Machine-learning based identification of undiagnosed dementia in primary care: a feasibility study

Area of Science:

• Gerontology
• Medical Informatics
• Machine Learning

Background:

• Undiagnosed dementia affects up to 50% of patients, hindering access to care.
• Routine clinical data may contain symptom profiles indicative of undiagnosed dementia.

Purpose of the Study:

• To develop a machine learning model for detecting dementia in primary care using routinely collected National Health Service (NHS) data.
• To aid in identifying individuals with undiagnosed dementia.

Main Methods:

• Case-control study analyzing 2 years of pseudonymized primary care data from 18 GP surgeries.
• Utilized Read-encoded data from over 26,000 patients aged 65+.
• Trained a machine learning classification model on Read codes associated with dementia risk.

Main Results:

• The developed model achieved a sensitivity of 84.47% and a specificity of 86.67% in identifying potential dementia cases.
• Demonstrated the feasibility of using routinely collected clinical data for dementia detection.

Conclusions:

• Routinely collected primary care data can effectively identify patients with undiagnosed dementia.
• This methodology offers a promising approach to improve dementia diagnosis rates within general practice settings.