Table 2.

Summary of key effects of generalized estimating equation (GEE) and linear mixed model (LMM) analyses.

Effect and metric		Modela	βb (95% CI)	P value (raw)	P value (FDRc)d	P value (global FDR)e
Language main effect (English vs Chinese)
	Accuracy	GEEf	0.00 (0 to 0)	—g	—	—
	Comprehensiveness	LMMh	0.01 (−0.32 to 0.35)	.94	.97	—
	Precision	LMM	0.01 (−0.33 to 0.35)	.94	.97	—
	Understandability	LMM	0.02 (−0.32 to 0.36)	.91	.96	—
Expert prompt (vs no-context)
	Accuracy	GEE	−0.39 (−0.93 to 0.16)	.17	.47	—
	Comprehensiveness	LMM	−0.56 (−0.89 to −0.22)	.001	.003	—
	Precision	LMM	−0.66 (−1.00 to −0.32)	.001	.001	—
	Understandability	LMM	−0.30 (−0.64 to 0.04)	.09	.22	—
DeepSeek×English (2-way interaction)
	Accuracy	GEE	−0.54 (−1.29 to 0.21)	.16	.47	—
	Comprehensiveness	LMM	−0.28 (−0.75 to 0.19)	.25	.46	.56
	Precision	LMM	−0.23 (−0.71 to 0.25)	.35	.50	.62
	Understandability	LMM	−0.73 (−1.21 to −0.25)	.003	.01	.02
Grok×English×expert (3-way interaction)
	Accuracy	GEE	−1.88 (−2.94 to −0.82)	.001	.02	—
	Comprehensiveness	LMM	−0.76 (−1.43 to −0.09)	.03	.07	.11
	Precision	LMM	−0.78 (−1.46 to −0.10)	.03	.06	.11
	Understandability	LMM	−0.79 (−1.47 to −0.11)	.02	.06	.10

^aReference categories: model—ChatGLM, language—Chinese, and prompt—no-context.

^bβ: regression coefficient on the log-odds scale (GEE) or the raw score scale (LMM).

^cFDR: false discovery rate.

^dBenjamini-Hochberg false discovery rate–corrected P value within each metric.

^eCorrection was applied jointly across all 4 metrics for the interaction terms.

^fGEE: binary accuracy, logit link, and exchangeable correlation clustered on question ID.

^gGlobal false discovery rate was not applicable.

^hLMM: comprehensiveness, precision, and understandability as separate dependent variables.