The UK has a low proportion of Intensive Care (ICU) beds and a correspondingly high sepsis-related mortality. Appropriate identification of deteriorating ward patients for ICU admission is a national priority area. Current methods use severity-based scores derived from routine ward observations (e.g. pulse, blood pressure). However, the current biological paradigm of sepsis emphasises its time-dependent nature progressing from a pro- to an anti-inflammatory cytokine milieu. Piggybacking a cohort study onto a national ICU audit, this project will firstly evaluate how delay to ICU admission affects outcome (hypothesising that the relationship will rest on these time-dependent sepsis phenotypes). Accordingly, delay should not simply lead to greater severity, but to different clinical states with different outcomes. Secondly, in a detailed prospective cohort of patients with severe pneumonia, daily organ dysfunction will be used to plot timing of ICU admission and organ support aga inst the duration and trajectory of illness. Correlation with clinical phenotypes defined by duration and trajectory will be sought using biological (cytokine and endocrine) markers. Using these longitudinal descriptions of illness, a multi-state model will be developed to assess the effect of altering criteria for, and timing of, ICU admission.