We conducted both traditional pairwise meta-analysis and network meta-analysis by systematically searching PubMed, Embase, Web of Science, Cochrane review, and Cochrane Trial from their inception to January 26, 2023, with language restrictions set to English. The search targeted previously published articles incorporating the terms “depression disorder” and the specified anti-inflammatory agents mentioned above ( Supplementary Table S1 ). We specifically sought double-blind, randomized controlled trials (RCTs) comparing anti-inflammatory agents with a placebo for the treatment of acute depression in adults of both sexes aged 18 years and older. These trials could involve either anti-inflammatory drugs alone (anti-inflammatory agent vs. placebo) or in combination with antidepressant drugs (anti-inflammatory agent + antidepressant drug vs. placebo + antidepressant drug). Inclusion criteria required studies to use established diagnostic criteria for identifying patients with major depressive disorder, such as Feighner criteria, any version of DSM, and ICD-10. We excluded incomplete trials, those involving participants with psychotic depression or seasonal depression, as well as trials with participants having severe endocrine, metabolic, or other diseases. Additionally, trials were excluded if 20% or more of participants had bipolar disorder. For a comprehensive search, we utilized ClinicalTrials.gov, using the following strategies: “studies with results”, “interventional studies”, “depression disorder”, and “anti-inflammatory agents”.

Our primary outcomes comprise efficacy and acceptability. Efficacy is measured by the response rate, indicating patients with a ≥50% reduction in the total score of standardized depression assessment scales. Acceptability is measured by the rate of patient dropouts due to all-causes. Secondary outcomes include remission rate, defined as MADRS ≤ 7, HAMD ≤ 7, GDS ≤ 11, or BDI-II ≤ 8 at the end of the trial, and the proportion of patients who dropped out due to adverse events (AE) ( Supplementary Table S2 ).

To retrieve relevant studies, Min Wang and Yushun Yan imported all retrieved studies into Endnote and removed duplicate studies. Then, Yue Du and Yikai Dou independently screened the titles and abstracts of each article and reviewed the full text based on our inclusion and exclusion criteria. In case of disagreement, Xiaohong Ma and Xiao Yang jointly reviewed and made the final decision. Huanhuan Fan and Ningdan Fan recorded study information, including author name, publication date, sample sizes, and patient characteristics such as age and gender. In addition, Xiao Yang recorded intervention details such as intervention classification, treatment duration, treatment efficacy, and other clinical outcomes.

We used Review Manager 5.3 for traditional pairwise meta-analysis to assess the efficacy and acceptability of anti-inflammatory agents. For network meta-analysis, we used a frequentist framework model in Stata (version 17) software. As all our results are binary variables, we calculated odds ratios (ORs) and 95% confidence intervals (CIs) to present the findings. Random-effects models were used for both traditional pairwise meta-analysis and network meta-analysis. Additionally, we conducted subgroup meta-analysis based on different inflammatory drugs and other feathers. To rank the efficacy and acceptability of different anti-inflammatory drugs, we used the surface under the cumulative ranking curve (SUCRA) values. Heterogeneity for each anti-inflammatory drug was quantified using the I² statistic and p-value ( 22). According to the Cochrane Handbook, heterogeneity values of 0–40% were deemed insignificant, 30–60% indicated moderate heterogeneity, 50–90% suggested essential heterogeneity, and 75–100% represented appreciable heterogeneity (23). The included RCTs were evaluated by using the Cochrane Risk of Bias (RoB 2) tool, version 2 (24), which includes five domains for assessing bias such as randomization process, deviations from intended interventions, missing outcome data, measurement of outcomes, and selection of the reported results. When assessing whether outcome data for all participants were completed, we set the cut-off value at 80% (25). Leave-one-out sensitivity analyses were performed, and meta-regression was conducted to adjust for the effect of publication year and treatment duration. To investigate published bias, we used comparison-adjusted funnel plots. Egger’s test was also conducted to test the published bias. We performed the trim and fill procedure to further assess the possible effect of publication bias in pairwise meta-analysis.

This study is registered with PROSPERO, number CRD42023422004, and was conducted in compliance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) extension statement ( Supplementary Table S3 ).

We initially retrieved a total of 14,254 relevant studies. After removing duplicate studies and screening titles/abstracts, we assessed the full text of 258 studies that potentially met the criteria. Finally, 48 randomized clinical trials were included for subsequent data analysis (26–73). The entire process of literature search and trial selection is shown in Figure 1 . The characteristics of the included trials are summarized in Supplementary Table S4 . The included studies were published between 1995 and 2022, and involved 3,394 participants, with 53.1% of participants being female. All these trails were placebo-controlled, and the 48 randomized clinical trials employed different interventions, including omega-3 fatty acids (n=19), nonsteroidal anti-inflammatory drugs (n=9), pioglitazone (n=3), minocycline (n=5), NACs (n=2), corticosteroids (n=4), statins (n=4), and monoclonal antibodies (n=2). Sample sizes ranged from 20 to 432, and study durations ranged from 2 days to 16 weeks. In all studies, twenty-eight (56.3%) trials reported efficacy (response rate), including omega-3 fatty acids (n=11), NSAIDs (n=5), pioglitazone (n=1), minocycline (n=4), NACs (n=1), corticosteroids (n=1), statins (n=3), and monoclonal antibodies (n=2), and 45 (93.8%) trials included acceptability of anti-inflammatory agents, including omega-3 fatty acids (n=18), NSAIDs (n=8), pioglitazone (n=3), minocycline (n=4), NACs (n=2), corticosteroids (n=3), statins (n=3), and monoclonal antibodies (n=2). Among all involved RCTs, ten trails were monotherapy (40, 41, 44, 47, 51, 53, 60, 68, 71, 72) and 38 RCTs were adjunctive therapy. Five RCTs focused on treatment-resistant depression (TRD) patient (28, 30, 34, 39, 53). Four RCTs only included female participants (35, 40, 47, 72). Two RCTs included overweight participants (Body Mass Index: BMI ≥ 25 kg/m²) (47, 68) and two studies included patients with low-grade inflammation (30, 68). One RCT included participants with preclinically metabolic dysfunction (56) and another included 9 (15%) bipolar disorder patients for exploratory purpose (53).

Among all studies, two trials (4.2%) showed a high risk of bias due to inadequate information regarding outcome measurement, and one (2.1%) trial might have a high risk of bias due to missing outcome data. Moreover, twenty trials (41.7%) raised some concerns, while 26 (54.2%) had a low risk. A summary assessment of the risk of bias is illustrated in Supplementary Figure S1 .

Among the included trials, 28 reported efficacy, reflecting the response rate of participants and the efficacy of anti-inflammatory drugs for MDD patients. A random-effects model was used, as shown in Figure 2 , showing a significant antidepressant effect of anti-inflammatory drugs when compared to placebo (OR=2.04, 95% CI: 1.41–2.97, p=0.0002). However, there was moderate heterogeneity among the trials (Chi² = 49.38, df=27, p=0.005, I² = 45%).

For acceptability, forty-three trials assessed the acceptability of anti-inflammatory agents in MDD patients. As shown in Figure 3 , there was no statistically significant difference in acceptability between anti-inflammatory drugs and placebo (OR=0.91, 95% CI: 0.74–1.11, p=0.35). There was no heterogeneity among the trials (Chi² = 37.07, df=37, p=0.4, I² = 0%).

Twenty-one studies reported remission rates, showing the remission rates of 7 anti-inflammatory drugs in participants. As shown in Supplementary Figure S2A , compared with placebo, the remission rate of MDD patients was higher after treatment with anti-inflammatory drugs (OR=1.89, 95% CI: 1.20–2.98, p=0.006). There was relatively low heterogeneity among the studies (Chi² = 30.82, df=19, p=0.04, I² = 38%).

Sixty studies reported dropout rates due to adverse events (AE). As shown in Supplementary Figure S2B , there was no significant difference in dropout rates between anti-inflammatory agents and placebo (OR=1.11, 95% CI: 0.66–1.87, p=0.68). There was no heterogeneity among the studies (Chi² = 12.74, df=15, p=0.62, I² = 0%).

In the subgroup analysis of different anti-inflammatory agents NACs (OR=1.97, 95% CI: 1.06–3.65, p=0.03) and statins (OR=4.41, 95% CI: 1.33–14.60, p=0.02) showed significantly antidepressant effects. Trials of NSAIDs (Chi² = 13.39, df=4, p=0.01, I² = 70%), minocycline (Chi² = 7.65, df=3, p=0.05, I² = 61%) and monoclonal antibodies (Chi² = 2.54, df=1, p=0.11, I² = 61%) showed essential heterogeneity ( Figure 2 ). All anti-inflammatory agents demonstrated the same acceptability as placebo ( Figure 3 ).

In the subgroup analysis of studies with different therapies (monotherapy or adjunctive therapy), different results were observed. In the adjunctive therapy group, anti-inflammatory agents showed a significantly antidepressant effect compared to placebo (OR=2.17, 95% CI: 1.39–3.37, p=0.0006). In the monotherapy group, anti-inflammatory agents didn’t demonstrate a better antidepressant effect compared to placebo (OR=1.71, 95% CI: 0.88–3.31, p=0.11). Trials of adjunctive therapy (Chi² = 46.04, df=22, p=0.002, I² = 52%) showed moderate heterogeneity, but trials of monotherapy (Chi² = 3.32, df=4, p=0.51, I² = 0%) showed insignificant heterogeneity ( Supplementary Figure S3A ). Both therapies showed the same acceptability as placebo ( Supplementary Figure S3B ).

In the subgroup analysis of studies included treatment-resistant depression patients, anti-inflammatory agents in TRD group showed the same antidepression effect as placebo (OR=0.54, 95% CI: 1.25–2.89, p=0.60). However, in studies only including MDD patients, anti-inflammatory agents demonstrated a better antidepression effect compared to placebo (OR=2.33, 95% CI: 1.53–3.54, p<0.0001). Trials including TRD patients (Chi² = 7.79, df=4, p=0.10, I² = 49%) showed moderate heterogeneity, but trials including MDD group (Chi² = 38.93, df=22, p=0.01, I² = 43%) also showed moderate heterogeneity ( Supplementary Figure S4A ). Both kinds of studies showed the same acceptability as placebo ( Supplementary Figure S4B ).

In the subgroup analysis for mixed sexes, anti-inflammatory agents in mixed sexes group showed a better antidepression effect compared to placebo (OR=2.00, 95% CI: 1.37–2.92, p=0.0003). Trials including mixed sexes (Chi² = 48.11, df=26, p=0.005, I² = 46%) showed moderate heterogeneity ( Supplementary Figure S5A ). The subgroup analysis for both sex types demonstrated similar acceptability to placebo ( Supplementary Figure S5B ). Only one RCTs, which included only female patients, reported the efficacy.

In the subgroup analysis of studies without limitation of BMI, anti-inflammatory agents in showed a better antidepression effect compared to placebo (OR=2.11, 95% CI: 1.43–3.11, p=0.0002). Trials without BMI limitation (Chi² = 49.27, df=26, p=0.004, I² = 47%) showed moderate heterogeneity ( Supplementary Figure S6A ). Studies without limitation of BMI showed the same acceptability as placebo ( Supplementary Figure S6B ). Only one RCTs, which included overweight patients, reported the efficacy.

In the subgroup analysis for studies without inflammatory patients, anti-inflammatory agents showed a better antidepression effect compared to placebo (OR=2.13, 95% CI: 1.43–3.18, p=0.0002) and exhibited moderate heterogeneity (Chi² = 49.26, df=25, p=0.003, I² = 49%) ( Supplementary Figure S7A ). The trials that excluded patients with low-grade inflammation demonstrated comparable acceptability to placebo ( Supplementary Figure S7B ).

In the subgroup analysis for studies without preclinically metabolic dysfunction patients, anti-inflammatory agents showed a better antidepression effect compared to placebo (OR=2.13, 95% CI: 1.43–3.18, p=0.0002) and exhibited moderate heterogeneity (Chi² = 49.38, df=27, p=0.005, I² = 45%) ( Supplementary Figure S8A ). The trials that excluded patients with preclinically metabolic dysfunction demonstrated comparable acceptability to placebo ( Supplementary Figure S8B ).

In the subgroup analysis for studies without bipolar disorder participants, anti-inflammatory agents showed a better antidepression effect compared to placebo (OR=2.12, 95% CI: 1.43–3.15, p=0.0002) and exhibited moderate heterogeneity (Chi² = 49.24, df=26, p=0.004, I² = 47%) ( Supplementary Figure S9A ). The trials that excluded bipolar disorder patients demonstrated comparable acceptability to placebo ( Supplementary Figure S9B ).

Excluding high-risk studies, anti-inflammatory agents still showed a better antidepression effect compared to placebo (OR=2.04, 95% CI: 1.41–2.97, p=0.0002) and exhibited moderate heterogeneity (Chi² = 49.38, df=27, p=0.005, I² = 45%) ( Supplementary Figure S10A ). The trials excluding high risk studies still demonstrated comparable acceptability to placebo ( Supplementary Figure S10B ).

Supplementary Figure S11 show the comparison network diagrams between different interventions. However, we found that in the included studies, no trial directly compared two anti-inflammatory drugs, and all comparisons were between anti-inflammatory agents and placebo. Therefore, network meta-analysis was used to conduct direct and indirect comparisons of anti-inflammatory drugs. Each network plot did not form a closed loop, so we did not test for inconsistency in the NMA and only selected the consistency model.

Supplementary Table S5 shows the results of efficacy and acceptability in the NMA. For efficacy (response rate), statins (OR=4.27, 95% CI: 1.04–17.56) were identified as more efficacious than placebo. Regarding acceptability (all-cause dropout), all anti-inflammatory drugs were found to be as acceptable as the placebo. Notably, more patients dropped out for all causes in minocycline trials compared to NSAIDs (OR=0.41, 95% CI: 0.17–0.98) and NACs (OR=0.45, 95% CI: 0.20–0.99). SUCRAs and cumulative probability plots are presented in Supplementary Table S6 .

The sensitivity analysis results also confirmed the stability of the analytical outcome, indicating that the results remained consistent when employing the leave-one-out method ( Supplementary Figure S12 ). Additionally, meta-regression analysis showed no significant correlations between publication year (p=0.08) and duration of treatment (p=0.43).

The funnel plot displayed an asymmetrical funnel shape ( Supplementary Figure S13 ). The Egger test (p<0.001) and Begg test (p=0.001) for studies reported efficacy also raised concerns about potential publication bias ( Figure 4A ). To explore this further, a trim and fill method was employed. The analysis resulted in a changed outcome, which indicated the instability of the outcome with publication bias (OR=1.39, 95% CI: 0.93–2.09) ( Figure 4B ). It’s noteworthy that studies reporting acceptability didn’t exhibit publication bias (Egger test: p=0.48; Begg test: p=0.63).

In subgroup analysis for adjunctive therapy, both the Egger test (p=0.001) and Begg test (p=0.008) for studies reporting efficacy suggested a high possibility of publication bias, and the outcome of the meta-analysis was changed in trim and fill analysis (OR=1.39, 95% CI: 0.87–2.21). However, monotherapy group didn’t show publication bias (Egger test: p=0.07; Begg test: p=0.03).

Similarly, in the subgroup analysis for studies including MDD patients, both the Egger test (p<0.001) and Begg test (p=0.001) for studies reporting efficacy suggested a high possibility of publication bias, and the outcome of the meta-analysis was changed in trim and fill analysis (OR=1.44, 95% CI: 0.91–2.28). However, studies included TRD patients didn’t show publication bias (Egger test: p=0.49; Begg test: p=0.81).