Critical care outcome prediction equation model, version 7

Résumé fourni par la source

Mortality prediction models are used for benchmarking, audit, research, and epidemiology. This report describes the development and validation methodology, for version 7 of the critical care outcome prediction equation for monitoring adult intensive care unit (ICU) mortality risk. This was a multiphase study incorporating data extraction, curation, aggregation, modelling, and validation applied to jurisdictional administrative datasets to model 90-day survival. The study was conducted across 28 public and 19 private hospitals in state of Victoria, Australia, with an adult population of 5.28 million. A total of 215,148 consecutive adult hospital separations receiving intensive care over 5 years, July 2019–June 2024, were included in the study. The main outcome measures included the case fatality rate (CFR) and standardised 90-day mortality ratio (SMR) at the provider level with model fit assessed for calibration (Brier score; Hosmer–Lemeshow statistic [H 10 ]), discrimination (area under the receiver-operator characteristic and precision-recall curves), classification (CFR and SMR results for ICU-years classified within ±3 standard deviation [SD]), and dispersion (dispersion value [ϕ]; SD random effect [τ]) characteristics. The CFR was found to be 12.0 (95% confidence interval: 11.8–12.1) per 100 separations and 16.1 (95% confidence interval: 15.7–16.4) per 100 persons. A total of 7771 (53.7%) admission diagnoses were aggregated into 24 clinical diagnosis groups and eight demographic (final) model variables. The following results were obtained: mean Brier score = 0.08; H 10 = 15.95; area under the receiver-operator characteristic curve = 0.85; area under the precision-recall curve = 0.43. A total of 126 (54.1%) CFR results and 220 (94.4%) SMR values were within ±3 SD. A total of 105 (45.3%) CFR-outlier ICU-years were reclassified as SMR inliers, with no CFR inlier reclassified as an SMR outlier. Overdispersion metrics: ϕ = 2.80; τ = 0.15. Critical care outcome prediction equation, version 7 is a parsimonious hospital mortality prediction model for adult intensive care admissions, derived from administrative data common to all jurisdictions.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.