Table 2.
Optimal parameters and performance per CMA method, based on 100 simulations of 1000 individuals.
| CMA9 | LCMA1 | LCMA1-thr | LCMA2 | LCMA2-thr | |
|---|---|---|---|---|---|
| Window size | – | 150 | 60 | 90 | 60 |
| Overlap | – | 80% | 70% | 0% | 0% |
| Mean cARI [95% CI] | 0.35 [0.33–0.36] | 0.65 [0.62–0.68] | 0.65 [0.62–0.68] | 0.72 [0.69–0.74] | 0.58 [0.55–0.62] |
| Classification accuracy [95% CI] | |||||
| Consistent adherence | 100% | 100% | 100% | 100% | 100% |
| [100–100%] | [100–100%] | [100–100%] | [100–100%] | [100–100%] | |
| Erratic adherence | 18% | 29% | 27% | 33% | 29% |
| [13–23%] | [23–35%] | [21–34%] | [28–39%] | [23–35%] | |
| Gradual decline | 53% | 74% | 65% | 85% | 77% |
| [45–62%] | [66–82%] | [57–74%] | [77–91%] | [70–85%] | |
| Intermittent adherence | 51% | 70% | 75% | 76% | 74% |
| [43–60%] | [62–77%] | [68–83%] | [69–83%] | [67–81%] | |
| Partial drop-offs | 51% | 81% | 84% | 86% | 89% |
| [43–59%] | [75–88%] | [77–90%] | [80–92%] | [85–95%] | |
| Non-persistence | 100% | 100% | 89% | 100% | 21% |
| [100–100%] | [100–100%] | [84–95%] | [100–100%] | [14–28%] | |
| Overall | 66% | 80% | 79% | 84% | 75% |
| [65–68%] | [79–82%] | [78–81%] | [83–86%] | [73–76%] | |
A classification accuracy of 100% indicates that the clustering algorithm correctly identified all individuals of a pre-allocated group. cARI: adjusted Rand Index for all groups, CI: confidence interval.