NU Learning Phase Three:

EXPLORE

In the third (but most important) phase of NU Learning, the outcomes researcher performs

Systematic Sensitivity Analyses!

Specifically, he/she explores the implications of using alternative ways to form clusters.  For example, what sorts of LTD distributions result from (simultaneously) varying the following three NU analysis decisions?

[1] choice of  number of clusters,

[2] choice of clustering algorithm / strategy, and

[3] choice of only a subset of the available patient X-characteristics.

This third phase can be quite tedious and repetitive if the researcher has to specify each alternative analysis to be performed.  Ideally, software for NU Learning would minimize the need for user input when performing sensitivity analyses ...via some form of “batch mode” processing.

The primary objective in Phase Three is to identify, say, the three most “interesting” LTD distributions that are typical (representative) of all of the salient LTD distributions that have been identified in sensitivity analyses.  For example, which distribution is typical of the salient distributions most favorable to a treatment?  Which distribution is most representative of the LTD distributions least favorable to a treatment?  And which LTD distribution is most typical of all salient LTD distributions?

Phase Four (REVEAL) analyses should concentrate upon these most typical LTD Distributions.