Study Combines NLP, Conventional Coding To Improve Dementia Detection

The National Centre for Healthy Ageing (NCHA), a research institute by Monash University and Peninsula Health in Australia, has developed a new AI-driven method to efficiently identify dementia in hospitals by combining traditional and AI-based case identification approaches. The NCHA research team created dementia-finding algorithms through two streams: a traditional stream using routinely collected structured data from electronic medical records (EMRs) on the Healthy Ageing Data Platform, and an AI stream analyzing unstructured text records powered by natural language processing (NLP) and guided by clinical experts. According to Dr. Taya Collyer, one of the research leads, special software was used to process large amounts of free text data to enable the application of NLP. Beyond standard dementia codes, the research also incorporated demographic information, socioeconomic status, medications, emergency and clinic health utilization, and in-hospital events.

These algorithms were tested in a study involving over 1,000 seniors aged 60 and above in Frankston-Mornington Peninsula, demonstrating high classification accuracy with 72.2% specificity and 80.6% sensitivity in identifying individuals with or without dementia. This highlights the potential to improve dementia detection, counting, and management in healthcare settings. The research was supported by grants from the National Health and Medical Research Council, the Medical Research Future Fund, and the Department of Health and Aged Care. The importance of this work is underscored by projections from the World Alzheimer Report that the number of people living with dementia will rise to 150 million worldwide by 2050. Accurate identification is crucial to understanding the scope of the issue nationally and to planning effective services, as noted by Monash University.

While algorithmic tools for dementia detection are widely available, the NCHA team points out that many rely on proxies like diagnostic codes or medications and often lack clinically meaningful case definitions. Their dual algorithm approach, combining analysis of both structured and unstructured EMR data with NLP, aims to unlock greater potential for accurate detection. The use of large language models (LLMs) is also expanding in detecting cognitive impairment, such as in South Korea, where an LLM-driven model demonstrated high accuracy in recognizing Alzheimer's disease. Additionally, gamification techniques have shown promise for screening mild cognitive impairment in Singapore. Professor Velandai Srikanth, NCHA director and project lead, emphasized that clinical recognition of dementia in hospital settings remains poor, and this new approach could enable earlier identification of patients for appropriate diagnostic and clinical care, ensuring many currently missing out on proper care receive it.