Do patients receiving extracorporeal membrane-oxygenation need antibiotic prophylaxis? A systematic review and meta-analysis on 7,996 patients

Abstract

Background: Patients undergoing Extracorporeal Membrane Oxygenation (ECMO) are particularly susceptible to infections: 42% experience sepsis and 26% develop a nosocomial infection (NI). Whether antibiotic prophylaxis is effective in reducing mortality and its effects on the rate of NIs is currently unclear.

Research question: Can antibiotic prophylaxis decrease 30-day mortality for patients on ECMO? Can antibiotic prophylaxis prevent the occurrence of NIs in these patients?

Study design and methods: A systematic review and meta-analysis was conducted. We searched PubMed, Scopus, and CINAHL libraries from inception to June 12, 2024. Two researchers were involved in abstract screening and three researchers were involved in full text selection.

Results: A pooled population of 7,996 patients is represented by 5 retrospective studies. Reported mortality ranges between 46 and 58% and the NIs rate is between 14 and 62%. Regarding 30-day mortality, the random-effects model (I² = 65%) indicates a non-statistically significant difference between the antibiotic prophylaxis group and the non-prophylaxis group (OR 0.76; 95%CI 0.37-1.59). For the NIs rate, a fixed-effect model (I² = 36%) shows an OR of 0.81 (95%CI 0.71-0.92) in favor of the antibiotic prophylaxis group, with a number-needed-to-treat (NNT) of 39.7 patients.

Conclusion: According to a very low degree of certainty, antibiotic prophylaxis appears to have no impact on the 30-day mortality rate of ECMO recipients. The risk of NIs seems to decrease with antibiotic prophylaxis, even though the NNT is high. Prospective high-quality studies that address these specific clinical questions are necessary.

Clinical trial registration: PROSPERO: International prospective register of systematic reviews, 2024, CRD42024567037.

Keywords: Antibiotic prophylaxis; ECMO; Meta-analysis; Mortality; Nosocomial infection.

Fig. 1

PRISMA flowchart. Eighty-one records were found during the initial identification process. Five studies were finally included and analyzed in the systematic review and quantitative synthesis, accounting for a pooled population of 7,996 patients

Fig. 2

Risk of bias in the included studies. For observational studies, the Robins-I was used to assess the risk of bias. The main source of risk of bias was in the definition of intervention (i.e. antibiotic prophylaxis). The two studies considered at risk of critical bias did not use an unambiguous definition, considering any antibiotic administered within the first 48 h of cannulation as an ‘antibiotic prophylaxis’

Fig. 3

Forest plots for 30-day mortality. The random effects model exhibited an OR of 0.76 (95%CI 0.37–1.59). Since the confidence interval crosses the unit, the difference was not statistically significant. We considered only the random-effect model since the high between-study inconsistency found (65%)

Fig. 4

Funnel plot for 30-day mortality. Egger’s test was not statistically significant (p = 0.595)

Fig. 5

Forest plots for nosocomial infections rate. The fixed-effect model showed an OR of 0.81 (95%CI 0.71–0.92) in antibiotic prophylaxis group compared to no prophylaxis group, with a number-need-to-treat (NNT) of 39.7 patients. The between-studies inconsistency was low (I² = 36%), so we only considered the fixed-effect model

Fig. 6

Funnel plot for nosocomial infections rate. Egger’s test was not statistically significant (p = 0.397)

Fig. 7

Multivariate meta-analysis model (SEM). Due to the small sample size, the results were inconclusive due to the wide confidence intervals. The fixed-effect model showed an estimated effect of -0.07 (95%CI -0.12 – -0.02; p = 0.005) for the 30-day mortality outcome; and -0.11 (95%CI -0.18 – -0.04; p = 0.001) for the nosocomial infections rate. The random-effects model showed an estimated effect of 0.03 and -0.003 for the two outcomes, respectively, but this model was non statistically significant (with a significant amount of heterogeneity: I² = 87% for the first outcome and 82% for the second outcome)