TABLE 4.
Algorithmic diagnoses Estimated prevalence (difference between estimated and true prevalences) Estimated prevalence ratio (ratio of estimated to true prevalence ratios) |
|||||||
---|---|---|---|---|---|---|---|
ADAMS (true) diagnoses | Hurd probit model | Expert Model | Gradient boosting model | Conditional random forests | LASSO | Super Learner | |
True and estimated dementia prevalence | |||||||
Non-Hispanic white | 12.4% | 14.9% (2.5%) | 17.2% (4.9%) | 17.8% (5.4%) | 17.8% (5.4%) | 18.7% (6.4%) | 19.0% (6.6%) |
Non-Hispanic black | 22.2% | 28.0% (5.8%) | 27.4% (5.2%) | 30.0% (7.8%) | 26.3% (4.2%) | 32.0% (9.8%) | 30.8% (8.6%) |
Hispanic | 10.8% | 16.7% (5.9%) | 23.1% (12.3%) | 25.8% (14.9%) | 27.2% (16.4%) | 32.8% (22%) | 32.6% (21.8%) |
True and estimated prevalence ratios | |||||||
Non-Hispanic black vs. white | 1.79 | 1.88 (1.05) | 1.59 (0.89) | 1.69 (0.94) | 1.48 (0.83) | 1.71 (0.95) | 1.62 (0.90) |
Hispanic vs. Non-Hispanic white | 0.87 | 1.12 (1.28) | 1.34 (1.54) | 1.45 (1.65) | 1.53 (1.75) | 1.75 (2.01) | 1.72 (1.97) |
Non-Hispanic black vs. Hispanic | 2.05 | 1.68 (0.82) | 1.18 (0.58) | 1.17 (0.57) | 0.97 (0.47) | 0.97 (0.47) | 0.94 (0.46) |