Background
Little is known regarding the reliability of hospital coding for race and ethnicity particularly for Hispanics, Asians, and American Indians. We assessed these questions using hospital and mortality data from a large, diverse state.
Material and Method
We used California hospital discharge data 1998--2000 and linked death data. We assessed crude agreement and kappas for race and ethnicity coding between admissions to different hospitals and between hospital admissions and death data.
Results
Overall kappas for race and ethnicity were higher between hospital and death data than between different hospitals. Kappas between hospital admission and death certificate data differed by race and ethnicity. Reasonable kappas were observed for Blacks (0.92), Asians and Pacific Islanders (0.88), Hispanics (0.77), and Whites (0.76), but not for American Indians (0.27) or others (<0.01). Crude agreement for race and ethnicity varied for Asians and Hispanics based on country of origin. It was highest for persons born in China (93%) and Southeast Asian (93%) and Mexico (82%) and lowest for persons born in India (38%) and for "Other Spanish/Hispanics" born in (42%) and outside the US (46%).
Conclusions
Race and ethnicity coding agreement between hospitals and between hospitals and death certificate is generally good for most major and ethnic groups and subgroups with the notable exception of American Indians, persons of other races, and selected Asians and Hispanics born in certain countries.

Keywords: Medical Records - statistics & numerical data, Mortality - ethnology, Forms and Records Control - methods, Death Certificates, Ethnic Groups - statistics & numerical data, Cross-Sectional Studies, Continental Population Groups - statistics & numerical data