Associations of dexamethasone's effect on morphine consumption after total knee arthroplasty—Subgroup analyses

The DEXamethasone twice for pain treatment after Total Knee Arthroplasty (DEX‐2‐TKA) trial showed that adding one and two doses of 24 mg intravenous dexamethasone to paracetamol, ibuprofen and local infiltration analgesia, reduced morphine consumption (primary outcome) within 48 h after TKA. We aimed to explore the differences in the effect of dexamethasone on morphine consumption in different subgroups. Quantile regression adjusted for site was used to test for significant interaction between the predefined dichotomised subgroups and treatment group. The subgroups were defined based on baseline data: sex (male/female), age (≤65 years/>65 years), American Society of Anaesthesiologists (ASA)‐score (ASA I + II/III), visual analogue score of preoperative pain at rest (≤30 mm/>30 mm), pain during mobilisation (≤30 mm/>30 mm), type of anaesthesia (spinal anaesthesia/general anaesthesia and spinal converted to general anaesthesia), and prior daily use of analgesics (either paracetamol and/or NSAID/neither). These analyses were supplemented with post hoc multivariate linear regression analyses. Test of interaction comparing sex in the pairwise comparison between DX2 (dexamethasone [24 mg] + dexamethasone [24 mg]) versus placebo (p = .02), showed a larger effect of dexamethasone on morphine consumption in male patients compared to females. Test of interaction comparing age in the pairwise comparison between DX1 (dexamethasone [24 mg] + placebo) versus placebo (p = .04), showed a larger effect of dexamethasone on morphine consumption in younger patients (≤65 years) compared to older. All remaining subgroup analyses showed no evidence of a difference. The supplemental multivariate analyses did not support any significant interaction for sex (p = .256) or age (p = .730) but supported a significant interaction with the type of anaesthesia (p < .001). Our results from the quantile regression analyses indicate that the male sex and younger age (≤65 years) may be associated with a larger analgesic effect of dexamethasone than the effects in other types of patients. However, this is not supported by post‐hoc multivariate linear regression analyses. The two types of analyses both supported a possible interaction with the type of anaesthesia.

(either paracetamol and/or NSAID/neither).These analyses were supplemented with post hoc multivariate linear regression analyses.Test of interaction comparing sex in the pairwise comparison between DX2 (dexamethasone [24 mg] + dexamethasone [24 mg]) versus placebo (p = .02),showed a larger effect of dexamethasone on morphine consumption in male patients compared to females.Test of interaction comparing age in the pairwise comparison between DX1 (dexamethasone [24 mg] + placebo) versus placebo (p = .04),showed a larger effect of dexamethasone on morphine consumption in younger patients (≤65 years) compared to older.All remaining subgroup analyses showed no evidence of a difference.The supplemental multivariate analyses did not support any significant interaction for sex (p = .256)or age (p = .730)but supported a significant interaction with the type of anaesthesia (p < .001).Our results from the quantile regression analyses indicate that the male sex and younger age (≤65 years) may be associated with a larger analgesic effect of dexamethasone than the effects in other types of patients.However, this is not supported by post-hoc multivariate linear regression analyses.The two types of analyses both supported a possible interaction with the type of anaesthesia.

Editorial Comment
In this secondary analysis of a previous randomised clinical trial, the opioid-sparring effect of two doses of dexamethasone on patient subgroups after hip arthroplasty was explored.Assessing the effect through different statistical analyses, indications of a higher opioid-sparing effect in male patients and age under 65 years were found, affected by the choice of anaesthesia.

| INTRODUCTION
It is generally suggested that dexamethasone as an adjuvant to analgesic treatments may reduce pain and thereby patient decides on morphine consumption after surgery. 1,2In addition, dexamethasone is used for the prevention of postoperative nausea and vomiting. 3,4wever, firm evidence for the use of perioperative dexamethasone has previously been lacking. 5The DEXamethasone twice for pain treatment after Total Knee Arthroplasty (DEX-2-TKA) trial 6 showed that one perioperative dose of intravenous dexamethasone (24 mg) as an addition to a basic perioperative analgesic regime (paracetamol, ibuprofen and local infiltration analgesia), reduced morphine consumption the first 48 h after TKA.7][8] Postoperative morphine consumption was chosen as the primary outcome and considered to mirror the total patient experienced pain during the intervention period. 6 effective strategy to alleviate postoperative pain is crucial to long-term outcomes. 9There is broad consensus on a multimodal approach for treating postoperative pain, but the optimal procedurespecific treatment regimens of analgesics, adjuvants and analgesic methods often remain uncertain. 5,10,11Furthermore, previous studies might miss identifying subgroups of patients with specific effects of interventions either beneficial or non-beneficial, so interventions are only introduced to patients who benefit. 12,135][16] A recent randomised clinical trial of patients undergoing TKA showed that the subgroup of high pain responders benefitted in regards to postoperative pain from a larger preoperative dose of dexamethasone (0.3 mg/kg vs. 1 mg/kg). 17This is important knowledge because this subgroup of patients suffers from severe postsurgical pain and is difficult to treat. 18 elucidate factors of relevance for dexamethasone's effect on postoperative morphine consumption, we decided post hoc to protocol, register and perform a subgroup analysis of the DEX-2-TKA trial data.We hypothesised the effect of dexamethasone would vary in selected subgroups (sex, age, American Society of Anaesthesiologists 0][21] As the defined subgroups are based on predictors of postoperative pain, sparse knowledge of the beneficial effect of glucocorticoids on pain in given subgroups exist besides the aforementioned trial. 17Few studies have shown that glucocorticoid injections for other conditions are more beneficial in patients with higher preoperative pain. 22,23udraCT (2018-001099-39, 8 June 2018).All participants provided written informed consent prior to enrolment.This planned post hoc subgroup analysis is based on a statistical analysis plan made publicly available before any analyses were carried out. 24The study is reported according to CONSORT guideline. 25

| DEX-2-TKA
The protocol, 7 the statistical analysis plan 8 and the main results from the trial have been previously published. 6In brief, we randomly allocated 485 patients having elective primary TKA to three intervention groups in a 1:1:1 ratio.The intervention groups were as follows: Group DX1 (dexamethasone [24 mg] + placebo); DX2 (dexamethasone [24 mg] + dexamethasone [24 mg]); or placebo (placebo + placebo).The trial medication was given preoperatively and after 24 h.All participants had patient-controlled analgesia (PCA) with morphine iv (2 mg per bolus, lock-out 6 min) for escape medication for 0-24 h and on-demand oral morphine tablets of 10 mg 24-48 h.The randomisation was stratified by site.Key exclusion criteria were daily use of opioids above 30 mg oral morphine equivalents, body mass index >40 and contraindications or allergies to trial medication.Anaesthesia was standardised (spinal anaesthesia using bupivacaine 0.5% heavy 2-3 mL, or general anaesthesia with propofol and remifentanil, and 0.3 μg/kg sufentanil at the end of surgery).No other pain medication (including peripheral regional anaesthesia, gabapentinoids, etc.) was allowed in addition to trial medication and PCA morphine.
The primary analysis was based on the intention-to-treat population defined as all participants who were randomised.The primary outcome was 0-to 48-h morphine consumption.The median consumptions in the three intervention groups were DX1 37.9 mg, DX2 35.0 mg and placebo 43.0 mg.We performed pairwise comparisons between the three groups (three comparisons).The main findings were that compared to the placebo, one dose of dexamethasone reduced morphine consumption (7.8 mg, 98.3% confidence interval [CI]: 0.7-14.7,p = .008)and two doses resulted in a further reduction (10.7 mg, 98.3% CI: 4.0-17.3,p < .001).

| Subgroups
Before data were analysed, we predefined the following subgroups: (1) sex (male/female), (2) age (≤65 years/>65 years), (3) ASA score (ASA I + II/III), ( 4) preoperative pain at rest measured using VAS (≤30 mm/>30 mm), (5) preoperative pain during mobilisation using VAS (≤30 mm/>30 mm), (6) type of anaesthesia (spinal anaesthesia versus general anaesthesia incl.spinal converted to general anaesthesia) and (7) prior daily use of paracetamol and/or NSAID (either paracetamol or NSAID/ neither).Type of anaesthesia was planned prior to randomisation but could, at the discretion of the anaesthetist, be changed post randomisation.It is the executed type of anaesthesia that we have registered.These subgroups were selected based on prior findings suggesting the higher need for opioids in female patients, younger patients, healthier patients, patients with higher preoperative pain, and patients having general compared to spinal anaesthesia. 19,26

| Outcome
The primary outcome was 0-48 h of morphine consumption. 6

| Statistics
A statistical analysis plan was made publicly available before conducting any analysis. 24Two statisticians (MHO and JCJ) performed the statistical analyses independently and, thereafter, agreed on the results.Their final statistical report can be found in the Supporting Information, the two original reports, and a discrepancies log.The final report was approved by all co-authors.
The statisticians performed pairwise comparisons of the three intervention groups (three comparisons).First, subgroup heterogeneity of the intervention effect was tested between the predefined dichotomised subgroups and treatment group using quantile regression adjusted for site to test the interaction between the subgroup and the treatment group.Significance level for the specific interactions was set to a threshold of p value less than .05.Second, if the quantile regression showed a significant effect on the predefined subgroup when compared between the two of the three intervention groups, pairwise comparisons (as the one used in the main article) on the effects of dexamethasone were assessed for each of the subgroups with separate analyses using van Elteren test adjusted for site.
For these analyses, we used the same threshold for statistical significance as we used in the main DEX-2-TKA publication ( p < .017). 6nally, the differences between the groups were then quantified using Hodges-Lehmann median difference with corresponding 98.3% confidence intervals.
Furthermore, post hoc supplemental multivariate linear regression analyses were performed testing the interaction of sex, age, ASA, preoperative pain at rest and during activity, type of anaesthesia and preoperative pain medicine.

| RESULTS
Between September 2018 and March 2020, 485 participants were enrolled in the DEX-2-TKA trial.Thirteen participants (2.7%) were not included in the subgroup analyses because of missing data for the primary outcome.The number of participants in each subgroup per intervention group and baseline characteristics is presented in Table 1 for the entire intention-to-treat population.Median morphine consumptions for all subgroups are presented in Table 2.

| Males compared to females
Test of interaction for sex (male/female) in the pairwise comparison between DX2 versus placebo showed evidence of a difference ( p = .015).When females were analysed separately, the van Elteren test showed no evidence of a difference (p = .094).The Hodges-Lehmann median difference was À6 mg morphine (98.3% CI: À14.7 to 2.0).When males were analysed separately, the van Elteren test showed evidence of a beneficial effect (p < .001).The Hodges-Lehmann median difference was À16.7 mg morphine (98.3% CI: À28.0 to À5.3).
Test of interaction for sex in the pairwise comparison between DX1 versus DX2 and DX1 versus placebo showed no evidence of a difference (p = .420and p = .173).p Values for the interaction tests and the forest plots for the pairwise comparisons in the sex subgroup are shown in Figure 1.

| Participants below 65 years compared to participants above 65 years
Test of interaction for age (≤65 years/>65 years) in the pairwise comparison between DX1 versus placebo showed evidence of a difference (p = .040).
Test of interaction for age in the pairwise comparison between DX1 versus DX2 and DX2 versus placebo showed no evidence of a difference ( p = .284and p = .120).p Values for test of interaction and the forest plot for the pairwise comparisons in the age subgroup are shown in Figure 2.

| Spinal anaesthesia compared to general anaesthesia and converted to general anaesthesia
Test of interaction for type of anaesthesia (spinal anaesthesia/general anaesthesia and converted to general anaesthesia) in the pairwise comparison between DX1 versus placebo showed evidence of a difference ( p = .038).However, none of the van Elteren tests showed evidence of a beneficial effect for either spinal anaesthesia ( p = .056)or general anaesthesia (p = .020).Hodges-Lehmann median differences were À6 mg morphine (98.3% CI: À13.3 to 1.3) and À16.7 mg morphine (98.3% CI: À33.3 to 0.0), respectively.
p Values for test of interaction and forest plots for the pairwise comparisons in the type of anaesthesia subgroup are shown in Figure S4.
None of the predefined remaining tests of interaction showed evidence of a difference.p Values for test of interaction and forest plots for all other subgroups are in the supplemented material (Figure S1-S3 and S5).

| Additional analyses
Analyses with a different cut of preoperative VAS scores (≤70 mm/ >70 mm) were decided upon and conducted after the statistical analysis plan was published.These analyses are also a part of and commented on in the supplemented material.Further, a multivariate linear regression of the defined subgroups was performed to possibly substantiate the findings of the quantile regression analyses.This analysis did not support any significant interaction for sex (p = .256)or age ( p = .730).However, it supported a significant interaction with the type of anaesthesia ( p < .001).Results for all subgroup interactions are embedded in the Supporting Information.

| DISCUSSION
With the present post hoc subgroup analysis of the randomised, multicentre clinical trial DEX-2-TKA investigating the addition of one and two doses of dexamethasone to a basic perioperative non-opioid analgesic regime in patients after TKA, we found that both the male sex and younger age (≤65 years) were associated with a larger analgesic effect of dexamethasone than the effects in other types of patients.
All remaining subgroup analyses showed no evidence of a difference.
As in the main study, morphine consumption was chosen as the outcome parameter comparator between the subgroups.This is the predominant choice when investigating postoperative analgesic interventions, 15,27 as the patient decided the use of morphine over time, which can be understood as a mirror of the cumulated burden of pain.In addition, any reduction of morphine consumption may lower the incidence of opioid-related adverse events and the risk of prolonged opioid use after rehabilitation.

| Factors affecting morphine consumption and pain
Our results indicate that the male sex may be associated with a larger analgesic effect of two doses of dexamethasone than females.A possible explanation is that it has previously been shown that morphine seems to have greater potency and slower analgesic onset and offset in females, 28 possibly providing a more stable analgesic effect of morphine, thus, resulting in a lower consumption in itself.A previous subgroup analysis of patients undergoing total hip arthroplasty and receiving combinations of paracetamol and ibuprofen showed no difference in effect on postoperative morphine consumption when assessing subgroups differences in regard to sex.However, this subgroup analysis might have been underpowered to show a difference. 29Further investigations should assess whether sex plays a specific role in glucocorticoids' modulation of the surgical stress response.
Our results indicate that younger age (≤65 years) may be associated with a larger analgesic effect of dexamethasone than the effects in patients older than 65 years.However, the subgroup difference was only observed in the comparison between DX1 versus placebo, and the interaction test in the pairwise comparison between DX2 versus placebo did not show evidence of a difference, which may indicate that the significant test of interaction may be a chance finding.

| Strengths and limitations
Our trial has several strengths.Data arise from a large, multicentre trial with few missing data and an overall low risk of bias. 30The subgroups (except preoperative pain) were defined in the statistical analysis plan for the main trial 8 and the statistical methodology and analysis plan were defined, registered and made publicly available before the analysis began. 24Two independent statisticians performed the analyses.The statistical significance threshold was adjusted according to the number of intervention groups.
Our trial also has limitations.First, low power is an inherent problem of subgroup analyses resulting in an increased risk of type II errors, and the number of additional outcome comparisons results in an increased risk of type I errors.Therefore, the present analyses' results are explorative and hypothesis generating regarding positive findings and cannot be used to discard possible clinically relevant subgroup effects that were not detected. 310][21] The subgroups defined and analysed in this paper do not constitute an exhaustive list of characteristics that may be associated with pain levels and opioid requirements postoperatively.Pain catastrophising or response to experimental pain are psychological factors that may play a role. 17,32,33Consequently, the pragmatic nature of the DEX-2-TKA trial did not entail such patient characteristics to be recorded.Therefore, substantial covariance regarding psychological factors or co-morbidities such as diabetes or obesity are not accounted for.Third, we dichotomised the subgroups regarding age, VAS scores, ASA scores, and prior use of analgesics.Other results may have been shown if other cut-offs for dichotomisation were made.Fourth, a subgroup analysis comparing patients with or without prior use of opioids would be interesting.However, the number of patients with opioid use before surgery was too low to conduct a meaningful analysis, and additionally our trial only included patients with opioid consumption below 30 mg of oral morphine equivalent dose before surgery, and therefore does not offer information for patients with higher daily consumption.Fifth, the equal distribution of all baseline characteristics in the compared groups may be compromised when subgrouping is not based on variables used as stratification variables in the randomisation process, thus, increasing the risks of random errors.Finally, except for a possible interaction with type of anaesthesia, our results were not supported by an additional multivariate linear regression.These limitations need to be considered when interpreting our results.

| CONCLUSION
Our results from the quantile regression analyses indicate that the male sex and younger age (≤65 years) may be associated with a larger analgesic effect of dexamethasone than the effects in other types of patients.However, this is not supported by post-hoc multivariate linear regression analyses.The two types of analyses both support a possible interaction with the type of anaesthesia.

The DEX- 2 -
TKA trial was a randomised, blinded, three-group multicentre clinical trial in five Danish hospitals investigating the effects of adding one and two doses of dexamethasone to a basic perioperative analgesic regimen with paracetamol, ibuprofen and local infiltration analgesia in patients after TKA.The trial was conducted in accordance with the Declaration of Helsinki and monitored by the Good Clinical Practice Units at Odense and Copenhagen University Hospitals.Ethics approval was granted by the Biomedical Research Ethics Committee of Region Zealand (SJ-695, 7 May 2018), and the trial was preregistered at ClinicalTrials.gov(NCT03506789, 13 April 2019) and

F I G U R E 1
Test of interaction for sex (male/female) and forest plot of pairwise comparisons of the sex subgroup of differences in 48-h morphine consumption.Significance level for sex as covariate between the intervention groups are p values to the left.In the forest plot, negative values indicate lower morphine consumption in the first group.Whiskers are 98.3% confidence intervals and p values to the right are significance of the comparison of the specific subgroups of the intervention groups.F I G U R E 2 Test of interaction for age (≤ 65 years/> 65 years) and forest plot of pairwise comparisons of the sex subgroup of differences in 48-h morphine consumption.Significance level for age as covariate between the intervention groups are p values to the left.In the forest plot, negative values indicate lower morphine consumption in the first group.Whiskers are 98.3% confidence intervals and p values to the right are significance of the comparison of the specific subgroups of the intervention groups.
T A B L E 1 Patient characteristics across the different subgroups.
T A B L E 2 Morphine consumption 0-48 h across the different subgroups.