Ethnicity, Equity, and AI study
OVERVIEW
The Covid pandemic has highlighted inequalities in health systems around the world. However, inequity is not limited to the pandemic – it is in fact a long-standing and multifaceted issue. In addition to socio-economic complexities, imbalances in healthcare technologies can worsen existing biases.
An example is the artificial intelligence technology behind clinical prediction models. If there are imbalances in the data used to train the models, or if there are algorithm biases within the analytical pipeline, the resulting models can be biased and result in mis-estimation of the health risks of patients in real-time. This in turn can lead to some groups of patients being under- or over-prioritised.
This research will develop prediction models that are based on bias-minimisation guidelines (developed by the Equator Centre UK housed in the Centre for Statistics in Medicine) and that are tailored to specific patient groups, including patients with different ethnic backgrounds, patients with rare conditions and patients with disabilities. By addressing any sources of bias in the data and in the analytical pipelines, prediction models can be made more targeted and equitable.
The project is conducted via Trusted Research Environments, such as NHS Digital and SAIL. The study uses routinely collected data from UK GDPPR/GPES, Hospital Episode Statistics (HES), and Office of National Statistics.
Patient and public engagement and involvement will be an important element of this research.
Study materials
Ethnicity, Data, Health Research poster - Punjabi
Ethnicity, Data, Health Research poster - Gujrati
Ethnicity, Data, Health Research poster - English
Ethnicity, Data, Health Research Project infographic
1/10 patients in England don't have an ethnicity record
Individuals with no ethnicity records tends to be younger and are more likely to be males than individuals with a recorded ethnicity.
Granularity of ethnicity concepts
Ethnicity data recorded in the National Health Service in the UK can be disaggregated from 6 the high-level ethnic groups (Asian, Black/African/Caribbean, White, Mixed, Other Ethnic Groups and Unknown), to 19 NHS ethnicity codes and up to 489 SNOMED-CT ethnic concepts.
Ethnicity breakdown N (%) in primary care data (England)
From primary care records (GDPPR data source), we observed than 9.8% of individuals self-identified as Asian/Asian British, 3.6% as Black/African/Caribbean/Black British, 77.3% as White, 2.2% as Mixed, 3.6% as Other Ethnic Groups and 3.2% as Unknown/Non-stated.