Using observational data to emulate a target trial, we here aimed to assess the effect of treatment with low-dose erythromycin on the outcome of critically ill patients with sepsis – a syndrome with major global impact but no targeted treatment options to date [2, 5]. We could not demonstrate an effect of low-dose erythromycin on 90-day mortality, nor on secondary outcomes indicative of duration of symptoms, occurrence of ICU-acquired complications, or levels of biomarkers reflective of the septic host response. These results, while perhaps limited by the low total dose and short duration of erythromycin treatment, do not argue in favor of using low-dose erythromycin as an adjunctive immunomodulatory therapy in this population, although more studies are needed to obtain more precise effect estimates.
Macrolides exert an array of immunomodulatory and other non-antibiotic effects in vitro and in vivo that could, at least in theory, benefit critically ill patients, including those with sepsis [9]. We chose primary and secondary surrogate or patient-important outcomes that could reflect these effects. Macrolides may reduce excessive inflammation and thereby prevent organ damage (including ventilator-induced lung injury) and expedite the return to immune homeostasis [44–47]; we did not find an effect on secondary outcomes indicative of duration of symptoms, incidence of inflammation-associated complications (AKI, ARDS), or biomarkers reflective of inflammation. Macrolides may stimulate key host immune defenses – including phagocytosis and intracellular killing, commonly impaired in sepsis-induced immune suppression [48] – and interfere with microbial virulence mechanisms such as biofilm formation [49–51]; we did not find a reduction in ICU-acquired secondary infections. Ultimately, we expected that the synergistic effect of these processes could, as it does in animal models [10–14], reduce mortality rates for critically ill patients with sepsis both in the short term (by preventing organ failure) and longer term (by preventing secondary infections); we did, however, not find an effect on mortality.
Previous clinical studies that investigated the immunomodulatory effects of macrolides in critically ill patients, while limited in number, both corroborate and contrast the findings presented here. Most published observational studies, both in sepsis and ARDS (often caused by sepsis and exhibiting similar immune disturbances), have reported lower mortality rates and reduced duration of symptoms in patients treated with macrolides [52–57]. Two RCTs have been published, in which patients with sepsis due to micro-organisms likely to be macrolide-resistant received clarithromycin (in antibiotic doses, 1g once per day for three days[20] or four days [58]). The first trial, in 200 patients with sepsis due to ventilator-associated pneumonia, reported no reduction in 28-day mortality [20], but a remarkable reduction in 90-day mortality in a follow-up study [21]. The second trial, in 600 patients with sepsis likely due to gram-negative bacteria, similarly did not find an effect on 28-day mortality (to the best of our knowledge, 90-day mortality data are not available for this trial [58]). In secondary analyses, both trials did present results consistent with a mortality benefit for clarithromycin in the most severely ill patients (those with septic shock and multiple organ dysfunction syndrome), and a reduction in duration of symptoms. While the results of our study regarding 30-day mortality are in line with these two trials, discrepancies in other outcomes may be explained by several factors: (1) a different drug, as different immunomodulatory macrolides may exhibit subtle differences in effects [59]; (2) a lower dose in our study, for a briefer and more varied duration; (3) differences in the study population, such as different sources of infection; (4) the presence of gastrointestinal dysmotility, which could still affect patient prognosis in ways not captured by baseline covariates; or (5) a slightly different time window, as our study could, by design, only assess outcomes occurring more than 72 hours after ICU admission. We eagerly await the results of a third trial (ClinicalTrials.gov Identifier: NCT03345992), which only included patients with multiple organ dysfunction syndrome, who were most likely to benefit from adjunctive macrolide treatment in the two previous trials (recruitment has concluded, but the results are not available at the time of writing).
We used a target trial emulation approach to reduce the influence of biases common to non-randomized studies of interventions [27–29]. This emulation is always performed within the constraints of the available observational data [27]. An important deviation that our data made from the target trial designed to assess the immunomodulatory effectiveness of low-dose erythromycin is that, in a trial, treatment with erythromycin would not be limited to patients with gastrointestinal dysmotility. We chose this indication to infer immunomodulatory effects from the absence (or negligible presence) of antibacterial effects, but we cannot exclude an effect of reduced gastrointestinal dysmotility on the outcomes – which could, in theory, either oppose or augment the immunomodulatory effects. Not having to account for this indication would both eliminate residual confounding by indication and any post-baseline effects that gastrointestinal dysmotility would have on the mortality (e.g. nutritional deficiencies, or intestinal bacterial translocation leading to new infections [36]). To illustrate this point: several observational studies have reported worse outcomes in patients with gastrointestinal dysmotility even after controlling for disease severity [36]. Nevertheless, we consider it unlikely that any (unmeasured) confounding variable would be strong enough to reject the null-hypothesis (no difference between groups), as indicated by the E-Values [43] for the primary analyses described in the Supplementary Information.
Another deviation from the target trial pertains to the large between-patient variation in total dose and duration of erythromycin treatment, because the necessary total dose and duration to achieve sufficient immunomodulatory effects in acute critical illness are unknown. For most patients in the erythromycin group, the cumulative dose and duration of the first course were low: a median of 800mg over a median of 42 hours (divided over a median of 5 administrations), whereas an antibiotic dose would commonly be up to 2000mg per day for several days. A per protocol treatment directly comparable with the two RCTs using clarithromycin[20, 58] would consist of 2000mg per day for 72 or 96 hours. Despite these considerations, the immunomodulatory effects of macrolides do occur at much lower doses than the antibiotic effects (e.g. 400-600mg erythromycin per day for diffuse panbronchiolitis; 500mg erythromycin twice daily for chronic obstructive pulmonary disease), although chronic use may be needed for some of these effects to occur [59–61].
Several strengths and limitations of this study – partly discussed in the preceding paragraphs – are worth emphasizing. Strengths include the target trial study design, the comprehensiveness of the available data, the use of a DAG to identify confounding covariates, and the robustness of the results to different analysis techniques (matching for the ATT, weighting for the ATE) and sensitivity analyses. Limitations include the aspects of study design that deviate from the target trial (e.g. indication of gastrointestinal dysmotility, uncertainty of the per protocol dose). Also, due to limitations of sample size, considerable statistical uncertainty remains in our effect estimates, making it impossible to exclude potentially meaningful benefits or harms of treatment. In addition, we cannot fully exclude prevalent user bias [62], because data on macrolide use prior to ICU admission were unavailable. We nevertheless considered this type of bias unlikely, as low-dose erythromycin is not commonly prescribed for adults in The Netherlands in outpatient, emergency department, or hospital ward settings, and patients receiving clarithromycin or azithromycin upon ICU admission were excluded. Furthermore, only including patients who survive the first 72 hours (to prevent immortal time bias) means our results cannot be generalized to patients who leave the ICU before this time window.