Introduction
Of three broad categories of infection prevention in the ICU patient group, selective oral decontamination/selective digestive decontamination (SOD/SDD) shows superior apparent benefit towards overall infection prevention within the ICU context versus anti-septic-based and non-decontamination-based prevention methods [
1‐
9].
The control of gut overgrowth (COGO) is one mechanism proposed to explain how SOD/SDD regimens might prevent ICU-acquired infection. In general, the antibiotics constituent within SOD/SDD regimens, such as topical polymyxin and aminoglycosides, specifically target Gram-negative bacilli including
Pseudomonas and
Acinetobacter bacteria whereas anti-septic- and non-decontamination-based prevention methods do not [
10].
The exact mechanism for how each of these methods prevents ICU-acquired infection, the basis for the apparent superiority of SOD/SDD among these methods, and even the optimal locus for decontamination, whether the gut or elsewhere, remains unclear despite > 200 studies among patients requiring prolonged mechanical ventilation (MV) or ICU stay [
11]. Moreover, the relative importance of the individual SDD components, topical (TAP), enteral (EAP), and protocolized parenteral antibiotic prophylaxis (PPAP; not contained within SOD regimens), versus other methods of infection prevention and versus other contextual exposures such as length of stay and being in a trauma ICU context remains unclear. In addition, concurrency, being the concurrent mixing of study and control patients within the ICU, as typically occurs within randomized concurrent control studies, is believed to influence the results of SOD/SDD studies versus studies without concurrency (i.e., concurrent versus non-concurrent control; CC versus NCC) [
10,
12].
The objectives here are threefold. Firstly, to recapitulate the evidence for overall ventilator-associated pneumonia (VAP) and bacteremia prevention among the three broad categories of infection prevention for which Pseudomonas and Acinetobacter infection data is available. Secondly, to develop and confront candidate models founded on COGO concepts using Pseudomonas and Acinetobacter infection data from these studies as well as studies without an intervention using GSEM modeling. Thirdly, to compare the relative impacts of the various group-level exposures and interventions on Pseudomonas and Acinetobacter gut overgrowth as latent variables within the optimal GSEM model.
Materials and methods
Being an analysis of published work, ethics committee review of this study was not required.
Study selection and decant of groups
The literature search and study decant used here (Fig
S1; see Electronic Supplementary Material for additional ESM tables, ESM figures, and ESM references) is in six steps which is described in full in the ESM and as described previously [
13].
Of note, studies undertaken in the context of an ICU outbreak [
14‐
16] were excluded. Due to the absence of eligible studies of TAP undertaken in Asia and Central and South America, together with the significant worldwide variation in both
Pseudomonas [
17] and moreso
Acinetobacter-associated VAP [
18], studies from these regions were excluded from this analysis. A snowballing search strategy [
19] using the “Related articles” function within Google Scholar was undertaken for additional studies not identified within systematic reviews.
All eligible studies were then collated, and any duplicate studies were removed and streamed into groups of patients from studies with or without an infection prevention intervention. Those studies without a study intervention provide observational groups.
The component groups were decanted from each study as either observational, control, or intervention groups. Within studies of TAP, any group receiving TAP in any formulation was regarded as an intervention group and all other groups were regarded as a control group regardless of other interventions. The control groups from studies of TAP were stratified into NCC and CC groups.
Outcomes of interest
The incidences of overall Pseudomonas and Acinetobacter VAP as well as the incidences of overall Pseudomonas and Acinetobacter bacteremia were extracted. These were each expressed as a proportion using the number of patients with prolonged (> 24 h) stay in the ICU as the denominator. Pseudomonas and Acinetobacter gut overgrowth are latent variables as defined within the GSEM models (see below).
Exposures of interest
The following were also extracted where available: the proportion of each group receiving MV, the proportion of admissions for trauma, and the mean length of ICU stay (LOS). An anti-septic exposure included agents such as chlorhexidine, povidone-iodine, and iseganan. All anti-septic exposures were included regardless of whether the application was to the oropharynx, by tooth-brushing or by body wash.
TAP is defined here as the application of topical antibiotic (TA) prophylaxis to the oropharynx without regard to the specific TA constituents nor to concomitant EAP, being the enteral applications of TA, or PPAP. Note that SOD generally consists of only TAP whereas SDD typically involves TAP together with both EAP and PPAP. A control group of an SOD/SDD study was classified as a CC control if the group was concurrent within the same ICU at the same time as intervention group patients were receiving TAP.
Visual benchmarking
Scatter plots of the overall and
Pseudomonas and
Acinetobacter VAP and bacteremia incidence data were generated to facilitate a visual survey of the entire data as derived from the literature. To facilitate this visual survey, a benchmark for each outcome of interest was generated from the groups of the observational studies using the “metan” command as described in the ESM. The caterpillar plots [
20] illustrating the derivation of each bacteremia benchmark are shown in the supplementary material.
Structural equation modeling
Seven candidate GSEM models were developed using Pseudomonas and Acinetobacter gut overgrowth as the central latent variables. Group exposure or not to the following factors served as binary indicator variables towards these two latent variables: non-decontamination-based prevention methods, anti-septic-based prevention methods, TAP-based prevention methods, exposure to PPAP, membership of a CC control group within a TAP intervention study, whether the majority of the group were trauma patients, whether more than 90% of patients of the group received more than 24 h of MV, and whether the mean (or median) length of ICU stay for the group was 7 days or more.
The VAP and bacteremia count data for each of
Pseudomonas and
Acinetobacter using the number of observed patients as the denominator served as the measurement component for the latent variables using a logit link function in each GSEM. In each model, the observations were clustered by a study identifier in order to generate a robust variance covariance matrix of the parameters of each coefficient estimate. The various exogenous variables were entered into each model without any preselection step to sequentially develop the seven candidate GSEM models using the “GSEM” command in Stata [
21]. The model with the lowest Akaike’s information criterion (AIC) score was selected as having parsimony and optimal fit from among the seven candidate models.
Availability of data and materials
All data generated or analyzed during this study are included in this published article and its supplementary information files (see ESM).
Discussion
Generally accepted risk factors towards the acquisition of Gram-negative bacilli in the ICU include LOS > 7 days, exposure to invasive devices such as MV, and exposure to antibiotics together with acquisition by cross infection within the ICU environment [
10]. A GSEM model founded on COGO concepts is used to evaluate these risk factors versus other group-level exposures. This GSEM model enables the component groups of studies of the various infection prevention methods to be considered as a natural experiment with various group-wide exposures among over two hundred ICU populations in the literature. This enables a novel perspective on the COGO concept that would not be possible within any one study examined in isolation nor within several studies examined collectively as within a systematic review [
22].
The data used here to confront the COGO model is drawn mostly from studies located in systematic reviews. The extracted data is provided in sufficient detail in the ESM to enable replication of the analysis. In this regard, the summary effect sizes here for each of the three broad categories of TAP, anti-septic, and non-decontamination methods, against both overall VAP and against overall bacteremia, are similar to prior published estimates [
1‐
10]. As has previously been noted, TAP (moreso when in combination with PPAP [
23]) appears to have the strongest prevention effect against both overall VAP and against overall bacteremia.
In confronting the COGO model with the Pseudomonas and Acinetobacter infection data, the COGO model is robust with several factors remaining consistent over the evolution through seven candidate versions of the GSEM. There are several expected observations. Length of stay and admission to a trauma ICU are strong positive factors, and non-decontamination interventions appear not to mediate significant effects on either Pseudomonas gut overgrowth or Acinetobacter gut overgrowth. TAP exposure is associated with a negative coefficient towards Pseudomonas gut overgrowth, albeit weaker than that associated with anti-septic interventions. These negative coefficients in association with TAP and anti-septic exposures towards Pseudomonas gut overgrowth reflect the generally lower Pseudomonas VAP among the intervention groups of these studies.
On the other hand, the various components of the SOD/SDD regimens, TAP, EAP, and PPAP, have mixed effects within the GSEM models. Neither TAP nor EAP has negative coefficients towards Acinetobacter gut overgrowth. This is surprising as in nearly all instances these contain polymyxin and/or an aminoglycoside. Moreover, PPAP is associated with a strong positive correlation with Pseudomonas bacteremia.
Finally, patient groups exposed to the full SDD regimen (i.e., all of TAP, EAP, and PPAP) have
Pseudomonas and
Acinetobacter bacteremia incidences that are either higher than or else not lower than patient groups receiving TAP alone. This is possibly not paradoxical as antibiotics used for PPAP typically lack activity against
Pseudomonas and
Acinetobacter. In this regard, the cumulative days of exposure to antibiotics without activity against
Pseudomonas have been reported as being a risk factor for acquiring
P. aeruginosa and
Acinetobacter in the ICU [
24‐
26]. Moreover, concomitant systemic antibiotic therapy fails to prevent the acquisition of respiratory tract colonization with Gram-negative bacteria [
27] and more than triples the risk of subsequent infection among ICU patients receiving an enteral decolonization regimen with gentamicin against KPC-producing
Klebsiella pneumonia [
28] and CRE-producing
Acinetobacter [
29].
The exact relationship between gut colonization, PPAP use, and subsequent bacteremia remains controversial amid conflicting reports that PPAP use may or may not be important for some Gram-negative bacteremias versus others [
30‐
33]. In studying the relative prevention effects of SDD versus SOD each versus standard care in the prevention of Gram-negative bacteremias (i.e., not limited to
Pseudomonas bacteremia), the majority of bacteremias occur after 4 days in the ICU (the typical duration of PPAP) and indeed the daily risk peaks after day 30 [
11,
31]. Moreover, among patients receiving SDD or SOD,
Pseudomonas accounts for one third of GN bacteremia episodes with most episodes not preceded by enteral colonization.
Defining the separate effects of EAP, TAP, and PPAP on the
Acinetobacter and
Pseudomonas bacteremia incidences is difficult as these exposures are confounded with each other among the multiple SDD/SOD regimens under investigation in the different studies. Also, the duration of the application of the regimens and the duration of follow up varied among the studies. In this regard, a non-significant increase in hospital-acquired infections post discharge from the ICU as great as 50% was noted in a small SDD sub-study [
34].
In critical care research, SEM is emerging as a method to model the relationships among multiple simultaneously observed variables in order to provide a quantitative test of any theoretical model proposed within the literature [
35]. The use of latent variables within the model enables the ability to test the validity of concepts that can only be indirectly quantified through their inferred relationship to observed variables [
36]. GSEM allows generalized linear response functions in addition to the linear response functions allowed by SEM.
Limitations
There are five key limitations to this analysis, the first being that this analysis is a group-level modeling of two latent variables, Pseudomonas gut overgrowth and Acinetobacter gut overgrowth, within a GSEM founded on the COGO construct. These latent variables and the coefficients derived in the GSEM are indicative and intended for internal reference only. They have no counterpart at the level of any one patient or study and cannot be directly measured. There was no ability nor purpose to adjust for the underlying patient-level risk. There was considerable heterogeneity in the interventions, populations, and study designs among the studies here as the inclusion criteria for the various studies have been intentionally broadly specified. In this regard, a strength of the analysis is that the heterogeneity among the studies here generally resembles that expected among ICU populations to which these interventions might be targeted.
The second limitation is that the analysis is inherently observational. Only a limited number of key group-level factors were entered into the GSEM models. Moreover, the GSEM modeling is deliberately simplistic with exposures entered as only binary variables and without the use of interaction terms. In reality, the relationships between exposures and outcomes will likely be complex and exposure interactions could have great importance.
Thirdly, the analysis is likely underpowered to examine the Acinetobacter infection data, being a relatively rare end point. Likewise, the incidences of resistant infections with Acinetobacter and Pseudomonas are of great interest. However, examination of the incidence of these resistant infections is difficult as these end points are generally uncommon or rare and have been inconsistently reported among these studies here.
Fourthly, only those studies for which
Pseudomonas and
Acinetobacter infection data were available were able to be included in this analysis. However, the effect of the interventions on overall VAP and bacteremia incidences among the studies included here (Fig
S2-S7) resembles that in the broader literature.
Finally, it should be noted that the various interventions among the studies here targeted a range of sites which may or may not have included the oropharynx and gastrointestinal tract. In this regard, it is surprising that the TAP and EAP interventions, which most directly target the oropharynx and gastrointestinal tract, had weaker effects than did anti-septic interventions, several of which, such as chlorhexidine body washes, target other sites.
Can the paradoxical findings of the GSEM model be reconciled with the apparent superior summary prevention effects of TAP against VAP and bacteremia? TAP exposure and control group concurrency have associations with
Pseudomonas gut overgrowth that are each similar in size but contrary in direction to each other. In this regard, the incidences of overall VAP, overall bacteremia and also mortality [
37] among the concurrent control groups within studies of SOD/SDD are as much as ten percentage points higher than the repsective incidences of these end points among the control groups within studies of equivalent ICU populations. This higher overall VAP incidence can partly be accounted for by incidences of VAP with specific bacteria such as
Acinetobacter [
38],
Pseudomonas [
39], and
Staphylococcus aureus [
40] being each 3 to 5 percentage points higher among CC (but not NCC) control groups and these incidences are generally each up to 2 percentage points higher for the intervention groups of SOD/SDD studies
.
Likewise, the higher overall bacteremia incidence can partly be accounted for by noting that the incidences of bacteremia with specific bacteria are generally 1 to 4 percentage points higher among CC (but not NCC) control groups. Even among intervention groups, these bacteremia incidences may on average be up to 3 percentage points higher for
Acinetobacter (Fig.
2),
Pseudomonas (Fig.
2) [
41],
Staphylococcus aureus [
42],
Enterococci [
43], and
coagulase-negative Staphylococci [
44]
.
In each case, the increased incidence within control groups of CC design studies of topical antibiotics remains apparent in meta-regression models adjusting for other recognized associations. The influence of topical placebo use, concurrent colonization with
Candida, and other influences may also have influences in this process [
45‐
47].
Hence, reconciling the findings of the GSEM model founded on COGO concepts on the one hand, with the apparent superior summary prevention effects of TAP against VAP and bacteremia, on the other, is possible by noting that the incidences of VAP and bacteremia are generally higher among CC (but not NCC) control groups of studies of TAP. These higher incidences within CC (but not NCC) control groups of studies of TAP remain to be explained.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.