Analysis of automated administrative and survey databases to study patterns and outcomes of care

Large computerized databases often arise from national surveys, insurance claims, and statewide health care registries. These databases are increasingly used to examine patterns of medical care and certain outcomes of care and may be helpful in planning clinical trials. They are highly representative of defined populations, but have limited clinical information. Methods have been developed to identify episodes of low back pain and to quantify the severity of unrelated, comorbid medical conditions. Pitfalls in analysis are discussed, including limitations of diagnosis and procedure coding, cross-sectional nature of most data, limited clinical detail, and the necessarily observational (not experimental) nature of any group comparisons. There is growing interest in expanding the clinical information in such databases, for both quality improvement and research purposes.