ADMINISTRATIVE DATA HAVE BEEN USED TO ASSESS quality of care and variations in health care delivery for nearly 3 decades. Indeed, administrative data have provided important insights on racial disparities in care, geographic differences in utilization, and variations in outcomes across hospitals. In the absence of national clinical registries, administrative data will continue to be an important source of information about health care delivery in the United States because of their ready availability, low cost, and ability to span multiple years and health care settings. Moreover, in contrast to data from randomized controlled trials, administrative data reflect real-world treatment settings and unselected populations. However, users must be aware of the inherent limitations of administrative data to avoid erroneous conclusions. In this issue of JAMA, Lindenauer and colleagues illustrate the complexity in deciphering simple temporal trends in hospital mortality for pneumonia. Using a nationally representative database—the Nationwide Inpatient Sample— the authors found that the rate of hospitalizations with a principal diagnosis of pneumonia decreased by a relative 27% from 2003 to 2009 while in-hospital mortality decreased by 28%. During the same period, rates of hospitalization for sepsis and respiratory failure with a secondary diagnosis of pneumonia increased by 178% and 9%, respectively. When these 3 groups were combined, the authors’ assessment of trends in pneumonia admissions and mortality changed substantially, with a 12% relative decline in pneumoniarelated admissions and a 6% increase in mortality during the study period. This study highlights the importance of understanding nuances and vagaries of administrative data to evaluate trends over time or compare clinician performance. Even relatively pure and simple measurements may be marred by a failure to appreciate the dynamic circumstances that affect how International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes are recorded. These issues go beyond pneumonia and may have affected many studies based on administrative data. For example, relative decreases in risk-adjusted hospital mortality between 2000 and 2007 for congestive heart failure (CHF), acute myocardial infarction (AMI), and stroke have also been reported—49%, 36%, and 26%, respectively. Given that this study identified patients using principal diagnoses, it is possible that the reported decreases also resulted from changes in coding practices in which patients with more severe disease were given other principal diagnoses (eg, pulmonary edema or respiratory failure for CHF). Nuances in the assignment of principal and secondary diagnoses can also affect assessments of hospital performance. Since 2005, the Centers for Medicare & Medicaid Services (CMS) has monitored the provision of recommended care to patients hospitalized with AMI, CHF, pneumonia, and chronic obstructive pulmonary disease (COPD). Because sampling of patients for specific performance measures is usually defined by principal diagnoses, the sequencing of ICD-9-CM codes may affect performance assessments. For example, a study by Wu et al in 2002 reported lower use of aspirin, heparin, -blockers, and primary angioplasty among AMI patients with CHF compared with AMI patients without CHF, although patients with CHF were more likely to receive angiotensionconverting enzyme inhibitors. If AMI is not recognized as the principal diagnosis, inadequate use of aspirin, heparin, -blockers, or primary angioplasty among these patients may be underreported. Because administrative data are derived from claims submitted by clinicians to receive payment, the selection of pri-
Read full abstract