Drug-specific effects of anxiety on high-dose oxycodone: age as a modifier for oxycodone response

Study setting and population

This study constitutes a retrospective analysis utilizing data from the MIMIC-IV database, which includes electronic health records of patients hospitalized at the Beth Israel Deaconess Medical Centre (BIDMC), a tertiary academic medical center in Boston, Massachusetts, USA, from 2008 to 2019. All data have undergone deidentification and are publicly available. Researchers can access it after completing human research training and signing a data usage agreement. The Institutional Review Board at BIDMC approved the waiver of informed consent and authorized the sharing of these research resources (Johnson et al., 2023). As a result, this study did not require additional ethical review or patient consent.

Refer to the ICD code for hip arthroplasty as published in the relevant literature (Wang et al., 2025). A total of 2286 patients were identified, with 105 cases of hip arthroplasty revision surgery and 2191 cases of primary hip arthroplasty. To minimize confounding factors such as repeated anesthetic drug use, the analysis focused mainly on primary hip arthroplasty patients (n = 2191). Subsequently, the following patients were excluded: those with dementia (n = 148), malignant tumors (n = 101), severe liver disease (n = 18), severe malnutrition (BMI < 18.5 kg/m²) or morbid obesity (BMI > 40 kg/m²) (n = 112), or in-hospital mortality (n = 14). Finally, this yielded 1763 patients over 25 years of age undergoing primary hip arthroplasty. Furthermore, given the variability in opioid prescribing patterns across patients (eg, some patients received morphine but not oxycodone, and vice versa), no additional exclusions were made for the absence of specific opioid administration records. Instead, each opioid type was analyzed separately using all available data for that particular drug.

Exposure

The primary exposure variable in this research was whether patients received an anxiety disorder diagnosis upon hospital admission. Based on this definition, patients were categorized into an anxiety group and a nonanxiety group. Anxiety diagnoses were based on the ICD codes (ICD-10: F41.1, F41.0, F40.10, F40.2, F06.4 and ICD-9: 300.02, 300.01, 300.23, 300.29, 293.84).

Opioid tolerance

Drug tolerance refers to the reduced effectiveness of a medication and increased pain sensitivity that can occur with long-term or repeated use, often requiring higher doses to achieve the same analgesic effect (Khan and Mehan, 2021). In this study, patients’ tolerance levels were determined by analyzing dosage data for three opioids: morphine, oxycodone, and tramadol. Patients were then categorized based on their dosage levels. Specifically, morphine and tramadol were divided into low- and high-dose groups, while oxycodone, which had a more concentrated dosage distribution, was further divided into low-, mid-, and high-dose groups.

Collected variables

The following variables were collected to control for confounding factors: (a) demographic characteristics: age, BMI, race, marital status, and gender; (b) healthcare coverage and comorbidities: health insurance, diabetes, hypertension, peripheral nerve disorder, perioperative complications, and anemia; (c) health assessment scores: Sequential Organ Failure Assessment at 24 h, Simplified Acute Physiology Score II, Systemic Inflammatory Response Syndrome, Oxford Acute Severity of Illness Score, and Charlson Comorbidity Index; (d) other medications used: warfarin, aspirin, heparin, lorazepam, and vancomycin; (e) surgical-related information: type of hip arthroplasty surgery; (f) hospitalization information: admission type, admission to ICU, and 7-day hospital length of stay; and (g) laboratory parameters: anion gap, blood urea nitrogen (BUN), creatinine, platelets, red blood cell distribution width, and white blood cell count.

Mechanistic exploration

To investigate the potential mechanisms underlying anxiety and differential opioid tolerance at the molecular level, this study utilized transcriptome data from the Gene Expression Omnibus (GEO) database. Given the absence of human transcriptional data related to opioid analgesia within the GEO repository, we selected a rat dataset (GSE53897) as the subject for preliminary mechanistic exploration. This dataset was derived from a study by Thibault et al. (2014), which established a vincristine-induced neuropathic pain model in Sprague–Dawley rats. Neuropathic pain was induced by intraperitoneal injection of vincristine (0.1 mg/kg/day) over two 5-day cycles, followed by intraperitoneal administration of morphine or oxycodone (3.33 mg/kg/day) or saline for 5 consecutive days. L2-L5 dorsal root ganglia were harvested for RNA extraction and microarray analysis. Detailed experimental procedures are available in the original publication (Thibault et al., 2014). To focus on molecular changes related to analgesic effects, this study selected and compared two groups treated with morphine or oxycodone following neuropathic pain induction: the ‘vincristine + morphine’ group and the ‘vincristine + oxycodone’ group, comprising four samples per group for a total of eight samples.

Initially, box plots were created to preliminarily assess heterogeneity among samples. Subsequently, expression data from RNA sequencing underwent log₂(X + 1) transformation for standardization. Using the ‘Limma’ package in R, key differentially expressed genes (DEGs) between the two groups were systematically identified based on the screening criteria of an adjusted P value less than 0.05 and |log₂(fold change)|(log₂ FC) > 0.5. Volcano plots were generated using GraphPad Prism 8 software to visually represent the distribution of DEGs. Finally, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis and Gene Ontology Biological Processes analysis were performed on the key DEGs. This aimed to elucidate the biological mechanisms underlying the functional heterogeneity of opioid tolerance.

Statistical analysis

For categorical variables, statistical description was performed using counts and composition ratios (percentages). For continuous variables, normality was first assessed using the Shapiro–Wilk test. Where data met normal distribution, description employed mean ± SD; otherwise, median and quartiles were used. For statistical inference, comparisons of categorical variables between groups were performed using χ² tests. Comparisons of continuous variables between groups were determined based on their distribution: independent samples t tests were used if data were normally distributed with homogeneity of variance; otherwise, the nonparametric Mann–Whitney U test was employed.

Multiple logistic regression models were constructed to investigate the association between anxiety and tolerance to three opioid medications: morphine, oxycodone, and tramadol. Model 1 represented an unadjusted crude model; model 2 adjusted for baseline differences in demographic characteristics; model 3 adjusted for baseline differences in laboratory data; model 4 adjusted for differences in surgery-related factors and other medications. To assess the robustness of our findings against potential unmeasured confounding, we calculated the E-value for each significant association. In addition, Johnson-Neyman analysis was employed to investigate potential interactions between anxiety and other variables on opioid tolerance. The Johnson-Neyman technique is a probing method for interactions between continuous and categorical moderators. Unlike traditional simple slope analysis that arbitrarily selects moderator values, the Johnson-Neyman technique identifies specific threshold values of the continuous moderator at which the effect of the categorical predictor transitions between statistical significance and nonsignificance, thereby providing a more objective and data-driven approach for interaction detection. Restricted cubic spline (RCS) analysis was then used to determine threshold nodes for interaction variables. Following stratification of patients based on this RCS threshold node, further stratified analyses were conducted to investigate the association between anxiety and key opioid tolerance. All statistical analyses were performed using R software and SPSS 26.0, with P < 0.05 indicating statistical significance.

Participant demographics

This study included 1763 patients undergoing primary hip arthroplasty, with a median age of 68.000 years. Among these, 198 patients (11.231%) were diagnosed with anxiety at admission. Comparing baseline characteristics between the anxiety and nonanxiety groups showed differences in age, race, gender, type of hip arthroplasty, lorazepam use, creatinine, and BUN (all P < 0.05). Patients in the anxiety group exhibited a lower age (63.000 vs. 68.000), male proportion (35.859 vs. 45.112%), BUN (15.000 vs. 16.000), and creatinine (0.800 vs. 0.900) than in the nonanxiety group. On the other hand, the anxiety group had a higher proportion of White race patients (87.879 vs. 79.553%), total hip arthroplasty (87.879 vs. 78.850%), and lorazepam use (30.303 vs. 18.530%) than the nonanxiety group (Table 1). Regarding opioid use, no significant differences were observed between groups in the use of morphine or tramadol, nor in their dose groups (low/high dose) (all P > 0.05) (Fig. 1a–d). Although the difference in oxycodone use between the anxiety and nonanxiety groups was marginal (55.556 vs. 48.179%; P = 0.050), a more pronounced difference was observed in oxycodone dosage groups (low/mid/high dose) (P = 0.007) (Fig. 1e and f).

Association between patient anxiety and opioid tolerance

This study aimed to clarify the link between anxiety and opioid tolerance by systematically evaluating the correlation between patients’ anxiety status and their tolerance to different opioids, while controlling for confounding variables through logistic regression analysis (Table 2). The results indicated that anxiety was consistently not linked to the use of high-dose morphine or tramadol in both unadjusted and adjusted models (all P > 0.05). However, the findings regarding oxycodone use exhibited greater complexity. In models 1, 3, and 4, anxiety was associated with high-dose oxycodone use [model 1: odds ratio (OR) = 2.025, 95% confidence interval (CI): 1.283–3.196, E-value = 3.180; model 3: OR = 1.786, 95% CI: 1.129–2.827, E-value = 2.580; model 4: OR = 1.679, 95% CI: 1.046–2.694, E-value = 2.275]. However, in model 2, which adjusted for demographic characteristics, anxiety was not linked to high-dose oxycodone use (OR = 1.370, 95% CI: 0.838–2.240), suggesting that demographic factors might significantly confound the relationship between anxiety and high-dose oxycodone use.

Age influences the association between anxiety and oxycodone dosage

Regarding the association between anxiety and oxycodone dose, we observed that this relationship disappeared after adjusting for demographic characteristics (particularly age) in model 2. Analysis demonstrated that age was the key confounding factor responsible for this disappearance. The association re-emerged when the age variable was removed from the model (Fig. 2a).

Further analysis indicated a significant interaction between age and anxiety in influencing oxycodone dose (interaction OR = 0.959, 95% CI: 0.918–0.998). Full adjustment Johnson-Neyman analysis revealed that age under 50 significantly influenced the association between anxiety and oxycodone dose (Fig. 2b). However, owing to the small sample size (n = 78) with complete data for those under 50 years in this study, including only 12 patients with anxiety, and the division of the dependent variable (oxycodone dosage) into three groups, anxiety patients were sparsely distributed across dosage groups, resulting in insufficient statistical power. As a result, we did not use 50 years as the grouping threshold. To address this, we used RCS analysis to identify age thresholds affecting oxycodone dose. The results indicated 75 years as the critical point for risk change (Fig. 2c). Consequently, we stratified the population using 75 years as the age boundary. The stratified analysis revealed that, among patients aged less than or equal to 75 years, anxiety status was an independent risk factor for oxycodone tolerance (P = 0.042). In patients aged more than 75 years, no significant association was observed between anxiety and oxycodone tolerance (P = 0.102) (Table 3).

Subsequently, within the cohort aged 75 years or younger, patients were further grouped into age brackets of less than or equal to 60 years and 5-year increments to investigate the presence of an age gradient effect, thereby enabling more precise identification of the critical window period for the association between anxiety and oxycodone dosage. Fully adjusted logistic regression analysis observed a trend of ‘decreasing OR values with increasing age’, suggesting that the strength of the association between anxiety and oxycodone dose may diminish with advancing age (Fig. 3a). Johnson-Neyman analysis demonstrated changes in OR values for the association between anxiety and oxycodone dose as age increased. This similarly confirmed the trend of declining OR values with increasing age (Fig. 3b).

Exploring mechanisms underlying heterogeneous associations between anxiety and oxycodone dose

The above findings indicated that, among three opioid medications, anxiety was only associated with high-dose oxycodone use. Further exploration of differences in pain responses to oxycodone vs. other opioids (morphine) via the GEO database revealed potential underlying mechanisms. Boxplots showed balanced samples with no noticeable batch effects (Fig. 4a). The ‘limma’ method identified 10 409 DEGs between the oxycodone group and the morphine group (Fig. 4b). Subsequently, key DEGs were filtered based on |log₂FC| > 0.5 and P < 0.05, yielding 255 DEGs. KEGG analysis revealed that these 255 DEGs were primarily enriched in the neuroactive ligand signaling pathway and the peroxisome pathway. The biological processes involved mainly comprised positive regulation of the mitogen-activated protein kinase (MAPK) activity, hydrogen peroxide metabolism, and response to interleukin-1 (Table 4).

Conclusion

This study demonstrated that preoperative anxiety disorders constituted an independent factor of increased postoperative oxycodone requirements, with this association exhibiting drug specificity. Age emerged as a critical effect modifier, with 75 years serving as a significant threshold for risk stratification. These findings not only enriched the ‘bio-psycho-social’ model of perioperative pain management but also advocate for individualized analgesic strategies stratified by age and psychological characteristics. By identifying high-risk individuals preoperatively and implementing enhanced multimodal analgesia alongside psychological support, it is anticipated that effective pain control can be achieved while minimizing opioid-related risks, thereby advancing accelerated recovery surgery and improving overall patient outcomes.

Conflicts of interest

There are no conflicts of interest.