Statistical Methods - Text Mining Techniques for Healthcare Provider Quality Determination - page 108

Figure 23. Kernel density estimation for prediction of mortality
dIscussIon
Given large datasets, and the presence of outliers, the traditional statistical methods are not always ap-
plicable or meaningful. Assumptions can be crucial to the applicability of the model, and assumptions
are not always carefully considered. In particular, great care must be taken when performing logistic
regression for a rare occurrence. Ideally, the data should be resampled so that the proportion of occur-
rences and non-occurrences is the same. We will discuss this in more detail in future chapters. However,
if we consider the model assumptions carefully, we can usually find a reasonable model that will fit the
data.
The data may not give high levels of correlation, and regression may not always be the best way to
measure associations. It must also be remembered that associations in the data as identified in regression
models do not demonstrate cause and effect. In observational studies for outcomes research, potential
confounders should always be considered. It will require considerable domain knowledge to be able to
develop a list of potential confounders that should be considered in the study. With the large datasets that
are typically used for this type of research, we can include many potential confounders in the analysis.
Many of these confounders can be examined through examination of diagnosis and procedure codes.
Up to this point, we have just examined a small list of both; in subsequent chapters, we will demonstrate
how all of these codes can be used in an outcomes analysis.
references
Anonymous-KY. (2008). Hospital Cost-to-Charge Ratio . Retrieved June, 2008, 2008, from http://www.
labor.ky.gov/workersclaims/medicalservices/hospitalcost/