Derivation and Validation of a Model to Predict Daily Risk of Death in Hospital

Bibliographic Data

ID9102053
AuthorsJenna Wong (Institute for Clinical Evaluative Sciences, corresponding author), Monica Taljaard (0000-0002-3978-8961, University of Ottawa, corresponding author), Alan J Forster (0000-0003-2942-2891, Institute for Clinical Evaluative Sciences, corresponding author), Gabriel J Escobar (0000-0003-2540-3327, Kaiser Permanente), Carl van Walraven (0000-0002-8390-0930, Ottawa Hospital, corresponding author)
Year2011
Volume49
Issue8
Pages734-743
Publication date2011-08-01
Peer ReviewedYes
Open AccessNo
TypeARTICLE
VenueMedical Care (JOURNAL)
Journal identifiersISSN: 0025-7079 • E-ISSN: 1537-1948
PublisherOvid Technologies (Wolters Kluwer Health) (PUBLISHER)
DOI10.1097/mlr.0b013e318215d266
PMID21478775
OpenAlexW2073286613
LanguageEN
Citations received2
References cited16

BACKGROUND: As electronic patient data from automated hospital databases become increasingly available, it is important to explore the ways in which these data could be used for the purposes other than patient care, such as quality assurance and improvement. OBJECTIVE: To determine if information from automated patient databases can be used to derive a model that can predict patients' daily risk of death in hospital. Such a model could be used to improve the ability to risk-adjust hospital mortality rates. STUDY DESIGN AND SETTING: Retrospective cohort study of 159,794 hospitalizations at The Ottawa Hospital between April 1, 2004 and March 31, 2009. The model was derived using time-dependent Cox regression methods on a random two-thirds of admissions. The model was validated by applying the coefficients to the other third of admissions. RESULTS: Inpatient mortality was 5%. The final model included: patient age; admission type; intensive care unit status; alternative level of care status; and separate scores for patient comorbidity, in-hospital procedures, and acute illness (using information from 14 laboratory tests). In the validation set, the model had excellent discrimination (c-statistic 0.879, 95% confidence interval: 0.872-0.886) and calibration in all risk strata over all admission days. CONCLUSION: We found that information from our hospital's automated patient databases could be used to accurately predict patients' daily risk of death in hospital. The predictions from this model could be used in quality of care analyses to more accurately risk-adjust hospital mortality rates and by hospitals to improve triage processes and patient flow

Comorbidity · Confidence interval · Intensive care medicine · Intensive care unit · Medical emergency · Proportional hazards model · Quality assurance · Retrospective cohort study · Statistic · Statistics · Electronic Health Records Systems · Emergency and Acute Care Studies · Emergency Medicine · Internal Medicine · Medicine · Sepsis Diagnosis and Treatment