How AI can improve dementia detection

Researchers from the National Center for Healthy Aging (NCHA), a partnership between Monash University and Peninsula Health, have developed a novel method for improving dementia detection in hospitals by combining traditional methods with artificial intelligence (AI).

Approximately 50 million people worldwide live with dementia, a number expected to triple by 2050, according to the World Alzheimer Report.

In Australia, there is still a need to substantially improve our methods for counting people with dementia. Accurate identification is critical to understanding the true size of the problem nationally, and to being able to effectively plan services. However, routine health data that are currently used for this purpose probably underestimate the numbers of people with dementia.

Regular health care contact and hospitalizations provide an important opportunity to address this issue. Currently, in hospitals, dementia is recorded based on the gathering of information in the medical records by medical coders, who find it difficult to look through the vast amount of written information in the records.

In a study involving more than 1,000 individuals aged 60 and above in the Frankston-Mornington Peninsula, algorithms using traditional data approaches with AI in electronic health records demonstrated high accuracy in identifying whether or not a person may have dementia.

Supported by national health bodies, the initiative could transform how dementia is identified, counted by national estimates, and managed in health care settings.

Given the global rise in dementia cases and the difficulty in accurately identifying patients through conventional medical coding, this approach has the ability to transform the Australian landscape in this field.

The research team based at Peninsula Health, involving NCHA’s Healthy Aging Data Platform group and clinicians from Australia and the U.S., have tackled this problem using AI, and found that a particular type of AI called natural language processing (NLP) applied to written text in medical records significantly enhances dementia identification capacity.

Their peer-reviewed paper, “Dual-Stream Algorithms for Dementia Detection: Harnessing Structured and Unstructured Electronic Health Record Data,” published in Alzheimer’s & Dementia showed that algorithms combining traditional methods with AI demonstrated very high accuracy for detecting the presence of dementia from information in electronic health records.

Lead author, Dr. Taya Collyer, said the study was based on people aged 60 and over with dementia diagnosed by specialists using gold standard methods, and a comparison group without dementia.

“Accessing high-quality curated electronic health records from our Healthy Aging Data Platform helped assemble the data efficiently to address this problem. Special software was used to harness the large amount of free text data in a way that NLP could then be applied,” Dr. Collyer said.

“We then developed dementia-finding algorithms through a traditional stream for usual structured data and an NLP stream for text records.”

For the traditional stream, in addition to standard codes for dementia, information was also obtained that reflected demographics, socioeconomic status, medications, emergency and clinic health utilization, and in-hospital events such as confusion or distressed behavior.

For the NLP stream, the team used clinical experts to guide the analysis to ensure its clinical relevance.

NCHA Director and project lead Professor Velandai Srikanth said the future impact of this novel approach is exciting, not only for the better counting of numbers of people with dementia, but also for the efficient identification of people with a high probability of dementia who may need care and support but who may get missed otherwise.

“Given that clinical recognition of people diagnosed with dementia presenting to hospitals is poor, using this new approach we could be identifying people earlier for appropriate diagnostic and clinical care. I am sure that many people are missing out on good care because we are not very good at identifying them or their needs,” Professor Srikanth said.

“This new method offers a novel digital strategy for capturing and combining clues in written text, such as descriptions of confusion or forgetfulness, or alerts for distressed behavior, to flag them for suitable care and support.

“Responsibly using AI in scientific research and dementia identification is potentially game-changing. The NCHA’s Healthy Aging Data Platform, an Australian-first initiative, has been able to bring together various sources of data from electronic health records, safety and governance, and the technical capacity to enable such high-value projects.”