Table 1.

Multilevel logistic regression analysis of radiologists’ diagnoses by specific intervention type: with versus without AI assistance

	No time pressure		2-minute time pressure
	OR (SE)	P-value	OR (SE)	P-value
Probability of correct diagnoses for all cases (Accuracy)
Intercept	2.88 (1.21)	.012*	0.81 (0.31)	.588
AI assistance	1.58 (0.31)	.021*	2.13 (0.44)	.000***
Probability of correct diagnoses for invasive cases (Sensitivity)
Intercept	4.96 (3.43)	.021*	0.80 (0.50)	.725
AI assistance	2.45 (0.72)	.002**	2.20 (0.69)	.012*
Probability of correct diagnoses for minimally/noninvasive cases (Specificity)
Intercept	2.06 (1.09)	.171	0.80 (0.34)	.602
AI assistance	1.09 (0.30)	.760	2.08 (0.57)	.007**
Probability of correct diagnoses for all cases by radiologists with low self-efficacy (Accuracy of radiologists with low self-efficacy)
Intercept	2.48 (1.13)	.046*	0.79 (0.35)	.590
AI assistance	1.55 (0.34)	.045*	2.02 (0.48)	.004**
Probability of correct diagnoses for all cases by radiologists with high self-efficacy (Accuracy of radiologists with high self-efficacy)
Intercept	5.43 (6.73)	.173	0.95 (0.55)	.923
AI assistance	1.35 (0.81)	.615	2.80 (1.51)	.055+

Note: Controlling for gender and years of experience in CT diagnosis.

^*** P < .001; **P < .01; *P < .05; +P < .1. OR: odds ratio; SE: standard error.