Fairness of NLP-supported clinical decision-making process in healthcare

Text mining and natural language processing (NLP) systems are proving very useful for clinical care and research. Several decisions on patient inclusion/exclusion and coding of key study variables in clinical studies are taken out of the hands of clinicians and put into the care of NLP systems. However, clinical care is not always equitable; for example, women present very different stroke symptoms from men, leading to underdiagnosis and undertreatment of female stroke victims. These inequities are inevitably encoded in the electronic health records (EHR) corpora mined by clinical data scientists.

The aim of this project is to study the extent to which existing healthcare inequities are propagated by the application of text mining/NLP to EHR records and to develop methods to assess and reduce the impact of the encoded inequities.