Session number: 1019
Abstract title: Predictive Validity of Comorbid Measures for Administrative Data in VA and Private Sector Inpatients
Author(s):
PJ Kaboli - Iowa City VAMC and University of Iowa
MJ Barnett - Iowa City VAMC
GE Rosenthal - Iowa City VAMC and University of Iowa
Objectives: Compare the predictive validity of a newly developed comorbidity measure for administrative data (Elixhauser) and a widely used comorbidity index (Charlson) in admissions to VA and private sector hospitals.
Methods: The sample included consecutive discharges from VA (n=228,356) and private sector (n=10,903,990) hospitals over a 4-year period (1996-1999) for 4 high-volume diagnoses: congestive heart failure (CHF); chronic obstructive pulmonary disease (COPD); pneumonia; and gastrointestinal hemorrhage (GIH). VA data were obtained from the Patient Treatment File. Private sector data were obtained from the National Hospital Discharge Survey, a nationally representative database of patients in non-federal hospitals. The Elixhauser method assesses 30 comorbidities that are unlikely to be hospital complications or related to admitting diagnosis. The Charlson method for administrative data assesses 11 specific conditions. Discrimination of the two methods for in-hospital mortality was compared using receiver operating characteristic (ROC) curve analysis.
Results: VA patients were younger than private sector patients (mean ages, 68 vs. 71 years;P<.001), were more likely to be male (98% vs. 44%;P<.001), and had higher mortality (4.9% vs. 3.9%;P<.001). For each of the four conditions, ROC curve areas in VA patients were higher (P<.001) for the Elixhauser method than for the Charlson method: CHF (0.68 vs. 0.59); COPD (0.67 vs. 0.56); pneumonia (0.69 vs. 0.64); GIH (0.73 vs. 0.64). ROC curve areas were also higher for the Elixhauser method in private sector patients: CHF (0.67 vs. 0.61); COPD (0.69 vs. 0.56); pneumonia (0.67 vs. 0.62); GIB (0.76 vs. 0.65). However, ROC curve areas for the Elixhauser method were generally similar in VA and private sector patients: CHF (0.68 vs. 0.67); COPD (0.67 vs. 0.69); pneumonia (0.69 vs. 0.67); GIH (0.73 vs. 0.76).
Conclusions: A recently proposed method of adjusting for comorbidity in administrative data—Elixhauser—demonstrated higher predictive validity than the previously developed and widely used Charlson method. The Elixhauser method demonstrated similar predictive validity in VA and private sector patients.
Impact statement: The Elixhauser method of adjusting for comorbidity may improve the attributional validity of outcomes studies in VA patients based on administrative data and may allow for wider use of administrative data in studies of the quality and efficiency of hospital care.