Selective decontamination of the digestive tract (SDD) consists of the application of topical, non-absorbable antibiotics in the mouth and digestive tract, most often accompanied by a short intravenous course of antibiotics without anaerobic spectrum, with the aim to eradicate potentially pathogenic bacteria (mostly Staphylococcus aureus and Gram-negative aerobic rods) and yeasts from the gut while preserving the resident anaerobic flora. The debate whether SDD should be universally adopted to prevent nosocomial infections and improve outcome of patients receiving invasive mechanical ventilation in the intensive care unit (ICU) has been ongoing for over 3 decades, as this intervention could theoretically be associated with increased antimicrobial resistance.

The two most recent important contributions to this field again produced conflicting results. The SuDDICU trial, a cluster, crossover, randomized clinical trial (RCT) carried out in the low antibiotic resistance setting of Australia and New Zealand (including 5982 adult ICU patients), failed to demonstrate a significant mortality benefit of SDD [1]. A systematic review and meta-analysis incorporating this trial as well as 29 previous RCTs (including 24,034 adult ICU patients) and using Bayesian analysis concluded for a 99.3% posterior probability that SDD reduced hospital mortality [2]. The beneficial association of SDD was evident in trials with an intravenous agent, but not in other trials. SDD was associated with reduced risk of ventilator-associated pneumonia (VAP) and ICU-acquired bacteremia. Of note, evidence regarding the effect of SDD on antimicrobial resistance was of very low certainty. The effectiveness of SDD in settings with a higher prevalence of multidrug antibiotic resistance and in specific patient populations remain uncertain.

In a post hoc explanatory analysis of the SuDDICU trial published in this issue of Intensive Care Medicine [3], the authors dichotomized the population in patients with versus without acute brain injuries (i.e., traumatic brain injury, stroke, intracerebral bleeding, cardiac arrest, and brain infection) or conditions (i.e., seizures or coma). In patients with acute brain injuries or conditions, the primary endpoint of in-hospital death within 90 days was significantly reduced by 5.7% in the SDD group, as compared to patients receiving standard of care (unadjusted odds ratio [OR] 0.76, 95% confidence interval [CI] 0.63 to 0.92). Consistency was added by the same observation of benefit for the four secondary endpoints: ICU mortality, days alive and free of mechanical ventilation, ICU, and hospitalization through 90 days. Of note, in patients without acute brain injuries or conditions, who made up around two-thirds of the SuDDICU trial population, no significant differences in any clinical outcome between SDD and standard care groups were observed.

This secondary analysis of the SuDDICU data trial indicates a heterogeneity of treatment effect associated with SDD. It identifies a subgroup of critically ill patients that may derive greater benefit from the intervention. Nevertheless, it is essential to acknowledge that this proposition is currently a hypothesis, formulated solely through the retrospective nature of the analysis. RCTs are considered the gold standard for assessing intervention effects, ensuring robust internal validity when adequately powered, especially in relation to the primary endpoint.

However, conducting RCTs for every clinical query or specific population is often impractical. Consequently, post hoc analyses are frequently employed. Since they increase, however, the risk of false discoveries with every post hoc test done, multiple comparison corrections are needed. It is crucial to recognize that each post hoc test increases the risk of false discoveries, necessitating correction for multiple comparisons. Unfortunately, determining the number of tests conducted is often challenging in many post hoc analyses intended for publication.

Despite these inherent challenges, post hoc analyses, included the one conducted by Young et al. [3], play a valuable role in generating new hypotheses and elucidating effect sizes. RCTs often have stringent participant eligibility criteria, limiting generalizability of findings to broader or target populations. Post hoc analyses offer readily accessible data, facilitating analyses for the exploration of specific hypotheses. Several methodologies, including matching, Inverse Probability Weighting (IPW), and doubly robust methods, have been developed to estimate causal effects from observational and post hoc data [4]. Machine learning approaches can complement post hoc analyses, enhancing the generalizability of results.

The chance of a false-positive result in a post hoc analysis diminishes when there is a strong theoretical or empirical basis for the expected effect a priori. In this context, exploring why SDD might be particularly beneficial in patients with acute brain injuries or conditions is relevant. According to the authors, the beneficial effect of SDD in this population might stem from preventing secondary neurological damage caused by infections. The conventional view of SDD suggests that it may primarily work by preventing VAP, a common issue in neurocritical patients due to the risk of aspiration, prolonged mechanical ventilation, and neuro-immune suppression. It may also reduce pathogen translocation, as indicated by lower bacteremia rates observed with the intervention [3]. However, among patients with acute brain injury, the impact of VAP on outcomes remains debated. Recent studies in traumatic brain injury patients and broader neurocritical care populations have shown that VAP is associated with extended duration of mechanical ventilation and ICU stay, but not with mortality [5, 6]. In another study conducted in ischemic stroke patients, pneumonia was independently associated with increased mortality [7]. Previous single-center trials conducted in brain-injured patients suggested that antibiotic prophylaxis could reduce early onset pneumonia [8, 9]. Data on multicenter RCTs evaluating prophylactic antibiotics in brain-injured patients are presented in Table 1. Specifically, among patients resuscitated from cardiac arrest, prophylactic antibiotics have been shown to reduce the incidence of early onset VAP, without affecting duration of mechanical ventilation, ICU stay, or neurologic outcomes [10].

Table 1 Multicenter trials conducted in brain-injured patients investigating the effect of antibiotic prophylaxis on infections and neurologic outcomes
Full size table

Conducting a patient-centered meta-analysis may offer valuable insights, although it may not yield conclusive evidence. The essential step of identifying patients who stand to gain the most from SDD is crucial for sha** future trials. These trials should ideally be conducted in settings characterized by a moderate-to-high prevalence of antimicrobial resistance. The focus of such trials should encompass patients with diverse forms of acute brain injuries or conditions, incorporating robust a priori defined subgroup analyses to precisely identify those who may benefit most from SDD. Prioritizing patient-centered outcomes, such as functional recovery, cognitive function, epilepsy, and long-term neuropsychological consequences, is imperative for a comprehensive assessment of the intervention's impact. Furthermore, documenting decisions regarding life-support limitations, which are linked to hospital mortality, should be meticulously undertaken [11]. Considering that brain-injured patients are often younger and have fewer comorbidities, their baseline risk for multidrug antibiotic resistance and potential harm from additional antibiotic selection might be lower, thus limiting the risk of harm. As such, it may be hoped that SDD will find its right place as an effective treatment for ICU patients.