Table 6.

Results by model and outcome class for the model applied to different cancer sites.

Model, test dataset, and metrics			Bio+ClinicalBERT														Clinical-BigBird						Clinical-Longformer
			2-class							3-class							2-class			3-class			2-class			3-class
			Head		Tail		Head and tail		Head			Tail		Head and tail
Colorectal cancer model applied to breast cancer test dataset
	Sensitivity, %	76.3		76.8		76.3		76.1			75.9		73.0		84.4 a			83.8			77.1			82.0
	PPVb, %	83.9		85.1		81.6		78.3			87.0		83.3		81.4			83.3			87.0			82.7
	AUCc	0.96		0.96		0.94		0.95			0.95		0.95		0.97			0.97			0.95			0.96
	Scaled Brier score	0.62		0.63		0.61		0.58			0.65		0.60		0.65			0.68			0.69			0.66
	Charts reduced, %	88.2		88.3		87.9		87.4			88.7		88.6		86.6			86.9			88.5			87.1
Breast cancer model applied to colorectal cancer test dataset
	Sensitivity, %	69.4		63.4		75.4		70.7			63.0		70.3		68.5			83.6			86.8			82.4
	PPV, %	80.6		82.5		76.3		80.9			82.9		77.7		86.8			74.6			70.8			74.0
	AUC	0.97		0.96		0.97		0.97			0.96		0.97		0.97			0.98			0.98			0.98
	Scaled Brier score	0.55		0.52		0.55		0.57			0.55		0.55		0.58			0.61			0.55			0.61
	Charts reduced, %	90.8		91.8		89.5		90.7			91.9		90.4		91.2			88.1			86.9			88.1

^aItalicization indicates the highest value.

^bPPV: positive predictive value.

^cAUC: area under the curve.