Lung fluid biomarkers for acute respiratory distress syndrome: a systematic review and meta-analysis

We manually searched PubMed, Embase, Wed of Science, and the Cochrane Library for studies on biomarkers for ARDS in lung fluid samples published prior to January 11, 2018. Details of the search strategy are listed in Additional file 1. We also searched the references of included studies. Two researchers screened and evaluated the eligibility of all studies independently, and a third reviewer intervened whenever there was a disagreement. The inclusion criteria were (1) original research report of adult with or at-risk of ARDS, (2) report of exact values of biomarker concentration in lung fluid related to a clinical outcome (diagnosis of ARDS in at-risk patients and/or mortality of ARDS), (3) description of demographic variables, and (4) written in English. The exclusion criteria were (1) written in languages other than English, (2) not related to ARDS/ALI, (3) not an original research, (4) in vivo/in vitro studies, (5) pediatric studies, (6) biomarker not measured in lung fluid, (7) biomarker used for treatment monitoring, and (8) only one article available for a specific biomarker for no mergeable effect size and low reliability.

We built Excel spreadsheets (Microsoft Corp., Redmond, WA) to extract data from the included studies, and the two researchers finished data extraction independently. The ARDS etiology and the mean or median level and standard deviation (SD) of the biomarker in the lung fluid were obtained. When a biomarker was measured sequentially, only the day 1 measurement was extracted. We extracted lung fluid biomarker levels from different subgroups as follows: patients with ARDS versus critically ill non-ARDS controls and survivors versus non-survivors in patients with ARDS. The mean value of a biomarker’s concentration was equal to the median level in this study, and standard error (SE) was converted to SD using an Excel formula. In addition, demographic variables (age, sex, and number of participants for each subgroup), diagnostic criteria for ARDS, ARDS mortality, the moment the lung fluid sample was retrieved, the sample type (BALF/other than BALF), sample retrieval location, and volume of BALF irrigation solution used were recorded. The recovery rate of BALF was also recorded, if provided.

All studies were assessed for quality according to the Quality Assessment of Diagnostic Accuracy Studies Score-2 (QUADAS-2), and the content was tailored according to the guideline of QUADAS-2 [6]. Details of the tailored QUADAS-2 are listed in Additional file 2. Risk of bias and an applicability concerns graph/summary was conducted using Review Manager version 5.3 (Cochrane Collaboration, Oxford, UK).

Meta-analysis was performed with Stata 13.1 (StataCorp LLC, College Station, TX). The ratio of means (RoM) was employed to assess the effect size [7‐9]. RoM is the mean value of a biomarker in the ARDS group divided by the mean value in the at-risk group (mean_ARDS/mean_at-risk) or the mean value of a biomarker in the non-survivors group divided by the mean value in the survivors group (mean_non-survivor/mean_survivor). RoM of each study was log transformed and pooled using the inverse-variance method to gain a pooled, transformed RoM, which was then back-transformed to determine the pooled RoM and 95% confidence interval, using the fixed effect model of the Stata software. The significance level for this meta-analysis model was set at p < .05. Forest plots were provided for biomarkers of which four or more studies were included in this meta-analysis. Biomarkers were ranked according to pooled RoM and statistical significance. We used the Q statistic to test the existence of heterogeneity; a p value of less than 0.10 was considered significant for heterogeneity. I² was employed to assess the proportion of total variability due to heterogeneity. An I² value of approximately 25% was regarded as low heterogeneity, 50% as medium, and 75% as high heterogeneity. Publication bias was assessed with Egger’s regression test [10], where a p value of less than 0.10 was considered significant for publication bias. Duval and Tweedie’s trim and fill was then conducted [11].

For the biomarkers with a significant RoM and existence of heterogeneity, we performed a subgroup meta-analysis on study type (case-control study versus another study type) or sample type (BALF versus other lung fluid), when three or more studies were included.

The total literature search yielded 1156 articles from the databases as follows: PubMed, 340 articles; Web of Science, 522 articles; Embase, 279 articles; Cochrane Library, 12 articles; and 3 articles from the reference lists of included studies. By reviewing the titles and abstracts, studies were mainly excluded due to the following: in vitro/animal studies (n = 434), duplication (n = 423), not an original research (reviews, editorials, or case reports, n = 90), and biomarkers not related with occurrence or mortality of ARDS (n = 30). After the initial screening, 95 articles remained for full-text review. Of these, 25 articles only reported on a specific biomarker, 16 articles contained insufficient data, and 4 articles had no full-text copy available, despite attempts to contact the authors. The remaining 49 articles were used for the meta-analysis [12‐60] (Fig. 1).

Demographic variables of the included studies are summarized in (Table 1). A total of 49 articles involving 2189 patients were included in this meta-analysis. ARDS/ALI was diagnosed according to the AECC criteria in 71% of the studies. Other criteria, such as edema fluid/plasma protein ratio [61], lung injury score [62], Fowler criteria [63], and clinical criteria, were used along with the AECC criteria. The mean age ranged from 37 to 70 years, mortality rate ranged from 15 to 77%, and lung fluid was collected between 30 min of intubation and 72 h of ARDS diagnosis. As for sample retrieval location, the right middle lobe and lingular lobe were the most common. Other locations were based on abnormal areas identified on chest radiographs, and a blind sampling of BALF was performed in two studies. In regard to sample type, 69% of the studies measured biomarkers in BALF with a certain volume of irrigation solution. Ten articles used pulmonary edema fluid, and four articles measured a biomarker in epithelial lining fluid. Only three articles provided the recovery rate of the irrigation solution; therefore, we could only assume a stable recovery rate between subgroups for this study.

The ARDS etiologies are summarized in Additional file 3. The most common cause of ARDS was sepsis (30.87%), followed by pneumonia (23.70%), trauma (10.94%), aspiration (8.53%), transfusion (4.23%), and major surgery (3.47%). Other etiologies included vasculitis, retroperitoneal hematoma-DIC, drug overdose, reperfusion injury, and diabetic ketoacidosis.

The quality assessment is displayed in Additional file 4, including the risk of bias and applicability of studies to the review question.

We performed a meta-analysis on 22 biomarkers in lung fluid associated with the diagnosis of ARDS in the at-risk population (Table 2); Fig. 2 shows the forest plots for biomarkers available in at least 3 studies. Pooled RoM values for total phospholipases A2 activity (total PLA2 activity) (17.995 [11.381, 28.454]), total protein (9.299 [7.575, 11.414]), albumin (6.544 [4.908, 8.725]), plasminogen activator inhibitor-1 (PAI-1) (5.525 [3.876, 7.877]), soluble receptor for advanced glycation end products (sRAGE) (4.901 [3.603, 7.673]), platelet activating factor-acetyl choline (PAF-AcH) (4.783 [3.495, 6.545]), soluble tumor necrosis factor-α receptors II (STNF-RII) (3.253 [1.765, 5.993]), hepatic growth factor (HGF) (3.199 [1.668, 6.135]), and interleukin-8 (IL-8) (3.008 [2.322, 3.896]) were the highest. The overall effect size ranged from 0.548 to 17.995, among biomarkers with significant RoM between subgroups, and decreased Club cell protein (CC16) (0.553 [0.369, 0.827]) and matrix metalloproteinases-9 (MMP-9) (0.548 [0.336, 0.893]) levels in lung fluid indicated a higher possibility of ARDS diagnosis in the at-risk population. However, a pervasive heterogeneity was displayed.

We performed an influence analysis to examine the sensitivity of the results. Influence analysis showed that the heterogeneity was possibly caused by the limited number of studies. By removing the studies with extreme RoM, we observed a robust effect on the biomarkers. The outcome of the influence analysis is displayed in Additional file 5.

Subgroup analysis was performed for biomarkers IL-8, total protein, albumin, sRAGE, PAF-AcH, and IL-6. Since most of the included studies were case-control studies, we excluded studies with other design types. Heterogeneity for total protein, albumin, and IL-6 was partly explained. However, heterogeneity for IL-8 was not clarified. Only one article was case-control study; therefore, the source of heterogeneity was not determined for PAF-AcH because of a limited number of study. We also excluded studies with ARDS that was not diagnosed using AECC criteria, which significantly reduced the heterogeneity for IL-6, but not for albumin. In addition, we assumed sample type may be a variable between studies because biomarker measurement in BALF was influenced by recovery rate and dilution. IL-8 remained significantly increased when BALF studies were excluded. As only one article measured biomarkers in lung fluid other than BALF, it was not evaluated. Results of the subgroup analysis are presented in Additional file 6.

We performed a meta-analysis on 11 biomarkers in lung fluid associated with ARDS mortality (Table 3); Fig. 3 shows the forest plots for biomarkers associated with ARDS mortality. Interleukin-1β (IL-1β) (4.617 [4.331, 4.921]), IL-6 (3.882 [3.270, 4.608]), IL-8 (3.679 [3.414, 3.964]), and Kerbs von Lungren-6 (KL-6) (3.178 [2.931, 3.446]) ranked the highest in biomarkers associated with ARDS mortality. The overall effect size ranged from 0.406 to 4.617. Among the biomarkers with a significant difference between survivors and non-survivors, decreased levels of interleukin-2 (IL-2) (0.828 [0.715, 0.959]) and CC16 (0.406 [0.362, 0.405]) were associated with a high mortality rate. Heterogeneity was displayed for many of the biomarkers, and influence analysis indicated that the heterogeneities were not likely caused by extreme RoM values. Due to the small number of studies, subgroup analysis based on design type or sample type could not be performed. Subgroup analysis for tumor necrosis factor-α (TNF-α) was performed when excluding patients with ARDS not diagnosed using AECC criteria. Results of the subgroup analysis are presented in Additional file 7.

Among the biomarkers associated with ARDS, Egger’s regression test demonstrated a p value of less than 0.10 for IL-6; furthermore, when we adjusted for possible publication bias by Duval and Tweedie’s trim and fill, the RoM remained significant for IL-6. Among the biomarkers associated with mortality, no publication bias was noted. The results of the publication bias analysis are presented in Additional file 8.