Table 5.
The performance metrics of the neural network model using stochastic gradient descent using the default threshold (50%) after matching.
| Sociodemographic groups | Recall | Specificity | Accuracy | AUC a | Differenceb | P value | |||
| Sex | 2.02 | <.001 | |||||||
|
|
Male | 50.00 | 97.67 | 97.47 | 92.94 |
|
|
||
|
|
Female | 51.85 | 97.84 | 97.46 | 96.10 |
|
|
||
| Marital status | 13.02 | <.001 | |||||||
|
|
Never been married | 58.82 | 97.74 | 97.54 | 95.06 |
|
|
||
|
|
Other groups of marital status | 45.83 | 97.77 | 97.38 | 94.83 |
|
|
||
| Working condition | 19.91 | <.001 | |||||||
|
|
Working ≥35 hours | 66.67 | 97.84 | 97.66 | 96.61 |
|
|
||
|
|
Other groups of working condition | 46.88 | 97.72 | 97.40 | 94.43 |
|
|
||
| Race | 60.16 | <.001 | |||||||
|
|
White | 58.62 | 97.46 | 97.21 | 95.31 |
|
|
||
|
|
Black | 0.00 | 99.00 | 98.51 | 84.33 |
|
|
||
| Income | 3.42 | <.001 | |||||||
|
|
An income of <US $20,000 | 50.00 | 96.54 | 96.33 | 94.54 |
|
|
||
|
|
Other groups of income | 51.61 | 98.35 | 98.02 | 95.48 |
|
|
||
aAUC: area under the curve.
bDifference between recall and specificity values.