BACKGROUND: Annually, more than 11 million patients are admitted to hospital overnight in England, but the UK is ranked 24 of 31 European countries with respect to per head provision of intensive care unit (ICU) beds. This lack of beds places strain on the capacity to admit patients from the ward because of high ICU occupancy. Such delay can cause harm, but the effect of such harm is difficult to measure. Prompt admissions are prompt precisely because these patients are severely unwell. Measured severity is unlikely to completely capture the clinical judgment used to allocate early admission, and therefore risk-adjusted outcomes will be biased against the early admission. We aimed to evaluate the effect of delayed admission to critical care without this treatment selection bias. METHODS: We did a prospective cohort study of deteriorating ward patients assessed for critical care admission in National Health Service hospitals in the UK. Early admission was defined as within 4 h of assessment. The primary endpoint was 90-day survival. We used critical care occupancy as an instrumental variable, assuming that a full ICU could only affect outcome of a ward patient by deflecting or delaying admission. FINDINGS: 12 495 patients from 48 hospitals were available for analysis of whom 3797 (30·4%) died within 90 days. 4494 (36·0%) patients were admitted to critical care of whom 2492 (55·5%) were admitted early. The median time to admission was 2 h (IQR 1-3) with a bedside decision to admit, and 12 h otherwise (5-29). 991 patients (7·9%) were assessed when the critical care unit was fully occupied. Compared with those assessed when more than one bed was available, these patients were admitted less often (odds ratio [OR] 0·37, 95% CI 0·28-0·48), experienced greater delays (median increase 2 h, IQR 1-3), and deteriorated further while waiting (1·4 ICNARC physiology points, 95% CI 0·4-2·5). Early admission reduced mortality (OR 0·49, 95% CI 0·27-0·89). When averaged across the full population, absolute mortality fell by 13·9% (95% CI 25-23·0). INTERPRETATION: Our study has shown that the deteriorating ward patient is vulnerable with a high short-term mortality (none of these patients had treatment limitations). Delays to admission were large and common, and arose both from our inability to perfectly triage these patients, and from limits to the capacity of the system. That these delays cause harm is very likely. FUNDING: Wellcome Trust, National Institute for Health Research Service Support Costs, Intensive Care National Audit & Research Centre.