A systematic literature search was conducted in four databases: Pubmed, Web of Science, Embase, and The Cochrane Library. This study exclusively focused on original human research articles published in English between 2003 and 8 May 2023. Relevant literature pertaining to the bacterial microbiota of COPD was limited to three different cavities: oral cavity, airway, and intestine. The search strategy was formed by medical subject headings (MeSH) words and their corresponding free words to minimize the risk of missing relevant literature. The MeSH words used in this article are ‘Microbiota’ [Mesh] and ‘Pulmonary Disease, Chronic Obstructive’ [Mesh]. The search strings are provided in Supplementary Table 1 .

To ensure the comprehensiveness of the literature search, we conducted a secondary search using the “backward snowballing” method to identify relevant articles that were not captured through equation recognition from the reference list of included studies. A total of four additional articles were discovered during the secondary search.

The titles and abstracts were independently screened by two investigators. Any discrepancies between the two researchers were resolved through consensus with a third researcher. To ensure consistent screening criteria, all three investigators underwent standardized training prior to conducting the formal literature screening. The inclusion criteria for studies are as follows (1): the study population comprises individuals aged over 18 years with COPD and HC (2); the study design is an observational case-control or cross-sectional study published in English (3); the article describes microbial characteristics in either oral cavity, airway, or intestine sites of COPD (4); metagenomic sequencing or 16S rRNA sequencing analysis was performed (5); reports on microbial diversity indices are provided in either the article or its Supplementary Materials (including tables and figures), with specific numerical values extractable from relevant charts; and (6) there is at least one control group consisting of HC or stable disease patients, as well as at least one case group comprising stable disease patients or exacerbation patients.

Studies meeting any of the following criteria will be excluded (1): studies conducted on non-human populations or individuals under 18 years old (2); studies lacked a clear representation of microbial diversity index or specific data extraction from microbial diversity charts (3); studies without a control group for comparison; and (4) article types such as abstracts, case reports, expert opinions, reviews, letters, or editorials.

All studies were selected based on the aforementioned eligibility criteria, with a primary focus on assessing bacterial alpha-diversity in different cavities of both COPD and HC. This encompassed evaluating metrics such as the Chao1 index, Richness, Shannon index, and Simpson index. Initially, articles were screened based on their titles and abstracts, followed by a thorough examination of the full text to identify those that ultimately met the inclusion criteria. The final selection was reached through consensus among all authors.

The necessary data from the included studies were extracted by two researchers and recorded in a self-designed Excel spreadsheet. The extracted data encompassed the following variables (1): study information (title, first author’s name, publication year, journal of publication, study design type, sample type) (2); population characteristics (sample size, different disease states of COPD, age) (3); community-level measurements of microbial composition in different cavities of the host body (oral cavity, airway, and intestine), with a primary focus on α-diversity followed by β-diversity. Taxonomic discoveries at the phylum and genus levels were also included (bacterial relative abundance) (4); sequencing information (NGS sequencing method, amplification region of 16S rRNA). Quantitative parameters for Alpha-diversity were obtained using Get Data Graph Digitizer software when necessary to extract specific data from charts.

The included studies were independently assessed by the two investigators using the Newcastle-Ottawa Scale (NOS), which was a validated tool for evaluating bias risk in non-randomized studies. The NOS assigned scores based on three key aspects of the included studies: selection, comparability, and outcome. In this assessment process, the selection section had a maximum score of 4 points, comparability had 2 points, and outcome had 3 points, resulting in a total possible score of 9 points. Studies achieving scores between 7-9 were considered high-quality research, while those scoring between 4-6 were classified as moderate-quality research. Studies with scores below 4 were regarded as low-quality research. The specific scores of each article according to the NOS are shown in Figure 1 ; Supplementary Table 2 .

Standardized mean differences (SMDs) were employed to summarize and analyze the variations in microbial diversity between observation groups and control groups for different study subjects using Revman 5.4 software. The differences were estimated with a 95% confidence interval (CI). For quantitative data, means (M) and standard deviations (SD) were used for description, while studies reporting median and interquartile range provided estimates of M and SD through conversion formulas, where SD was calculated as the interquartile range divided by 1.35. Regarding qualitative data such as relative abundance of microbial communities, results from multiple studies were pooled together. Forest plots were generated using random effects models to visualize the disparities in microbial community structures among samples from different study subjects. Sensitivity analysis was performed using Begg’s test and Egger’s test in Stata 12.0 software.

We retrieved a total of 4,715 articles from four databases (Pubmed, n=1005; Web of Science, n=1589; Embase, n=1867; The Cochrane Library, n=254). Using EndNote software and manual screening to eliminate duplicate studies, we identified 1,983 unique articles. Subsequently, we screened the remaining 2,732 articles based on their titles and abstracts. During this stage, 2,677 articles were excluded primarily due to their lack of relevance to our study or its outcomes (n=2296) and inconsistency with the study type (n=354). After this round of screening, we conducted a full-text review of 55 studies and determined that 16 studies met our inclusion criteria for systematic review. Additionally, employing a “ backward snowballing “ approach, we included an additional four studies. Ultimately, a total of 20 studies were incorporated into our literature review. Figure 2 provides an overview of the entire process involved in study selection.

The characteristics of the studies included in the meta-analysis were presented in Table 1 . Among them, two studies focused on investigating the oral microbiota of COPD, utilizing oropharyngeal swabs and periodontal plaques as sample types. In terms of examining the respiratory microbiota of COPD, sixteen studies were identified, employing sputum (n=10), bronchoalveolar lavage fluid (BALF) (n=5), and lung tissue (n=1) as sample types. Regarding the investigation of intestinal microbiota in COPD, two studies utilized fecal samples. Study design-wise, two studies adopted a cross-sectional approach while the remaining seventeen studies were case-control studies. Additionally, there was heterogeneity observed in bacterial sequencing methods employed across these studies; specifically, eleven studies utilized Illumina^© systems (primarily MiSeq platform) for sequencing which had the highest representation. Furthermore, six studies targeted V3-V4 regions which was found to be most commonly used for amplification of 16S rRNA gene sequences.

When characterizing the microbial attributes of different cavities of the host body, including the oral cavity, airway, and intestine, researchers predominantly employed Alpha diversity indices. Analysis of aggregated data revealed significant variations in Alpha diversity indices across different study populations. However, notable heterogeneity existed within studies conducted on the same population with regarded to Alpha diversity indices, which could be attributed to factors such as sample size, types of samples collected, and sampling procedures. ( Supplementary Tables 3 – 5 ).

Quality was assessed using the NOS. After assessment, a total of two high-quality studies, sixteen medium-quality studies, and two low-quality studies were incorporated into this article. The specific scores obtained for each study based on the NOS are presented in Figure 1 .

A total of twenty studies, including 337 HC, 511 SCOPD, and 154 AECOPD, were included in this meta-analysis. In this study, observed species richness and Chao1 index were selected to evaluate microbial richness, while the Shannon index and Simpson index were used to assess microbial evenness. Figures 3 , 4 present the standardized mean differences and their confidence intervals between the control group and observation group. Among these studies, two focused on oral microbiota using the Chao1 index and Shannon index as indicators of sample alpha-diversity (8, 9). For respiratory microbiota analysis, out of sixteen studies included in this meta-analysis, five reported the Chao1 index, six reported the Richness, thirteen reported the Shannon index, and six reported the Simpson index. Only one out of two studies investigating intestinal microbiota provided data on HC and SCOPD for the Chao1 index, Richness, Shannon index, and Simpson index (26). Additionally, in our search results we found one study reporting both Chao1 index and Shannon index for different severities of SCOPD.

Based on the forest plot, it was evident that a majority of studies indicated lower Alpha-diversity indices in the observation group compared to the control group. Only a few studies reported higher Alpha-diversity indices in the intervention group (11, 18, 21). However, there were no statistically significant differences in Alpha-diversity among most comparison groups. Two studies provided Chao1 and Shannon indices for oral microbiota in 42 SCOPD and 37 HC. The pooled estimate of Chao1 index exhibited substantial heterogeneity (I ² = 95%, p<0.0001), with no significant difference observed between groups (SMD=-1.91, 95%CI -4.62 to 0.80) ( Figure 3A ). Similarly, the pooled estimate of Shannon index also demonstrated high heterogeneity (I ² = 95%, p<0.0001), with no significant difference found between groups (SMD=-1.92, 95%CI -4.42 to 0.57) ( Figure 3B ).

In terms of airway microbiota diversity, there were no significant differences observed in the comparisons of the Chao1 index and Shannon index among the groups ( Figures 4A, C ). However, regarding the Simpson index, data from two studies including 62 AECOPD patients and 46 HC individuals showed high heterogeneity (I² = 68%, p = 0.02), with significant differences between groups (SMD = -0.93, 95% CI -1.72 to -0.15). On the other hand, no significant differences were found in the comparisons of the Simpson index for other groups ( Figure 4D ). Three studies provided Richness data for a total of 77 AECOPD patients and 41 SCOPD patients without observing any significant heterogeneity in the pooled analysis (I² = 22%, p = 0.28), and there were significant differences in terms of Richness (SMD = -0.59, 95% CI -1.09 to -0.10). Whereas no significant difference was found in Richness when comparing SCOPD and HC individuals Figure 4B ).

Two studies were identified that characterized the diversity of intestinal microbiota between SCOPD and HC. Bowerman et al. observed disparities in intestine microbial composition between SCOPD and HC; however, specific data on diversity indices were not provided in the Supplementary Materials , impeding our ability to obtain precise diversity indices. Nevertheless, this study reported no significant differences in diversity levels between SCOPD and HC (pShannon = 0.329, pSimpsonInverse = 0.291) (27). Conversely, Wu et al. revealed lower alpha-diversity indices for both SCOPD and AECOPD when compared to HC; AECOPD exhibited even lower diversity than SCOPD (26). Furthermore, Chiu et al., through further investigation, compared Chao1 and Shannon indices of intestinal microbiota among mild, moderate, and severe COPD patients. They discovered a gradual decrease in the Chao1 index with increasing disease severity while observing that the Shannon index was lowest among mild COPD patients and highest among moderate COPD patients (28) ( Supplementary Table 5 ).

In all the included articles, 16 studies reported on β-diversity, with the most commonly measured being PCoA analysis based on weighted UniFrac distance and Bray-Curtis dissimilarity. Table 2 reveals that regarding oral microbiota, one study supported differences in microbial composition between SCOPD and HC (8), while another study found no significant differences (9). Concerning airway microbiota, four studies supported differences in microbial structure between SCOPD and HC (11–13, 15), while one study found no significant differences (16); three studies supported differences in microbial composition between AECOPD and HC (12, 16, 17); two studies supported differences in microbial composition between AECOPD and SCOPD (12, 16), while three studies found no significant differences (14, 18, 24). Regarding intestinal microbiota, two studies supported differences in microbial composition between COPD and HC (26, 27).

Most studies had presented graphical representations of the relative abundance of microbial phyla and genera in both the experimental and control groups. We observed significant alterations in microbial communities depicted in these images, and synthesized these research findings ( Figure 5 ). Compared to HC, Prevotella, Streptococcus, Actinomyces, and Veillonella of oral microbiota in SCOPD were reduced at the genus level. Most studies supported that Firmicutes, Bacteroidetes, Actinobacteria, and Fusobacteria of airway microbiota at phyla level in SCOPD were decreased. Haemophilus, Lactobacillus, and Pseudomonas were increased, but Veillonella, Prevotella, Actinomyces, Porphyromonas, and Atopobium were decreased at the genus level in SCOPD. However, the abundance of Bacteroidetes phylum decreased, while the Proteobacteria and Actinobacteria phyla exhibited an increase in the airway microbiota of AECOPD compared to HC. And the abundance of Haemophilus, Lactobacillus and Pseudomonas genera exhibited an increase, whereas Actinomyces and Porphyromonas showed a decrease. We also observed that Lachnospira of intestinal microbiota in SCOPD was reduced at the genus level.

The heterogeneity assessment results did not meet our expectations. The forest plot of this meta-analysis revealed high heterogeneity in all summary results, except for the comparison between AECOPD and SCOPD which showed low heterogeneity in Chao1 index, Simpson index, and Richness index. Among the 20 included studies on airway microbiota, Shannon index was the most frequently reported alpha-diversity index with 11 studies reporting differences between SCOPD and HC. Additionally, 5 studies reported differences in Richness index between SCOPD and HC. Sensitivity analysis was conducted to determine sources of heterogeneity for these 11 and 5 studies by sequentially removing each study from the analysis to observe result stability and accuracy. Results remained stable after removing each study one by one. ( Supplementary Figures 1A, B ).

To further validate the reliability of the Shannon index results for airway microbiota, we conducted Begg’s test and Egger’s test to evaluate the potential presence of publication bias. The outcomes revealed that all p-values associated with these indicators exceeded 0.05, indicating a lack of evidence supporting publication bias and providing additional support for the robustness of the conclusions derived from our meta-analysis. ( Supplementary Figures 1C, D )