Why standard methods miss ethnic differences in patient mortality risk

Patient risk varies by ethnicity and clinical subgroup, but standard statistical methods struggle with this—they're tripped up by model misspecification and regularization bias. This paper uses Neyman orthogonality, a technique that insulates estimates from nuisance parameter errors, to reliably detect risk heterogeneity across populations. On simulated data, the method cuts bias substantially; on real ICU data, it uncovers ethnicity-specific mortality patterns that likelihood-based methods miss.