134 Chapter 6 Figure 1 Overall mean adherence trajectory estimated from 52 imputed datasets (thick grey), mean adherence trajectory estimated from each of the 52 imputed datasets separately (thin grey) and mean adherence trajectory of the imputed dataset (#24) with the smallest mean deviation from the overall mean adherence trajectory (black). Models with 1 to 4 classes were estimated using LCGA. The addition of a quadratic term for time did not increase the fit of the linear models. The 4-class solution showed optimal performance based on the maximum log-likelihood criterion, with a value of −3115.1, while the 2-class and 3-class models yielded lower log-likelihood scores of −3126.3 and − 3119.1, respectively. Similarly, the Akaike Information Criterion favored the 4-class solution with a value of 6254.2, compared to the 2-class (6264.7) and 3-class (6256.3) models. Conversely, the Bayesian Information Criterion favored the 2- class model with a lower value of 6283.6, in contrast to the 3-class (6284.6) and 4-class (6292.0) models. Furthermore, the entropy measure indicated a better fit for the 2-class model (0.61) compared to the 3-class (0.49) and 4-class (0.54) models. However, the 2-class model displayed 2 nearly parallel trajectories (Fig. 2), suggesting limited clinical significance and within the 4-class model, the fourth class contained less than 5% (4.6%) of the patient population. Therefore, the k-1 model (model 3) was chosen instead (Fig. 3).
RkJQdWJsaXNoZXIy MTk4NDMw