Table 1.
Performance evaluation of machine-learned models.
| Disease and data set | Algorithm | AUCa (95% CI) | Optimal SNb/SPc | SP at SN=0.8 | Sample size, nd | Cases, n (%) | |||||
| DKDe | |||||||||||
|
|
SEEDf | LRg | 0.743 (0.698-0.787) | 0.674 | 0.54 | 2653 | 517 (19.5) | ||||
|
|
SEED | LASSOh | 0.832 (0.794-0.871) | 0.757 | 0.703 | 2666 | 532 (20) | ||||
|
|
SEED | GBDTi | 0.838 (0.801-0.874) | 0.751 | 0.709 | 2668 | 529 (19.8) | ||||
|
|
UKBBj | LR | 0.691 (0.683-0.699) | 0.635 | 0.472 | 5236 | 345 (6.6) | ||||
|
|
UKBB | LASSO | 0.791 (0.784-0.797) | 0.723 | 0.636 | 5090 | 333 (6.5) | ||||
|
|
UKBB | GBDT | 0.738 (0.719-0.756) | 0.666 | 0.535 | 5543 | 366 (6.6) | ||||
| DRk | |||||||||||
|
|
SEED | LR | 0.764 (0.722-0.806) | 0.696 | 0.594 | 2597 | 653 (25.1) | ||||
|
|
SEED | LASSO | 0.779 (0.736-0.822) | 0.708 | 0.596 | 2514 | 628 (25) | ||||
|
|
SEED | GBDT | 0.790 (0.748-0.831) | 0.709 | 0.616 | 2598 | 655 (25.2) | ||||
|
|
UKBB | LR | 0.760 (0.755-0.765) | 0.707 | 0.571 | 5492 | 336 (6.1) | ||||
|
|
UKBB | LASSO | 0.778 (0.773-0.782) | 0.716 | 0.617 | 4678 | 280 (6) | ||||
|
|
UKBB | GBDT | 0.778 (0.769-0.786) | 0.715 | 0.604 | 4833 | 296 (6.1) | ||||
aAUC: area under the receiver operating characteristic curve.
bSN: sensitivity.
cSP: specificity.
dFinal sample size after variable selection and missing data removal.
eDKD: diabetic kidney disease.
fSEED: Singapore Epidemiology of Eye Diseases.
gLR: logistic regression.
hLASSO: Least Absolute Shrinkage and Selection Operator.
iGBDT: gradient-boosting decision tree.
jUKBB: UK biobank.
kDR: diabetic retinopathy.