Statistical Methods - Text Mining Techniques for Healthcare Provider Quality Determination - page 112

Figure 1. Predictive modeling of patient outcomes
There is still limited use of predictive modeling, with the exception of regression models, in medical
studies. Most of the use of predictive modeling is fairly recent. (Sylvia et al., 2006) While most predictive
models are used for examining costs (Powers, Meyer, Roebuck, & Vaziri, 2005), they can be invaluable
in improving the quality of care. (Hodgman, 2008; Tewari et al., 2001; Weber & Neeser, 2006; Whitlock
& Johnston, 2006) One recent study does indicate that predictive modeling can be used to target the most
high risk patients for more intensive case management. (Weber & Neeser, 2006) It has also been used to
examine workflow in the healthcare environment. (Tropsha & Golbraikh, 2007) Some studies focus on
particular types of models such as neural networks. (Gamito & Crawford, 2004) In many cases, admin-
istrative (billing) data are used to identify patients who can benefit from interventions, and to identify
patients who can benefit the most. Most of the use of predictive modeling is fairly recent.
This chapter is not intended to be a complete discussion of predictive modeling; it provides a basic
introduction to its use in defining severity indices and risk models. For a more complete discussion,
the reader is referred to Cerrito (2007, 2008). However, we will briefly examine two commonly used
predictive models: neural networks and decision trees.
Neural networks act like black boxes. There is no definite model or equation, and the model is not
presented in the concise format available for regression. Its accuracy is examined similar to the diagnostics
of the regression curve, including the misclassification rate, the AIC (Akaike's Information Criterion),
and the average error. The simplest neural network contains a single input (an independent variable) and
a single target (a dependent variable) with a single output. Its complexity increases with the addition of
hidden layers and additional input variables (Figure 2).