Supernatant from endobronchial ultrasound-guided transbronchial needle aspiration samples for molecular profiling in NSCLC: a systematic review and meta-analysis.

Introduction
Lung cancer remains the leading cause of cancer-related mortality worldwide [1], with nonsmall cell lung cancer (NSCLC) representing the most prevalent subtype [2, 3]. The diagnosis of lung cancer remains a challenge due to the asymptomatic nature of early-stage disease and the nonspecific symptoms that emerge later, often leading to delayed diagnoses [4, 5]. As a result, most patients are diagnosed at advanced stages when surgical resection is no longer an option [5, 6]. Advances in NSCLC management have highlighted the critical role of molecular diagnostics in enabling personalised therapies. ∼50% of NSCLC cases harbour targetable genetic alterations that influence the response to treatment [3, 7, 8]. These findings have propelled the adoption of precision medicine strategies, improving survival, reducing treatment toxicity, and enhancing patients’ quality of life [9–11]. However, implementing molecular diagnostics in routine clinical practice requires high-quality biological samples [12].
Endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA) has emerged as a cornerstone technique for the diagnosis and staging of NSCLC, offering high sensitivity and specificity [13, 14]. Its minimally invasive nature makes it an optimal choice for sample acquisition in patients with advanced disease [12, 15]. Traditionally, tissue and cytology samples obtained via EBUS-TBNA have been used for histopathological and molecular testing [15]. Nevertheless, the growing demand for increasingly detailed molecular profiling to guide therapeutic decisions places additional pressure on these samples, which are often limited in quantity and quality [12]. Furthermore, the standard diagnostic workflow prioritises histology, with molecular testing conducted sequentially, resulting in prolonged time to treatment initiation [16, 17].
These challenges have sparked interest in using the supernatant phase from EBUS-TBNA samples as an alternative source for molecular testing. In fact, circulating tumour DNA (ctDNA) is already established as an alternative source for molecular testing and has demonstrated utility in various biological fluids [18–20]. However, ctDNA often presents limitations in DNA quantity, compromising its reliability for molecular analysis [19].
To address this challenge, researchers have proposed utilising discarded biological material, such as supernatant from EBUS-TBNA, as a novel source of ctDNA for molecular profiling. Early studies suggest that supernatant-based molecular diagnostics may offer comparable results to traditional tissue samples while preserving the cellular components and potentially improving diagnostic efficiency and patient outcomes [21–28].
This systematic review aims to evaluate the feasibility, diagnostic yield and molecular performance of EBUS-TBNA supernatant as a source for molecular testing in NSCLC. By comparing its results with conventional tissue samples, we seek to determine its potential role in the evolving landscape of minimally invasive diagnostics and precision medicine in NSCLC.

Material and methods

Study design
This systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [29] to ensure methodological rigour and transparency. The protocol was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO) (identifier CRD42024600046). The review was structured according to the PICO framework, as follows.
Population (P): patients with suspected or confirmed NSCLC undergoing EBUS-TBNA.

Intervention (I): molecular profiling performed on supernatant from EBUS-TBNA.

Comparison (C): molecular results from supernatant samples versus a reference standard of conventional tissue or cytology-based specimens.

Outcomes (O): measures of feasibility, diagnostic yield, molecular concordance between sample types, turnaround time.

Search strategy
A comprehensive search of peer-reviewed literature was performed across major biomedical databases, including PubMed, Embase, Cochrane Library, Connected Papers, and ClinicalKey, covering studies published from January 2000 to the present. This timeframe was chosen to reflect the period since the introduction of EBUS-TBNA in 2002, to ensure the inclusion of all studies that investigated the use of supernatant obtained from EBUS-TBNA samples for molecular profiling in NSCLC. No language restrictions were applied; non-English articles were translated as needed.
Additional searches were conducted in trial registries such as www.ClinicalTrials.gov, ISRCTN, and the Australian New Zealand Clinical Trials Registry (anzctr.org.au) to identify unpublished or ongoing studies. Grey literature, including conference abstracts, theses, and dissertations, was explored using platforms such as OpenGrey (www.opengrey.eu) and OAIster (www.oclc.org/en/oaister.html). Reference lists of included studies were screened manually to identify additional relevant publications.
The search strategy included combinations of terms such as (“nonsmall cell lung cancer” OR “lung cancer”) AND (“EBUS-TBNA” OR “endobronchial ultrasound”) AND (“molecular profiling” OR “molecular testing” OR “NGS”) AND (“supernatant”). Full details of the search terms and Boolean operators are provided in the supplementary material (appendix S1).

Eligibility

Inclusion criteria
Studies were included according to the following criteria: 1) research studies published from 2000 onward; 2) studies involving NSCLC patients undergoing EBUS-TBNA; 3) comparison of molecular findings from EBUS-TBNA supernatant versus reference standard; 4) studies reporting diagnostic outcomes, including sensitivity, specificity, predictive values or molecular yield (e.g. detection of EGFR, ALK or KRAS mutations); 5) observational studies (prospective and retrospective cohorts, case–control studies), diagnostic accuracy studies and randomised controlled trials; and 6) full-text articles and abstracts containing sufficient data for analysis.

Exclusion criteria
Studies excluded were 1) studies from before the year 2000; 2) studies involving samples obtained exclusively by methods other than EBUS-TBNA; 3) studies focusing on cancers other than NSCLC; 4) narrative reviews, editorials or commentaries unless those presenting novel data.

Outcomes

Primary outcome
The primary outcome was the feasibility of molecular profiling from EBUS-TBNA supernatant, defined as the ability to obtain DNA of sufficient quantity and quality for molecular analysis, compared to reference standard. Diagnostic performance was also assessed (sensitivity, specificity, predictive values), using reference standard specimens for comparison.

Secondary outcomes
Pre-analytical factors: impact of preservation medium, storage temperature and centrifugation on feasibility and molecular yield.
Molecular concordance: agreement between molecular results obtained from supernatant and reference standard, quantified as percentage concordance and using Cohen's κ coefficient when applicable.
Detection of actionable mutations: proportion of supernatant samples with clinically relevant genetic alterations, as defined by current guidelines [3, 30].
Turnaround time (TAT): time, in days, required to obtain molecular results from supernatant compared to reference standard.

Screening and selection process
References were imported into Rayyan [31], a web-based review tool, to remove duplicates. Screening was conducted in two stages. First, two independent reviewers (L. Vaz Rodrigues and J. Oliveira) assessed all titles and abstracts for relevance. Second, potentially eligible studies were retrieved in full and independently evaluated against the inclusion criteria by the same reviewers.
Discrepancies in study selection were resolved through discussion and consensus. If disagreements persisted, a third reviewer (T. Maricoto) was consulted. Selection flow and exclusion reasons are detailed in the PRISMA 2022 diagram (figure 1).

Data extraction
Two independent reviewers (L. Vaz Rodrigues and J. Oliveira) extracted and synthesised data using a standardised form, piloted on a subset of studies to ensure consistency. The extracted data included the following.
1) Study characteristics: authors, year of publication, study design and location.

2) Patient characteristics: sample size, age, sex, NSCLC subtype, stage, smoking history and comorbidities (if reported).

3) Diagnostic methods: EBUS-TBNA technique (needle gauge, number of passes), collection methods for supernatant and reference standard samples.

4) DNA and RNA concentration (if reported) in supernatant and reference standard, preserving solution and pre-processing details (storage time, storage temperature, centrifugation speed and duration). Molecular techniques applied and detected mutations were also recorded.

5) Outcomes: feasibility; concordance between supernatant and reference standard results; detection of actionable mutations and TAT.

Disagreements were resolved by consensus or, if needed, by a third reviewer (T. Maricoto). Data were compiled into a central database for descriptive and inferential analysis.

Risk-of-bias assessment
The methodological quality of the included studies was assessed using the Cochrane Risk of Bias Tool (RoB 2.0) [32] for RCTs and the Newcastle–Ottawa Scale (NOS) for observational studies [33]. Studies were rated as low, moderate or high risk based on selection, performance, detection, attrition and reporting domains. Assessments were conducted independently by L. Vaz Rodrigues and J. Oliveira, with disagreements resolved by consensus or, if necessary, consultation with a third reviewer (T. Maricoto).

Data synthesis and analysis
A qualitative synthesis was performed to summarise the included studies. Study characteristics, methods and outcomes were summarised in a descriptive table (table 1). Narrative synthesis was used to compare results and explore factors influencing molecular profiling using EBUS-TBNA supernatant.
Most studies reported outcomes as median (range). To allow data aggregation, means and standard deviations were estimated from medians using the method by Hozo
et al. [34]. Using the estimated means, standard deviations and sample sizes, a restricted random-effects meta-analysis was performed with the metafor package in R software (version 4.4.1) [35].
Subgroup analyses by “preservative solution” and “storage temperature” were conducted to evaluate differences.
Between-study heterogeneity was assessed using the Q statistic, I2 index, and the τ2 estimate. Meta-regression was used to test subgroup differences with preservative solution and temperature as moderators.
Agreement between supernatant and reference standard was assessed by direct comparison, defined as the percentage of identical molecular alterations in both samples. Cohen's κ was computed for subgroups with sample size ≥30 samples and κ-values were interpreted using standard thresholds: values >0.80 indicated almost perfect agreement, 0.61–0.80 substantial agreement, and 0.41–0.60 moderate agreement.
Additionally, a detailed analysis of diagnostic accuracy was conducted by extracting the number of true positives, false positives, false negatives and true negatives from each study when available. All pooled estimates are reported with 95% confidence intervals.

Results

Overview of eligible studies
A total of 469 articles were initially identified through database searches based on the pre-defined eligibility criteria. After removing 41 duplicates, 428 articles remained for screening. Titles and abstracts were reviewed by two investigators (L. Vaz Rodrigues and J. Oliveira). Studies were excluded if they were unrelated to NSCLC molecular profiling with EBUS-TBNA supernatant, had ineligible design, or fell outside the defined timeframe. After this initial screening, 23 articles were selected for full-text review.
Following detailed assessment, 16 studies were excluded. One was a narrative review with no original data. 10 analysed needle lavage samples, but not supernatant. Five included heterogeneous sample populations without clearly distinguishing NSCLC or EBUS-TBNA-derived samples.
Seven studies met eligibility criteria and were included. The PRISMA flow diagram (figure 1) summarises the selection process. Table 1 provides an overview of the methodological features of the included studies. All seven studies were published between 2018 and 2024 and were mostly prospective observational studies, conducted in the USA (six studies) and China (one study). The reference standard varied across studies, with most relying on cell-block preparations, while others utilised core biopsies, smears or previous surgical biopsy specimens for comparison.

Risk-of-bias assessment
As all included studies were observational, risk of bias was assessed using the NOS. Overall, six studies were rated as low risk, while one had a moderate risk, as comparator samples were frequently derived from different anatomical sites or collected at different times potentially limiting the reliability of molecular concordance. Assessment was conducted independently by two reviewers (L. Vaz Rodrigues and J. Oliveira), with full agreement. A detailed breakdown is provided in supplementary table S1.

Data synthesis

Study population and diagnostic methodologies
Across the included studies, there was a cumulative total of 506 patients (range 17–214 per study) with a median age ranging from 63 to 72 years. The proportion of male patients varied from 37% to 60%, while female representation ranged from 40% to 63%. Smoking status was not detailed in any of the included studies. Only Wu
et al. [26] reported tumour staging using the 8th edition of the TNM (tumour, node, metastasis) classification [36]. In that cohort (n=214), most patients were stage IV (n=160), followed by stage III (n=28), II (n=10) and I (n=16). EBUS-TBNA was the primary diagnostic method applied in all studies, often combined with computed tomography-guided transthoracic biopsy or other bronchoscopy techniques. Histological classification varied, with adenocarcinoma being the most common diagnosis, followed by squamous cell carcinoma and NSCLC not otherwise specified. Needle gauges (G) for EBUS-TBNA ranged from 19 G to 25 G, and rapid on-site evaluation was reported in four studies. Only Hannigan
et al. [24] reported the number of needle passes, averaging two to three per case. Next-generation sequencing (NGS) was used in all studies: alone in four, combined with droplet digital (dd)PCR in two, and with amplification refractory mutation system (ARMS)-PCR in one. The key methodological details of the included studies are summarised in table 1. Based on these data, the following sections provide a detailed analysis of the diagnostic performance, molecular testing feasibility and the impact of methodological variables on the results.

Feasibility of supernatant based molecular testing
All seven studies confirmed that EBUS-TBNA supernatant is a viable source for molecular testing, with feasibility rates ranging from 87% to 100%. This finding was consistent throughout the different studies, despite variability in sample processing protocols, including storage conditions (−80°C to 5°C), centrifugation speeds (reported as 400–10 000×g or 600–16 000 rpm), and preservation media (CytoLyt, RPMI, physiological saline or ThinPrep). While no clear trend was observed between storage temperature and feasibility, studies employing multistage centrifugation protocols consistently reported 100% feasibility, whereas those using single-step centrifugation had the lowest feasibility rates (87–90%).
Feasibility was also evaluated in relation to molecular testing methods. Most studies employed NGS as the primary analysis technique, while some incorporated ddPCR [23, 24] or ARMS-PCR [26]. Among studies using NGS exclusively, feasibility ranged from 87% to 100% [25, 27, 28]. In studies combining NGS with ddPCR or ARMS-PCR, feasibility was between 90% and 100%. Given the narrow range of feasibility across all molecular methods, no meaningful differences were observed.

DNA yield from supernatant samples
Six of the seven included studies reported quantitative DNA yields from supernatant, showing substantial variability. Median values ranged from 0.77 ng·µL−1 [26] to 85.5 ng·µL−1 [21]. Gokozan
et al. [27] did not report DNA yield.
Notably, high-speed centrifugation protocols, such as those used by Wu
et al. [26] (10 000 rpm) and Jager
et al. [28] (16 000 rpm), were associated with lower DNA yields (0.77–4.7 ng·µL−1), compared to moderate-speed processing (e.g. 600×g Hannigan
et al. [24]), which yielded 6.8 ng·µL−1.
Differences in preservative solutions and storage conditions were evident. While most studies utilised CytoLyt or physiological saline, Hannigan
et al. [24] employed RPMI medium, and Tafoya
et al. [21] used ThinPrep. Storage temperatures varied from −80°C [23] to 5°C [21], while storage durations ranged from <1 day [28] to a median of 14 days [25]. Table 2 summarises the protocols for sample handling and DNA yield from EBUS-TBNA supernatant, illustrating considerable heterogeneity in centrifugation, storage and preservation media across studies.
Pooled DNA yields were calculated across six studies and stratified by preservation medium and storage temperature.
For preservation medium, studies using CytoLyt (k=2) showed a pooled mean DNA yield of 24.36 ng·µL−1 (95% CI 11.94–36.77), with high heterogeneity (I2=88.7%; Q=8.84; p<0.001; τ2=71.21). Studies using other media (k=4) had a pooled mean of 61.92 ng·µL−1 (95% CI −23.82–147.66; I2=99.8%; Q=53.19; p<0.001; τ2=7174.25). Combining all six studies, the overall pooled mean was 40.10 ng·µL−1 (95% CI 1.25–78.94; I2=99.5%; Q=69.96; p<0.001; τ2=2139.56). Despite high intragroup heterogeneity, subgroup differences were not statistically significant (Q=0.33; p=0.5664). Figure 2a shows the forest plot stratified by preservation medium.
A separate analysis by storage temperature is shown in figure 2b. Studies using 4°C (k=3) had a pooled mean yield of 19.26 ng·µL−1 (95% CI 7.38–31.13; I2=93.3%; Q=34.24; p<0.0001; τ2=102.39). Studies using other temperatures (k=3) showed a pooled mean of 84.05 ng·µL−1 (95% CI −35.39–203.49; I2=99.5%; Q=17.27; p=0.0002; τ2=10 450.95). The overall model again indicated significant heterogeneity (I2=99.5%; Q=69.96; p<0.0001; τ2=2139.56), with a combined mean of 40.10 ng·μL−1 (95% CI 1.25–78.94). As with preservation medium, no significant difference was found between temperature subgroups (Q=1.15; p=0.2834).

Diagnostic accuracy of supernatant
All included studies utilised either cytology cell-blocks or histological tissue biopsies as the reference standard for molecular analysis. Sensitivity of supernatant molecular analysis compared to reference standard was consistently high, ranging from 0.95 to 1.00, with full concordance in several studies (supplementary table S2). Guibert
et al. [23] and Janaki
et al. [25] reported 100% agreement between supernatant and reference standard for all mutations analysed. Wu
et al. [26] reported a sensitivity of 0.98, with six discordant cases (three false positives and three false negatives). Specificity, when measurable, ranged from 0.80 to 1.00. However, specificity could not be determined in Guibert
et al. [23] and Janaki
et al. [25] due to the absence of true negative cases. Due to heterogeneity in study design, variations in sample size, and the absence of true negative cases in some studies, pooled sensitivity and specificity estimates were not calculated to prevent potential bias. Instead, individual study results were reported in detail to ensure accuracy and preserve the integrity of the findings (supplementary table S2).

Comparison of molecular profiling results in supernatant and reference standard
The agreement between supernatant and reference standard samples was consistently high, with concordance rates ranging from 83% to 100%, depending on study design and sample comparison criteria (table 1).
Most studies reported full agreement in mutation profiling between supernatant and reference standard. However, Hannigan
et al. [24] noted variability based on biopsy conditions: concordance was 97% when samples were collected from the same site and day, decreasing to 89% when obtained from a different site on the same day, and to 83% when from a different site and day.
Across all studies, Cohen's κ was 0.947 (95% CI 0.905–0.989), indicating almost perfect agreement between supernatant and reference standard molecular results.
Regarding the detection of clinically actionable mutations, a total of six studies provided disaggregated mutation-level data enabling direct comparison between supernatant and reference standard results (supplementary table S3). One of these studies [26] focused exclusively on EGFR mutations in a cohort of 214 patients and did not report data on other genes.
EGFR mutations were reported in 169 individuals, corresponding to 33.4% of the total cohort. In one study, involving 214 patients, EGFR mutations were identified in 120 (56.1%) cases, although specific mutation subtypes were not detailed [26]. In the remaining studies, where individual EGFR variants were specified, the most frequently reported were L858R (18 cases, 3.6%), followed by T790M (eight cases, 1.6%) and exon 19 deletions (five cases, 1.0%). The concordance between supernatant and reference standard for EGFR mutations was 96.4% (163 out of 169 cases), with six discordant results (1.2% of the total cohort). Cohen's κ for EGFR was 0.974 (95% CI 0.948–0.999), indicating almost perfect agreement.
KRAS mutations were identified in 39 patients across the included studies, corresponding to 13.4% of the 292 patients tested for this gene. The most frequently reported mutations were G12C (14 cases, 4.8%), G12D (eight cases, 2.7%), and G12A (five cases, 1.7%). Concordant detection occurred in 38 (97.4%) out of 39 cases, with a Cohen's κ of 0.971 (95% CI 0.913–1.000), indicating almost perfect agreement.
ALK rearrangements were reported in four (1.4%) patients across the included studies, all of which were consistently detected in both supernatant and reference standard samples, corresponding to 100% concordance. ERBB2 alterations, including exon 20 insertions and gene amplifications, were identified in seven (2.4%) patients, also with complete agreement between supernatant and reference standard. MET exon 14 skipping was reported in a single case (0.3%) in the study by Janaki
et al. [25], with concordant detection in both sample types. BRAF mutations were detected in 11 (3.8%) patients and included variants such as V600E, V600R, D594N, N581S, K601N and G466E. Of these, nine were identified in both supernatant and reference standard, while two were detected only in the reference standard, resulting in a concordance of 81.8%. Other molecular alterations with potential clinical or prognostic relevance (such as those involving TP53, STK11, PIK3CA, IDH1, PTEN, FGFR family genes, CDKN2A and SMAD4) were also frequently reported across studies, with high rates of agreement between supernatant and reference standard (supplementary table S3). Due to the low sample size of the subgroups, κ-statistics were not performed for ALK, ERBB2, MET, BRAF and other less frequent molecular alterations.

Turnaround time
The use of supernatant derived from EBUS-TBNA samples for molecular characterisation of NSCLC demonstrated a consistent reduction in TAT across several studies. Guibert
et al. [23] highlighted that supernatant offers an immediate source of fresh DNA, potentially decreasing TAT compared to traditional cell-block material. Hannigan
et al. [24] reported a significant reduction, with tissue-based NGS requiring 5–7 days versus 2–3 days for supernatant samples. Similarly, Janaki
et al. [25] observed a decrease from 10±5.5 days to 3–5 days with the use of supernatant. Gokozan
et al. [27] also reported a reduction in TAT from 12.2±5.3 days (reference standard) to 8.5±1.8 days (supernatant). Additionally, Jager
et al. [28] noted a qualitative reduction in TAT by 1 day when using supernatant, compared to the 4–10 days required for the reference standard. These findings are summarised in table 3. Due to the limited number of studies reporting TAT with consistent metrics and the heterogeneity in how TAT was measured, no statistical pooling or meta-analysis was conducted.

Discussion
This systematic review and meta-analysis demonstrates that supernatant obtained from EBUS-TBNA samples is a feasible and reliable source for molecular profiling in NSCLC. Across the seven included studies, feasibility rates ranged from 87% to 100%. Despite heterogeneity in sample handling protocols, DNA yields were consistently sufficient for downstream molecular techniques. Moreover, the diagnostic performance of supernatant samples was comparable to that of conventional tissue or cytology-based specimens. Molecular concordance between supernatant and reference standard ranged from 83% to 100%, with a pooled Cohen's κ of 0.947, reflecting almost perfect agreement. Notably, actionable mutations, including EGFR, KRAS, ALK, ERBB2, MET and BRAF, were detected in supernatant with high accuracy when compared to reference standard, further supporting its utility in therapeutic decision-making. Additionally, most studies reported a reduction in TAT for molecular results when using supernatant, with differences ranging from 2 to 7 days. These findings reinforce the potential of supernatant as a robust and time-efficient alternative or complementary sample in molecular diagnostic workflows.
Minimally invasive diagnostic approaches play a critical role in the molecular characterisation of NSCLC, with EBUS-TBNA now established as a cornerstone technique [37–39]. Beyond its pivotal role in mediastinal staging, EBUS-TBNA has proven highly effective for obtaining diagnostic material in NSCLC, allowing simultaneous histopathological confirmation and molecular profiling from a single procedure [39–41]. However, limited sample volume and cellularity may constrain biomarker testing, particularly in advanced disease where comprehensive profiling is essential for therapeutic decisions [41, 42].
To address the issue of sample exhaustion, liquid biopsy has emerged as a complementary strategy [18]. While plasma-derived ctDNA is widely used in advanced NSCLC, its sensitivity varies significantly by mutation type [20, 43]. Alternative biological sources beyond plasma have therefore been investigated to enhance the diagnostic sensitivity of liquid biopsy approaches, including pleural fluid in patients with malignant pleural effusions [44] and bronchial lavage fluid obtained during bronchoscopy [45]. In the context of liquid biopsy strategies, supernatant derived from different cytological sample preparations is emerging as a valuable, reliable and most often underutilised source for molecular testing, as was recently emphasised in a narrative review by Roy-Chowdhuri [46].
Among these alternatives, supernatant derived from EBUS-TBNA samples offers several advantages. Unlike peripheral blood, which reflects DNA shed systemically, EBUS-TBNA supernatant is collected directly from the tumour microenvironment during sample processing. This proximity may result in a higher concentration and greater representativeness of tumour-specific genetic material. Our systematic review supports this hypothesis, with feasibility rates between 87% and 100% across all included studies. Interestingly, one study documented higher DNA concentrations in supernatant than in the corresponding cellular pellet, suggesting that the acellular phase may retain significant amounts of tumour-derived nucleic acids, possibly due to the accumulation of DNA-rich cellular debris [21]. However, the DNA yield from supernatant samples showed notable variability, which appears to be influenced by pre-analytical factors such as sample processing protocols, storage conditions, centrifugation speeds and preservation media.
The studies included in this review revealed substantial heterogeneity in the pre-analytical processing of EBUS-TBNA supernatant samples, particularly regarding storage temperature, preservative media and centrifugation protocols (table 2). This variability significantly affected the comparability of studies and the strength of pooled results. Despite the methodological heterogeneity, subgroup analyses for storage temperature and preservative solutions were feasible and showed no statistically significant differences in DNA yield. However, the associated high heterogeneity limits the generalisability of these findings (figure 2). The impact of methodological heterogeneity was even more pronounced for other variables, particularly centrifugation protocols. Speeds, g-forces and processing times were inconsistently reported across studies. Due to this lack of standardisation, no formal meta-analysis could be performed for this parameter. Nevertheless, individual studies suggested that high-speed centrifugation may negatively impact DNA yield. Wu
et al. [26], using a high-speed centrifugation protocol (10 000 rpm), reported the lowest DNA concentration (0.77 ng·µL−1), whereas Hannigan
et al. [24], employing a more moderate protocol (1500 rpm), achieved higher yields (6.8 ng·µL−1). One possible explanation is that excessive centrifugal force may induce DNA fragmentation, reducing yield, an effect described in other contexts, including spermatozoa [47] and Escherichia coli DNA studies [48]. While this hypothesis is biologically plausible, the findings must be interpreted with caution. The wide methodological variability between studies precludes firm conclusions regarding optimal pre-analytical protocols and limits the broader applicability of individual observations.
Additionally, this heterogeneity prevented a formal meta-analysis of diagnostic accuracy metrics, such as sensitivity and specificity. Differences in the definition of reference standard, variability in the molecular targets assessed, and frequent absence of true negative cases made pooled estimates methodologically unsound. Available individual data are presented in supplementary table S2, but should be interpreted cautiously and within the methodological context of each study. Overall, these inconsistencies underscore the need for prospective studies adopting standardised workflows to enable reproducibility and improve data comparability.
Despite the methodological heterogeneity, this review found consistently high concordance between the molecular profiling results obtained from EBUS-TBNA supernatant and reference standard. Specifically, actionable mutations such as EGFR and KRAS were reliably detected in supernatant, with agreement rates ranging from 83% to 100% and pooled Cohen's κ-values nearing 1.0, indicating almost perfect agreement. Discordant cases were rare and may be attributed to several biological and technical factors, such as low tumour cellularity in individual samples [28], variability in DNA shedding related to tumour stage, burden and vascularisation [49], or differences in sequencing platforms and amplicon design [24]. Temporal and spatial mismatches in sample acquisition also affected agreement. This was illustrated in the study by Hannigan
et al. [24], where lower concordance was observed when reference standard samples were obtained from different sites and at different times. Such discrepancies probably reflect not only technical factors, but also biological variability, as spatial and temporal heterogeneity in solid tumours can lead to distinct molecular profiles and introduce bias when comparing samples [50, 51]. In addition, the limited number of studies restricted subgroup analyses for less frequent mutations such as ALK, ERBB2, BRAF and MET. Although numerical concordance was high, their low cumulative frequency prevented the calculation of Cohen's κ, underscoring the need for larger datasets and standardised reporting to support robust comparisons for these targets.
The molecular alterations analysed in this review are among the most clinically relevant in NSCLC and are prioritised in current treatment algorithms by both the National Comprehensive Cancer Network [3, 17] and the European Society for Medical Oncology [30, 52]. Importantly, the molecular landscape of NSCLC continues to evolve, driven by the discovery of new predictive biomarkers, resistance mechanisms and therapeutic targets [53]. As precision oncology advances, the scope and complexity of molecular testing is expected to expand further. In this dynamic context, sample exhaustion has emerged as a growing barrier to timely and complete molecular profiling [15, 42]. This underscores the importance of maximising the utility of all available biological specimens, particularly those obtained through minimally invasive techniques such as EBUS-TBNA. In this context, the supernatant, routinely discarded during conventional processing, emerges as a valuable and underutilised resource. As highlighted in our review, supernatant consistently provided sufficient nucleic acids and delivered high diagnostic performance. These findings align with previous studies and reinforce its potential to reduce the need for additional procedures [54].
One of the most promising and clinically meaningful advantages of using supernatant from EBUS-TBNA samples is the potential to reduce TAT for molecular results. Although heterogeneity among the included studies precluded data pooling, five studies [23–25, 27, 28] reported meaningful reductions in TAT, ranging from 1 to 7.5 days when compared to reference standard. This gain is largely attributed to the immediate availability of ctDNA in supernatant and the ability to initiate molecular workflows on the same day of sample collection, bypassing delays associated with tissue processing. Reducing TAT is not a trivial matter, as delays in initiating systemic therapy can negatively impact prognosis, particularly in patients with rapidly progressing disease [55]. Interestingly, although clinical guidelines recommend a TAT of <10 working days [16], real-world studies often report longer intervals, reflecting logistical and institutional barriers [56–58]. These delays represent a well-recognised obstacle to the implementation of precision oncology [58, 59] and have been associated with reduced survival in large cohort studies [60]. These discrepancies between molecular testing guideline recommendations and real-world clinical practice highlights the need for diagnostic strategies that are not only biologically effective but also operationally efficient, such as reflex test protocols, or the early incorporation of liquid biopsy approaches [58, 59]. Alternatively, integrating supernatant from EBUS-TBNA samples into molecular workflows may offer a pragmatic solution capable of reducing TAT while simultaneously minimising the need for additional sample acquisition or repeat biopsies. These advantages may be particularly valuable in resource-limited settings, where access to invasive procedures, histological processing, or comprehensive molecular platforms is often constrained [37, 61]. In such contexts, the use of supernatant may help reduce the need for re-biopsies and accelerate molecular diagnostics using existing resources, although this was outside the scope of the present review and remains to be demonstrated in dedicated studies.
This study has some limitations that should be acknowledged. First, only seven studies met our inclusion criteria, which were specifically restricted to NSCLC and supernatant derived from EBUS-TBNA samples. While this narrow scope allowed for focused analysis, it also limited the breadth of available evidence and reduced the statistical power of subgroup analyses. Additionally, heterogeneity in the methodologies used across studies (centrifugation speed, storage time, preservation solutions and DNA extraction protocols) hindered direct comparisons and constrained our ability to identify the most effective practices for optimising DNA yield. Consequently, subgroup analyses were feasible only for a limited number of parameters (specifically, storage temperature and preservation medium) whereas the greater variability observed in others, such as centrifugation protocols, precluded formal meta-analysis. As a result, no definitive conclusions can be drawn about optimal pre-analytical workflows, and the generalisability of DNA yield findings remains limited.
The limited number of included studies also restricted the ability to perform mutation-specific concordance analyses, particularly for alterations with lower frequency. While EGFR and KRAS mutations were sufficiently represented to support more robust statistical evaluation, other clinically relevant targets, such as ALK, BRAF, ERBB2 and MET, were reported too infrequently to allow pooled analysis.
Despite these constraints, to our knowledge, our systematic review and meta-analysis represents the first comprehensive synthesis focused on the use of EBUS-TBNA supernatant for molecular profiling in NSCLC. The findings are encouraging, demonstrating high feasibility, strong agreement with conventional samples, and potential operational advantages, including reduced TAT. These results support the integration of supernatant into diagnostic workflows as a complementary or rescue material, especially valuable when tissue is limited or exhausted.
Future studies should aim to standardise supernatant processing protocols and validate their use in larger, prospective cohorts. In addition to evaluating technical performance, future research should also explore clinical implementation pathways and assess real-world impact on turnaround times, treatment decision-making, and patient related outcomes. Establishing best practices will be essential to support broader adoption and fully harness the diagnostic potential of this minimally invasive, high-yield biospecimen in precision oncology.

Points for clinical practice
The supernatant phase from EBUS-TBNA samples is a reliable and underutilised source for molecular profiling in NSCLC, with successful DNA extraction reported in 87% to 100% of cases and near-perfect concordance with tissue-based reference standards (Cohen's κ up to 0.974).

Supernatant consistently provides sufficient DNA for next-generation sequencing and other molecular platforms, even when pre-analytical conditions vary.

Supernatant use may reduce the need for re-biopsies, optimise available material, and accelerate turnaround time for molecular results, with reported reductions of up to 7.5 days.

This practice may further enhance the strategic role of EBUS-TBNA as a cornerstone in minimally invasive precision oncology, aligning diagnostic efficiency with the growing complexity of biomarker testing.

Questions for future research
What are the optimal pre-analytical protocols (e.g. centrifugation speed, storage temperature, preservation medium) to maximise DNA yield from EBUS-TBNA supernatant?

Could the integration of supernatant analysis consistently reduce time-to-treatment and improve survival in prospective clinical pathways?

What are the cost-effectiveness and scalability implications of adopting supernatant as a standard component in NSCLC molecular diagnostics?