Potential for bias in severity adjusted hospital outcomes data: analysis of patients with rheumatic disease

To examine the predictive validity of MedisGroups, a widely used method of measuring severity of illness, among patients with rheumatic disease and identify determinants of hospital outcomes, after adjusting for severity of illness.

Adult medical and surgical patients with rheumatic disease (5421) admitted to an academic medical center in 1988-90 were studied using a retrospective cohort design. Sociodemographic, clinical, and financial data were obtained from computerized hospital information systems. Severity of illness on admission was determined for each patient using MedisGroups, which classifies patients into groups of increasing severity.

MedisGroups admission severity groups were highly related (p < 0.001) to inhospital mortality rates, which were 0.4, 0.8, 5.1 and 16.1%, respectively among patients in 4 groups of increasing severity. Controlling for MedisGroups admission severity using logistic regression, age, admission from the emergency room, and transfer from an acute care hospital were found to be additional independent predictors of mortality. MedisGroups severity groups were also directly related (p < 0.001) to length of stay and total hospital charges. Controlling for admission severity using linear regression, length of stay, and charges were independently related to several other variables; for example, length of stay was greater for patients admitted from the emergency room or transferred from other hospitals and for nonwhites, women, and older patients. Finally, within common individual diagnoses, these factors substantially increased the amount of variance in length of stay and charges explained by MedisGroups alone.

Our findings demonstrate that after adjusting for severity of illness using MedisGroups, several other easily measured variables were associated with hospital outcomes in patients with rheumatic disease. Thus, generic severity systems, such as MedisGroups, may not adequately adjust outcomes among patients with rheumatic disease. Comparative hospital data based on these systems may be subject to bias.