Machine-learning vs. logistic regression for preoperative prediction of medical morbidity after fast-track hip and knee arthroplasty—a comparative study

Table 2 Performance of the models with a predefined positive prediction fraction of 20% for primary outcome

Positive prediction fraction 20%	TP/FP	FN/TN	Sensitivity/Precision %	MCC %	AUROC %	AUPRC %	Brier %	P (sensitivity) %
Full machine-learning model	106 / 676	76 / 3055	58.2 / 13.6	21.1	76.3	15.5	4.19	-
Full logistic regression model	97 / 685	85 / 3046	53.3 / 12.4	18.4	74.7	15.6	4.32	17.2
Parsimonious machine-learning model	100 / 682	82 / 3049	54.9 / 12.8	19.3	75.9	17.3	4.34	26.4
Parsimonious logistic regression model	90 / 692	92 / 3039	49.5 / 11.5	16.3	73.8	15.8	4.33	4.86
Age-only model	87 / 676	95 / 3055	47.8 / 11.4	15.8	69.7	12.1	38.8	3.55

TP true positives, FP false positives, FN false negatives, TN true negatives, MCC Matthews correlation coefficient, AUROC area under the operating receiver curve, AUPRC area under the precision recall curve P(sensitivity): probability that a model performs better than the full machine-learning model relative to sensitivity

ISSN: 1471-2253