Development and validation of a disease-specific risk adjustment system using automated clinical data

Area of Science:

• Medical Informatics
• Health Services Research
• Clinical Epidemiology

Background:

• Accurate inpatient mortality risk adjustment is crucial for quality assessment and resource allocation.
• Existing models often rely on complex data or manual abstraction, limiting efficiency.

Purpose of the Study:

• To develop and validate a disease-specific automated system for inpatient mortality risk adjustment.
• To evaluate the contribution of computerized laboratory data, administrative data, and manually abstracted clinical data.

Main Methods:

• Developed 39 automated clinical models using demographics, admission laboratory findings, and diagnosis codes for 1,271,663 discharges (2000-2001).
• Augmented models with manually abstracted clinical data.
• Validated models using 1,178,561 discharges (2004-2005), comparing discrimination and calibration.

Main Results:

• Overall mortality was 4.6% (derivation) and 4.0% (validation).
• Automated models achieved an average c-statistic of 0.83; adding manual data improved it to 0.85.
• Numerical laboratory results were the strongest predictors of mortality.

Conclusions:

• A limited set of numerical laboratory results and administrative data effectively risk-adjust inpatient mortality across diverse conditions.
• Automated systems offer a viable approach for accurate and efficient mortality prediction.