Table. Results of the Multinomial Logistic Regression Mixed-Effects Model to Estimate Odds of a Higher Pain Rating^a.

Variable	Odds ratio (95% CI)	P value	Interpretation
Sex of the rater	1.10 (0.88-1.37)	.42	No significance between age or sex of the rater and pain ratings
Age of the rater	0.99 (0.94-1.05)	.77
Medical training year	1.28 (1.11-1.47)	.001	Trainees with higher training levels were more likely to give a higher pain rating
Scenario in the case (kidney stone vs fracture)	0.46 (0.37-0.58)	<.001	Cases with fracture pain were more likely to receive higher pain ratings than cases with a kidney stone
Rater = Gemini Pro vs trainee	0.70 (0.47-1.03)	.07	No significant difference was found between the pain ratings from LLMs vs trainees
Rater = GPT-4 vs trainee	1.07 (0.45-2.53)	.88
(Race of the patient in the case = Black vs White) × false beliefs	0.09 (0.01-0.64)	.02	A significant interaction was found between race of the patient and presence of false beliefs, with Black patients less likely to receive a higher pain rating in the presence of false beliefs; this association is independent of the rater (ie, humans vs LLMs)
(Rater = Gemini Pro vs trainee) × false beliefs	0.71 (0.12-4.37)	.71
(Rater = GPT-4 vs trainee) × false beliefs	0.14 (0.00-1312.63)	.68

Abbreviations: GPT-4, Generative Pre-trained Transformer 4; LLM, large language model.

^a The variables of race and false beliefs are not presented by themselves due to noninterpretability of their results in the presence of significant interaction.