Clinical Trials Versus Health Outcomes Research: SAS/STAT Versus

Clinical trials typically involve a random selection of subjects to receive an experimental treatment or a control. They often use a minimal sample size, a short term rather than a long term period of study, and clinical trials often use surrogate endpoints. Inclusion/exclusion criteria are used to create a fairly homogeneous set of subjects so that the assumption of normality is relatively valid. Because of the nature of clinical trials, the database is designed to optimize the statistical analysis; most of the techniques used are available in SAS/STAT and there is relatively little preprocessing required prior to the statistical analysis. In contrast, health outcomes typically use data collected in the course of patient care with databases that are not designed for statistical analyses. The data are extremely messy and require considerable preprocessing; They are also observational and require consideration of potential and/or actual confounding factors. Typically, these databases are extremely large, containing thousands to millions of records. The data can be used to investigate real rather than surrogate endpoints in a longitudinal setting as well as to investigate rare occurrences. The subjects are heterogeneous, so the assumption of normality is not reasonable and creates problems when using regression models. In addition, the effect size in any analysis is virtually zero, so the p-value has no real meaning; other measures must be used to determine the adequacy of the model. Many of the statistical tools required to analyze health outcomes data are readily available in SAS Enterprise Miner. We will compare and contrast the differences and similarities between clinical trials and health outcomes research.