The graph illustrates that almost everyone (except for possibly one) is getting worse. A question of some interest to researchers is to predict who is getting worse, in other words, to analyze the slopes with respect to some covariates.

An immediate question is whether the slopes are correlated to the intercepts. Then, for example, a subject with a low score at their first visit to a clinician would be likely to have a steep slope. As it turns out, this question is not terribly precise, since there is some artifactual correlation. A better question, then, is whether there is any correlation beyond the artifactual correlation.

There is still room for disagreement as to whether this is a useful question. The intercepts are defined as being the score at first visit at the ADRC, and the clinical usefulness of this score is debateable. There has been some effort spent on builing models around a definition of the intercept as the score at inception, and much of this work is based on similar work in HIV modelling. However, the time of inception is more difficult to define in AD than in HIV, and certainly more difficult to estimate. And its our position that we would prefer to work with a well-defined value of debateable clinical meaning rather than a poorly estimated value (and possibly poorly defined ) with very strong clinical meaning.

But this discussion has nothing to do with BLUPs, per se, and more rightly belongs on the AD page. However, during the construction process, I expect these two pages to get shuffled together quite a bit.