Identifying people with potentially undiagnosed dementia with Lewy bodies using natural language processing.

Abstract

Natural language processing (NLP) can expand the utility of clinical records data in dementia research. We deployed NLP algorithms to detect core features of dementia with Lewy bodies (DLB) and applied those to a large database of patients diagnosed with dementia in Alzheimer's disease (AD) or DLB. Of 14,329 patients identified, 4.3% had a diagnosis of DLB and 95.7% of dementia in AD. All core features were significantly commoner in DLB than in dementia in AD, although 18.7% of patients with dementia in AD had two or more DLB core features. In conclusion, NLP applications can identify core features of DLB in routinely collected data. Nearly one in five patients with dementia in AD have two or more DLB core features and potentially qualify for a diagnosis of probable DLB. NLP may be helpful to identify patients who may fulfil criteria for DLB but have not yet been diagnosed.