The search process in four databases generated 9626 articles (PubMed 3964, Embase 2622, Web of science, 2678, Cochrane of library 362), of which 120 potential articles were identified. After 68 records excluded screening full texts, 52 were considered eligible for the analysis at last[13-64]. The study selection process is presented in Figure 1. Reasons for exclusion were showed in Additional File (eTable2).
The characteristics of the eligible studies are presented in Table 1 and Additional File (eTable3). Five studies were RCTs[25, 45, 46, 49, 50], 47 were NRSIs[13-24, 26-44, 47, 48, 51-64]. Thirty-two studies reported infection due to CRE[13-15, 17-21, 23, 24, 26, 28-31, 34-37, 39, 43, 44, 52-58, 60, 61, 63]. Fifty studies reported mortality[13-16, 18-28, 30-64], and sixteen studies reported clinical response[13, 16, 17, 22, 29, 30, 38, 41, 45-47, 50, 51, 53, 55, 61], and fifteen studies reported microbiological response[13, 20, 25, 29, 30, 38, 41, 45-51, 61]. Twenty-eight studies evaluated patients with only BSI[13, 14, 16, 19, 21, 22, 26-28, 33, 34, 36, 37, 39, 40, 42-44, 48, 52, 54, 55, 57-61, 63], whereas five studies patients with HAP/VAP[18, 23, 45, 46, 51], one study patients with IAI[64], two studies patients with UTIs[17, 20], two studies patients with pneumonia[47, 62], one study not reported[41] and the remaining thirteen studied patients with several types of infections[15, 24, 25, 29-32, 35, 38, 49, 50, 53, 56]. Twenty-three studies focused on a particular organism, nineteen with K. pneumoniae[13, 14, 21, 23, 28, 31, 34-37, 43, 44, 52, 55-58, 61, 63], fourteen with A. baumannii[16, 25, 32, 38, 40-42, 45-47, 49, 51, 62, 64]. The remainder included multiple species of Gram-negative bacilli[15, 17-20, 22, 24, 26, 27, 29, 30, 33, 39, 48, 50, 53, 54, 59, 60]. Among eight studies all the participants were admitted to ICU[22, 23, 46, 48, 51, 52, 59, 62].
22 studies for treatment of Gram-negative bacteremia were definite, of which 14 compared colistin monotherapy with colistin-based combination therapy[16, 23, 25, 33, 38, 41, 42, 45-47, 49-51, 59], 4 compared ceftazidime-avibactam with ceftazidime-avibactam-based combination therapy[30, 53, 55, 56], one for aminoglycosides[63], one for fosfomycin[17], one for cefoperazone/sulbactam[40], and one for tigecycline[64]. Several different antimicrobial classes were used in the included studies for treatment of Gram-negative bacteremia; however, the detail provided in each study regarding the specific antimicrobial used varied greatly[64].
Mortality
50 studies reported all mortality, of which ten reported 14-day mortality[16, 19, 22, 34, 42, 47, 50, 58, 63, 64], eleven reported 28-day mortality[14, 22, 25, 33, 40, 46, 50-52, 56, 57], twenty reported 30-day mortality[13, 15, 18, 20, 26, 28, 31, 32, 36, 37, 39, 41-44, 49, 55, 59-61, 63, 64], one reported 90-day mortality[53], seven reported infection-related mortality[13, 15, 22, 24, 25, 45, 49], and ten reported in-hospital mortality[13, 19, 23, 27, 30, 35, 45, 47, 48, 64].
Monotherapy groups had a higher mortality than monotherapy for treating CRGNB (a total of 6524 patients, OR=1.38, 95%CI=1.17-1.64, I2=50%) (Figure2). Funnel plot analysis showed no asymmetry
(Additional File.eFigure2). Publication bias was not detected.
Subgroup analyses found that 30-day overall mortality was significantly higher among trials with monotherapy (a total of 2767 patients, OR=1.60, 95%CI=1.23-2.67). The infection-related (866 patients, OR=1.20,95%CI=0.69-2.07), in-hospital (902 patients, OR=1.14,95%CI=0.73-1.78), 14-day (1790 patients, OR=1.25,95%CI=0.95-1.65), 28-day (1364 patients, OR=1.27,95%CI=0.83-1.96) all-cause mortality rate increased compared with combination therapy, although there was no significant difference (Additional File.eFigure3). Monotherapy was associated with significantly more mortality in 45 NRSIs pooling a total of 5688 patients (OR=1.41,95%CI=1.17-1.70). Five RCTs enrollled 719 patients more mortality on monotherapy but no statistically significantly difference (OR=1.06, 95%CI=0.80-1.41) (Additional File.eFigure4). Monotherapy was more mortality among the 753 patients admitted to ICU in eight studies, although there was no significant difference (OR=1.30,95%CI=0.80-2.10) (Additional File.eFigure5).
Similarly, monotherapy groups had a significantly higher mortality than monotherapy for treating CRE (a total of 3765 patients, OR=1.58, 95%CI=1.22-2.06) (Figure3).
Two subgroup analyses regarding organism and type of infection were performed. Patients with Infection due to CRKP (2398 patients, OR=1.86,95%CI=1.38-2.51) and CRAB (1779 patients, OR=1.27.95%CI=1.04-1.55) who received monotherapy had higher mortality than those receiving combination therapy (Additional File.eFigure6). Type of infection from included studies were mainly BSI, HAP/VAP, IAIs, UTIs, pneumonia. The remaining studies were several types of infection. Monotherapy showed higher mortality for BSI (3644 patients, OR=1.73,95%CI=1.39-2.15) and HAP/VAP (257 patients, OR=1.89,95%CI=1.08-3.30). On the contrary, there was no statistically significantly difference between monotherapy and combination therapy for IAIs, UTIs, pneumonia, several types of infection (Figure 4).
Specific treatments we found were colistin-based and CAZ-AVI-based regimens. Patients who received colistin had higher mortality than receiving colistin-based combination (1954 patients, OR=1.23,95%CI=1.02-1.47). However, Patients who received CAZ-AVI had lower mortality than receiving CAZ-AVI -based combination (244 patients, OR=0.51, 95%CI=0.27-0.95) (Additional file. eFigure7).
Clinical success
16 studies consisting of 2001 patients showed that monotherapy was associated with lower clinical success for treating CRGNB(OR=0.63,95%CI=0.45-0.89) (Additional File.eFigure8). Likewise, monotherapy was associated with lower clinical success for treating CRE (485 patients, OR=0.36,95%CI=0.24-0.53) (Additional File.eFigure9). Monotherapy was associated with lower clinical success for treating CRKP (343 patients, OR=0.29,95%CI=0.18-0.48). Clinical success showed a decreasing trend in monotherapy for treating CRAB, but there was no statistically significantly difference(896 patients, OR=0.94,95%CI=0.66-1.34)(Additional File.eFigure10). Patients who received colistin in 8 studies seemed to have a lower clinical success than colistin-based combination, although there was no significantly difference (1341 patients, OR=0.77,95%CI=0.51-1.16). The same situation was observed for CAZ-AVI-based regimens compared with carbapenem or sulbactam regimens (97 patients, OR=0.59, 95% CI=0.25-1.38).
Microbiological eradication
Fifteen studies, including 2164 patients reported microbiological eradication. Monotherapy was lower microbiological eradication than combination therapy for treating CRGNB(OR=0.60,95%CI=0.43-0.84) (Additional File.eFigure11). Subgroup analyses occurred different outcomes. Monotherapy was lower microbiological eradication in NRSIs (1380 patients, OR=0.62,95%CI=0.42-0.92), but no statistically significantly difference in RCTs (784 patients, OR=0.51,95%CI=0.24-1.09).
Five studies, in which 514 patients reported microbiological eradication for treating CRE, suggested that patients with monotherapy had a significantly lower mortality than those with combination therapy (OR=0.36,95%CI=0.18-0.72) (Additional File.eFigure12).
Microbiological eradication showed a decreasing trend in monotherapy for treating CRKP (343 patients, OR=0.25,95%CI=0.12-0.52) and CRAB (1160 patients, OR=0.65,95%CI=0.44-0.96). (Additional File.eFigure13).
Patients receiving colistin seemed lower microbiological eradication than receiving colistin-based combination therapy, although it was no significantly difference (1268 patients, OR=0.73,95%CI=0.47-1.16). Only one studies reported microbiological eradication (60 patients, OR=1.18,95%CI=0.41-3.41).
Publication Bias
Funnel plot analysis showed no asymmetry (Additional File.eFigure2). Publication bias was not detected. Substantial statistical heterogeneity was present in the main analyses (I2>50%, p<0.01). We were not able to detect publication bias, as tested using the Egger method (P=0.486, 95%CI=-0.625-1.296). The shape of the funnel plot was symmetrical. We assumed that the source of bias is not related to publication, but is related to the severity of the patient's disease, underlying disease, type of infection, etc.