Prediction of risk of death using 30-day outcome

A practical end point for quality auditing in intensive care

Petra L. Graham*, David A. Cook

*Corresponding author for this work

Research output: Contribution to journalArticle

22 Citations (Scopus)

Abstract

Study objective: To validate the APACHE (acute physiology and chronic health evaluation) III unadjusted and similar hospital mortality estimate models on 30-day mortality, and to propose a simple approach to modeling local 30-day in-hospital mortality of critically ill hospitalized adults for quality management and risk-adjusted monitoring. Design: Noninterventional, observational study. Patients: A total of 5,278 consecutive eligible hospital admissions between January 1, 1995, and December 31, 1999. Measurements: Prospective collection of demographic, diagnostic, physiologic, laboratory, and hospital admission and discharge data. Results: The APACHE III mortality predictions exhibited excellent discrimination (receiver operating characteristic [ROC] curve area) for 30-day outcome (ROC area, 0.89) and hospital outcome (ROC area, 0.89). Calibration curves and Hosmer-Lemeshow statistics demonstrated good calibration of all models on 30-day outcome, except for the unadjusted APACHE III model. New, simplified risk adjustment models showed good discrimination and calibration on development and test data. ROC areas were 0.88 (developmental data) and 0.87 (test data), and the new model calibration was equivalent to the APACHE III model. Conclusion: For quality audit, 30-day in-hospital mortality can be used as an alternative outcome to survival to hospital discharge. New logistic regression models provide evidence that local models, possessing good calibration and discrimination, may be built from a few explanatory variables.

Original languageEnglish
Pages (from-to)1458-1466
Number of pages9
JournalChest
Volume125
Issue number4
DOIs
Publication statusPublished - Apr 2004
Externally publishedYes

Keywords

  • Logistic regression
  • Quality monitoring
  • Risk adjustment
  • Risk assessment

Fingerprint Dive into the research topics of 'Prediction of risk of death using 30-day outcome: A practical end point for quality auditing in intensive care'. Together they form a unique fingerprint.

Cite this