Table 2

Validation of models’ performance. Hold-out and external validation performance scores for multiclass and binary predictions. External validation was conducted on patients from the POMA study

Multiclass predictions
Validation	Model	N features	AUC-PRC (95% CI)	AUC-ROC (95% CI)	F1 score (95% CI)	Precision (95% CI)	Recall (95% CI)
Hold-out	AP1_mu	11	0.649 (0.646 to 0.652)	0.847 (0.847 to 0.847)	0.529 (0.528 to 0.530)	0.559 (0.532 to 0.586)	0.633 (0.633 to 0.633)
	AP5_mu	304	0.665 (0.661 to 0.669)	0.842 (0.841 to 0.843)	0.488 (0.485 to 0.491)	0.500 (0.469 to 0.531)	0.614 (0.612 to 0.616)
	AP5_top5_mu	5	0.650 (0.647 to 0.653)	0.846 (0.845 to 0.847)	0.552 (0.549 to 0.555)	0.558 (0.537 to 0.579)	0.629 (0.627 to 0.631)
External	AP1_mu	11	0.727 (0.726 to 0.728)	0.881 (0.881 to 0.881)	0.593 (0.592 to 0.594)	0.637 (0.631 to 0.643)	0.670 (0.670 to 0.670)
External	AP5_top5_mu	5	0.589 (0.583 to 0.595)	0.838 (0.836 to 0.840)	0.563 (0.555 to 0.571)	0.566 (0.552 to 0.580)	0.596 (0.585 to 0.607)

Binary predictions
Validation	Model	N features	AUC-PRC (95% CI)	AUC-ROC (95% CI)	F1 score (95% CI)	Precision (95% CI)	Recall (95% CI)
Hold-out	AP1_bi	11	0.637 (0.636 to 0.638)	0.700 (0.699 to 0.701)	0.631 (0.630 to 0.632)	0.669 (0.666 to 0.672)	0.666 (0.666 to 0.666)
	AP5_bi	304	0.598 (0.590 to 0.606)	0.675 (0.674 to 0.676)	0.622 (0.619 to 0.625)	0.636 (0.633 to 0.639)	0.648 (0.645 to 0.651)
	AP5_top5_bi	5	0.618 (0.613 to 0.623)	0.693 (0.689 to 0.697)	0.660 (0.656 to 0.664)	0.675 (0.670 to 0.680)	0.680 (0.676 to 0.684)
External	AP1_bi	11	0.764 (0.762 to 0.766)	0.780 (0.780 to 0.780)	0.714 (0.713 to 0.715)	0.733 (0.732 to 0.734)	0.726 (0.726 to 0.726)
External	AP5_top5_bi	5	0.688 (0.683 to 0.693)	0.702 (0.691 to 0.713)	0.653 (0.615 to 0.691)	0.669 (0.641 to 0.697)	0.663 (0.626 to 0.700)

AP, AutoPrognosis; AUC-PRC, area under the precision-recall curve; AUC-ROC, area under the receiver operating characteristic curve; POMA, Pivotal Osteoarthritis Initiative MRI Analyses.