The study protocol was registered at the International Prospective Register of Systematic Reviews (number CRD42022362280).

PubMed, Embase, Web of Science, and Cochrane Library were searched from their inception to 21 August 2022. We used the following search strategy to find all studies evaluating anti-T2s, including IL-5, IL-4, IL-9, IL-13, IL-25, IL-33, IgE, TSLP, and TLR9 for patients with eosinophil-associated COPD and asthma: (mepolizumab OR reslizumab OR benralizumab OR tralokinumab OR lebrikizumab OR dupilumab OR anti-interleukin OR MEDI-528 OR GSK679586 OR omalizumab OR tezepelumab OR AZD1419 OR CYT003 OR itepekimab OR XKH001) AND (asthma OR chronic obstructive diseases). The detailed search strategy is shown in Supplementary Table 1 . Languages had no restrictions. The pertinent review articles and their citations were also checked.

Endnote X9 software was adopted to manage the eligible studies during the literature screening and automatically remove duplicate documents. The following particular inclusion criteria were met (1): Participants: individuals (6 years of age or older) with asthma or COPD who met one or more criteria for eosinophilic inflammation at study enrolment or within the previous year. (2) Interventions: with anti-IL-5, anti-IL-4, anti-IL-13, anti-IL-9, anti-IL-25, anti-IL-33, anti-TSLP, anti-IgE or TLR9 agonist therapy at any dose. (3) Randomized placebo-controlled trials. (4) Reporting the following outcomes: exacerbation rate and change in pre-bronchodilator forced expiratory volume in 1 s (FEV1) from baseline.

Excluded criteria were as follows: (1) Studies did not involve eosinophilic endotype. (2) Interventions were not related to type 2 inflammation. (3) Studies did not assess the exacerbation rate or FEV1. (4) Not RCTs or literature types were reviews, letters, second analysis, or conferences.

The source data, together with the rate ratio (RR) or mean difference (MD), are given or can be computed from the data. All references were independently reviewed by two authors (YW and MH) following the selection criteria. Any disagreements were resolved through conversation or by a third author (JL).

The preferred reporting items for systematic reviews and meta-analyses (PRISMA) statement was followed (25) (see Supplementary Table 5 ). Two authors (YW and MH) independently retrieved data from eligible studies using Excel 2019 in a standardized data extraction form in a blinded manner based on the authors, publication year, research design, patient characteristics (age, gender, etc.), the type of anti-T2s used, the dosage, the length of the therapy, the definitions of the outcomes, the exacerbation rate, and the change in FEV1. A third author (JL) was consulted to settle disagreements. Furthermore, we evaluated the risk of bias using Cochrane Risk of Bias Assessment Tool 1.0, which included sufficient sequence generation, allocation concealment, blinding of participants and staff, inadequate outcome data, selective reporting, and additional bias (26). Two senior researchers (RY and YL) evaluated the reliability of the evidence using the GRADE-profiler software (V.3.6, The GRADE Working Group, 2010), items including the risk of bias, inconsistency, indirectness, imprecision, and publication bias, and the evidence was assessed as 4 levels: high quality, moderate quality, low quality, and very low quality.

We conducted a series of meta-analysis to compare the efficacy of anti-T2s with a placebo. For dichotomous data (exacerbation rate), intervention effects were reported using RR and 95% confidence intervals (CI), whereas standard mean differences (SMD) and 95%CI were used for continuous data (FEV1). Following the Cochrane Handbook, we aggregated two or three intervention groups into a single intervention group when research demonstrates more than two arms (26). The chi-squared test and the I² statistic were used to measure heterogeneity. Significant heterogeneity is indicated by an I² value of more than 50% (27). When there was significant heterogeneity, a random effect model was applied, and meta-regression analyses were performed to investigate the possible origins of heterogeneity. Planned considerations included ages, risk of bias, and demographics (exacerbation history, type 2 inflammatory criteria, and so forth). In comparisons involving at least 10 trials, publication bias was examined using a funnel plot and Eggers’ test (28, 29). The influence of publication bias was estimated using the trim-and-fill method (30). Sensitivity analyses were carried out to assess the robustness of the overall effect sizes by removing one study at a time. Review Manager (V.5.4.1, The Cochrane Collaboration, 2020) and Stata (V.15.1) were used for all statistical analyses. A two-sided P value of 0.05 was considered statistically significant.

A total of 7569 potentially pertinent articles were found. 3530 duplicate records from among all the potential studies were eliminated, leaving 4039 papers for screening. We found and obtained 131 papers in full text for review after examining the titles and abstracts. Ninety-three of these publications were excluded due to the following reasons: improper population (n = 39) (31–69), incorrect intervention (n = 4) (70–73), improper outcomes (n =29) (74–102), non-RCT (n = 13) (103–115), conferences (n = 3) (116–118), and second analysis (n = 5) (119–123). In the end, this meta-analysis included 38 articles with 41 studies ( Figure 1 ).

Thirty-eight articles, covering forty-one trials with 17,115 individuals, were included ( Supplementary Table 2 ). The number of subjects in the studies ranges from 61 to 1545. Eleven of these studies employed benralizumab (124–133), three reslizumab (23, 134, 135), two dupilumab (21, 136), one lebrikizumab (137), eight mepolizumab (22, 138–143), seven omalizumab (111, 144–149), two tezepelumab (150, 151), one astegolimab (152), one itepekimab (153), one AZD1419 (154), one quilizumab (155), one CYT003 (156) and two tralokinumab (157). The duration of the treatment ranges from 12 to 56 weeks, and the follow-up was 12 to 84 weeks. Six trials administered the monoclonal antibody by intravenous infusion (IV), thirty-three studies by subcutaneous (SC), one by inhalation, and one study comprised both IV and SC arms. Thirty-four studies included patients with asthma, whereas seven researches included those with COPD.

All included patients with COPD who had an exacerbation history. In thirty-four studies with asthma patients, seven studies included severe asthma, five studies included moderate to severe asthma, one study included mild to moderate asthma, twelve studies included refractory, uncontrolled, or persistent asthma, and the remaining studies did not specify asthma severity; patients in eight studies required medium to high dose ICS plus long-acting β2-agonists (LABA), two studies required at least medium ICS, one study required 6-month maintenance treatment with systemic glucocorticoids, and one study required not receiving ICS; seventeen studies required exacerbation history.

The definition of ‘type 2 inflammation’ varied among studies. Four studies were defined by FeNO levels, three studies were defined by a sputum eosinophil count, twenty-two studies were defined by baseline blood eosinophil counts, one study was defined by eosinophil counts in blood or sputum, one study was defined by baseline blood eosinophil counts or IgE levels, five by IgE levels, and one study defines by combinatorial biomarkers, including FeNO levels, eosinophil counts in blood or sputum, whereas four studies did not specify the criteria.

In chronic airway illnesses associated with eosinophils, we contrasted anti-T2s with a placebo. The primary outcome was the exacerbation rate. A COPD or asthma exacerbation was defined as a clinical worsening for at least three days, a temporary increase in the ICS background dosage, the need for systemic corticosteroid treatment, the consumption of antibiotics, hospitalization, or mortality resulting from an airway disease. The secondary endpoint was the change in FEV1 from baseline measured by spirometry. Since the number of studies on asthma is much higher than studies on COPD, to eliminate the influence, the population was divided into the asthma group and COPD group for meta-analysis, respectively.

A total of 25 researches (61.0%) adequately explained the randomization process. In 16 researches (39.0%), the random allocation was acknowledged, while in 25 studies (61.0%), allocation concealment was unclear. Research blinding was used in all of the investigations. 11 studies (26.8%) were at low risk for the outcome assessment’s blinding. There was a low risk to the integrity of the outcome data in 23 trials (56.1%). There was a low risk of selection bias for 37 (90.2%). The data from each study was insufficient to determine if the risk of other biases was low or high ( Figures 3A, B ). In 32 studies included to assess the exacerbation rate, 20 studies in total (62.5%) provided a comprehensive explanation of the randomization procedure. In 11 investigations (34.4%), the random allocation was confirmed, while allocation concealment was uncertain in 21 studies (65.6%). In each study, research blinding was performed. 7 studies (21.9%) had a low risk of blinding in outcome assessment. In 19 trials (59.4%), there was a low risk in the integrity of the outcome data. For 28 studies (87.5%), the risk of selection bias was low. In 26 studies included to assess FEV1, 18 studies (69.2%) in total provided a thorough explanation of the randomization procedure. In 12 investigations (46.2%), the random allocation was acknowledged, and allocation concealment was ambiguous in 14 studies (53.8%). In each study, research blinding was applied. 9 studies (34.6%) had a low risk of blinding outcome assessment. In 11 trials (42.3%), there was a low risk of the integrity of the outcome data. 25 studies (96.2%) had a low risk of selective bias.

By removing one study at a time, sensitivity analyses were utilized to examine the impact of each study on the combined results. The outcome demonstrated that there had been no appreciable changes to the results’ stability ( Figures 3C–F ), which supported the validity and dependability of our analysis.

Because of the considerable heterogeneity and inconsistent findings across trials, the evidence received a low-quality level in exacerbation rate in asthma and change in FEV1 from baseline in COPD, and a moderate-quality level in exacerbation rate in COPD and change in FEV1 from baseline in asthma ( Supplementary table 4 ).