Table 3 Predictive performance of the individual dementia type-specific models.
From: Machine learning models identify predictive features of patient mortality across dementia types
Dementia type (sample size [n] in internal test set/external test set) | Internal test set | External test set | ||||
|---|---|---|---|---|---|---|
Accuracy (95% CI) | AUC-ROC (95% CI) | AUC-PR (95% CI) | Accuracy | AUC-ROC | AUC-PR | |
No dementia (n = 8427/11800) | 0.834 (0.827–0.844) | 0.873 (0.859–0.879) | 0.513 (0.475–0.546) | 0.696 | 0.842 | 0.336 |
Alzheimer’s disease (n = 7598/7351) | 0.774 (0.766–0.783) | 0.854 (0.845–0.863) | 0.790 (0.769–0.805) | 0.683 | 0.827 | 0.695 |
Missing/unknown (n = 1264/191) | 0.808 (0.773–0.834) | 0.862 (0.813–0.889) | 0.504 (0.477–0.629) | 0.838 | 0.794 | 0.263 |
Frontotemporal lobar degeneration (n = 858/1058) | 0.714 (0.683–0.745) | 0.796 (0.760–0.821) | 0.810 (0.735–0.819) | 0.695 | 0.772 | 0.677 |
Lewy body dementia (n = 637/561) | 0.719 (0.696–0.769) | 0.796 (0.780–0.842) | 0.806 (0.771–0.846) | 0.717 | 0.807 | 0.763 |
Vascular brain injury or vascular dementia (n = 458/507) | 0.751 (0.727–0.803) | 0.839 (0.796–0.871) | 0.752 (0.681–0.815) | 0.712 | 0.797 | 0.640 |
Cognitive impairment for other specified reasons (n = 273/495) | 0.780 (0.744–0.828) | 0.832 (0.795–0.900) | 0.665 (0.469–0.731) | 0.786 | 0.833 | 0.421 |
Depression (n = 271/235) | 0.815 (0.790–0.889) | 0.800 (0.732–0.895) | 0.408 (0.228–0.606) | 0.813 | 0.785 | 0.421 |
