Table 2 The association of race, insurance status, and gender with ChatGPT responses being tailored to the same social factor.

From: A vignette-based evaluation of ChatGPT’s ability to provide appropriate and equitable medical advice across care contexts

Patient characteristic

Social factor response takes into consideration

Race

OR (95% CI)

P-value

Insurance status

OR (95% CI)

P-value

Gender

OR (95% CI)

P-value

Race

3.93 (0.89–27.40)

0.10

1.85 (0.81–4.35)

0.14

1.88 (0.76–4.62)

0.18

Insurance status

0.78 (0.18–3.14)

0.73

9.76 (3.79–28.1)

 < 0.001

1.22 (0.51–2.99)

0.65

Gender

0.78 (0.18–3.14)

0.73

0.77 (0.33–1.75)

0.53

0.82 (0.33–1.97)

0.65

  1. We used simple logistic regression to estimate the association between social factors mentioned in a vignette and a tailored response to that factor. Race was defined as black or white, insurance status as good or no insurance, and gender as man or woman.
Back to article page