Table 1.

ROC analyses: model statistics at a tangent for the change in the m-BPI-WPS score.

Pain score change (type)	PGIC	AUC	Sensitivity (%)	Specificity (%)	Value (change in the pain score)∗	Total accuracy (%)
Raw change	Very much improved	0.801	73.7	77.7	−4	77.1
Raw change	Much or very much improved	0.814	70.3	77.8	−3	74.5
Raw change	Minimally, much, or very much improved	0.784	69.2	74.1	−2	70.5
Percentage change∗	Very much improved	0.820	75.4	74.9	−50.0	75.0
Percentage change∗	Much or very much improved	0.823	76.2	72.8	−33.3	74.3
Percentage change∗	Minimally, much, or very much improved	0.790	72.7	73.3	−20.0	72.9

Percentage change = raw change in the BPI worst pain score/baseline pain score. ^∗The value of change in the pain score is defined by the intersection of a 45° tangent line with each ROC curve, which is mathematically equivalent to choosing the point at which sensitivity and specificity are the closest to being equal. AUC, area under the curve; m-BPI-WPS, modified Brief Pain Inventory-worst pain severity; PGIC, patient global impression of change; ROC, receiver operating characteristic.