All available randomized controlled trials (RCTs) using NAC in ARDS patients were identified from the following data sources: PubMed/MEDLINE, EMBASE and Cochrane Library (inception until October 2015). The following keywords were used: (‘N-acetylcysteine’ OR ‘NAC’ OR ‘acetylcysteine’) AND (‘acute respiratory distress syndrome’ OR ‘adult respiratory distress syndrome’ OR ‘ARDS’ OR ‘acute lung injury’ OR ‘ALI’). In addition, the reference lists of the eligible studies were manually searched. No language restrictions were applied.

The retrieved studies were assessed by two reviewers. They identified the titles, abstracts and citations independently. Based on the criteria presented below, the reviewers assessed all the retrieved studies for inclusion. Disagreements in all phases were resolved by discussing with a third reviewer. The reviewers selected the eligible studies which satisfied the inclusion and exclusion criteria. Neither ethical approval nor patient consent were required in this meta-analysis, as all the studies included had already been published.

Inclusion criteria were as follows: i) RCTs, which were reviewed either published in full or in abstract form; ii) trials with additional use of NAC in patients with ALI or ARDS; iii) the definition or the diagnostic criteria of ARDS were clear and specific, and diagnosis was required to be based on at least the following criteria: a) Bilateral infiltration evidence on chest radiography, and b) hypoxemia evidence via arterial partial pressure of oxygen/fraction of inspired oxygen (PaO₂/FiO₂); iv) short-term mortality (30-day or hospital or ICU mortality), length of ICU stay, ratio of PaO₂/FiO₂ or adverse events were presented.

Exclusion criteria were as follows: i) The study was not an original research article; ii) the study was not an RCT; iii) the trial was limited to animals or cells; iv) no extractable outcome was included in the study.

The modified Jadad quality score was used to evaluate the methodological quality of included trials (19). It evaluates the method of randomization, allocation concealment, double-blinding and information on withdrawals and drop-outs to follow-up. According to these criteria, the studies were divided into two groups: High-quality group (score, ≥4) and low-quality group (score, ≤3).

The primary outcomes of the present meta-analysis were short-term mortality and 30-day mortality. The secondary outcomes were duration of ICU stay and PaO₂/FiO₂ ratio. The following data were extracted and recorded according to a pre-designed form: Name of first author, year of publication, number of patients, dosage of NAC, PaO₂/FiO₂ ratio, application of positive end expiratory pressure (PEEP), diagnostic criteria and adverse events. Additional information, including sex and age of patients, the number of cases of organ dysfunction and days of mechanical ventilation was also recorded. The data were extracted by two independent reviewers. Disagreement was resolved by discussion with the third reviewer when necessary.

According to the design, the weighted mean difference (WMD) was determined to analyze continuous variables and the 95% confidence interval (CI) was used to indicate the effect size. Similarly, the relative risk (RR) was used to analyze dichotomous data and the effect size was also indicated by the 95% CI. I² statistics were used to measure statistical heterogeneity (20). The fixed-effects model was applied if there was no obvious heterogeneity among studies (I²<50%) (21) and otherwise, a random-effects model was applied (I²>50%) (22). Publication bias was investigated via Begg's funnel plot method (23). P-values were 2-tailed P<0.05 was considered to indicate a statistically significant difference. The Review Manager version 5.3 (the Cochrane Collaboration, Copenhagen, Denmark) was used to perform statistical analyses. There was no registered protocol in the present meta-analysis.

Initially, 1,131 articles were found to be eligible according to the inclusion criteria applied to the search strategy. After removal of duplicates, 182 trials were excluded. A total of 102 trials remained after a preliminary screen, and 97 trials were excluded for the reasons listed in Fig. 1, which included editorials and review articles (n=33), animal studies (n=54) and studies not relevant to the present analysis (n=10). Five RCTs met the inclusion criteria and were subjected to further analysis (15,18,24–26).

There was no obvious difference in age, gender or any other basic information among all of the included trials. There were 183 patients in total, including 94 in the experimental group and 89 in the control group. Table I illustrates the characteristics of patients in the five studies (15,18,24–26), including basic information, intervention strategies, PEEP, oxygenation index and diagnostic criteria Table II presents the results of the quality evaluation. The studies by Bernard et al (24) and Domenighetti et al (25) were considered to be of high quality and the other three were of low quality. As <10 articles were included, publication bias was estimated using Begg's funnel-plot method. As presented in Fig. 2, no significant publication bias was found among those articles according to Begg's test (P=0.086).

All of the five trials reported on short-term mortality (Fig. 3). As there was no significant heterogeneity (I²=0%; P=0.58), the fixed-effects model was used to merge the data. After pooling of the results from all RCTs, the meta-analysis indicated that NAC did not reduce short-term mortality (RR=0.73; 95% CI: 0.50–1.07; P=0.10). Three of the trials contained data on 30-day mortality (Fig. 4). The fixed-effects model was used according to the test of heterogeneity (I²=0%; P=0.86). The meta-analysis revealed that additional use of NAC did not contribute to reduce 30-day mortality as compared with conventional therapy (RR=0.72; 95% CI: 0.44–1.19; P=0.20).

Four trials reported on the duration of ICU stay (Fig. 5). The fixed-effects model was used according to the test of heterogeneity (I²=25%; P=0.26). The meta-analysis revealed that NAC treatment significantly shortened the duration of ICU stay when compared with that in the control treatment group [WMD, −4.56; 95% CI: (−7.32 to −1.80); P=0.001].

Three trials reported on changes in oxygenation in ARDS patients. There was obvious heterogeneity between the three trials (I²=91%, P<0.0001), so a random-effects model was used (Fig. 6). There was no significant difference in the ratio of PaO₂/FiO₂ between the two groups [WMD, 54.34; 95% CI: (−30.50 to 139.17); P=0.21]. The heterogeneity may have been caused by inconsistent intervention measures and different stages of disease. As the number of included trials was small, no subgroup analysis was performed.

All of the five trials reported that no adverse events were caused by the study medication.