Abstract:
|
Missing data is an unavoidable problem of data collection in clinical research and many methods have been developed to handle missing data problem. Multiple imputation(MI) is advanced and commonly used method to ensure unbiased and valid estimates. One of the MI's pitfalls is violation of the normal distribution assumption. In the real world, there are some biochemical patient’s characteristics that can strongly violate this assumption(e.g.NT-pro-BNP). Therefore, the main goal of this research is to compare the results of the few practical ways to perform MI for the strongly right-skewed data, with a long tail in SAS software. Since SAS is one of the most commonly used statistical software for data analysis. The access to the original data is still a topic of the discussion with the sponsor. Therefore, at the moment artificial data were populated to address problem of skewed missing data in MI application. There were two major foundling. Firstly, trying to set up minimum and maximum ranges for variables cause syntax errors. Secondly, the FCS statement with REGPMM option is the best method to properly handle the problem of MI application for data with strongly skewed distributions.
|