No difference in microbial diversity between bronchoalveolar lavage and tracheal sampling: a systematic review and meta-analysis

Abstract

Introduction

The respiratory microbiome has a vital role in maintaining respiratory health and preventing pathogen colonisation, but traditional diagnostic methods fail to capture a complete picture of it. Metagenomic sequencing has improved our understanding of microbial ecosystems in both acute and chronic pathologies. However, its results have not been systematically compared between different respiratory sampling techniques, as has been done with traditional methods. Our study aims to compare the microbial diversity in bronchoalveolar lavage (BAL) and tracheal samples using microbiome sequencing.

Methods

A systematic search was conducted in Medline, Embase and CENTRAL databases to identify studies where lower respiratory tract microbiome specimens were collected simultaneously using BAL and tracheal sampling and diversity was analysed postsequencing. Risk of bias was assessed with our specifically tailored tool. A random-effects model was used for data synthesis, analysing pooled Shannon, Chao1 and Simpson indices.

Results

We screened 1050 potentially relevant publications, 10 of which were included. No significant difference was found in microbial alpha diversity between BAL and tracheal samples. The subgroup analysis of tracheal sample types, including sputum and endotracheal aspirate, revealed no significant differences compared with BAL.

Conclusions

Tracheal sampling methods offer a viable and less invasive alternative to BAL for characterising microbiome alpha diversity in clinical or research settings where segmental sampling is not required. However, further high-quality comparative studies are needed to confirm these findings.

PROSPERO registration number

CRD42023436934.

WHAT IS ALREADY KNOWN ON THIS TOPIC.

WHAT THIS STUDY ADDS

• Analysing the scarce evidence we have, we found that endotracheal aspirates from intubated patients and carefully collected induced sputum samples reflect lower respiratory microbiome diversity similar to bronchoalveolar lavage samples.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• This implies that in routine cases without isolated pulmonary pathologies, the more invasive and resource-intensive bronchoalveolar lavage may be avoided when analysing alpha diversity.

Introduction

The respiratory microbiome has a pivotal role in safeguarding respiratory system functionality and avoiding colonisation of external pathogens.1,3 Perturbations in the lower respiratory tract (LRT) microbiome are often seen both in chronic pulmonary pathologies⁴ and acute diseases.⁵ LRT infections that lead to serious deterioration of both the healthy and the chronically ill are a major global health concern.⁶ Traditional diagnostic methods, primarily culture-based and specific micro-organism PCRs, provide a framework for pathogen identification and clinical management. These techniques prioritise bacterial species that are virulent and grow under standard laboratory conditions, omitting species that are hard to culture or inactivated easily.7,9 In a range of respiratory pathologies, such as cystic fibrosis, chronic obstructive pulmonary disease^{4 10} or pneumonia,^{5 11} traditional approaches fail to accurately depict the pathological changes and interactions.^{12 13}

Advances in culture-independent diagnostic methods, particularly those using nucleic acid sequence signatures through 16S rRNA gene sequencing, have expanded our understanding of bacterial ecosystems^{2 14 15} and have the potential to change the way we diagnose infectious diseases.^{9 16} Mapping the whole bacteriome in airways can feature the diversity of all resident bacteria: virulent and non-virulent, pathogenic and non-pathogenic, culturable and non-culturable, living and inactivated.^{2 11}

This diversity of a microbial ecosystem is often described by its species richness and the evenness of how the many organisms distribute. This is called alpha diversity, a critical measure of community complexity and health. Alpha diversity indices such as Shannon, Chao1 and Simpson are the most commonly reported parameters in microbiome research. The Shannon index evaluates community richness and evenness, the Chao1 index serves as a richness estimator and Simpson index highlights species dominance, contrasting the richness-centric measures.¹⁷

For respiratory sample collection, invasive methods are available, such as bronchoalveolar lavage (BAL) and less invasive strategies, such as endotracheal aspiration or sputum induction and collection. Several previous studies compared these methods based on the culturing results. Tracheal microbial sampling methods are safer and easier to perform with significantly higher patient tolerance. BAL is more invasive and resource-intensive, but on the other hand, it appears to offer advantages in detecting viable pathogens and improving targeted antibiotic therapy.18,20

For 16S rRNA gene sequencing and shotgun sequencing, however, viability is not a concern. Despite its growing potential, sequenced microbiome profile differences have not been systematically analysed between BAL and tracheal sampling.

Methods

Our study aimed to detect any microbial diversity or composition pattern differences between BAL and tracheal samples.

We report our systematic review and meta-analysis based on the recommendations of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines and PRISMA checklist²¹ (online supplemental material 1), while we followed the Cochrane Handbook.²² This study was conducted in the framework of Academia Europaea’s position on the cycle model of translational medicine for community healthcare benefit.^{23 24} The study protocol, registered on PROSPERO (ID: CRD42023436934), was closely adhered to, except for employing beta-diversity as an outcome measure, due to data unavailability.

Eligibility criteria

We followed the PIRD structure as a framework to identify eligible papers.²⁵ Human studies reporting on patients whose LRT microbiome was sampled simultaneously using BAL and tracheal sampling, and the diversity of the samples was analysed after sequencing were included. Papers assessing neonatal microbiome and conference abstracts were excluded.

Search strategy

An initial systematic search was performed on 15 June 2023, using the MEDLINE, Embase and Cochrane CENTRAL databases. To ensure the comprehensiveness and relevance of the evidence, the search was updated on 26 March 2025. The search key was constructed to identify all articles assessing BAL and tracheal sampling and diversity (online supplemental material 2).

Selection and data collection

The selection was performed by two independent review authors (DMK and MP) using Rayyan, a reference management software designed for systematic reviews.²⁶ After automatic and manual duplicate removal, reviewers screened titles, abstracts and full texts against predefined eligibility criteria (online supplemental material 3). Data were collected independently by two authors (DMK and MP) on a predeveloped data extraction spreadsheet (online supplemental material 4). Individual diversity results were precisely²⁷ collected from graphs with WebPlotDigitizer (V.4.6).²⁸

Risk of bias assessment and certainty of evidence

A novel Risk of bias (RoB) tool was developed based on the Cochrane guidelines²² to address the primary sources of bias when comparing microbiome sampling techniques. A detailed explanation of the four domains, development, item checklist, grading and interpretation is provided separately (online supplemental material 5 and online supplemental material 6). In the first superdomain, consistency and objectivity of the applied sampling techniques were assessed. Bias in these domains directly affects the magnitude of differences in diversity. In the second superdomain, patient selection, population differences, variations in sample processing, sequencing data handling and amplicon analysis methods were analysed and compared. Bias in these domains does not affect the magnitude of difference between samples taken from the same patient via different methods, but could potentially skew the microbial composition and affect the accuracy of diversity index scales. To address this bias and normalise variations in index scales, standardised mean differences (SMDs) were calculated and compared as part of the data synthesis process. Two authors (DMK, MP) performed the RoB assessment independently. Disagreements were discussed and resolved with the help of a third author (CT). We used GRADEPro to assess the quality of evidence.²⁹

Synthesis methods

We collected all reported diversity indices, ensuring to include Chao1, Shannon and Simpson’s if multiple were reported. In cases where an inverse Simpson index was published, we recalculated it to obtain the Simpson’s diversity index for consistency. If the mean and SD were not reported, we either transformed the available medians and IQRs using the R package ‘metamedian’ or recalculated these values directly from original datapoints. To address the minor scale discrepancies in these indices due to variations in sequencing methods and bioinformatics pipelines used, we calculated SMDs with 95% CIs to measure the effect size. We used random-effects models to pool effect sizes using the inverse variance weighting method. For subgroup analysis, we used a fixed-effects ‘plural’ model (aka. mixed-effects model). We assumed that subgroups had different τ² values as we anticipated differences in between-study heterogeneity in the subgroups, although for practical reasons (subgroup size was five or less), a common τ² assumption was used following the recommendations of Harrer et al³⁰ to estimate τ², and we used the Paule-Mandel method and the Q profile method to calculate the CI of τ². Heterogeneity was assessed using Higgins and Thompson I² statistics. Outlier and influence analyses were carried out following the recommendations of Harrer et al and Viechtbauer and Cheung.^{30 31} Forest plots were used to graphically summarise results. We also reported the prediction intervals (ie, the expected range of effects of future studies) of results following the recommendations of IntHout et al.³² In two studies,^{13 33} mean and SD values were estimated from median and quartiles.

All statistical analyses were performed with (R Core Team 2023, V.4.2.3), using the meta (Schwarzer 2023, V.6.2.1) package for basic meta-analysis calculations and plots, and dmetar (Cuijpers, Furukawa and Ebert 2023, V.0.0.9000) package for additional influence analysis calculations and plots.

Patient and public involvement

No patients or members of the public were involved in the design, conduct or reporting of this research. All included studies were previously published and had received appropriate ethical approval.

Results

Search and selection

Our updated systematic search identified 1050 studies. After the selection process, 10 articles1333,41 were included in the meta-analysis. The PRISMA 2020 flow diagram shows the process of study selection (online supplemental file 7). Basic characteristics of included studies are collected in online supplemental material 8.

Microbial diversity in different sample types

We collected and pooled the three main indices that are used to describe microbial diversity.

There was no statistically significant difference between BAL and tracheal sample diversity depicted by the Shannon diversity index (SMD−0.01, 95% CI −0.62 to 0.59, n=191). Detailed results are shown in the first forest plot (figure 1).

Figure 1. Forest plot of Shannon diversity index results in tracheal and BAL samples. Standardised mean differences (SMD), their 95% CI and weights for individual trials are shown. Subgrouping is based on the type of tracheal sampling (endotracheal aspiration, endotracheal brush or sputum collection). BAL, bronchoalveolar lavage.

The Q test revealed no significant differences between the tracheal subgroups (χ²=0.81, p=0.668). Neither the endotracheal aspirates subgroup (SMD=−0.3, 95% CI −0.73 to 0.13, n=111) nor the sputum subgroup (SMD=0.1, 95% CI −1.21 to 1.41, n=72) showed diversity difference from BAL. The only study comparing tracheal brush and BAL samples found no difference in Shannon diversity (SMD=−0.05, 95% CI −1.03 to 0.93, n=8).

Similarly, there was no difference between the two methods in diversity represented by the Simpson index (SMD=−0.22, 95% CI −1.9 to 1.46, n=118; figure 2), and no difference in sample richness measured by the Chao1 index (SMD=−0.8, 95% CI −4 to 2.4, n=130; figure 3).

Figure 2. Forest plot of Simpson diversity index results in tracheal and BAL samples. Standardised mean differences (SMD), their 95% CI and weights for individual trials are shown. For Feng et al,³⁵ and Rogers et al¹³, the extracted (inverse) results were converted into a common Simson index format. BAL, bronchoalveolar lavage.

Figure 3. Forest plot of Chao1 diversity index results in tracheal and BAL samples. Standardised mean differences (SMD), their 95% CI and weights for individual trials determined from three articles. BAL, bronchoalveolar lavage.

RoB and certainty of evidence assessment

The results of the RoB assessment are summarised in the RoB results table (online supplemental material 9). Within the first superdomain, which examined the overall RoB in the sampling procedure, no significant outliers were identified, except for the study by Cheng et al,³³ which showed a high risk due to inadequate reporting. Higher RoB was generally associated with tracheal sampling compared with BAL, primarily due to inconsistencies in sampling techniques. In contrast, BAL procedures were typically well-documented and standardised. In 5 of the 10 studies, a potential bias was identified in the second superdomain in relation to differences in biomass and diversity measurements, affecting both sample types. Each decision is carefully documented and reported (online supplemental material 10). Funnel plots were visually inspected for small study bias, but due to the low number of included articles, it was not feasible to determine publication bias; Egger’s test (t=–0.13, df=8, p=0.9001) did not indicate asymmetry, though this result should be interpreted with caution (online supplemental material 11). Certainty of evidence²⁹ was graded generally low for all outcomes and sampling subgroups (online supplemental material 12).

Discussion

In the current meta-analysis, we systematically collected all publicly available data on alpha diversity measurements from parallel BAL and tracheal sampling. This comprehensive approach allowed us to construct the first preliminary evidence supporting the possible interchangeability of tracheal sampling and BAL for assessing LRT alpha diversity. None of the three diversity indices showed significant differences between the sampling techniques. Due to the limited number of included articles, subgroup analysis of tracheal sampling procedures (sputum or endotracheal suctioning) was only feasible with the Shannon index. The absence of subgroup differences there also suggests generally comparable diversity. Omitting each study individually did not result in a significant change in the overall effect, further indicating that the pooled estimate is robust and not disproportionately influenced by any single study (online supplemental material 13). The influence analysis did not show patterns associated with RoB. Most studies had low influence on the meta-analysis results and no single study disproportionately altered the overall pooled estimate (online supplemental material 14).

As present data showed no difference between BAL and tracheal sampling, it is reasonable to assume that there is no clinically meaningful difference. This is in line with the conclusions of other authors.1333,38 Although Rogers et al¹³ found differences in sensitivity for pathogens, the microbial diversity in their samples did not significantly differ. One of their supplementary figures (original Suppl. Fig. S2) illustrates the similarity of diversity results. Samples obtained by sputum and BAL from the same patients are plotted against each other. In all indices, the orientation and the extent of differences in measured diversity were completely scattered, with neither BAL nor sputum showing systematically different microbial patterns.¹³ Despite the limited evidence, mixed use of both tracheal and BAL techniques in LRT microbiome studies is already widespread^{42 43}; our findings support this approach.

Alpha diversity does not capture differences in taxonomy. Ideally, beta diversity would be the method of choice for comparing microbial communities between samples, but the studies included in our review did not provide sufficient information to allow mathematical reanalysis. In the work of Weiser et al, the Bray-Curtis dissimilarity measure, commonly used for beta diversity analysis, showed differences in some regards, but a general overlap between BAL and sputum samples.³⁸ Miao et al also showed high beta similarity when analysing BAL and endotracheal samples of COVID-19 patients.³⁶ Besides, Cheng et al found that microbiota profiles in tracheal and BAL samples showed similar diversity, composition and community correlations in their mechanically ventilated ICU (intensive care unit) patient cohort. The antimicrobial therapy employed was also rarely altered based on the additional BAL conducted alongside routine tracheal screening.³³

Differentiating between anatomical localisation of pathogens can be important in selected cases. Differences in spatial distribution of lung microbiome and isolated location of infections cannot be detected by tracheal sampling.⁴⁴ However, in healthy individuals and in most cases of respiratory infection, isolated differences and marked inhomogeneity of the microbiome are not pronounced.^{2 38} In the context of identifying LRT pathogens or capturing the near alveolar core microbiome without oral or gastric flora contamination, BAL has been known to have a slight advantage in terms of sensitivity.^{18 20 45} However, our analysis indicates that the impact of this cross-contamination seems to be overrated when assessing diversity with any alpha diversity metrics.

BAL, however, is associated with a number of additional safety issues, such as transient fever, mild airway bleeding, (transient) hypoxaemia, risk of introducing an infection or exacerbating an existing pulmonary infection, arrhythmias, bronchospasm and pneumothorax.^{46 47} Hence, it requires sedation or deepening of sedation, and patients with unstable cardiovascular status are also at higher risk of BAL-related complications.

Limitations

The absence of disease-specific subgrouping is a limitation, as heterogeneous study populations and lack of individual participant data made such analyses unfeasible. Illness severity correlated with sampling route (ETA from ICU patients, sputum from ambulatory patients), but no subgroup differences were seen when grouping by sample type.

Due to the uncontrolled cross-sectional nature of the studies included, certainty of evidence is generally low. Further standardisation of the sampling procedure and microbiome result analysis with a much larger sample size may overcome the suspected imprecision and could address this uncertainty. A persistent challenge in metagenomics analysis is the lack of standard frameworks. Minor methodological differences can significantly affect the comparability of results. Variations in pre-analytical steps, such as sample collection and handling, or during the ‘wet lab’ phase, can influence the measured bacterial biomass and potentially result in the underrepresentation of low-copy micro-organisms.⁴⁸ Even healthy lungs exhibit a small yet detectable microbiota,^{2 35} where variations may be tied to pulmonary function.^{5 49} Distinguishing between pathogenic entities and commensal microbiota is also essential to guide targeted therapy,^{1 50} but if a sample is generally low in bacterial DNA, relative abundances may be misleading. The ‘healthier’ the ecosystem, the lower the biomass yield, which makes PCR amplification more challenging, whereas sampling itself can cross-contaminate oral, tracheal and lower respiratory flora. Upper airway contamination is a real concern in sputum samples, where bacteria from the upper tract can distort the observed community structure more than in BAL or ETA samples. This may introduce bias because the different tracheal sampling methods are prone to contamination to varying degrees. As an example of another type of potentially induced contamination, Miao et al³⁶ found Ralstonia to be one of the most abundant genera. Ralstonia is commonly known as part of the contaminating kitome.^{51 52} Kitome contamination may cause a greater impact on low-biomass tracheal samples, where background DNA can dominate sequencing results. Negative controls and rigorous bioinformatic filtering should therefore be standard components of future study designs to minimise this source of bias. Some included studies reported mitigation measures, like presampling oral rinsing^{34 35 39} or referring to community guidelines as their guidance,¹³ but only Chiyaka et al³⁹ performed contamination control sampling and used software (decontam) to flag potential low-biomass contaminants. Future studies should address contamination bias through simple countermeasures and adhere to community guidelines when performing sampling. As highlighted in the third domain of our RoB tool, the low success rate in creating gene libraries can introduce selection bias. This could be due to inadequate sampling, collection errors, the presence of active DNAse or simply low levels of bacterial DNA. This problem is evident in some of our included studies, with sample loss reaching as high as 63% in the work of Rogers et al.¹³

Furthermore, the ‘dry lab’ phase of analysis, including factors such as the choice of reference gene database, differences in alignment strategies (clustering first or assignment first), and variations in similarity thresholds, can have an even greater impact on composition and diversity outcomes.⁴⁸ Running the same sample in different laboratories often yields different results.⁵³ Variability in the reporting depth of sampling procedures and amplicon analysis contributes to inconsistency.

The statistical estimation of heterogeneity was high in all but one of our syntheses. Comparing the Shannon index in the endotracheal aspiration sampling group, we found that the I² statistic was low, indicating a high degree of similarity between the measured populations. Unsurprisingly, studies contributing to this finding^{33 36 37} involved only critically ill ICU patients. In contrast, due to our broad inclusion criteria for the patient populations, we cannot confidently say that we measured the same effect in other comparisons.

An additional limitation is that paired samples were not consistently analysed as dependent in the primary studies. While Rogers et al¹³ applied appropriate paired-sample methods and Cho et al³⁴ used a paired non-parametric test (Wilcoxon signed-rank), other comparative studies^{37 39 41} treated paired samples as independent. Five studies did not compare alpha-diversity between sampling methods directly. As a meta-analysis can only build on the methods of its included studies, this inconsistency may affect the robustness of pooled comparisons.

Implications for future research

Standardisation of the microbial community sequencing procedures and reporting of microbiome analysis results with improved correlation should be targeted to overcome the noted imprecision. As next-generation metagenomics moves ever closer to everyday application in healthcare practice, we should reevaluate and recalibrate the tools and strategies employed in infectology to ensure the highest possible patient safety while maintaining sample quality.

Conclusions

In clinical or research settings where segmental sampling is not required, tracheal sampling methods may offer a viable and less invasive alternative to BAL for characterising microbiome alpha diversity. However, further high-quality comparative studies are needed to confirm these findings.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information.