Data representations and -analyses of binary diary data in pursuit of stratifying children based on common childhood illnesses.

Abstract

In this article we analyse diary reports concerning childhood symptoms of illness, these data are part of a larger study with other types of measurements on childhood asthma. The children are followed for three years and the diaries are updated, by the parents, on a daily basis. Here we focus on the methodological implications of analysing such data. We investigate two ways of representing the data and explore which tools are applicable given both representations. The first representation relies on proper alignment and point by point comparison of the signals. The second approach takes into account combinations of symptoms on a day by day basis and boils down to the analysis of counts. In the present case both methods are well applicable. However, more generally, when symptom episodes are occurring more at random locations in time, a point by point comparison becomes less applicable and shape based approaches will fail to come up with satisfactory results. In such cases, pattern based methods will be of much greater use. The pattern based representation focuses on reoccurring patterns and ignores ordering in time. With this representation we stratify the data on the level of years, so that possibly yearly differences can still be detected.