Determine whether subgroup performance differences reflect genuine disparities or sampling imprecision

Determine whether the comparatively low AUROC observed for the Hispanic/Latino subgroup in PRISM reflects a genuine performance difference or sampling imprecision.

Background

The Hispanic/Latino subgroup has the lowest point-estimate AUROC among the reported ethnicity groups, but its confidence interval is wide and overlaps the point estimate of every other ethnicity group. Consequently, the paper does not resolve whether the observed difference represents a real subgroup effect or uncertainty caused by limited sampling.

References

Whether this reflects a genuine difference or sampling imprecision cannot be settled here, and we flag it for further evaluation before deployment.

— Retrieval-Augmented Multimodal Language Model with Heterogeneous Evidence Integration for Clinical Prognosis  (2610.13510 - Shafiq et al., 9 Oct 2026) in Appendix I, Section 'Performance by Patient Subgroup'