Multivariable Analysis in Diagnostic Accuracy Studies: What are the Possibilities?
Buntinx F., Aertgeerts B., Aerts M., Bruyninckx R., Knottnerus JA., Van Den Bruel A., Van Den Ende J.
Diagnostic research requires multivariable analytical approaches to take the contributions of different tests to a diagnosis simultaneously into consideration. Tree-building methods, logistic regression analysis, and neural networks can provide solutions to this challenge. Latent class analysis adds a method that can be used in situations without a normal reference standard. For each method, we provide a short description, an overview of advantages and disadvantages, and a real-life example. Researchers should concentrate on either logistic regression analysis or classification and regression tree (CART) type methods, try to master it in detail and consequently use it, always keeping in mind that alternatives are available, each with their own advantages and disadvantages. © 2009 Blackwell Publishing.