Machine learning to improve analysis of disability in electronic health records: an untapped opportunity for health inequities research.

Electronic Health Records (EHRs) are a leading source of epidemiological data, but often lack information on a patient's disability status. This gap hampers our ability to analyse the full scope of health inequities faced by people with disabilities. Current approaches to identify disability within EHRs have limitations because of inadequate proxies for disability or issues linking data sources. Machine learning (ML) offer unprecedented opportunities to create disability markers within EHRs, such as through unsupervised learning to classify disability groups and Natural Language Processing to extract relevant information from clinical notes. These methods have the potential improve disability-disaggregated analyses within EHRs to uncover patterns and provide a more comprehensive understanding of healthcare pathways and outcomes for people with disabilities. Leveraging these approaches to improve disability data in EHRs is a critical step towards improving health inequities research, though require strong adherence to ethical guidelines and validation of these new approaches.