Table 2.

Performance of R classifiers when using all the available clinician grades as features

Features	# Samples	AUC±SD	P value vs Referent
Telephonic screening interview
Manual grades	2756	62.8±0.4	Referent
Manual Grades+Gender+BMI		66±0.5	<0.001
Best automated		63.8±0.2	0.15
Manual+Automated		65.6±0.3	<0.001
Removing widespread pain	1374	61±0.2	0.51
Clinic
Manual grades	2756	66.4±0.2	Referent
Manual Grades+Gender+BMI		68.8±0.2	<0.001
Best automated		65.6±0.9	0.29
Manual+Automated		63±0.3	0.01
Removing widespread pain	1374	61±0.2	0.02
Self-assessed questionnaire (HOME)
Manual grades	2756	66.7±0.3	Referent
Manual Grades+Gender+BMI		68.9±0.4	<0.001
Best automated		67.7±0.3	0.30
Manual+Automated		68±0.2	0.05
Removing widespread pain	1374	69±0.2	0.10
Consistent pain (answered yes to pain at all time points)
Manual grades	1066	73.9±0.5	Referent
Manual Grades+Gender+BMI		76.1±0.2	0.01
Best automated		73.1±0.7	0.97
Manual+Automated		75.6±0.6	0.14
Removing widespread pain	565	78±1	0.04

The p values compare the AUCs with the referent in that pain group. For example, for telephone screening, compared with manual grades, none of the other approaches was significant.* Sometimes these p values show significantly worse AUCs than the referent.

*Comparison eliminating participants with widespread pain was performed using manual grades+Gender+BMI features.

Bold values correspond to the best results.