Clinical decision support app for dementia
Research type
Research Study
Full title
Development of a hybrid computational model based app to standardise and improve the quality of dementia assessment data used to support clinical decision-making
IRAS ID
230077
Contact name
Kongfatt Wong-Lin
Contact email
Sponsor organisation
Ulster University
Duration of Study in the UK
0 years, 8 months, 29 days
Research summary
Summary of Research
More timely and accurate diagnosis of dementia can lead to better and earlier interventions, which promote independence. Additionally, improved diagnosis permits planning for end-of-life care and reduces institutionalisation, thereby reducing direct and indirect health and social care costs burdens. It has been estimated that identification of risks and early diagnosis, which lead to the proper treatments or interventions, can delay the onset of dementia by 2-5 years and reduce the prevalence by 20-50%. The aim of this project is to make use of clinical data from N. Ireland to refine our hybrid computational approach to standardising and improving the quality of dementia assessment data, and thus to develop an app to support clinical decision-making. To address this aim a databank will be created of key clinical data linked with the assessment and diagnosis of dementia from patients currently living with diagnosed dementia (mild cognitive impairment, Alzheimer’s dementia, and non-Alzheimer’s dementia) and attending the N. Ireland Memory Assessment Service (WHSCT). Data mining methods will be employed to identify the strengths of the assessment tools used in predicating the dementia diagnosis. The information will be incorporated into an app for use by patients within the clinical setting, providing clinical decision-makers with standardised, high-quality and prompt dementia assessment information.
Summary of Results
The background research underpinning the research arose from investigating datadriven, AI-based feature selection methods on various existing datasets (Ding et al., 2018; Bucholc et al., 2019; McCombe et al., 2020; McCombe et al., 2021; McCombe et al., 2022a-d; Haridas et al., 2024). Further, it built on an intellectual property (IP) agreement in 2019 between Ulster University, NHSCT, Nightingale Analytics and AgeNI.
A common theme of the underpinning research was the development of practical AI based algorithms, which was comprehensively reviewed with respect to UK’s dementia care pathways (Wong-Lin et al., 2020; Wong-Lin et al., in press). In particular, McCombe et al (2021) made use of the dataset in WHSCT’s memory clinic to motivate and identify optimal AI-based algorithms for imputing missing data and enhance diagnostic accuracy. Another notable study was the development of practical AI algorithms and associated human-in-the-loop AI-based app to optimise for diagnostic accuracy, administrative time and financial costs (McCombe et al., 2022a, 2022b).
In terms of direct application to NHSCT data, the project first led to the collation of a comprehensive dataset for dementia assessment. The highly characterised database contains longitudinal data with 108 data features from 2962 memory service referrals, including demographic, dementia assessment and diagnosis information, all based on clinical expertise. The database is easily shared and has extensive potential from a clinical, research and commercial perspective, providing opportunities for future integration of the data to further our understanding of dementia and to support dementia care. The specific contributions are as follows.
• An algorithm to support dementia diagnosis: A highly predictive dementia diagnosis algorithm based on extensive modelling of 108 features of patient data with the ability to deal with missing data.
• Digitisation of existing dementia assessment tools: Digitisation of demographics & general information data collection. Digitisation of Addenbrooke’s Cognitive Examination (ACE)-III dementia diagnosis tool.
• Development of a Dementia Online Decision Support Tool: The app developed includes the ability to collect the current memory assessment data plus the additional 61 features collected in memory service interviews, including medical & clinical history, family history, education, employment and occupation, social circumstances, appearance, behaviour and physical health risk factors, onset of symptoms and referral details, cognitive capabilities and difficulties, impairment in activities of daily living, current mental health stressors, medication. The app is compliant with Trust security requirements, has the ability to document additional notes within the app during interviews, the ability to complete interviews in a non-linear manner and contains the digitisation of Zarit Burden Interview & Bristol Activities of Daily Living Scale dementia diagnosis tools as additional assessments tools.
• The feedback on the app was obtained from in-depth interviews with NHSCT Memory Service Staff.Key Findings for the NHSCT Data:
• It was found that 40.7% of people attending the Memory Service did not have dementia. Sharing information about the profile of people who are likely to have a dementia with referral agents will help them to make more informed decisions about who to refer to specialist Memory Services. This has benefits for the person by avoiding the unnecessary stress of a referral for assessment of dementia. It also offers savings to the health economy by reducing inappropriate referrals.
• Greater understanding of the cognitive profile and relative strengths and weakness of the different types of dementia enables appropriate advice to be provided and the opportunity for the development of compensation strategies.
• Greater understanding of MCI which is likely to develop into dementia offers the opportunity to provide early intervention and also informs decisions about who should be offered follow up assessment.
• This data has facilitated better planning of services to accommodate the number of people presenting to the Memory Service who are likely to have dementia.
• ACE-III is a useful tool for discriminating between dementia, MCI and controls, but it is not reliable for discriminating between dementia subtypes. Nonetheless, the ACE-III is still a reliable tool for clinicians that can assist in making a dementia diagnosis in combination with other factors at assessment.REC name
HSC REC B
REC reference
17/NI/0142
Date of REC Opinion
16 Aug 2017
REC opinion
Favourable Opinion