Table 3.
Group | Studies, Na | Sample size, N | Accuracy (%), range | Pooled mean accuracy, mean (95% CI) | Heterogeneity measures | Test for subgroup differences (P value) | |||||||||||||
|
|
|
|
|
τ2 | Q (P value) | I2 (%) |
|
|||||||||||
Algorithms | .07 | ||||||||||||||||||
|
Support vector machine | 7 | 21,413 | 0.50-0.99 | 0.82 (0.67-0.94) | 0.0520 | 819.0 (<.001) | 99.3 |
|
||||||||||
|
Random forest | 6 | 22,132 | 0.56-0.99 | 0.83 (0.68-0.94) | 0.0426 | 1187.6 (<.001) | 99.6 |
|
||||||||||
|
Decision tree | 4 | 21,785 | 0.70-0.99 | 0.87 (0.68-0.98) | 0.0585 | 1164.3 (<.001) | 99.7 |
|
||||||||||
|
Multilayer perceptron | 3 | 504 | 0.71-0.87 | 0.81 (0.70-0.90) | 0.0087 | 8.3 (.02) | 75.8 |
|
||||||||||
|
Logistic regression | 2 | 93 | 0.70-0.71 | 0.71 (0.61-0.80) | 0.0000 | 0.0 (.98) | 0.0 |
|
||||||||||
|
XGBoost | 2 | 1239 | 0.55-0.67 | 0.62 (0.50-0.73) | 0.0070 | 12.7 (<.001) | 92.1 |
|
||||||||||
|
Long short-term memory networks | 2 | 10,695 | 0.67-0.69 | 0.67 (0.66-0.69) | <0.0001 | 1.2 (.27) | 17.7 |
|
||||||||||
|
Ensemble model | 2 | 605 | 0.91-0.94 | 0.92 (0.89-0.94) | 0.0003 | 1.4 (.24) | 28.6 |
|
||||||||||
|
K-nearest neighbor | 2 | 61,022 | 0.62-0.99 | 0.88 (0.32-1.00) | 0.1672 | 15.0 (<.001) | 93.4 |
|
||||||||||
Aims of AIb | .33 | ||||||||||||||||||
|
Detectionc | 33 | 143,800 | 0.50-0.99 | 0.84 (0.72-0.91) | 0.2857 | 62,108.0 (<.001) | 99.9 |
|
||||||||||
|
Prediction | 7 | 6109 | 0.55-0.81 | 0.72 (0.66-0.78) | 0.0082 | 117.9 (<.001) | 94.9 |
|
||||||||||
Status of WDd | .91 | ||||||||||||||||||
|
Commercialc | 27 | 130,279 | 0.55-0.99 | 0.82 (0.68-0.91) | 0.3345 | 28,205.8 (<.001) | 99.9 |
|
||||||||||
|
Noncommercialc | 11 | 16,610 | 0.67-0.95 | 0.85 (0.71-0.92) | 0.0471 | 1363.3 (<.001) | 99.4 |
|
||||||||||
WDs | .12 | ||||||||||||||||||
|
Musec | 6 | 279 | 0.71-0.88 | 0.77 (0.67-0.85) | 0.0000 | 9.0 (.11) | 46.2 |
|
||||||||||
|
Empatica E4c | 5 | 121,048 | 0.86-0.99 | 0.97 (0.00-0.99) | 1.0715 | 1722.8 (<.001) | 100 |
|
||||||||||
|
Fitbit | 3 | 393 | 0.56-0.89 | 0.70 (0.45-0.89) | 0.0453 | 52.4 (<.001) | 96.2 |
|
||||||||||
Data sources | .59 | ||||||||||||||||||
|
WD-basedc | 27 | 141,516 | 0.50-0.99 | 0.81 (0.64-0.90) | 0.3498 | 59,871.8 (<.001) | 99.9 |
|
||||||||||
|
WD-based and othersc | 13 | 8393 | 0.67-0.95 | 0.86 (0.75-0.92) | 0.0552 | 622.7 (<.001) | 98.4 |
|
||||||||||
Data types | .48 | ||||||||||||||||||
|
Activity datac | 8 | 18,619 | 0.67-0.94 | 0.88 (0.62-0.96) | 0.1133 | 1041.3 (<.001) | 99.7 |
|
||||||||||
|
Activity data and othersc | 12 | 7675 | 0.55-0.95 | 0.78 (0.57-0.90) | 0.1492 | 573.6 (<.001) | 99.1 |
|
||||||||||
|
EDAe data and othersc | 9 | 122,650 | 0.71-0.99 | 0.92 (0.55-0.99) | 0.6718 | 5870.6 (<.001) | 99.9 |
|
||||||||||
|
EEGf datac | 6 | 279 | 0.71-0.88 | 0.78 (0.67-0.85) | 0.0000 | 9.0 (.11) | 46.2 |
|
||||||||||
Reference standards | .80 | ||||||||||||||||||
|
STAIb,g | 8 | 398 | 0.58-0.88 | 0.73 (0.61-0.82) | 0.0087 | 19.5 (.006) | 61.7 |
|
||||||||||
|
CIDIh | 2 | 529 | 0.55-0.94 | 0.77 (0.33-1.00) | 0.1099 | 117.5 (<.001) | 99.1 |
|
||||||||||
|
DAMSi | 2 | 296 | 0.56-0.89 | 0.75 (0.39-0.98) | 0.0691 | 27.6 (<.001) | 96.4 |
|
||||||||||
Validation methods | .41 | ||||||||||||||||||
|
K-foldc | 24 | 129,113 | 0.55-0.99 | 0.86 (0.70-0.94) | 0.3875 | 24,618.6 (<.001) | 99.9 |
|
||||||||||
|
Hold-outc | 10 | 18,959 | 0.50-0.92 | 0.76 (0.57-0.87) | 0.0303 | 901.7 (<.001) | 99.3 |
|
||||||||||
|
Leave-one-out | 2 | 582 | 0.56-0.74 | 0.66 (0.49-0.82) | 0.0141 | 7.1 (.008) | 86.0 |
|
||||||||||
All studiesc | 40 | 149,909 | 0.50-1.00 | 0.82 (0.71-0.89) | 0.2713 | 75,900.5 (<.001) | 99.9 | N/Aj |
aMany studies were included more than once in each meta-analysis given that they assessed the performance of more than one algorithm.
bAI: artificial intelligence.
cAccuracy was pooled using the multilevel meta-analysis method.
dWD: wearable device.
eEDA: electrodermal activity.
fEEG: electroencephalogram.
gSTAI: State-Trait Anxiety Inventory.
hCIDI: Composite International Diagnostic Interview.
iDAMS: Depression and Anxiety Mood Scale.
jN/A: not applicable.