Exhibit Z. Data Verification and Validation

HCUP Data

Because administrative data on inpatient stays were not created for research purposes, there may be problems with the reliability and validity of certain data elements. Green and Wintfield (1993) summarized the literature on coding errors for hospital administrative data and described a decline in error rates during the 1970s and 1980s. Fisher, Whaley, Krushat, et al. (1992) reported that the accuracy of principal diagnosis and procedure has improved since 1983, when such information became important for determining reimbursement by Medicare and other payers. Green and Wintfield (1993) reported the results of a reabstraction study using records from the California Office of Statewide Health Planning and Development. Information on age and sex was most reliable (error rates less than 1 percent), and principal diagnosis was inaccurate in 9 percent of records.

Subsequent studies have shown over 90 percent agreement between hospital administrative data and other sources of data for serious conditions and for in-hospital procedures (Baron, et al., 1994; Pinfold, et al., 2000; Du, et al., 2000). A Veterans Affairs study compared administrative data to medical records and found adequate reliability for demographics, length of stay, and selected diagnoses (Kashner, 1998). A study that compared the accuracy of Medicare claims data to tumor registry data in identifying procedures performed for cancer found that claims data are accurate for studying surgical treatment but are less accurate in identifying diagnostic procedures (Cooper, et al., 2000). However, questions have been raised about the accuracy of administrative data for some conditions such as trauma, specifically splenic injury and thoracic aorta injury. Type of injury, injury severity, use of specific procedures, and complications were all under-reported in administrative data compared with trauma registry data (Hunt, et al., 1999; Hunt, et al., 2000).

Other problems inherent in hospital inpatient data include missing data, underreporting of socially stigmatized conditions such as alcoholism and drug abuse, and underreporting of minor procedures. One study found that analyses limited to principal diagnoses and procedures will produce an underestimate of diagnoses that tend to appear in secondary positions such as hypertension, osteoporosis, and Alzheimer's disease (May, Kelly, Mendlein, et al., 1991). However, another study concluded that while administrative data may underestimate the presence of comorbidities, there is a high degree of agreement between administrative data and medical records for symptomatic comorbid conditions (Humphries, et al., 2000).

Return to Exhibits
Proceed to Exhibit AA