Transformation of non-normally distributed outcomesView Presentation Claudine Jurkovitz,
Paul Kolm, Christiana Care Health System
Keywords: Data Transformation,PCI,Generalized Estimating Equation
Introduction: Clinical researchers often analyze variables that are not normally distributed, e.g. health care expenditures, length-of-hospital-stay, waiting times. These outcomes are non-negative measurements with a positively skewed empirical distribution. A common strategy is to transform the data before analysis. However, this does not always result in normally distributed data. Methods: Early Percutaneous Coronary Intervention(PCI) is the recommended treatment of acute myocardial infarction(AMI). During PCI, physicians inflate a balloon to dilate a narrowed coronary artery. Door-to-balloon time is one of the metrics used to evaluate the quality of care for treatment of AMI. We provide an example comparing gender differences of right-skewed minutes from door-to-balloon by choosing a natural logarithmic transformation and a generalized estimating equation model assuming a gamma distribution of the log transformed outcome. The results are then back-transformed for presentation. Conclusion: In modeling data that are not normally distributed, it is crucial to assess the distribution of the transformed data, and to then apply the appropriate distribution to the statistical model.