Four databases were systematically searched by two authors (YW Fu and Q Liu) independently: PubMed, Embase, Cochrane Library, and Web of Science. The last search was conducted on 26 September 2022. English search terms were as follows: “Opioid” OR “Opioids” OR “Opiate” OR “Opiates” OR “Opium” OR “Papaveretum” OR “Omnopon” OR “Pantopon” OR “Morphine” OR “Morphia” OR “MS Contin” OR “Oramorph SR” OR “Duramorph” OR “Hydromorphone” OR “Dihydromorphinone” OR “Hydromorphon” OR “Dilaudid” OR “Hydrocodone “OR “Hydrocodon” OR “Dihydrocodeinone” OR “Dicodid” OR “Hycodan” OR “Oxymorphone” OR “Numorphan” OR “Oxycodone” OR “Dihydrone” OR “Eucodal” OR “Oxycontin” OR “Fentanyl” OR “Fentanil” OR “Phentanyl” OR “Fentanest” OR “Sublimaze” OR “Fentora” OR “Sufenta” OR “Heroin” OR “Meperidine” OR “Pethidine” OR “Dolosal” OR “Dolcontral” OR “Dolantin” OR “Demerol” OR “Lidol” OR “Lydol” OR “Operidine EPJ I” OR “Levorphanol” OR “Levorphan” OR “Codeine” OR “Tramadol” OR “Tramundin” OR “Zydol” OR “Adolonta” OR “Contramal” OR “Methadone” OR “Dolophine” OR “Metadol” OR “Symoron” OR “Methadose” OR “Phenadone” OR “Physeptone” OR “Amidone” OR “Methaddict” OR “Buprenorphine” OR “Buprenex” OR “Subutex” OR “Pentazocine” OR “Talwin” OR “Fortral” OR “Butorphanol” OR “Moradol” OR “Stadol” OR “Torbugesic” OR “Nalbuphine” OR “Nubain” OR “Dezocine” AND “acute trauma*” OR “trauma pain*” OR “traumatic pain*” OR “traumatic injur*” OR “acute pain*” OR “acute injur*” OR “acute wound*“. The references of relevant meta-analyses were manually retrieved to avoid missing qualified studies. Titles and abstracts of the retrieved studies were screened primarily, and full texts were used for subsequent screening. This network meta-analysis was performed following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) extension statement for network meta-analyses (Hutton et al., 2015).

Inclusion criteria included: 1) studies on patients with traumatic pain in the emergency department (population); 2) studies on opioids such as morphine, fentanyl, tramadol, oxycodone, sufentanil, and codeine (intervention and comparator); 3) studies on the following outcomes: pain relief, adverse events (dizziness, hypotension, pruritus, sedation), and rescue analgesia (outcome); 4) RCTs, cohort studies, and non-randomized controlled trials (study design). Pain relief was defined as the difference in pain scores between before treatment and after treatment, indicating the degrees of pain reduction.

Exclusion criteria included: 1) studies on a mixed population, such as a population of trauma patients and patients with other acute pain; 2) studies with a control group using non-opioid drugs alone, such as acetaminophen, ketamine, lidocaine, and ibuprofen; 3) studies on animal experiments; 4) studies with incomplete or un-extractable data; 5) case reports, meeting abstracts, letters, reviews, meta-analyses; 6) non-English publications.

Two independent authors (YW Fu and Q Liu) extracted the following data from the included studies: first author, year of publication, country, study design, study population (indication), drug administration and type, dose, number of participants, sex (male/female), age (years), pain score, trauma type, follow-up time (FU), and outcome. Disagreements were resolved through discussion with another author (H Nie).

For the quality evaluation of RCTs, the Cochrane Collaboration’s tool for risk of bias assessment was used (Higgins et al., 2011). Six domains (selection bias, performance bias, detection bias, attrition bias, reporting bias, and other bias) were evaluated by this tool, and the risk of bias was classified as low, unclear or high. The Methodological Index for Non-Randomized Studies (MINORS) was applied to measure the quality of non-randomized studies (Slim et al., 2003), which had 12 evaluation items in total, each with a score of 0–2 (0: not reported; 1: reported but inadequate; 2: reported and adequate). A MINORS score of 1–12 was regarded as low quality and 13–24 as high quality. The quality of cohort studies was assessed with the modified Newcastle-Ottawa scale (NOS). This scale had a total score of 9, with 0–3 as poor quality, 4–6 as fair quality, and 7–9 as good quality (Stang, 2010). The evidence quality of the network meta-analysis was estimated using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach (Guyatt et al., 2011), which evaluated the overall quality of evidence across the following domains for each outcome: risk of bias, inconsistency, indirectness, imprecision, and other considerations. The quality of evidence was graded as high, moderate, low or very low.

The network meta-analysis allows assessment of multiple interventions in a single analysis via combining direct and indirect evidence. Direct evidence means evidence from studies directly comparing interventions A and B (head-to-head comparison) and is an estimate of relative effects between A and B. Indirect evidence refers to the evidence obtained from paths with one or more common comparators. For instance, when there are no studies that directly compare A and B (direct evidence), an indirect comparison between A and B can be made if both are compared with C in studies (indirect evidence) (Rouse et al., 2017; Mahil et al., 2020; Papakonstantinou et al., 2020; Devall et al., 2021). This network meta-analysis was carried out using a Bayesian framework and a Monte Carlo Markov Chain (MCMC) model. The number of model chains was 4, the number of initial iterations was 20,000, the number of updated iterations was 50,000, and the step size was 1. Heterogeneity referred to the overall degree of difference in the same pair of comparisons. The I² statistic was the main indicator of statistical heterogeneity, with values < 25%, 25%–50% and >50% signifying low, moderate and high heterogeneity, respectively. Consistency referred to the statistical consistency between direct and indirect effect sizes for the same comparison. The deviation information criterions (DICs) of the consistency model and the non-consistency model were compared, and a small value suggested a better fit. The absolute value of the difference in the DICs within 5 denoted consistency between indirect and direct evidence.

Network plots were drawn to exhibit direct and indirect comparisons for each outcome. A larger node indicated a larger sample size for the medication represented by the node, while a thicker line indicates a larger number of studies for comparison of the medications at both ends of the line. Forest plots and league tables were depicted to illustrate the effects of opioids on the outcomes. For pain relief, weighted mean differences (WMDs) and 95% credibility intervals (CrIs) were reported; for adverse events and rescue analgesia, relative risks (RRs) and 95%CrIs were displayed. Rank probabilities were utilized to rank the efficacy of different opioids for each outcome. STATA 15.1 (Stata Corporation, College Station, TX, United States), RevMan 5.4 (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark) and R 4.1.3 (R Foundation for Statistical Computing, Vienna, Austria) software was adopted for statistical analysis.

A total of 19,720 were retrieved from PubMed (n = 3,577), Embase (n = 5,113), Cochrane Library (n = 8,024), and Web of Science (n = 3,006). After deleting duplicates (n = 7,891), and excluding ineligible studies based on titles and abstracts (n = 11,698) and on full texts (n = 111), 20 studies (Turturro et al., 1998; Vergnion et al., 2001; Miller and Ernst, 2004; Soysal et al., 2004; Marco et al., 2005; Hewitt et al., 2007; Bounes et al., 2010; Shear et al., 2010; Jalili et al., 2012; Smith et al., 2012; Wenderoth et al., 2013; Shervin et al., 2014; Zare et al., 2014; Chang et al., 2017; Chew and Shaharudin, 2017; Eizadi et al., 2018; Pan et al., 2018; Blancher et al., 2019; Vahedi et al., 2019; Bijur et al., 2021) of 3,040 patients were eventually included in this network meta-analysis. Figure 1 shows the flow chart of study screening. These included studies were published from 1998 to 2021, covering 7 countries. The basic characteristics of the included studies are illustrated in Supplementary Table S1. Fourteen medications were involved. There were 17 RCTs, 1 cohort study, and 2 non-randomized controlled trials.

For 17 RCTs, the overall risk of bias was low. Figure 2 exhibits the assessment of risk of bias. One cohort was of medium quality, and two non-randomized controlled trials were of high quality. The quality of evidence was moderate for pain relief and dizziness, and was low for hypotension, pruritus, sedation, and rescue analgesia. Quality of evidence evaluation for this network meta-analysis is shown by Supplementary Table S2 in detail.

The effect of opioid medications on pain relief was investigated in 13 studies of 2,225 patients, involving 11 medications: buprenorphine, odeine + acetaminophen, fentanyl, fentanyl + tramadol, hydrocodone + acetaminophen, morphine, oxycodone, oxycodone + acetaminophen, sufentanil, tramadol, and tramadol + acetaminophen. Morphine and hydrocodone + acetaminophen were used in more patients than other medications. There were more studies for direct comparisons between morphine and fentanyl, and between oxycodone + acetaminophen and hydrocodone + acetaminophen (Supplementary Figure S1).

According to the forest plot, sufentanil relieved significantly more pain than morphine (pooled WMD = 0.70, 95%CrI: 0.01, 1.40) (Supplementary Figure S2). The league table showed that in contrast to buprenorphine (pooled WMD = -2.86, 95%CrI: −4.51, −1.21) or fentanyl (pooled WMD = −2.69, 95%CrI: −4.22, −1.15), fentanyl + tramadol was significantly less efficacious in pain relief. Hydrocodone + acetaminophen had a significantly reduced effect on pain relief than buprenorphine (pooled WMD = −0.98, 95%CrI: −1.91, −0.05) or fentanyl (pooled WMD = -0.81, 95%CrI: −1.52, −0.09). Oxycodone was inferior to buprenorphine (pooled WMD = −1.17, 95%CrI: −2.32, −0.03) or fentanyl (pooled WMD = −1.00, 95%CrI: −1.98, −0.02) in alleviating pain. Sufentanil exhibited a significantly greater impact on pain relief than codeine + acetaminophen (pooled WMD = 1.23, 95%CrI: 0.23, 2.23), fentanyl + tramadol (pooled WMD = 3.27, 95%CrI: 1.59, 4.95), hydrocodone + acetaminophen (pooled WMD = 1.38, 95%CrI: 0.40, 2.37), morphine (pooled WMD = 0.70, 95%CrI: 0.015, 1.40), and oxycodone (pooled WMD = 1.57, 95%CrI: 0.39, 2.76). Tramadol was significantly less effective in pain relief than buprenorphine (pooled WMD = −3.88, 95%CrI: −5.32, −2.43), fentanyl (pooled WMD = −3.71, 95%CrI: −5.02, −2.39), or sufentanil (pooled WMD = −4.28, 95%CrI: −5.76, −2.81). Tramadol + acetaminophen reduced significantly less pain than sufentanil (pooled WMD = −1.16, 95%CrI: −2.25, −0.06) (Table 1). Based on the rank probability, the top three analgesic medications for pain relief may be sufentanil (78.29%), buprenorphine (48.54%) and fentanyl (53.25%) in sequence (Table 2).

Data on rescue analgesia were provided by 11 studies with 1,755 patients, involving 8 opioid medications: butorphanol, codeine + acetaminophen, fentanyl, hydrocodone + acetaminophen, morphine, oxycodone, oxycodone + acetaminophen, and sufentanil. More studies investigated the direct comparison between morphine and fentanyl, and more patients used these two opioids (Supplementary Figure S1).

The forest plot demonstrated no significant difference between the presented medications (Supplementary Figure S2). According to the league table, compared with patients receiving codeine + acetaminophen (pooled RR = 0.02, 95%CrI: 0.00, 0.71) or hydrocodone + acetaminophe (pooled RR = 0.02, 95%CrI: 0.00, 0.69), patients receiving oxycodone had a significantly reduced need for rescue analgesia. The need for rescue analgesia following oxycodone + acetaminophen was significantly greater than that following oxycodone (pooled RR = 48.20, 95%CrI: 1.35, 3,694.76) (Table 1). The rank probability showed that patients who received oxycodone (83.64%), butorphanol (38.31%) and fentanyl (25.91%) were least likely to need rescue analgesia (Table 2). Soysal et al. (Soysal et al., 2004) reported that patients receiving meperidine + midazolam needed rescue analgesia more frequently than those receiving fentanyl + midazolam (21% vs. 9%).