Table 3.
Performance of predicting lifetime history of MDD.
| No. of subjects (no. of MDD subjects) | No. of features | F1-score | Sensitivity | Specificity | PPV | NPV | |
|---|---|---|---|---|---|---|---|
| Single modality of digital features | |||||||
| Subjective happiness levela | 81 (40) | 1 | 0.73 | 0.83 | 0.59 | 0.66 | 0.77 |
| Actigraphyb | 74 (36) | 17 | 0.71 | 0.78 | 0.61 | 0.65 | 0.74 |
| Facial expressionb | 75 (37) | 5 | 0.69 | 0.86 | 0.37 | 0.57 | 0.74 |
| Voiceb | 69 (33) | 4 | 0.77 | 0.88 | 0.64 | 0.69 | 0.85 |
| NLPb | 70 (34) | 2 | 0.75 | 0.85 | 0.61 | 0.67 | 0.81 |
| All digital modalitiesb | 60 (29) | 29 | 0.81 | 0.86 | 0.74 | 0.76 | 0.85 |
| All digital modalities + HADS-Db | 56 (26) | 30 | 0.77 | 0.85 | 0.70 | 0.71 | 0.84 |
| HADS-Da | 76 (36) | 1 | 0.76 | 0.75 | 0.80 | 0.77 | 0.78 |
NLP natural language processing, HADS-D Hospital Anxiety and Depression Scale—Depression subscale.
aCutpoint optimization for continuous one-dimensional feature.
bArtifical Neural Networks.