Table 2.

Final multivariate pretreatment pain-prediction models and performance metrics^a

	Responders (N = 297/94)b			Future pain intensity (N = 289/94)b
Independent variables	OR	(95 % CI)	P-value		β	(95 % CI)	P-value
Dose (per 6 spinal manipulation visits)	1.27	(1.08, 1.49)	0.004		−1.86	(−3.35, −0.38)	0.014
Pain/Disability
Pain intensity					4.77	(1.85, 7.70)	0.001
Pain unpleasantness					3.29	(0.35, 6.24)	0.028
General Health
Comorbidities	0.81	(0.67, 0.97)	0.025
EuroQol – VAS					−2.20	(−4.00, −0.39)	0.017
EuroQol 5D – self-care (1–3)	0.64	(0.41, 0.99)	0.044
Objective Physical Exam
LBP: sum for 4 lumbar ROM pain scores	0.81	(0.67, 0.97)	0.024		2.89	(0.61, 5.16)	0.013
Performance metricsc	AUC	(95 % CI)		RMSE	(95 % CI)	R2	(95 % CI)
Training set	0.624			17.4		.268
Test set	0.479	(0.387, 0.575)		19.4	(17.0, 21.6)	.065	(−10.5, 21.9)

OR Odds ratio, PC part correlation, β regression coefficient, VAS visual analogue scale, AUC Area under the curve (receiver operating characteristic curve), RMSE root mean squared error (SD of prediction error), R ² coefficient of determination, LBP low back pain

^aVariables were selected into the regression models using forward selection among variables with p < .05 in the univariate analysis; dose was forced into the models. Independent variables were standardized except for dose (scale unit = 6 visits) and self-care (scale unit = 1 on a 1–3 scale). Lower scores were favorable for pain and self-care; higher scores for EuroQol VAS

^bThe first number is the sample size for the model in the training set and the second number is the N for the test set

^cChance performance is indicated by 0.5 for AUC. RMSE is the standard deviation of the error in prediction of future pain intensity evaluated on the 0 – 100 pain scale. R ² is the proportion of the variance in pain intensity explained by the independent variables in the model. Confidence intervals for the performance metrics are given for the test set only