Table 2.
Optimal clinical predictive and mechanistic models against the benchmark of 30% behavioral improvement
| Model type | Drug | Optimal features | (C, γ) | Accuracy | Significancea |
|---|---|---|---|---|---|
| Clinical | Atomoxetine | L mean diffusivity,b levodopa equivalent dose, R fractional anisotropy, L fractional anisotropy | (25, 23) | 76.5% | p < 0.05 |
| Clinical | Citalopram | R fractional anisotropy, age, R mean diffusivity, MMSE | (210, 20.5) | 79.4% | p < 0.05 |
| Mechanistic | Atomoxetine | R caudate nucleus, L caudate nucleus, R pre‐SMA | (1, 27) | 85.3% | p < 0.05 |
| Mechanistic | Citalopram | L caudate nucleus, R putamen, R pre‐SMA | (26, 22) | 85.3% | p < 0.05 |
a
Statistical significance measured as p‐values from permutation tests (5000 randomizations, p < 0.05 corrected for multiple comparisons).
b
Values of fractional anisotropy and mean diffusivity were extracted from the anterior internal capsule. L, left; R, right; pre‐SMA, presupplementary motor area; MMSE, mimi mental state examination.