Veterans Affairs intensive care unit risk adjustment model: validation, updating, recalibration.

three-pronged approach—improvements in the deliveryof healthcare, alignment ofperformance incentives, andengaging consumers in the quest—mightenhance the quality of health care. Thecreation of registries using informationtechnology could improve health care de-livery by providing feedback about perfor-mance to providers (1). The Departmentof Veterans Affairs (VA) for Health Care,with its nationally recognized electronic(2) medical record and 459,000 acutecare inpatient admissions annually, is po-sitioned to create such a registry. Previ-ously, using computerized reminders andorganizational performance incentives,the VA improved patient safety, preven-tive care, and disease-specific measuresfor veterans (3–6). In 2004, the VA beganto measure and report the risk-adjustedmortality and length of stay for patientsin its intensive care units (ICUs). Thismeasurement system relies on data ex-tracted from the electronic medicalrecords of each hospital to identify out-comes and construct an Acute Physiol-ogy and Chronic Health Evaluation(APACHE)–like risk adjustment mea-sure. This computer-based risk adjust-ment model was originally developed inan ICU cohort housed in moderate andlarge, but not small, ICUs during 1996and 1997, with good discrimination andcalibration (validation cohort: 11,646cases, C statistic 0.885; Hosmer-Lemeshow C statistic [goodness-of-fit]chi-square 25.1) (7, 8).However, measurement of risk-ad-justed outcomes in complex populationscan be tricky and, if not done carefully,can incorrectly label a healthcare systemor instigate costly “wild goose chases”(9). Other ICU severity measures haveshown variation in both calibration anddiscrimination when applied to new pop-