Table 2. Significance of Associations for Version and Subjective Numeracy with Ordinal PMV.

Model #	Effect	c*	Wald χ2	df	p-value
0-a	Version	0.562	8.22	4	0.08
0-b	Version	0.616	9.43	4	0.05
	Chosen physician		17.05	4	0.002
0-c	Subjective numeracy	0.611	20.99	1	<0.0001
1	Version	0.657	10.67	4	0.03
	Subjective numeracy		20.94	1	<0.0001
	Chosen physician		15.43	4	0.004
2	Version	0.676	11.50	4	0.02
	Subjective numeracy		19.01	1	<0.0001
	Subjective numeracy*version		13.16	4	0.01
	Chosen physician		13.81	4	0.01

Models for 3-level ordinal outcome PMV are proportional-odds logit models. The “risks > benefits” category is the PMV reference level. Model 1: The main effect terms in the model are subject’s chosen physician (to control potential confounding effects), randomly assigned version, and subjective numeracy. Subjective numeracy as a continuous variable was associated with PMV (OR (95% CI) for Model 0-c = 0.64 (0.53, 0.78) and for Model 1 = 0.63 (0.52, 0.77). Model 2: A version-by-subjective numeracy interaction term was included to examine whether subjective numeracy modified the effect of the specified communication strategy on PMV.

^*c is the ‘concordance index’, a measure of association of predicted probabilities and observed responses, calculated as the area under the ROC curve.