A new risk prediction model for critical care: the Intensive Care National Audit & Research Centre (ICNARC) model.
Harrison, David A;
Parry, Gareth J;
Carpenter, James R;
Short, Alasdair;
Rowan, Kathy;
(2007)
A new risk prediction model for critical care: the Intensive Care National Audit & Research Centre (ICNARC) model.
Critical care medicine, 35 (4).
pp. 1091-1098.
ISSN 0090-3493
DOI: https://doi.org/10.1097/01.CCM.0000259468.24532.44
Permanent Identifier
Use this Digital Object Identifier when citing or linking to this resource.
OBJECTIVE: To develop a new model to improve risk prediction for admissions to adult critical care units in the UK. DESIGN: Prospective cohort study. SETTING: The setting was 163 adult, general critical care units in England, Wales, and Northern Ireland, December 1995 to August 2003. PATIENTS: Patients were 216,626 critical care admissions. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: The performance of different approaches to modeling physiologic measurements was evaluated, and the best methods were selected to produce a new physiology score. This physiology score was combined with other information relating to the critical care admission-age, diagnostic category, source of admission, and cardiopulmonary resuscitation before admission-to develop a risk prediction model. Modeling interactions between diagnostic category and physiology score enabled the inclusion of groups of admissions that are frequently excluded from risk prediction models. The new model showed good discrimination (mean c index 0.870) and fit (mean Shapiro's R 0.665, mean Brier's score 0.132) in 200 repeated validation samples and performed well when compared with recalibrated versions of existing published risk prediction models in the cohort of patients eligible for all models. The hypothesis of perfect fit was rejected for all models, including the Intensive Care National Audit & Research Centre (ICNARC) model, as is to be expected in such a large cohort. CONCLUSIONS: The ICNARC model demonstrated better discrimination and overall fit than existing risk prediction models, even following recalibration of these models. We recommend it be used to replace previously published models for risk adjustment in the UK.