Related Experiment Video
Updated: Aug 19, 2026
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
Using classification and regression trees (CART) to investigate unknown ethnicity records in NHS emergency care data
1NHS England, Leeds, UK.
Objective:
High-quality ethnicity data are necessary for tackling health inequalities. This study aims to support the improvement of ethnicity recording in emergency care data by (a) investigating variation in unknown ethnicity records, (b) exploring patterns of missingness between ethnicity and other variables and (c) identifying the variables most important in the recording of an unknown ethnicity.
Design:
Secondary analysis of the Emergency Care Data Set (ECDS).
Setting And Participants:
National Health Service (NHS) and independent sector organisations who provide emergency care services in England. Includes all anonymised ECDS records from the financial year 2023/2024, excluding those with a 'Null' anonymised pseudo-NHS number or those patients announced as dead on arrival (24 167 154 records after exclusions).
Results:
Differences in the percentage of unknown ethnicity records were seen across ages, organisation types and attendance characteristics. An increase in the percentage of missing values was observed for 62 variables when ethnicity was unknown, with the largest increases shown by variables related to investigations and treatments received by the patient (15.31% and 14.74%). Provider site code was identified as the most important variable in recording ethnicity as unknown. Further analysis highlighted a subset of acute trusts and independent sector organisations that disproportionately contributed to unknown ethnicity records.
Conclusions:
This study has improved understanding of unknown ethnicity recording by exploring variation across demographics and organisation types and demonstrating an increase in missingness across other ECDS variables when ethnicity is unknown. Unknown ethnicity recording appears to be a site-specific problem, which should be addressed through improvements to the data collection processes of individual providers. These insights should be used to improve the quality of ethnicity coding within emergency care data, necessary for tackling health inequalities.
