There’s No Such Thing as Normal Clinical Trials Data, or Is There?

Clinical Trials data comes in all shapes and sizes depending upon the therapeutic area, indication and phase of the trial. Quite a bit of the data arriving on Case Report Forms (CRFs) is fairly standard (e.g. Demography, Adverse Event, Medications, Laboratory, etc.), and hence can be stored in fairly standard data structures. Designing clinical data structures for data entry is important, but it should be done with some understanding of the analysis that will be performed. Once an appropriate clinical data structure is arrived at for data entry, it is important to then determine how to best use the data in the analysis environment. This paper will discuss data normalization and what it means along with the effects it has on SAS programs used to list and summarize clinical trials data. Side by side examples of programs and output from standard data structures versus normalized data structures will provide proof of the importance of normalization. NORMALIZATION The term “normalization ” can mean different things in different environments. It is used within this paper to describe a data structure which describes one element of data per observation within a data set. This means that there may be one or more context variables describing where the data belongs (e.g. GROUP, PATNUM, VISIT, etc.), along with one or more variables describing specific data points (e.g. VAR, VAL, VALC, VALN, FLAG, etc.). Another term used when referring to a normalized data structure is a vertical data structure. This implies the data exists in a long fashion with many observations, fewer variables