Table IV.

Statistical difference analysis between different areas under the receiver operating characteristic curve of the best machine learning and non-machine learning method.

PROM	Best ML model	AUC	Comparison 1		Comparison 2
			Logistic regression (AUC)	p-value*	Pre-surgery PROM scores (AUC)	p-value*
Knee arthroplasty
EQ-5D-5L	RF	0.80	0.74	0.012‡	0.76	0.052†
EQ-VAS	Elastic net	0.76	0.76	0.401	0.75	0.519
KOOS-PS	Elastic net	0.76	0.76	0.186	0.74	0.355
Hip arthroplasty
EQ-5D-5L	GBM	0.81	0.81	0.745	0.79	0.242
EQ-VAS	LASSO	0.84	0.84	0.597	0.80	0.034‡
HOOS-PS	Ridge	0.71	0.67	0.017‡	0.58	0.011‡

^*p-value for statistical difference of the AUCs of the compared models.

^†Indicates statistical significance at the 10% level.

^‡Indicates statistical significance at the 5% level.

^§Indicates statistical significance at the 1% level.

AUC, area under the curve; EQ-5D-5L, EuroQol five-dimension five-level questionnaire; GBM, gradient-boosting model; HOOS-PS, Hip disability and Osteoarthritis Outcome Score-Physical Function Short Form; KOOS-PS, Knee injury and Osteoarthritis Outcome Score-Physical Function Short Form; LASSO, least absolute shrinkage and selection operator; ML, machine learning; PROM, patient-reported outcome measure; RF, random forest; VAS, visual analogue scale.