Abstract
One of the biggest problems in dealing with healthcare data is that there can be multiple patient events for the same individual. Without finding some way to combine these many observations into one row in the dataset, it is very difficult to analyze the data by patient. When dealing with cost and reimbursement information, this can be relatively simple since a total sum of costs or reimbursements can be used to examine any one patient. However, if we are looking at patient conditions, or physician decisions in sequence, this consolidation of patient information can be especially difficult. When using the technique of survival data mining, as we will discuss in later chapters, such consolidation of information is absolutely essential. In this chapter, then, we will discuss some of the requirements for combining patient information into one data row.
TopBackground
A search of Medline using the keywords “preprocessing data” and “health outcomes” returns exactly two results. There are a few textbooks on the market as well that are focused on health outcomes. (Arslanian, 2001; Fink, 2004; Kane, 2005) However, these books tend to be more concerned with devising measurement indices than they are in examining the data analysis and preprocessing requirements. In particularly, there is very little basic information on converting a many-to-one or a many-to-many relationship into a relationship that is one-to-one so that it can be used for the analysis of outcomes. (Austin & Austin, 2008a; Dilba, Bretz, Guiard, & Hothorn, 2004; Hothorn & Hothorn, 2006; Nakagawa & Nakagawa, 2005; Schaarschmidt, et al., 2009)