Does natural language process offer a viable dementia screening tool in primary care?

This study used natural language processing to analyze free-text fields in primary care electronic medical records (EMRs) to determine the accuracy of this potential identification tool for dementia.
More than 900 dementia-related features were identified and subsequently grouped into 8 themes such as symptoms, social, function, and cognition.
For individuals prior to dementia diagnosis, model performance was found to be poor with a sensitivity of 8.3% and specificity of 99.9%. The authors concluded that, at this time, natural language processing of EMRs should not be used as a dementia screening tool.