Through database and manual search (Fig. 1), 2220 records were retrieved. We screened both titles and abstracts according to the inclusion and exclusion criteria, leaving 38 records that were deemed suitable for inclusion. After reviewing the full texts, 12 records [18-30] met the inclusion criteria and were finally included (Table 1). Studies were excluded for the following reasons: duplication (n = 24), non-RCTs (n = 1) .
The characteristics of each trial are listed in Table 1. There were 8 trials exploring the effects of AA combined with co-intervention, such as thiamine [18-24], N-acetylcysteine , and α-tocopherol . The longest intervention course of AA lasted for 10 days , while the shortest was only 1 day ; only 1 records lack of specific data of the treatment course . The articles included was published in a long span from 1997 to 2020. The sample sized ranged from 20 to 211. The included trials enrolled a total of 1232 patients of which 571 received AA administration, and 661 were controlled subjects.
Risk of bias and quality of evidence
The risk of bias was summarized in Fig. 2. The trial performed by Balakrishnan  and Jose  were deemed a high risk of bias as incomplete outcome data over mortality. Studies by Fujii , Wani , and Chang  were not performed in a double-blinded way and were rated as high risk. In total, 7 trials were defined as an unclear risk and 5 trials were deemed high risk.
IV AA and mortality
The meta-analysis, recruited 12 RCTs, indicated that IV AA may have no influence on 28-day mortality of sepsis patients (OR = 0.81; 95% CI (0.54-1.23); p = 0.326; I2 = 44.7%, Additional files 3: Fig. 3A). A random (M-H heterogeneity) model was employed due to clinical heterogeneity. Subgroup analysis demonstrated that high (＞ 10 g/d) doses of AA administration was related with a lower 28-day mortality rate (OR = 0.50; 95% CI (0.27-0.92); p = 0.025; I2 = 0.0%, Fig. 3B), however, the medium dose (3-10 g/d) of AA had no influence (OR = 0.92; 95% CI (0.59-1.42); p = 0.698; I2 = 34.9%, Fig. 3B). Long-term course (≥ 3 d) of AA administration could not reduce the 28-day mortality of sepsis patients (OR = 0.75; 95% CI (0.48-1.18); p = 0.215; I2 = 47.9%, Fig. 3C). Initiation of treatment within 6 hours may associated with a lower 28-day mortality rate (OR = 0.49; 95% CI (0.27-0.89); p = 0.019; I2 = 0.0%, Fig. 3D). Additional subgroup analysis suggested that IV AA alone could decrease the 28-day mortality rate compared with co-interventions (OR = 0.36; 95% CI (0.15-0.86); p = 0.020; I2 = 28.6% vs. OR = 1.10; 95% CI (0.79-1.54); p = 0.572; I2 = 0.0%, Fig. 3A).
Upon AA application, no decreased was found over ICU or in-hospital mortality (OR = 0.95; 95% CI (0.66-1.36); p = 0.780; I2 = 0.0% vs. OR = 1.11; 95% CI (0.79-1.56); p = 0.550; I2 = 0.0%, Additional file 3: Fig. S2A).
Length of ICU stay
Three trials [21,29,30] included compared the length of ICU stays between AA and control groups. The results suggested that AA administration did not shorten the length of ICU stay in sepsis patients (WMD = -1.52; 95% CI (-5.06-2.03); p = 0.401; I2 = 79.0%, Additional file 3: Fig. S2B).
Duration of vasopressor requirement
Our analysis among trials [21,23,29,30] suggested that AA administration may decline the duration of vasopressor requirement (WMD = -17.41; 95% CI (-31.16- -3.65); p = 0.013; I2 = 89.4%, Additional file 3: Fig. S2B).
Change in SOFA score
Four trials [19,21,23,27] were included to assess the influence of the change in SOFA score on which AA had no impact (WMD = 0.73; 95% CI (-0.28-1.75); P = 0.158; I2 = 60.2%, Additional file 3: Fig. S2B).
The incidence of AKI and RRT
From analysis among trials [20-22], AA intervention did not increase the incidence of AKI (OR = 1.28; 95% CI (0.80-2.05); p = 0.303; I2 = 0.0%, Additional file 3: Fig. S2A).
No difference was observed in the incidence of RRT among trials [20,21,29] when applying AA (OR = 0.78; 95% CI (0.35-1.75); p = 0.548; I2 = 39.0%, Additional file 3: Fig. S2A).
Publication bias and sensitivity analysis
A begger test was employed to assess the publication bias of the included trials, and the result suggested a minimal publication bias (p = 0.174).
Sensitivity analysis was performed through the removal of each individual trial and reanalysis the remaining trials. When excluding the trial performed by Wani , the impact of AA on change of SOFA score altered (WMD = 1.19; 95% CI (0.49- 1.89); p = 0.001; I2 = 0.0%).
Overall certainty of evidence
The evidence for 28-day mortality was downgraded to moderate certainty, due to study limitations (including high or unclear risk of bias over studies, clinical heterogeneity), and serious concerns about impression (including TSA results) (Table 2). Future evidence is likely to change the estimated effect.
Trial sequential analysis
TSA were performed based on included RCTs using a random-effects model, and were constructed for a heterogeneity adjusted information size of 2589 patients corresponding to a relative risk reduction of 20% with an α = 0.05 and β = 0.20 (power 80%). TSA indicated lack of solid evidence for a beneficial effect on AA for mortality and further RCTs are needed to testify the effects (Fig. 4).