Cluster analysis identifies patients at risk for long-term mortality in community-acquired pneumonia in CAPNETZ

Rationale Community-acquired pneumonia (CAP) remains a disease with high morbidity and mortality. While tools and scores to predict short-term prognosis of CAP, e.g. the CRB65-Score, have been developed, there is urgent need for early identification of patients at risk for adverse mid- or long-term outcome. Here, we report cluster analysis of CAP-patients clinical attributes at hospital admission with subsequent curation of an attribute panel predicting adverse clinical 31-to-180-day outcome.

Methods Data from the German prospective CAPNETZ cohort was extracted and clinical attributes of patients (n=7,248) at hospital admission clustered by means of self-organising-maps (SOM). Differential clinical attributes of clusters identified by outcome-based clustering were identified and used for attribute-based clustering. Clusters identified by attribute-based clustering were analysed for further differences in clinical attributes.

Main Results SOM-clustering identified a panel of eleven easily accessible clinical attributes to predict 30-day as well as 31-to-180-day mortality in 7,248 CAP patients. Clustering based on these attributes yielded 15 clusters, identifying differential clinical phenotypes with high risk of mortality in the 31-to-180-day timeframe, the 30-day timeframe, both periods, or very low overall mortality, respectively.

Publication History

Article published online:
01 March 2024