Rebound Inverts the Staphylococcus aureus Bacteremia Prevention Effect of Antibiotic Based Decontamination Interventions in ICU Cohorts with Prolonged Length of Stay

Abstract

Could rebound explain the paradoxical lack of prevention effect against Staphylococcus aureus blood stream infections (BSIs) with antibiotic-based decontamination intervention (BDI) methods among studies of ICU patients within the literature? Two meta-regression models were applied, each versus the group mean length of stay (LOS). Firstly, the prevention effects against S. aureus BSI [and S. aureus VAP] among 136 studies of antibiotic-BDI versus other interventions were analyzed. Secondly, the S. aureus BSI [and S. aureus VAP] incidence in 268 control and intervention cohorts from studies of antibiotic-BDI versus that among 165 observational cohorts as a benchmark was modelled. In model one, the meta-regression line versus group mean LOS crossed the null, with the antibiotic-BDI prevention effect against S. aureus BSI at mean LOS day 7 (OR 0.45; 0.30 to 0.68) inverted at mean LOS day 20 (OR 1.7; 1.1 to 2.6). In model two, the meta-regression line versus group mean LOS crossed the benchmark line, and the predicted S. aureus BSI incidence for antibiotic-BDI groups was 0.47; 0.09-0.84 percentage points below versus 3.0; 0.12-5.9 above the benchmark in studies with 7 versus 20 days mean LOS, respectively. Rebound within the intervention groups attenuated and inverted the prevention effect of antibiotic-BDI against S. aureus VAP and BSI, respectively. This explains the paradoxical findings.

Keywords: Staphylococcus aureus; antibiotic-based decontamination; bacteremia; intensive care unit; meta-regression; rebound; selective digestive decontamination; spill-over; structural equation model.

Figure 1

Meta-regression (red line with shaded 95% confidence intervals) of effect size (as log odds ratio, with individual studies appearing as blue circles with size proportional to the inverse variance) of antiseptic-BDI (a) and antibiotic-BDI (b) in preventing S. aureus BSI versus group mean LOS (logarithmic scale). Meta-regression plots for non-decontamination category of interventions in preventing S. aureus BSI (Figure S5) and all three intervention categories in preventing S. aureus VAP are presented in the online supplement as figures (Figures S4, S6 and S8) and as a summary table (Table S5).

Figure 2

Meta-regression of S. aureus VAP (○, broken regression line) and BSI (▲, solid regression line) incidence among groups of ICU patients within studies of various prevention interventions versus group mean LOS (logarithmic scale). The dotted horizontal lines at 1 and 3 percent are the day 7 and day 20 intercepts for the S. aureus BSI (as derived in Figure S11) and the dotted horizontal lines at 3 and 7.6 percent are the day 7 and day 20 intercepts for the S. aureus VAP (as derived in Figure S10), each as derived in the meta-regression model for observational groups. The regression model coefficients are presented in Table S6. The NCC (non-concurrent control) category includes control groups that, being non-concurrent to antibiotic-BDI or antiseptic-BDI groups, were not exposed to spillover effects from these interventions. Note that the y-axis is a logit scale and the x-axis is the group mean LOS transformed to a logarithmic scale after dividing by 7 such that the model intercept values correspond to group mean day 7 estimates as derived in the meta-regression models.

Figure 3

GSEM of Staphylococcus colonization. Staphylococcus col (oval) is a latent variable representing Staphylococcus colonization. The variables in rectangles are binary predictor variables representing the group-level exposure to the following: trauma ICU setting (trauma50), mean or median length of ICU stay transformed by dividing by 7 and logarithmic transformation (lnLOS7), exposure to a topical antiseptic-BDI (a_S), exposure to an antibiotic-BDI (tap being topical antibiotic prophylaxis as a component of antibiotic-BDI), exposure to a non-decontamination-based prevention method (non-D), exposure to protocolized parenteral antibiotic prophylaxis (ppap) and more than 90% of the cohort receiving mechanical ventilation (mvp90). In this model, the effect of spillover equates to the effect of concurrency of a control group with an antiseptic or antibiotic-based BDI intervention group (CC), and the effect of rebound equates to the interaction term between LOS and exposure to antibiotic- (1.tap#c.lnlos7) or antiseptic-BDI (1.a_S#c.lnlos7). Note that the model factorizes exposures from compound regimens (e.g., SDD and SOD, which combine topical antibiotic prophylaxis [TAP] with or without PPAP as an antibiotic-BDI) into singleton TAP and PPAP exposures. The circle represents the model error term. The three-part boxes represent the binomial proportion data for Staphylococcus VAP (v_sr_n) and BSI (b_sr_n) counts with the number of patients as the denominator which is logit transformed using the logit link function in the generalized model.