Use of administrative data or clinical databases as predictors of risk of death in hospital: comparison of models
Key result
Risk prediction models using administrative data predicted in-hospital death with similar discrimination to clinical databases across multiple procedures (e.g., ROC 0.77 vs 0.78 for isolated CABG).
Why the study?
Do risk prediction models based on administrative data perform comparably to those based on clinical databases for predicting in-hospital death in patients undergoing major surgery?
Population
341,379 patients admitted to NHS hospitals in England for isolated coronary artery bypass graft, repair of…
Comparison
Risk prediction models for death in hospital… vs Risk prediction models based on national…
Design
Cohort
Follow-up
In-hospital
Authors
Discussion
Loading...
Member takes
Implication
Administrative data models may support mortality prediction in surgery; leaves open need for prospective validation before practice change.
Study Design
Type
Observational (n=341,379)
Multicenter
Yes
Structured PICO
Do risk prediction models based on administrative data perform comparably to those based on clinical databases for predicting in-hospital death in patients undergoing major surgery?
P
Population
341,379 patients admitted to NHS hospitals in England for isolated coronary artery bypass graft (CABG) (n=152,523), repair of ruptured (n=12,781) or unruptured (n=31,705) abdominal aortic aneurysm, or colorectal excision for cancer (n=144,370) from 1996-7 to 2003-4.
I
Intervention
Risk prediction models for death in hospital based on an administrative database (inpatient hospital episode statistics).
C
Comparator
Risk prediction models based on national clinical databases (national cardiac surgical database, national vascular database, and colorectal cancer study).
O
Outcome
Deaths in hospital.hard clinical
Routinely collected administrative data can predict in-hospital mortality risk with similar discrimination to dedicated clinical databases for major surgical procedures.
Cite This Study
Aylin et al. (2007) conducted an observational in Isolated CABG, abdominal aortic aneurysm repair, and colorectal excision for cancer (n=341,379). Administrative database risk prediction models vs. Clinical database risk prediction models was evaluated on Deaths in hospital. Risk prediction models using administrative data predicted in-hospital death with similar discrimination to clinical databases across multiple procedures (e.g., ROC 0.77 vs 0.78 for isolated CABG).