PURPOSE: To estimate the effect of prompt admission to critical care on mortality for deteriorating ward patients. METHODS: We performed a prospective cohort study of consecutive ward patients assessed for critical care. Prompt admissions (within 4 h of assessment) were compared to a 'watchful waiting' cohort. We used critical care strain (bed occupancy) as a natural randomisation event that would predict prompt transfer to critical care. Strain was classified as low, medium or high (2+, 1 or 0 empty beds). This instrumental variable (IV) analysis was repeated for the subgroup of referrals with a recommendation for critical care once assessed. Risk-adjusted 90-day survival models were also constructed. RESULTS: A total of 12,380 patients from 48 hospitals were available for analysis. There were 2411 (19%) prompt admissions (median delay 1 h, IQR 1-2) and 9969 (81%) controls; 1990 (20%) controls were admitted later (median delay 11 h, IQR 6-26). Prompt admissions were less frequent (p < 0.0001) as strain increased from low (22%), to medium (15%) to high (9%); the median delay to admission was 3, 4 and 5 h respectively. In the IV analysis, prompt admission reduced 90-day mortality by 7.4% (95% CI 1.7-18.5%, p = 0.117) overall, and 16.2% (95% CI 1.1-31.3%, p = 0.036) for those recommended for critical care. In the risk-adjust survival model, 90-day mortality was similar. CONCLUSION: After allowing for unobserved prognostic differences between the groups, we find that prompt admission to critical care leads to lower 90-day mortality for patients assessed and recommended to critical care.