The added value that increasing levels of diagnostic information provide in prognostic models to estimate hospital mortality for adult intensive carepatients

Objective:To investigate in a systematic, reproducible way the potential of adding increasing levels of diagnostic information to prognostic models fo r estimating hospital mortality. Design: Prospective cohort study. Setting: Thirty UK intensive care units (ICUs) participating in the ICNARC Case Mix Programme. Patients: Eight thousand fifty-seven admissions to UK ICUs. Measurements and results: Logistic regression analysis incorporating APACHE II score, admission type and increasing levels of diagnostic information w as used to develop models to estimate hospital mortality for intensive care patients. The 53 UK APACHE II diagnostic categories were substituted with data from a hierarchical, five-tiered (type of condition required surgery o r not, body system, anatomical site, physiological/pathological process, co ndition) coding method, the ICNARC Coding Method. The inter-rater reliabili ty using the ICNARC Coding Method to code reasons for admission was good (k appa = 0.70). All new models had good discrimination (AUC = 0.79-0.81) and similar or better calibration compared with the UK APACHE II model (Hosmer- Lemeshow goodness-of-fit H = 18.03 to H = 26.77 for new models versus H = 6 3.51 for UK APACHE II model). Conclusion: The UK APACHE II model can be simplified by extending the admis sion type and substituting the 53 UK APACHE II diagnostic categories with n ine body systems, without losing discriminative power or calibration.