations; MD, Mean Difference; CI, Confidence Interval. Post-hoc analyses confirmed that the absolute reduction in ISI scores from baseline was significantly greater in the combined group at both evaluated time points. At 48 hours, the combined therapy yielded a profoundly larger improvement in sleep quality compared to opioids (Estimated MD in reduction, 9.85; 95% CI, 7.50 to 12.20; P < 0.001), as detailed in Table 5 and Figure 3B. Safety Profile and Adverse Events Adverse events were systematically monitored in all patients throughout the analgesic treatment period, in accordance with the predefined safety checklist. Overall, both treatment regimens were generally well-tolerated, and no grade 3 or higher severe adverse events—such as gastrointestinal bleeding, acute hepatic or renal dysfunction, or major cardiovascular events—were reported in either cohort. Notably, the combined therapy demonstrated a lower incidence of opioid-related gastrointestinal adverse events, driven primarily by nausea and vomiting. The incidence of nausea and vomiting was significantly lower in the combined group compared to the opioid group (0% vs 20.0%, P = 0.024). Additionally, constipation was observed in 10.0% of patients in the opioid group, whereas no cases were reported in the combined group, although this difference did not reach statistical significance. No significant differences were observed between the two groups regarding central nervous system side effects (such as dizziness and somnolence) or pyrosis/indigestion (Table 6). Table 6Comparison of Adverse Events Between the Two Groups After TreatmentAdverse EventsOpioid Group (n = 30)Combined Group (n = 30)P Value*Nausea and vomiting, n (%)6 (20.0)0 (0.0)0.024Constipation, n (%)3 (10.0)0 (0.0)0.237Pyrosis / indigestion, n (%)2 (6.7)2 (6.7)1Dizziness, n (%)1 (3.3)2 (6.7)1Somnolence, n (%)2 (6.7)1 (3.3)1Gastrointestinal bleeding, n (%)0 (0.0)0 (0.0)1Hepatic or renal dysfunction, n (%)0 (0.0)0 (0.0)1Major cardiovascular events, n (%)0 (0.0)0 (0.0)1Notes: Categorical variables for adverse events are presented as absolute numbers and percentages n (%). *Because the expected frequencies in several cells were less than 5, P values for all between-group comparisons were strictly calculated using Fisher’s exact test (two-sided) to avoid Type I error inflation. All P values are reported to three decimal places; values of 1.000 indicate an exact P value of 1.000. Comparison of Adverse Events Between the Two Groups After Treatment Notes: Categorical variables for adverse events are presented as absolute numbers and percentages n (%). *Because the expected frequencies in several cells were less than 5, P values for all between-group comparisons were strictly calculated using Fisher’s exact test (two-sided) to avoid Type I error inflation. All P values are reported to three decimal places; values of 1.000 indicate an exact P value of 1.000. Rescue Analgesic Consumption and Opioid-Sparing Effect The cumulative consumption of rescue analgesics over the first 48 hours was converted to oral morphine equivalents (OME) for both groups using standard conversion factors7 (oral morphine 10 mg = extended-release oxycodone 6.67 mg; rescue morphine hydrochloride 10 mg = 10 mg OME). The combined group required significantly less rescue medication compared to the opioid group (median OME: 10.0 mg [IQR, 5.0–15.0] vs 25.0 mg [IQR, 15.0–35.0]; Z = −3.842, P < 0.001), representing a median reduction of 60.0% in rescue opioid requirements. When including the scheduled extended-release oxycodone (20 mg OME per day for the opioid group), the total 48-hour OME was 20 + 25 = 45 mg for the opioid group versus 0 + 10 = 10 mg for the combined group, representing a 77.8% reduction in total opioid exposure. Detailed OME consumption data are presented in Table 7. Table 7Rescue Analgesic Consumption (Oral Morphine Equivalents) Over 48 hoursParameterOpioid Group (n=30)Combined Group (n=30)Z ValueP ValueTotal OME (mg), Median (IQR)25.0 (15.0–35.0)10.0 (5.0–15.0)−3.842< 0.001OME reduction in combined group, %/60.0% (40.0–75.0)//Notes: OME, oral morphine equivalents. OME was calculated using standard conversion factors: oral morphine 10 mg = extended-release oxycodone 6.67 mg (as per institutional protocol). Values represent cumulative rescue analgesic consumption only; scheduled extended-release oxycodone in the opioid group (10 mg q12h = 20 mg OME per day) is not included in this table but is discussed in the text. P value derived from Mann–Whitney U-test. Rescue Analgesic Consumption (Oral Morphine Equivalents) Over 48 hours Notes: OME, oral morphine equivalents. OME was calculated using standard conversion factors: oral morphine 10 mg = extended-release oxycodone 6.67 mg (as per institutional protocol). Values represent cumulative rescue analgesic consumption only; scheduled extended-release oxycodone in the opioid group (10 mg q12h = 20 mg OME per day) is not included in this table but is discussed in the text. P value derived from Mann–Whitney U-test. Discussion Head and facial pain, particularly the persistent and severe pain driven by tumor invasion of cranial nerves and skull base bones, profoundly impairs the quality of life in patients with NPC. In this context, NPC-related head and facial pain should be interpreted with careful consideration of cranial nerve and skull-base involvement, as atypical cranial neuropathies accompanied by severe headache may reflect local tumor invasion or perineural spread.19 These anatomic and pathophysiological characteristics support the clinical relevance of targeting both neuropathic (via gabapentin) and inflammatory (via celecoxib) pain pathways simultaneously in this patient population.20 The primary objective of this exploratory retrospective analysis was to evaluate the association of the combination of celecoxib and gabapentin with improved pain management and improve sleep architecture in this population, thereby reducing dependence on opioid medications. Robustly supported by GEE modeling, our findings demonstrated that the combined regimen yielded a significantly greater absolute reduction in NRS scores at 48 hours compared to conventional opioid monotherapy. Furthermore, patients receiving the combination therapy exhibited a more rapid onset of action, a substantial opioid-sparing effect through reduced rescue analgesic requirements, and a highly favorable tolerability profile. These results are consistent with a potential additive or complementary interaction between the two agents in managing complex tumor-related pain in NPC, although this interpretation remains speculative given the observational study design. Nonsteroidal anti-inflammatory drugs remain a cornerstone for tumor-related pain management, with celecoxib being a primary representative. Celecoxib exerts its analgesic effect by selectively inhibiting COX-2, leading to decreased release of inflammatory mediators and the interruption of pain signaling at the peripheral level.21 Conversely, while the precise mechanism of gabapentin is multifaceted, it is generally believed to mitigate central sensitization-induced pain by inhibiting voltage-gated calcium channels and reducing overall neuronal excitability.15 Consequently, the concurrent administration of these two agents theoretically blocks both peripheral nociceptive and central neuropathic pain pathways, resulting in enhanced overall analgesic efficacy. Due to the intricate pain mechanisms associated with cranial nerve involvement and skull base erosion, NPC patients often endure refractory pain that necessitates complex opioid dose titrations and carries the risk of addiction. Therefore, based on their complementary pharmacological profiles, this study is the first to report on the specific application of celecoxib and gabapentin for managing local invasion-induced head and facial pain in NPC. This complementary mechanism has been suggested by evidence from various animal models;22 however, robust clinical validation within cancer pain populations remains historically been limited. Previous literature has suggested that gabapentin and celecoxib can individually serve as effective strategies for treating hemicrania continua and new daily persistent headache.23,24 Another investigation highlighted that the combination of these two drugs was effective in controlling postoperative pain while maintaining a low incidence of adverse events, ultimately reducing patients’ opioid requirements.25 Regarding safety in oncology settings, a systematic review and meta-analysis confirmed the efficacy and tolerability of celecoxib when combined with antitumor therapies.26 Similarly, another review indicated that gabapentin effectively enhances pain control in opioid-treated neuropathic cancer pain and can mitigate unplanned treatment interruptions in head and neck cancer patients undergoing radiotherapy or chemoradiotherapy.27 Despite these insights, direct clinical evidence regarding their combined utility specifically for tumor-related pain remains remarkably scarce. Our findings should be viewed within the broader context of multimodal supportive care strategies being developed for head and neck cancer patients. Recent evidence has highlighted the importance of comprehensive symptom control, including pain management, in optimizing oncologic outcomes and quality of life in this population. Specifically, in head and neck oncology, electrochemotherapy has been shown to improve pain control and quality of life in patients without other curative options, underscoring the broader relevance of supportive care interventions that target symptom burden beyond tumor-directed therapy.28 The present opioid-sparing approach aligns with these broader efforts to minimize treatment-related morbidity and improve the overall supportive care experience for head and neck cancer patients undergoing active treatment. Our longitudinal results are consistent with prior hypotheses that combination therapy may offer advantages over monotherapy, particularly for complex pain syndromes harboring both nociceptive and neuropathic components. Importantly, this study provides real-world evidence that this non-opioid-based regimen offers a effective alternative, aligning with modern oncological efforts to reduce opioid reliance and mitigate dependency risks. Quantitative comparison with previous multimodal analgesia studies. To contextualize the magnitude of our findings, it is instructive to compare the observed effect sizes with those reported in prior studies of multimodal analgesic strategies. A systematic review of multimodal analgesia in head and neck surgery reported postoperative NRS reductions of 0.5 to 1.5 points at 24–48 hours when non-opioid adjuvants were added to opioid-based regimens.12 Similarly, a network meta-analysis of non-opioid analgesics for cancer pain found that most combination regimens yielded NRS reductions of less than 1.0 point compared with opioids alone.13 In contrast, the present study observed an adjusted between-group difference of 2.15 points in NRS reduction at 48 hours—a magnitude that exceeds the minimal clinically important difference (MCID = 1.0–1.5) by a substantial margin and is larger than what is typically reported in the literature. This unusually large effect size warrants careful consideration of several potential sources of amplification. First, the baseline pain intensity in both groups was moderate-to-severe (median NRS 6.5), leaving substantial room for improvement; studies with lower baseline pain levels may have less capacity to detect large between-group differences. Second, the opioid monotherapy group received a relatively conservative starting dose of extended-release oxycodone (10 mg every 12 hours), which may have resulted in suboptimal pain control in the comparator arm, thereby inflating the apparent benefit of the combined regimen. Third, the non-blinded, retrospective design introduces the potential for ascertainment bias, as both patients and treating physicians were aware of the treatment assignment, which may have influenced pain reporting, particularly for subjective outcomes such as NRS and ISI. Fourth, treatment allocation was based on physician judgment and patient preference rather than randomization, introducing the potential for selection bias; for instance, patients with more favorable prognostic factors or better treatment adherence may have been preferentially assigned to the combined regimen. These considerations do not negate the observed associations but suggest that the true effect size in a controlled, blinded setting may be more modest than what we observed. Accordingly, our findings should be interpreted as preliminary and hypothesis-generating, and prospective randomized trials with blinded outcome assessment are needed to confirm the magnitude of the treatment effect. Furthermore, in cancer patients, sleep quality is inextricably linked to levels of depression and anxiety, and inadequate pain control is a primary driver of sleep deterioration.29 Our GEE analysis revealed a significant treatment-by-time interaction for ISI scores, demonstrating that the combination therapy provided significantly greater improvement in ISI scores from baseline compared with opioid therapy at both 24 and 48 hours (estimated MD in reduction: 9.20 and 9.85, respectively; P < 0.001 for both). This finding indicates that the combined regimen restored normal clinical sleep significantly faster and more effectively than opioid monotherapy when evaluated as change scores adjusted for baseline values. Finally, exact statistical testing confirmed that the combination regimen was associated with a significantly lower incidence of nausea and vomiting compared with opioid monotherapy, while other adverse events including constipation, dizziness, and somnolence showed no statistically significant between-group differences. This suggests that the tolerability advantage of the combined regimen is primarily driven by reduced gastrointestinal adverse effects related to opioid-sparing. Despite these compelling results, several limitations inherent to the retrospective design must be carefully considered. First, treatment allocation was not randomized, introducing the potential for selection bias. Although our statistical models adjusted for key baseline covariates (such as TNM stage and invasion sites), residual confounding from unmeasured variables cannot be entirely ruled out. Second, the relatively small sample size from a single institution limited our statistical power to fully explore heterogeneity across different patient subgroups. Third, and most importantly, the 48-hour observation window evaluated only short-term analgesic efficacy and safety. No conclusions can be drawn regarding medium- or long-term pain control, opioid consumption trajectories, or cumulative safety outcomes beyond the initial 48-hour period. Extended follow-up data are essential to determine the durability of the observed benefits and the potential for delayed adverse events. Fourth, the Insomnia Severity Index (ISI) is a validated instrument for assessing sleep quality over a two-week recall period; its administration at 24 and 48 hours post-treatment required adaptation of the recall window to the past 24 hours. Although this modified recall approach has been used in prior perioperative and acute pain studies, it has not been formally validated for short-term ISI administration. This measurement limitation should be considered when interpreting the ISI improvement observed in the combined group. Given these methodological constraints, our findings should be interpreted as preliminary and hypothesis-generating rather than definitive. To definitively address these limitations, future research should employ prospective, multi-center, randomized controlled designs to validate our findings and establish standardized protocols for this combination therapy. Such prospective studies should include longer follow-up periods, larger sample sizes, and predefined subgroup analyses to better characterize the patient populations most likely to benefit from this opioid-sparing strategy. Conclusion This study represents the first exploratory analysis to report the clinical application of a celecoxib and gabapentin combination regimen in newly diagnosed NPC patients. The findings suggest substantial clinical efficacy in alleviating head and facial pain secondary to local tumor invasion within the 48-hour observation period. Supported by robust longitudinal modeling, this combined regimen was associated with more favorable short-term outcomes compared to traditional opioid-based therapy, offering a faster onset of pain relief, significantly accelerated recovery of sleep quality, and a reduced incidence of nausea and vomiting within the 48-hour observation period. These data provide preliminary support for the celecoxib-gabapentin combination as a potentially viable, opioid-sparing alternative for comprehensive cancer pain management in this cohort, but the findings should be considered preliminary and hypothesis-generating. Future prospective randomized trials are warranted to optimize dosing strategies and treatment durations, alongside biomarker studies to identify the specific patient populations most likely to benefit from this combination approach. Funding Statement This work was supported by the Gansu Provincial Health Commission Research Project (Grant No. GSWSKY2025-53). Data Sharing Statement The datasets used and analysed during the current study available from the corresponding author on reasonable request. Ethics Approval and Consent to Participate This retrospective observational study was approved by the Ethics Committee of Lanzhou University Second Hospital (IRB No.: 2024A-959). The requirement for informed consent was waived by the committee due to the retrospective nature of the study. All patient data were anonymized and de-identified prior to analysis to ensure strict confidentiality. The study was conducted in full compliance with the principles of the Declaration of Helsinki. Author Contributions All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work. Disclosure The authors declared that they have no conflicts of interest regarding this work. References Associated Data Data Availability Statement The datasets used and analysed during the current study available from the corresponding author on reasonable request.