Abstract
AIM: Our objective was to increase our understanding of the effects of the time course of metabolic alterations on the risk of hospital-acquired pneumonia (HAP) and response to treatment in critically ill patients. METHODS: We first studied the blood metabolome at day 1, day 3-4 and day 6-7 of patients from a prospective, observational cohort of brain-injured patients in two French centres. We classified the metabolic response by unsupervised longitudinal consensus clustering. To evaluate the robustness of metabolic patterns, a Fast-and-Frugal Tree trained on the discovery cohort was applied to a replication dataset from the PREV-HAP randomised clinical trial testing interferon gamma-1b for the prevention of HAP in critically ill patients. The primary outcome was the association of metabolic response patterns with HAP. FINDINGS: Of the 128 patients analysed (330 samples), 57 (45%) had developed HAP and 21 (16%) acute respiratory distress syndrome (ARDS). Based on metabolites involved in fatty acid metabolism, we identified three metabolic response patterns associated with the risks of HAP (24%, 60% and 78% of HAP, respectively) and ARDS (6%, 16% and 43%, respectively). In a replication dataset (315 samples from 105 patients), we found similarities regarding blood metabolite temporal courses and associations with HAP (18%, 28% and 40% of HAP, respectively). Moreover, the probability of being discharged alive from the ICU with interferon gamma-1b was decreased in patients with low-risk metabolic response patterns and increased in patients with high-risk metabolic response patterns. CONCLUSIONS: Longitudinal clustering classified metabolic response patterns into low-, moderate-, and high-risk groups for HAP and can help identify responders and non-responders to interferon gamma-1b after a single treatment injection. TRIAL REGISTRATION: Number clinicaltrial.gov NCT02003196 and NCT04793568.