Information Technology Reference
In-Depth Information
Figure 46. Results for total charges as target
future trends
Predictive modeling already includes all regression models. Therefore, it will be used much more
often when analyzing health outcomes than it has been used in the past. It needs to be brought more
commonly into the curriculum for students specializing in health outcomes research before predictive
modeling will become more common. Departments of Biostatistics and Informatics need to recognize
the availability of data mining tools and their use in health outcomes research. The process of predic-
tive modeling should be substituted for the now common use of regression models. Misclassification
and cost should be used instead of p-values and odds ratios to have more accurate results generally in
health outcomes research.
dIscussIon
The process of data mining automatically incorporates important components that are not generally a
part of more traditional statistical methods. These components include sampling, partitioning, and model
comparison. In addition, they include a component for scoring new data and defining a more meaningful
definition of a “best” model. Best does not always mean the most accurate. Often in healthcare, we can
Search WWH ::




Custom Search