Results

Included studies

In total, 166 articles were initially retrieved, and 144 articles were excluded due to duplication and were irrelevant based on title and abstract screening. After careful screening of the full texts of the 27 trials, 17 articles were excluded for following reasons: article was commentary (n=1) (Schott and Huang, 2012), study was the part of another study (n=1) (Pacht et al., 2003), studies did not have defined outcomes (n=2) (Nelson et al., 2003, Theilla et al., 2007). Full text was not available (n=1) (Tang et al., 2008), studies without sufficient data (n=1) (Elamin et al., 2005), studies conducted in children (n=4) (Al-Biltagi et al., 2017, Covar et al., 2010, Hamilton and Trobaugh, 2011, Jacobs et al., 2013), articles were not RCT (n=4) (Cohen et al., 2013, Gadek et al., 1998, Lev and Singer, 2012, Pontes-Arruda, 2005), or were not conducted on critically ill patients (n=3) (Kalantar-Zadeh et al., 2005, Remans et al., 2004, Matsuda et al., 2017). In total, 10 clinical trials (1166 participants) were eligible to be included in the present systematic review and meta-analysis. The Figure 1showed the study screening steps.
Three studies were conducted in USA (Elamin et al., 2012, Rice et al., 2011, Gadek et al., 1999), two in Israel (Kagan et al., 2015, Singer et al., 2006), two in Japan (Shirai et al., 2015, Tsukahara et al., 2014), two in Brazil (Pontes-Arruda et al., 2006, Pontes-Arruda et al., 2011) and one in Spain (Grau-Carmona et al., 2011). The duration of the intervention ranged from 7 to 28 days. The mean age of participants ranged from 42.72-72.5 years. All trials were parallel and conducted in both gender. Underlying diseases included sepsis, ARDS, ALI, severe multiple trauma, head and neck cancer with surgery. In all studies, participants in intervention group received immunomodulatory diet contains omega-3 fatty acid, γ-linolenic acid and antioxidants. In 4 studies, participants in the control group received high- fat, low- carbohydrate formula (Elamin et al., 2012, Kagan et al., 2015, Gadek et al., 1999, Singer et al., 2006) and in 6 studies received standard, iso-caloric formula (Grau-Carmona et al., 2011, Pontes-Arruda et al., 2011, Rice et al., 2011, Shirai et al., 2015, Tsukahara et al., 2014, Pontes-Arruda et al., 2006). The characteristics of the studies included in the meta-analysis are summarized in Table 1 .

Assessment of risk of bias

Five studies included in the meta-analysis had a good quality based on Cochrane Collaboration’s tool (Grau-Carmona et al., 2011, Kagan et al., 2015, Pontes-Arruda et al., 2011, Rice et al., 2011, Elamin et al., 2012), four were classified as poor (Pontes-Arruda et al., 2006, Singer et al., 2006, Gadek et al., 1999, Shirai et al., 2015) and one study was fair (Tsukahara et al., 2014). In three studies, method of random sequence generation and method to conceal the allocation of participants were unclear (Gadek et al., 1999, Pontes-Arruda et al., 2006, Shirai et al., 2015). In a study by Kagan et al (Kagan et al., 2015) the method to conceal the allocation of participants was not clear. One study did not report blinding (Tsukahara et al., 2014), two studies were not double blind (Shirai et al., 2015, Singer et al., 2006). Blinding of outcome assessment were not reported for three studies (Pontes-Arruda et al., 2006, Pontes-Arruda et al., 2011, Tsukahara et al., 2014). Two studies did not mention incomplete outcome data (Pontes-Arruda et al., 2006, Gadek et al., 1998) and one had bias in this section (Singer et al., 2006). No studies reported selective outcome bias (Supplementary Table 2 ).

Meta-analysis

3.3.1 Level of oxygenation, duration of mechanical ventilation and ventilator free- days and

Six studies with 709 participants reported the effect of the immunomodulatory formula on PaO2/FiO2. Meta-analyses comparing this immunomodulatory formula vs control formula showed no differences between groups in the effect on PaO2/FiO2 (WMD = 27.40, 95% CI: -7.84, 62.65, P=0.13). The heterogeneity was high (Q statistic=55.44, Cochrane Q test, P<0.001, I2 =91.0%)(Supplementary Figure 1). In patients with ALI and ARDS, enhancement in level of oxygenation was statistically significant and heterogeneity decreased. Both level of oxygenation and heterogeneity decreased in patients aged <60 years. Heterogeneity in subgroups of good quality RCTs and in ≥ 14 days intervention decreased(Supplementary Table 3) .
Seven studies with 667 participants reported the effect of the immunomodulatory formula on duration of mechanical ventilation. This immunomodulatory formula significantly shortened the duration of mechanical ventilation in the intervention group compared to control group (mean difference = -2.20 days, 95% CI: -4.29, -0.10, P=0.04)(Supplementary Figure 2) . The heterogeneity among the included studies was high (Cochrane Q test=25.86, P<0.001, I2 =76.8%). In septic and ARDS patients the immunomodulatory formula significantly reduced duration of mechanical ventilation compared to other subjects. Also, in patients older than 60 years and in the studies that control group consumed the standard formula significant reduction of duration of mechanical ventilation was observed. The reduction of duration of mechanical ventilation in poor quality RCTs was significant. Underlying disease, age and duration of intervention were the source of heterogeneity. (Supplementary Table 4).
Four studies with 629 participants reported the effect of the immunomodulatory formula on ventilator free- days. There was non-significant pooled effects of omega-3 fatty acid, γ-linolenic acid and antioxidant supplementation on ventilator-free days (WMD = 3.37 days, 95% CI: -2.20, 8.85, P=0.24) with high heterogeneity (Q statistic=31.74, Cochrane Q test, P<0.001, I2 =90.5%).

3.3.2. Duration of ICU stays and ICU free days

Eight studies with 901 participants reported the effect of the immunomodulatory formula on duration of ICU stays. The immunomodulatory formula shortened the duration of ICU stays by 2.97 days compared to control group (95% CI: -5.59, -0.35, P=0.02) (Figure 2) , between-study heterogeneity was significantly high (Cochrane Q test=62.55, P<0.001, I2 =88.8%). The reduction in duration of ICU stays was evident in ALI, ARDS and septic patients and heterogeneity decreased in these subgroups. Moreover, the reduced in duration of ICU stays was evident in the studies where control group received standard formula and in patients more than 60 years old. In the subgroup where the duration of the intervention was more ≥ 14 days, heterogeneity decreased and the reduction in ICU stays was statistically significant. Underlying diseases and the quality of studies were sources of heterogeneity (Supplementary Table 5).
Five studies with 735 participants reported the effect of the immunomodulatory formula on ICU- free days. This immunomodulatory diet extended 4.06 ICU- free days in the intervention group than the control group (95%CI: 0.02, 8.09, P=0.05). The heterogeneity was also high (Cochrane Q test=39.63, P<0.001, I2 =89.9%)