Invasive group A streptococcus infections in the intensive care unit: an unsupervised cluster analysis of a multicentric retrospective cohort.

backgroundInvasive group A streptococcus (iGAS) infection incidence is rising. These infections have been studied as a whole but can be associated with critical illness in a population with a wide array of underlying conditions, sites of infection and clinical presentations. Using an unsupervised clustering approach, we aimed to identify specific clinical phenotypes regarding presentation, management and outcome.

methodsThis was a retrospective multicentric study including all patients admitted to one of 9 ICUs of Paris University Hospitals for an iGAS infection between 01/03/2018 and 01/08/2023. iGAS infection was defined as GAS growth in any microbiological sample from a sterile site. Patients were grouped according to a clustering algorithm (k-prototypes) using a comprehensive set of clinical and biological variables available upon ICU admission. Clusters were described and clinical presentation, management and outcome were compared.

results148 patients were included. According to the Silhouette criterion, patients were grouped in 3 clusters, and 7 patients remained unclassified. Cluster 1 (n = 73) comprised a greater proportion of less severely-ill female patients with painful skin and soft tissue infections, a quarter of whom had taken non-steroidal anti-inflammatory drugs. Cluster 2 (n = 42) was characterized by a high rate of respiratory infections with frequent viral co-infections. Cluster 3 (n = 26) included mostly socially deprived patients with high rates of chronic alcohol consumption and psychiatric illness, with severe organ dysfunction related to otherwise pauci-symptomatic skin and soft tissue infections. There was no significant difference in time to source control across clusters (0 [0-0] vs 0 [0-0] vs 0 [0-1] days, p = 0.12). Patients included in cluster 2 less frequently received antitoxin antibiotics than patients from clusters 1 and 3 (79% vs 45% vs 69%, p < 0.001) and tended to more frequently require ECMO support (3% vs 12% vs 0%, p = 0.07), while those from cluster 1 were less likely to receive invasive mechanical ventilation (48% vs 74% vs 77%, p = 0.005). There was no difference in ICU-mortality between clusters (19% vs 29% vs 31%, p = 0.32).

conclusionsBased on simple and readily available clinical admission characteristics of critically ill patients with iGAS, unsupervised clustering analysis identified three specific patient populations that differed regarding ICU management. Whether tailoring management would affect outcome warrants further research.