Wij = (1, t;j, (tij - 2)+), where t;j E {0,2,6, 12, 18} and z+ = max(z,O). The three columns of W; correspond to individual-level intercept, slope, and change in slope following the change-point, respectively. We account for the effect of covariates by including them in the fixed-effect design matrix Xi. Specifically, we set

X; = (W;, d;W;, a; W;), (17.2) where d; is a binary variable indicating whether patient i received ddI (d; = 1) or ddC (d; = 0), and a; is another binary variable telling whether the patient was diagnosed as having AIDS at baseline ( a; = 1) or not (a; = 0). Notice from (17.2) that p = 3q = 9; the two covariates are being allowed to affect any or all of the intercept, slope, and change in slope of the overall population model. The corresponding elements of the a vector then quantify the effect of the covariate on the form of the CD4 curve. In particular, our interest focuses on the a parameters corresponding to drug status, and whether they differ from 0. The /3; are of secondary importance, though they would be critical if our interest lay in individuallevel prediction.