Assessing the Practicality of Designing a Comprehensive Intelligent

Conversation Agent to Assist in Dementia Care

Ritwik Raj Saxena

and Arshia Khan

Department of Computer Science, University of Minnesota, Duluth Campus, Duluth, Minnesota, U.S.A.

Keywords: Dementia, Generative Artificial Intelligence, Deep Learning, Machine Learning, Natural Language Processing,

Intelligent Conversation Agents, Multi-Stakeholder Approach, Clinical Decision Support Systems, Data Privacy.

Abstract: Tools and techniques powered by artificial intelligence (AI) and its subfields like machine learning (ML) and

natural language processing (NLP) have pervaded most disciplines across the global technological,

economical, and sociocultural landscapes. In most areas, the permeation of AI has shown exceptional promise.

Medicine and healthcare constitute a domain which has not remained aloof from the positive implications of

harnessing AI. AI-driven tools, for instance in neuroimaging and health monitoring, have painted a tapestry

of encouraging possibilities in this province. Such tools have found application in fields like assisting

diagnosis, disease progression tracking, and patient management in many subjects within medicine. Intelligent

conversation agents, more informally referred to as AI-based chatbots, form one of the most prevalent

applications of AI. AI-fueled chatbots like ChatGPT have made rampant inroads into the lives of countless

people around the world, easing innumerable routine tasks they are responsible for. This article offers a

systematic but succinct overview of dementia, and, in this backdrop, explores the potential efficacy of a

proposed intelligent conversation agent aimed at sufficing the fulfilment of the care-associated requirements

of various stakeholders in dementia care. We provide an outline and a critical assessment and suggest future

directions on the adoption of such a tool. We conclude that a smart conversation agent has the potential to

positively overhaul the extant worldwide paradigm of dementia care.

1 INTRODUCTION

Dementia is a broad term. Instead of effectively

referring to a disorder, it describes a set of symptoms

that affect the general functioning of the brain.

Cognitive and motor functions, inter alia, memory,

reasoning, communication, and the ability to perform

daily activities, including motor abilities in the later

stages of the disease, are affected in dementia.

Dementia has many common embodiments.

Alzheimer’s disease is the most frequently occurring

one, with about two-thirds of cases with dementia

presenting with it. Dementia primarily affects older

adults. Notwithstanding this fact, dementia is neither

an inherent part of aging nor is it restricted to

senescent humans. A specific disease called young-

onset dementia (early-onset dementia) exists, and

young-onset Alzheimer’s disease is the most common

form of it (Sim et al., 2022).

https://orcid.org/0009-0001-7876-3193

https://orcid.org/0000-0001-8779-9617

The World Health Organization (WHO) estimates

that over 55 million people worldwide are living with

dementia, and the World Alzheimer’s Report, 2009,

estimates that more than 65 million people will be

affected by dementia by 2030 (Gulland, 2012). As the

population ages, the burden of dementia is being felt

not only by those diagnosed and other stakeholders

active (participating) in dementia care (SDCs), but

also our society.

Caring for persons with dementia (PwD)

introduces plentiful challenges, most of which

intensify alongside the progression of the disease.

Dementia affects almost every function of the brain,

from memory and cognition to motor functions and

coordination. PwD may even face sensory struggles

in certain cases (National Institute on Aging, 2023).

The progression of dementia is commonly divided

into three phases, early, middle, and late (Hol et al.,

2024). In the early phase of the disease, PwD

Saxena, R. R. and Khan, A.

Assessing the Practicality of Designing a Comprehensive Intelligent Conversation Agent to Assist in Dementia Care.

DOI: 10.5220/0013191600003911

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 18th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2025) - Volume 2: HEALTHINF, pages 655-663

ISBN: 978-989-758-731-3; ISSN: 2184-4305

655

experience mild memory lapses and confusion. The

symptoms begin with patients forgetting names and

not being able to recall the task they barely arrived at

a certain location for.

As the severity of the disease advances, its

progression enters the second phase. PwD gradually

start losing the ability to communicate, recognize

loved ones, and carry out basic self-care tasks like

combing their hair and brushing their teeth. Mobility

issues arise in the second phase (middle stage), with

shuffling, hesitation, gait apraxia and festination in

gait becoming apparent (Elble, 2007). In the final

stage, severe cognitive issues are seen, including far-

reaching amnesia with regards to otherwise

effortlessly recognizable faces and objects. Urinary

incontinence and clinical-level mobility issues arise.

The Global Deterioration Scale (GDS), also

known as the Reisberg Scale, is a clinical tool used to

evaluate the progression of cognitive decline in PwD.

This scale categorizes the disease into seven stages

(Dementia Care Central, 2020). There is very slight

overlap between these stages and the phases in the

three-phase model. The Clinical Dementia Rating

(CDR) scale is another tool to assess the severity of

dementia through evaluation of cognitive and

functional performance (Morris, 1993). The stages

are CDR-0 (no impairment), CDR-0.5 (very mild

cognitive decline), CDR-1 (mild cognitive decline),

CDR-2 (moderate cognitive decline), CDR-3

(moderately severe cognitive decline), CDR-4

(severe cognitive decline), CDR-5 (very severe

cognitive decline). These closely correspond to the

GDS stages.

Considerations associated with such progression

places immense emotional, physical, and financial

strain on various SDCs, particularly the family

members and caregivers (Schulz and Sherwood,

2018). They are prone to burnout due to the

demanding and unrelenting nature of the care

required.

Role overload is the psychological feeling –

which may be real or perceived – of being left

exhausted owing to the duties and demands of

caregiving. Role captivity is the psychological feeling

of being trapped in the role of caregiving and the

resultant erosion of autonomy (Liu et al., 2019). Both

phenomena have a considerable impact on the well-

being of dementia caregivers affecting their mental

health and quality of life (QoL).

Dementia patients face behavioural changes,

such as aggression and wandering off. Caregiver

stress, as explained by the model proposed by

Pearlin et al., has some key factors associated with

it (Pearlin et al., 1990). They include the caregiver’s

personal background and circumstances, including

factors such as the caregiver’s socioeconomic status,

social support networks, and other life stressors; the

demands of the role which exacerbate with the

severity of the patient’s illness; the strains which

arise from the caregiver's other roles and

responsibilities, such as family conflicts, social

isolation and a lack of intermingling with friends,

and financial strain (caregiving might become a full-

time role, or require too much energy, leaving the

caregiver with little or no chance to engage in more

monetarily remunerating employment); and internal

factors such as the caregiver’s own personality (a

caregiver may be timid by nature, or lack the energy

to be an effective caregiver), perceived

incompetence, and other coping mechanisms like

self-pity (by convincing oneself of their role

captivity) (Brodaty and Donkin, 2019).

With these factors in mind, researchers and

healthcare providers enable themselves to empathize

with caregivers. They are, thus, more spurred to

develop useful policies to aid caregivers that mitigate

the negative effects of stress on them.

The healthcare system struggles to meet the needs

of dementia patients. Medical interventions are

largely focused on symptom management and

palliative therapeutics rather than cure (Walsh et al.,

2021). In this context, caregivers, whether informal

ones (Brodaty and Donkin, 2019) or hired

professional aides (Ferretti et al., 2021), have become

cornerstones of patient support. Informal caregivers

are often seen traversing this journey with limited

guidance and respite and may benefit from suitable

training (Birkenhäger‐Gillesse et al., 2020).

The motivation for writing this article stems from

the recognition of these critical challenges and the

urgent need for innovative technological solutions

that can alleviate the burden on not only patients and

their caregivers but also various other SDCs.

AI and NLP applications like chatbots assume the

role of a promising avenue for addressing the needs

of SDCs. Such tools provide aid in daily tasks and

tender cognitive stimulation to PwD, mitigate social

isolation by offering emotional support to those SDCs

who may require it, provide information as instructed

and as necessitated, as also education and edification

to those freshly involved in dementia care or those

already involved in dementia care but participating in

some kind of transition where it behooves them to be

reskilled for continued effective functioning.

However, the development of such tools requires

careful consideration of not only the needs of but also

the challenges faced by SDCs. Most SDCs,

particularly PwD and their caregivers, face unique

HEALTHINF 2025 - 18th International Conference on Health Informatics

656

challenges depending on the stage of dementia, and

any solution must account for these differing needs to

be truly effective (Sideman et al., 2022).

2 LITERATURE REVIEW AND

CURRENT APPLICATIONS

A PRISMA-guided literature review was conducted

to mine publications which would help us gather

knowledge regarding what has been done and what is

currently happening within the domain of intelligent

conversation agents being applied to facilitate and

advance dementia care. The systematic screening and

selection processes inherent to PRISMA left us with

a total of 8 articles to be considered, out of the 126 we

began with (Table 1).

T. Igarashi et al. compared the effects of human-

AI interaction on the communication patterns of

elderly people attending community centres and

found that AI-based options can be a viable solution

for routine conversational engagement with

cognitively healthier elderly individuals, that is,

people with early-stage dementia (Igarashi et al.,

2024). F. de Arriba-Pérez et al. used a ML-based

chatbot framework to dynamically predict cognitive

decline (de Arriba-Pérez and García-Méndez, 2024).

M. Boiting et al. have demonstrated a virtual,

interned-based interaction and service framework

that features an embedded chatbot which is intended

to provide organized, comprehensible information to

informal caregivers (Boiting et al., 2024).

C. Müller et al. drew insights from existing

research and from interviews at dementia care

institutions and developed a chatbot prototype to

facilitate facilitates caregiver-patient interaction

(Müller et al., 2022). D. Schmitz and B. Becker have

presented the design of an information platform

accompanied by a chatbot specializing in aiding

informal caregivers (Schmitz and Becker, 2024). M.

R. Lima et al. proposed leveraging an amalgamation

of conversational AI and Internet of Things (IoT) to

monitor older adults, specifically PwD, at home, for

identification of behavioral patterns (Lima et al.,

2023).

Table 1: Literature Review Table.

Author(s) Title Published Brief Description of the Publication

Igarashi et al.

Detailed Analysis of Responses from Older

Adults through Natural Speech:

Com

arison of AI vs. Humans

2024

Analyzes older adults’ responses to AI vs.

human interactions in dementia care.

de Arriba-

Pérez &

García-

Méndez

Leveraging Large Language Models

through NLP for Real-Time Mental

Deterioration Predictions

2024

Explores using large language models and

NLP for real-time predictions of mental

deterioration

Boiting et al.

eDEM-CONNECT: An Ontology-Based

Chatbot for Family Caregivers of People

with Dementia

2024

Presents an AI chatbot integrated into the

eDEM-CONNECT platform.

Schmitz &

Becker

Chatbot-Mediated Learning for Caregiving

Relatives of People with Dementia

2024

Investigates chatbot-mediated learning for

dementia caregivers

Lima et al.

Discovering Behavioural Patterns Using

Conversational Technology for In-Home

Health Monitoring

2023

Introduces a cutting-edge IoT and

conversational AI setup

Maia et al.

Empowering Preventive Care with GECA

Chatbot

2023

Introduces the GECA chatbot for preventive

care in dementia

Müller et al. Care: A Chatbot for Dementia Care 2022

Discusses a chatbot which facilitates

caregiver-patient contact

Kouroubali

et al.

Developing an AI-Enabled Integrated Care

Platform for Frailty

2022

Discusses building a care platform for

weakness in dementia

Assessing the Practicality of Designing a Comprehensive Intelligent Conversation Agent to Assist in Dementia Care

657

Currently, AI applications like AI-driven serious

games show a massive potential to reform dementia

care. AI-fueled smart conversation agents constitute

one such application. They have the potential to

enhance evidence-based, personalized patient care

plans (Maia et al., 2023). These chatbots can provide

services like medication reminders, automated

symptom tracking, and tailored event tracking and

symptom monitoring support (Clark and Bailey,

2024). They can also act as virtual assistants and

serve as communication bridges between dementia

patients and their caregivers (Müller et al., 2022). A

suitably trained chatbot can act as a virtual listening

ear for PwD to express their needs and emotions

while receiving companionship and emotional

support (Denecke et al., 2021).

AI-driven games are associated with AI-powered

chatbots and contribute to cognitive stimulation of

dementia patients (Irfan et al., 2024). AI (ML)

algorithms are used for early detection of cognitive

decline via analysis of data from sources like

historical data, electronic health records (EHR) and

wearables (Graham et al., 2020). This facilitates early

intervention which, in most cases, is able to arrest the

pace of dementia progression. AI tools can be used

for remote monitoring and continuous observation of

dementia patients. This creates conducive conditions

for the prompt detection of circumstance changes and

administering the necessary interventions (Ahmed et

al., 2020). This also obviates the need for frequent in-

person visits.

3 STAKEHOLDERS IN

DEMENTIA CARE

The list of SDCs begins with PwD themselves, who

vary in their cognitive abilities across the stages of

dementia, their caregivers, who often play a pivotal

role in managing daily routines and medical

appointments, healthcare providers, who must make

informed decisions regarding treatment and symptom

management, grassroots, voluntary, and social care

organizations participating in dementia care,

researchers in dementia care, physiotherapists who

assist with mobility challenges faced by PwD,

policymakers active in the areas of general

healthcare, especially those in mental healthcare, and

technologists who implement solutions based on the

outcomes of researchers’ activities and policymakers’

guidelines.

Pharmaceutical companies, insurance companies,

government agencies involved in funding dementia

research and regulating care services, social workers,

occupational therapists who help patients adapt to

ADLs, speech-language therapists who address

communication and swallowing difficulties,

community organizations which provide respite care,

support groups, and educational programs for

caregivers as well as for PwD, technology

manufacturers who focus on developing assistive

technologies and devices for PwDs, advocacy groups

who raise dementia awareness, nutritionists and

dieticians for PwD, pharmacists who assist caregivers

in managing medications for PwDs, home care

providers who provide in-home assistance with daily

tasks, legal advisors who help families with estate

planning and guardianship (particularly in case of the

families of elderly asset holders with end stage

dementia), advance directives, living wills, medical

power of attorney, and other legal matters and

memory care specialists who deliver specialized care

that is tailored to the unique cognitive needs of

dementia patients, together form a host of varied

stakeholders involved in dementia care, some more

key to the whole enchilada of dementia care, and

some comparatively less. An AI-based solution must

be able to satisfy the needs of these stakeholders to be

termed comprehensive and as a solution oriented

towards a multi-stakeholder view (Patel et al., 2021).

To inform the development of a user-centric

intelligent conversation agent in dementia care, the

needs of SDCs must be enunciated with clarity. Early-

stage PwD would benefit most from a chatbot,

because advanced stage dementia creates hurdles for

PwD (severe cognitive impairment) to efficiently

utilize such technology. PwD in early stages would

benefit from a chatbot that offers cognitive exercises,

serious games, along with comfort, while those in

advanced stages need an agent which could remind

them about daily tasks, warn them not to leave their

homes (for they may lose their way), and respond to

simple cues. The chatbot should be designed to adapt

its conversational style and complexity as dementia

progresses. This would ensure that PwD feel

supported without being overwhelmed and confused.

This adaptability is key to maintaining engagement

and promoting a sense of independence in the early

stages.

Caregivers manage daily routines, medical

appointments, and give emotional support to PwD.

An ideal chatbot must assist a caregiver with tracking

the patient’s symptoms, medication adherence, and

behavior management. It should provide timely

reminders, alert the caregiver to any potential issues,

and offer tailored advice based on the PwD’s current

cognitive state. The chatbot must also provide

HEALTHINF 2025 - 18th International Conference on Health Informatics

658

caregivers with educational and stress management

resources which would help them adapt better in their

role as caregivers and improve their skills.

Healthcare providers would find a chatbot that

analyzes, tracks, and presents health-related data to

aid informed decision-making helpful. It should also

provide them with insights from the data about trends

and patterns which would inform their service

delivery and improve it. The AI-solution should

monitor cognitive changes through the linguistic

input by the patient or their caregiver, discerning their

behavior patterns, and predicting any adverse

happening like a reaction to a medication. The system

can transmit, in real-time, the analytics to healthcare

providers, who can opt for the best course of action.

Physiotherapists engaged in dementia care could

need the chatbot to remind PwD to perform their

prescribed exercises and monitor their progress, and

also query it for help in dealing with unique

situations they may not have encountered before, for

example, a person with dementia revulsive to being

touched. With respect to social care organizations,

the AI solution should be able to connect PwD and

their caregivers to local resources, such as support

groups, educational programs, and respite care

services. The AI system must act as a bridge

between these entities and the patients they aim to

help. It should be able to identify the specific needs

of each patient and his or her caregiver and provide

tailored recommendations. It should also facilitate

access to the right resources at the right time.

The chatbot could provide researchers,

technologists, policymakers, and other SDCs, inter

alia pharmaceutical companies, insurance

companies, thinktanks, and advocacy groups, with

resources like insights and cognizant

recommendations based on updated anonymized

data of patients and other sources of relevant

knowledge on dementia care that the underlying

language model has been trained or finetuned on or

has access to through techniques like Retrieval

Augmented Generation (RAG), web scraping, or in-

context learning (ICL). The chatbot itself should

gather data relying on SDCs’ interaction with it; this

data pertains to, among other details, engagement

levels and patient health outcomes. The insights can

inform future research to improve treatment

strategies, developing new medications and

technologies, and designing dementia care policies

and strategies, outreach and marketing plans, and so

on, all of which are aimed at enhancing dementia

care.

4 FULFILLING THE NEEDS OF

SDCs THROUGH AN

AUTOMATED

CONVERSATION AGENT

Chatbots have been operative in the domain of mental

health for a long time. Weizenbaum’s ELIZA (1966)

and Colby’s PARRY (1972) were the earliest

chatbots in this field which were exclusively or

primarily based on rule-based models, that is, they

were preliminary models which relied on predefined

scripts (Saxena, 2024a, 2024b). There are several

types of chatbots that have evolved since then. The

types of chatbots include text-based, speech-based,

and multimodal systems. Text-based chatbots, like

Bard, interact through written text. Speech-based

chatbots, like Alexa and Siri, interact using speech.

Multimodal Chatbots, like Gemini, interact through

multiple modes including text and speech, and can

analyze images, audio, video, tabulated data etc. to

provide insights to the user. Some of them can also

generate data in various formats (multimodal) like

computer programs, audio, video, images, and so on.

We propose the design of a comprehensive

chatbot which can use novel AI techniques to

maximize the fulfillment of the needs of the SDCs.

The chatbot should specialize sufficiently in the

domain of dementia care but not so much that it

overfits and is incapable of processing unique, unseen

conditions or queries, expressly those which are less

directly related to dementia but have a bearing on it.

To build such a chatbot, the underlying model must

have access to diverse data. Exhaustive clinical

datasets which exemplify the various stages of

dementia, including patient profiles categorized by

the stage of dementia, cognitive abilities, memory

retention, speech patterns, motor functions,

behavioral issues like aggression and wandering,

cognitive assessment reports. and interventional

paradigms and medications utilized for patients

clustered based on gender, stage of dementia, age, and

other criteria, along with the evaluated efficacy of the

varied treatment approaches, patient outcomes, and

service utilization rates, must be used.

Access to EHRs, anonymized patient data, and

clinical trial data enable a chatbot to provide data

analytics, real-time insights, and intervention advice.

These insights will include detecting trends in

cognitive function, identifying adverse reactions to

medications, or predicting when interventions might

be necessary. Learning from bulk longitudinal

symptom progression datasets and from detailed

anonymized multi-patient clinical history adequately

Assessing the Practicality of Designing a Comprehensive Intelligent Conversation Agent to Assist in Dementia Care

659

empowers the model to recommend appropriately to

SDCs. Similar comprehensive datasets on caregiver

case studies, trends, patterns, analytics, and insights

on caregiver stress and other aspects associated with

caregivers and PwD, patient-caregiver conversation

transcripts and other transcripts of informative

conversations between various combinations of

SDCs must also be used to train the model. Caregiver

training manuals and comprehensive standardized

training materials commonly used in skilling,

upskilling, and reskilling various other SDCs.

A wide-ranging group of datasets consisting of

dementia- and mental health-related medical and

general texts, articles, chapters, and so on, existing

policies on dementia care, pertinent

recommendations from expert groups and advocacy

groups, publications and research findings, reports

and reviews on dementia, statistics associated with

global, national, and local dementia care data

collection and analytics efforts, industrial data to help

SDCs like pharma and insurance firms, and a

comprehensive description of latest developments

and proposals in the field must also be used during

model development.

To continually improve itself, one of the tactics

that the model must utilize is to augment its

knowledge base by leveraging the context as well as

the prompts provided to it and the queries posed to it.

The PwD’s responses to prompts and frequency of

interactions and inputs from caregivers provide

feedback loops that would refine its interaction. To

improve its comprehensiveness, the chatbot should be

able to process multimodal data inputs. Multimodal

models based on advanced neural networks like

Gated Graph Recurrent Neural Networks (GRNNs),

Convolutional Neural Networks (CNNs), RNNs, and

Transformers must be trained using these multi-

format datasets (Saxena and Saxena, 2024).

5 STRATEGIES THAT MUST BE

APPLIED WHEN BUILDING

THE SPECIALIZED

CONVERSATION AGENT

Domain-specific data collection and curation is an

essential tool to build specialized models. The

language model must be exposed to datasets that are

highly relevant to the field of dementia care as already

described. The language can be pretrained on the data

or can be a preexisting model finetuned on it. In this

way, transfer learning and finetuning become relevant

strategies where we start with an open-source LLM

like Mistral or BERT, which has already been

pretrained on large volumes of general language data.

This model is then finetuned on domain-specific data

related to dementia care to inform an efficient

chatbot. Finetuning allows the model to specialize by

focusing on labeled data. The process adjusts the

model’s internal weights so that it aligns with

dementia-specific requirements.

Prompt engineering can be applied to not only the

already rendered specialized chatbot for more

efficient responses but also to general-purpose

language models to adapt them for specific SDC

needs without retraining the entire model. Prompt

engineering involves users themselves, who must

carefully craft queries and frame – with explicitness,

straightforwardness, and elaboration – such problems

which will be served as prompts to the chatbot. This

helps an agent generate highly meaningful outputs

which show exceeding appropriateness to the user’s

needs.

Incorporating external knowledge bases also

offers a competent enhancement to the model.

Integrating healthcare ontologies like ICD-10 (Cooke

et al., 2011), dementia-specific databases like

International Alzheimer's Disease Research Portfolio

(IADRP) (Liggins et al., 2014), clinical care

guidelines (Shaji et al., 2018), evidence-based care

guidelines, multi-omics data and so on, associated

with dementia (Saxena et al., 2023), ensures accuracy

and holism in the chatbot’s responses.

Constructing knowledge graphs to represent the

relationships between various concepts and words

that a model is supposed to learn is another way to

train a model of surpassing systematicness. Using

specialized tokenization and vocabulary will help the

chatbot accurately interpret dementia-associated

medical abbreviations and jargon (such as PwD).

Task-specific training objectives enable the model to

focus on specific outcomes important to SDCs (Li et

al., 2020).

Custom loss functions can train an efficient

dementia care model. Active learning and human-in-

the-loop feedback are strategies where experts

periodically review an ML model’s output and

provide constructive feedback which is then used by

technologists to improve the model’s performance

over time.

To keep up with the most updated information

relevant to dementia care and to be resilient to the

evolving nature of the arena, incorporating continual

learning is essential (Wang et al., 2023). Leveraging

RAG is vital in scenarios where real-time, accurate

information is required. A model with the ability to

scrap the web for up-to-the-minute dementia

HEALTHINF 2025 - 18th International Conference on Health Informatics

660

research, clinical advances, recommendations,

policy, and technology ensures that newly available

pertinent information can be accessed and processed

as well as integrated into the conversational agent’s

responses.

6 DISCUSSION AND

CHALLENGES

Decision support should be a major aspect of the

chatbot. Systems which offer decision support are

termed clinical decision support systems (CDSSs)

(Bruun et al., 2019). To accelerate the

implementation of anticipatory approaches while

strengthening multimodal and inclusive care for

seniors, shared care planning smoothens

communication between various SDCs. A

multimodal AI model for an efficient paradigm of

dementia care must also incorporate diagnostic

pathology algorithms applicable to dementia,

including computer vision DL algorithms that enable

it to analyze pathological images like MRIs (Saxena

et al., 2019). This augments CDSSs and gives the

underlying model diagnostic capabilities.

A chatbot is an obvious example of the rapidly

burgeoning field of human-machine interaction

(human-computer interaction) (Saxena et al., 2023b).

A major concern in developing a chatbot that is useful

to a broad variety of stakeholders is creating a user-

centric design with a user-friendly interface (Saxena

et al., 2024). Considering the cognitive impairments

associated with dementia, it must be even more

intuitive, facile to navigate, and adaptable to the

varying abilities of its users.

A significant challenge that arises in case of a

complex ML model is its potential lack of

interpretability and explainability. It is vital for

various SDCs to have at least some understanding

regarding how the chatbot arrives at its

recommendations. Clear explanations help build trust

in the system. Ambiguous decisions cause confusion

and may result in misuse of an AI-based system.

Explainable AI techniques (XAI) like LIME and

SHAP should be applied.

Patient data privacy is one of the paramount

concerns in the realm where AI and any aspect of

healthcare or society in general intersect. Ensuring

the anonymity and confidentiality of patient data is

essential to comply with regulations like Health

Insurance Portability and Accountability Act

(HIPAA) and General Data Protection Regulation

(GDPR).

The right data is crucial for any ML model to be

trained or finetuned well. So, data availability is

another challenge to consider. In the case of a

comprehensive chatbot in dementia care, there is no

scope for a hallucinatory response to any query,

(hallucination implies the generation of incorrect and

nonsensical responses for queries that the chatbot or

the model is unable to, despite an extensive search

through its vector databases, find a germane answer

to), for misleading and inaccurate outputs may lead to

harmful decisions by SDCs, and may even endanger

lives.

The chatbot must be constantly and efficiently

connected to emergency response services and

healthcare providers. The system must be capable of

expeditiously propagating any warning to emergency

response teams and healthcare professionals in case

PwD encounter any crisis.

7 FUTURE DIRECTIONS AND

CONCLUSION

Dementia is a significant global health issue. The

WHO (2003) reports that dementia accounts for

11.2% of years lived with disability in individuals

past 60 years of age. Dementia has been able to

surpass the impact of global impact of stroke,

cardiovascular disease, and cancer. Advances in AI

over the past few years have demonstrated great

potential for enhancing dementia care and

management.

In this context, The AI-driven intelligent

conversation agent we propose shows promise to

portray the capability to significantly alleviate the

dementia-associated emotional, physical, and

financial burdens that fall on the shoulders of

different SDCs.

However, it must be taken care that such a chatbot

is designed to become progressively better at

accommodating the diverse and evolving needs of

PwD and other SDCs. Simplicity, accessibility, and

adaptability must also be ensured. Efficient

collaboration between various SDCs can lead to the

development of an effective intelligent conversation

agent in dementia care and the resolution of any

roadblocks in its implementation and deployment.

Further future directions include the possible releasing

of the agent as freeware in the best interests of the

society, ensuring continued funding for incremental

enhancements in its performance, and building similar

across-the-board systems for other disciplines in the

field of medicine as well as in other spheres.

Assessing the Practicality of Designing a Comprehensive Intelligent Conversation Agent to Assist in Dementia Care

661

REFERENCES

Sim, J., Li, H., Hameed, S., & Ting, S. K. S. (2022). Clinical

Manifestations of Early-Onset Dementia With Lewy

Bodies Compared With Late-Onset Dementia With

Lewy Bodies and Early-Onset Alzheimer Disease.

JAMA Neurology. https://doi.org/10.1001/jamaneurol.

2022.1133.

Gulland, A. (2012). Number of people with dementia will

reach 65.7 million by 2030, says report. BMJ, 344,

e2604–e2604. https://doi.org/10.1136/bmj.e2604.

National Institute on Aging. (2023). Take care of your

senses: The science behind sensory loss and dementia

risk. National Institute on Aging. https://

www.nia.nih.gov/news/take-care-your-senses-science-

behind-sensory-loss-and-dementia-risk.

Hol, E., Van Noort, O., Schotanus, F., & Van Montfort, G.

(2024). Definition and design of client profiles for

dementia clients in the Dutch home care sector.

Retrieved from https://essay.utwente.nl/65895/1/

Hol_MA_MB.pdf.

Elble, R. J. (2007). Gait and dementia: moving beyond the

notion of gait apraxia. Journal of Neural Transmission,

114(10), 1253–1258. https://doi.org/10.1007/s00702-

007-0743-3.

Dementia Care Central. (2020). Seven Stages of

Dementia | Symptoms, Progression &

Durations. Dementiacarecentral.com. https://www.

dementiacarecentral.com/aboutdementia/facts/stages/.

Morris, J. C. (1993). The Clinical Dementia Rating (CDR):

Current version and scoring rules. Neurology, 43(11),

2412. https://doi.org/10.1212/wnl.43.11.2412-a.

Schulz, R., & Sherwood, P. R. (2018). Physical and mental

health effects of family caregiving. AJN, American

Journal of Nursing, 108(Supplement), 23–27.

https://doi.org/10.1097/01.naj.0000336406.45248.4c.

Liu, Y., Dokos, M., Fauth, E. B., Lee, Y. G., & Zarit, S. H.

(2019). Financial Strain, Employment, and Role

Captivity and Overload Over Time Among Dementia

Family Caregivers. The Gerontologist, 59(5).

https://doi.org/10.1093/geront/gnz099.

Pearlin, L. I., Mullan, J. T., Semple, S. J., & Skaff, M. M.

(1990). Caregiving and the Stress Process: An

Overview of Concepts and Their Measures. The

Gerontologist, 30(5), 583–594. https://doi.org/10.1093/

geront/30.5.583.

Brodaty, H., & Donkin, M. (2019). Family Caregivers of

People with Dementia. Dialogues in Clinical

Neuroscience, 11(2), 217–228.

https://doi.org/10.31887/DCNS.2009.11.2/hbrodaty.

Walsh, S. C., et al. (2021). Palliative care interventions in

advanced dementia. Cochrane Database of Systematic

Reviews, 2021(9). https://doi.org/10.1002/14651858

.cd011513.pub3.

Ferretti, C., Nitrini, R., & Brucki, S. M. D. (2021). Virtual

Support in Dementia: A Possible Viable Strategy for

Caregivers. Frontiers in Neurology, 12. https://doi.org/

10.3389/fneur.2021.662253.

Birkenhäger‐Gillesse, E. G., Achterberg, W. P., Janus, S. I.

M., Kollen, B. J., & Zuidema, S. U. (2020). Effects of

caregiver dementia training in caregiver‐patient dyads:

A randomized controlled study. International Journal of

Geriatric Psychiatry. https://doi.org/10.1002/gps.5378.

Sideman, A. B., et al. (2022). Facilitators and Barriers to

Dementia Assessment and Diagnosis: Perspectives

From Dementia Experts Within a Global Health

Context. Frontiers in Neurology, 13, 769360.

https://doi.org/10.3389/fneur.2022.769360.

Igarashi, T., Iijima, K., Nitta, K., & Chen, Y. (2024).

Detailed Analysis of Responses from Older Adults

through Natural Speech: Comparison of Questions by

AI Agents and Humans. International Journal of

Environmental Research and Public Health, 21(9),

1170. https://doi.org/10.3390/ijerph21091170.

de Arriba-Pérez, F., & García-Méndez, S. (2024).

Leveraging large language models through natural

language processing to provide interpretable machine

learning predictions of mental deterioration in real time.

Arabian Journal for Science and Engineering.

https://doi.org/10.1007/s13369-024-09508-2.

Boiting, M., et al. (2024). eDEM-CONNECT: An

Ontology-Based Chatbot for Family Caregivers of

People with Dementia. Lecture Notes of the Institute for

Computer Sciences, Social Informatics and

Telecommunications Engineering, 73–91.

https://doi.org/10.1007/978-3-031-60665-6_6.

Müller, C., Paluch, R., & Ali, H. A. (2022). Care: A chatbot

for dementia care. Dl.gi.de, 10.18420/muc2022-mci-

src-442. Retrieved from https://dl.gi.de/items/

00956d93-3952-4fa3-a2e2-ffe817934a16.

Schmitz, D., & Becker, B. (2024). Chatbot-Mediated

Learning For Caregiving Relatives of People With

Dementia: Empirical Findings and Didactical

Implications For Mulitprofessional Health Care.

Journal of Multidisciplinary Healthcare, 17, 219–228.

https://doi.org/10.2147/jmdh.s424790.

Lima, M. R., et al. (2023). Discovering Behavioural

Patterns Using Conversational Technology for in-

Home Health and Well-being Monitoring. IEEE

Internet of Things Journal, 10(21), 18537–18552.

https://doi.org/10.1109/jiot.2023.3290833.

Maia, E., Vieira, P., & Praça, I. (2023). Empowering

Preventive Care with GECA Chatbot. Healthcare,

11(18), 2532. https://doi.org/10.3390/healthcare

11182532.

Clark, M., & Bailey, S. (2024). Chatbots in Health Care:

Connecting Patients to Information: Emerging Health

Technologies. Ottawa (ON): Canadian Agency for

Drugs and Technologies in Health. Retrieved from

https://www.ncbi.nlm.nih.gov/books/NBK602381/.

Denecke, K., Abd-Alrazaq, A., & Househ, M. (2021).

Artificial Intelligence for Chatbots in Mental Health:

Opportunities and Challenges. In Multiple Perspectives

on Artificial Intelligence in Healthcare (pp. 115–128).

https://doi.org/10.1007/978-3-030-67303-1_10.

Irfan, B., Kuoppamäki, S., & Skantze, G. (2024).

Recommendations for designing conversational

companion robots with older adults through foundation

models. Frontiers in Robotics and AI, 11, 1363713.

https://doi.org/10.3389/frobt.2024.1363713.

HEALTHINF 2025 - 18th International Conference on Health Informatics

662

Graham, S. A., et al. (2020). Artificial intelligence

approaches to predicting and detecting cognitive

decline in older adults: A conceptual review. Psychiatry

Research, 284, 112732. https://doi.org/10.1016/j.

psychres.2019.112732.

Ahmed, Z., Mohamed, K., Zeeshan, S., & Dong, X. (2020).

Artificial intelligence with multi-functional machine

learning platform development for better healthcare and

precision medicine. Database, 2020. https://doi.org/10.

1093/database/baaa010.

Patel, N. K., et al. (2021). Engaging multi-stakeholder

perspectives to identify dementia care research

priorities. Journal of Patient-Reported Outcomes, 5(1).

https://doi.org/10.1186/s41687-021-00325-x.

Saxena, R. R. (2024a). Examining Reactions about

COVID-19 Vaccines: A Systematic Review of Studies

Utilizing Deep Learning for Sentiment Analysis.

Authorea Preprints, Techrxiv. https://doi.org/10.36227/

techrxiv.171262869.98334885/v1.

Saxena, R. R. (2024b). Beyond Flashcards: Designing an

Intelligent Assistant for USMLE Mastery and Virtual

Tutoring in Medical Education (A Study on Harnessing

Chatbot Technology for Personalized Step 1 Prep).

arXiv.org. https://arxiv.org/abs/2409.10540..

Saxena, R. R., & Saxena, R. (2024). Applying Graph

Neural Networks in Pharmacology. Authorea

Preprints, Techrxiv. https://doi.org/10.36227/techrxiv.

170906927.71541956/v2.

Cooke, D. T., Gelfand, G. A. J., & Broghammer, J. A.

(2011). Billing, Coding, and Credentialing in the

Thoracic Surgery Practice. Thoracic Surgery Clinics,

21(3), 349–358. https://doi.org/10.1016/j.thorsurg.

2011.04.003.

Liggins, C., et al. (2014). International Alzheimer’s Disease

Research Portfolio (IADRP) aims to capture global

Alzheimer’s disease research funding. Alzheimer’s &

Dementia, 10(3), 405–408. https://doi.org/10.1016/j.

jalz.2013.12.013.

Shaji, K. S., Sivakumar, P. T., Rao, G. P., & Paul, N.

(2018). Clinical Practice Guidelines for Management of

Dementia. Indian Journal of Psychiatry,

60(Suppl 3), S312–S328. https://doi.org/10.4103/0019-

5545.224472.

Saxena, R., Carnevale, K., Yakymovych, O., Salzle, M.,

Sharma, K., & Saxena, R. R. (2023a). Precision,

Personalization, and Progress: Traditional and

Adaptive Assessment in Undergraduate Medical

Education. Innovative Research Thoughts, 9(4), 216–

223. https://doi.org/10.36676/irt.2023-v9i4-029.

Li, J., Zhang, Z., & Zhao, H. (2020). Dialogue-adaptive

Language Model Pre-training From Quality Estimation.

arXiv.org. Retrieved from https://arxiv.org/abs/2009.

04984.

Wang, L., Zhang, X., Su, H., & Zhu, J. (2023, June 10). A

Comprehensive Survey of Continual Learning: Theory,

Method and Application. arXiv.org. https://arxiv.org/

abs/2302.00487.

Bruun, M., et al. (2019). Impact of a Clinical Decision

Support Tool on Dementia Diagnostics in Memory

Clinics: The PredictND Validation Study. Current

Alzheimer Research, 16(2), 91–101. https://doi.org/

10.2174/1567205016666190103152425.

Saxena, R., Carnevale, K., Saxena, R. R., Moyer, C.,

Sahota, N., Gillett, R. L., & Saxena, A. R. (2019).

Thinking, Creating, Innovating: Utilizing Algorithms in

PBL Setup to Enhance Pathology Education. American

Journal of Clinical Pathology, 152(Supplement_1),

S100–S100. https://doi.org/10.1093/ajcp/aqz119.003.

Saxena, R. R., Nelavala, S. S., & Saxena, R. (2023b).

MuscleDrive – Electromyographic Navigation of a

Four-wheeled Remote-Operated Vehicle: A Novel

Proof of Concept Based on End-to-End Software

Integration. FMDB Transactions on Sustainable Health

Science Letters, 1(2), 107–117. Retrieved from

https://www.fmdbpub.com/uploads/articles/17034055

3189494.%20FTSHSL-70-2023.pdf.

Saxena, R. R., Voss, W., & Mahmood, S. (2024).

Empowering Learners Globally: A User Interface and

User Experience Design Case Study Describing the

Development of a Mobile Application That Bridges the

Educational Divide Through Accessible Instructional

Resources. Authorea Preprints, Techrxiv. https://

doi.org/10.36227/techrxiv.172175659.90097054/v2.

Assessing the Practicality of Designing a Comprehensive Intelligent Conversation Agent to Assist in Dementia Care

663