This meta-analysis followed the ‘Preferred Reporting Items for Systematic Reviews and Meta-Analyses’ (PRISMA; Appendix 1). The study was registered in the ‘International Prospective Register of Systematic Reviews’ (PROSPERO; https://www.crd.york.ac.uk/PROSPERO, no. CRD42020166141).

Two authors (SK and KK) searched PubMed/MEDLINE, Embase, the Cochrane Central Register of Controlled Trials, and Web of Science databases. The search terms included variants of terms, such as ‘ibuprofen,’ ‘intravenous,’ ‘postoperative pain,’ ‘analgesia,’ ‘opioid,’ ‘fentanyl,’ ‘morphine,’ and ‘patient controlled analgesia,’ as well as Medical Subject Heading or Embase Subject Heading terms (Appendix 2).

There was no limitation on the year of publication, but we limited the search to randomized controlled trials conducted on humans. The language of the article was limited to English and Korean. The date of the last search was May 12, 2020.

After searching the articles from the databases listed above, two authors (SK and KK) selected the studies independently. Selection consisted of the following three steps: the two authors first selected the articles based on the title and then the abstract, and for the remaining articles, the two authors reviewed the full text of each article for the final selection. In case of disagreement, the two authors discussed the final selection of the articles until an agreement was reached.

The inclusion criteria were as follows: (1) patients under general anesthesia; (2) ibuprofen was administered intravenously; (3) ibuprofen was administered preoperatively, which is defined as before surgical incision; (4) control group using placebo was reported with results; and (5) primary outcomes of original articles were postoperative pain scores or opioid consumption.

Studies were excluded based on the following criteria: (1) articles not written in English or Korean; (2) patients not under general anesthesia; (3) ibuprofen was not administered intravenously; (4) ibuprofen was administered after skin incision or multiple times; (5) did not include appropriate postoperative outcomes; (6) non-randomized clinical trials; (7) non-human studies; and (8) did not compare with the appropriate control group. We also excluded articles that were not available in full text.

Based on the ‘risk of bias’ of the Review Manager software (RevMan, version 5.3; The Cochrane Collaboration, UK), two authors (SK and KK) independently evaluated the quality of articles. A third author (SL) was included to resolve disagreements when needed. Seven categories were included to assess the risk of bias: random sequence generation and allocation concealment for detecting selection bias, blinding of participants for performance bias, blinding of outcome assessor for detection bias, incomplete outcome data for attrition bias, selective reporting for reporting bias, and other biases that were not covered by the above categories. We specified the adequacy of the sample size calculation as ‘other bias,’ the seventh category. The risk of bias was rated as ‘high,’ ‘low,’ or ‘unclear’ in each original article. The agreement of two independent raters regarding the risk of bias for the seven categories was evaluated using Cohen’s kappa. The authors interpreted the Cohen’s kappa values based on Cohen’s suggestions as follows: (1) below 0.00, no agreement; (2) 0.00-0.20, slight agreement; (3) 0.21-0.40, fair agreement; (4) 0.41-0.60, moderate agreement; (5) 0.61-0.80, substantial agreement; and (6) 0.81-1.00, almost perfect agreement.

Two authors (SK and KK) extracted data from the articles and cross-checked the data to avoid missing any information or extracting incorrect information. The extracted information included patient age, study design, publication year, authors’ first name, type of surgery, timing and dosage of study drug, and measured outcomes. The measured outcomes were as follows: postoperative pain score, postoperative analgesic regimen, and analgesic consumption. Two authors (SK and KK) independently extracted the data from the text, tables, and graphs. Cohen’s kappa was used to assess the agreement between the two authors on the extracted data, and the values were interpreted in a manner similar to the risk of bias.

We extracted pain scores at postoperative 1, 4-6, and 24 h to reflect immediate, early, and late postoperative pain, respectively, for the analysis of pain scores. If the pain score was not measured at postoperative 6 h in the original articles, we used pain scores measured at postoperative 4 h as a replacement [13,15–17]. Cumulative opioid consumption up to postoperative 4-6 h and 24 h was included. We extracted data on cumulative opioid consumption at postoperative 6 h, and if data were not available at 6 h, we extracted data measured at postoperative 4 h as a replacement [16,17].

To analyze the intensity of postoperative pain, pain scores measured using the visual analog scale (VAS) or numerical rating scale (NRS) were extracted from each study. When the studies evaluated pain scores during movement and resting state simultaneously, we only used scores assessed in the resting state. If the studies used opioids other than fentanyl, we converted them into fentanyl equivalents [14,18].

The authors obtained 1,534 articles after an initial database search of PubMed (n = 154), Embase (n = 880), the Cochrane Library (n = 348), and Web of Science (n = 152). We excluded 350 duplicate articles. Two authors reviewed the articles independently, and subsequently excluded 109 reports based on the title and 1011 articles based on the abstract. Final full-text reviews were performed on the remaining 64 articles. Of the 64 articles, 58 were excluded based on exclusion criteria. Details of the reason for the exclusion are described in Fig. 1.

The characteristics of the six studies are presented in Table 1. All articles were randomized clinical trials. The types of surgeries were diverse, including laparoscopic cholecystectomy [16,17], septorhinoplasty [13,18], pancreaticoduodenectomy [14], and thyroidectomy [15]. Ibuprofen was administered intravenously in all the studies; however, the doses were different: one study administered 400 mg of ibuprofen [17], while other studies injected 800 mg of ibuprofen. Furthermore, opioids used in these studies were diverse and included fentanyl [13,15–17], morphine [14], and tramadol [18]. The timing of pain score measurement and opioid consumption varied among the studies.

The risk of bias assessment indicated that all included studies had low bias (Figs. 2 and 3). All trials were evaluated as having a low risk of random sequence generation. Of all included studies, 66.6% in allocation concealment, 16.6% in blinding of participants and personnel, 16.6% in blinding of outcome assessment, 66.6% in selective reporting, and 16.6% in other bias categories were assessed as ‘unclear’. Only two trials [14,18] were clear about allocation concealment, while other trials (66.6%) did not describe the allocation concealment method in detail. With respect to blinding of patients, one study [18] did not mention the term ‘double-blinded’ or describe its method of blinding. Furthermore, it did not specify blinding of the outcome assessor method, so we assessed it as ‘unclear’. Only two studies [14,16] provided information regarding protocols written in advance. All studies were rated as ‘low’ in other bias, except for two trials: one study [15] without calculation of required sample size and power analysis rated as ‘unclear,’ and the other [18] with inappropriate power analysis rated as ‘high.’

Agreement between the two raters for assessing the risk of bias was moderate (Cohen’s kappa = 0.52).

One study [14] reported the effects of intraoperative remifentanil infusion along with preoperative ibuprofen on postoperative pain and opioid consumption. Patients in this study received intraoperative remifentanil infusion targeting 4 ng/ml or 1 ng/ml as effect-site concentration, with or without preoperative ibuprofen administration. In this study, we extracted the data from subgroups with low remifentanil infusion only, because most of the other articles included in our meta-analysis did not use opioid infusion or used a low infusion rate of opioids during surgery.

Pain scores at postoperative 4 h were extracted from four reports [13,15–17], since data at 6 h were not available. Opioid consumption of one study [15] was not included in the analysis of opioid consumption owing to a lack of information, and in two studies [16,17], opioid consumption at postoperative 4 h was used because there were no data available at postoperative 6 h.

Agreement between the two raters for data extraction was substantial (Cohen’s kappa = 0.63).

A meta-analysis of six studies [13–18] (n = 366; 185 in the ibuprofen group and 181 in the control group) showed that pain scores measured at 1 h postoperatively were significantly reduced in the preoperative ibuprofen group (MD: -1.64, 95% CI [-2.56, -0.72], P < 0.001, I² = 95%) (Fig. 4A).

In the sensitivity test, the pain score measured at 1 h postoperatively was lower (MD: -1.94, 95% CI [-2.30, -1.57], P < 0.001, I² = 29%) than the estimated pooled effect with consistent direction and significance after excluding the outlier study [14]. It can be concluded that the results were robust. In addition, there was an effect of a reduction in heterogeneity to an acceptable level.

A meta-analysis of six studies [13–18] (n = 366; 185 in ibuprofen group and 181 in control group) showed that the preoperative ibuprofen group reported lower pain intensity with a statistical significance (MD: -1.17, 95% CI [-2.09, -0.26], P < 0.001, I² = 94%) (Fig. 4B).

In the sensitivity analysis, the effects of pain intensity at 4-6 h were significantly reduced (MD: -0.79, 95% CI [-1.29, -0.30], P = 0.0016, I² = 61%) compared to the pooled effect after excluding one trial [18], which was an outlier in this analysis. However, the results were reliable, because the direction and significance were maintained. Although the heterogeneity decreased, it was still of moderate intensity; hence, the results should be interpreted carefully.

A meta-analysis of data from six studies [13-18] (n = 366; 185 in the ibuprofen group and 181 in the control group) demonstrated that pain scores of the ibuprofen group were lower than those of the control group (MD: -0.58, 95% CI [-0.99, -0.18], P < 0.001, I² = 90%) (Fig. 4C).

In the sensitivity analysis, the pain score of the ibuprofen group measured at 24 h postoperatively was still significantly more effective (MD: -0.75, 95% CI [-1.17, -0.32], P < 0.001, I² = 88%) than the estimated effect after excluding one trial [14], an outlier. As the direction and significance of the results were maintained, the results were regarded as robust.

A total of four studies [14,16–18] (n = 275; 139 in the ibuprofen group and 136 in the control group) showed data on the accumulative opioid consumption at postoperative 4-6 h. Preoperative ibuprofen administration significantly reduced opioid consumption, which was presented as fentanyl equivalents (MD: -56.35 μg, 95% CI [-101.10, -11.60], P < 0.001, I² = 91%) (Fig. 5A). Sensitivity analysis did not show any outliers.

Meta-analysis of five studies [13,14,16–18] (n = 326; 165 in the ibuprofen group and 161 in the control group) showed that cumulative opioid consumption at 24 h postoperatively was lower in the ibuprofen group (MD: -131.39 μg, 95% CI [-224.56, -38.21], P < 0.001, I² = 95%) (Fig. 5B).

In the sensitivity analysis, the effect size of cumulative opioid consumption at 24 h increased (MD: -170.70 μg, 95% CI [-265.63, -75.76], P < 0.001, I² = 95%) after excluding one trial [14], which was indicated to be an outlier.