Phase 1 involves adapting, translating, and validating 4 instruments into the Malay language: Dementia Knowledge Assessment Tool Version 2 (DKAT-2) [22], Dementia Attitudes Scale (DAS) [23], Caregiver Task Inventory—25 items (CTI-25) [24,25], and the Short-form Zarit Burden Interview—12 items (ZBI-12) [26-28]. It is scheduled from December 2024 to July 2025.

There will be forward and backward translations of the questionnaires by two independent translators. Two independent translators will translate the original questionnaires into Malay and then back into the original language. This process helps identify and rectify any discrepancies or inconsistencies in meaning. The drafted Malay versions of the KAP questionnaires will be rigorously validated, encompassing content validation, face validation, construct validation, and an assessment of reliability (Figure 1). Eligibility criteria for Phase 1 are given in Textbox 1.

Content validity will be evaluated by a panel of 6 experts comprising public health physicians, geriatricians, and 1 informal caregiver of people living with dementia who will serve as an experiential expert to ensure contextual and practical relevance for the target population. Each panel member will independently assess all the items for relevance, clarity, simplicity, and the absence of ambiguity, following established content-validity assessment procedures [29-32]. The ratings provided will be used to compute item-level and scale-level content validity indices (CVIs).

Face validity will be examined among 30 dementia caregivers who will assess the clarity and comprehensibility of each item. Their feedback will be used to calculate the face validity index in accordance with recommended guidelines [33-35]. This process will ensure that the instrument is clear and understandable to the target population.

Construct validity will be assessed using exploratory factor analysis (EFA). Based on the recommended participant-to-item ratio of at least 5:1, a minimum of 330 participants is required, with an additional 10% to account for potential attrition, yielding a total target sample of 363 participants. EFA will be performed to evaluate the underlying factor structure and to determine whether the items load appropriately onto the intended domains [36,37]. Participants will be recruited through convenience sampling from the Geriatric Clinic at Universiti Malaya Medical Center (UMMC), Kuala Lumpur, and Hospital Kuala Lumpur. This sample size is expected to provide adequate power and stability for robust factor extraction and model interpretation.

Reliability will be assessed through internal consistency and test-retest procedures. Internal consistency will be evaluated using Cronbach α for each domain and for the overall scale. Test-retest reliability will be examined over a 2-week interval using intraclass correlation coefficient, following established guidelines for reliability assessment [38-41]. This combined approach will ensure stability and consistency of the instrument across time.

In Phase 2, the research assesses the effectiveness of the DemensiaKITA mHealth app intervention among dementia caregivers in Selangor, Malaysia. This phase uses an NRCT with 2 parallel arms (intervention and control) and a 1:1 allocation ratio. The study includes 3 assessment points: baseline (T0), 1 month (T1), and 3 months (T2). The intervention period spans from August 2025 to February 2026. Participants will be blinded to group allocation (single-blinded). Hospital-level allocation is applied to minimize contamination (Figure 2). Eligibility criteria for Phase 2 are given in Textbox 2.

The trial will be conducted in the geriatric clinics of 4 public hospitals in Selangor, Malaysia: (1) Hospital Tengku Ampuan Rahimah, Klang; (2) Hospital Sungai Buloh; (3) Hospital Selayang; (4) Hospital Al-Sultan Abdullah, Universiti Teknologi MARA (UiTM) Puncak Alam. These settings were chosen due to their established geriatric subspecialty services and continuous caseload of dementia caregivers. To prevent contamination, hospitals were allocated based on geographic distribution and patient-flow patterns—Intervention arm: Hospital Tengku Ampuan Rahimah, Klang, and Hospital Al-Sultan Abdullah, UiTM Puncak Alam; Control arm: Hospital Selayang and Hospital Sungai Buloh (Figure 2).

Because follow-up care rarely overlaps across hospitals, this site-based allocation minimizes cross-exposure. The DemensiaKITA app was installed only by the research team on participants’ smartphones in the intervention arm. Caregivers were advised not to share the app or its contents during the study period.

Participants will be identified using purposive sampling among caregivers of registered people living with dementia attending the 4 participating geriatric clinics. Screening and recruitment will be conducted by the research team. Eligible caregivers will receive a full explanation of the study objectives, procedures, confidentiality safeguards, and follow-up requirements.

Because hospitals were preassigned to their respective study arms, no individual-level randomization will be implemented. The study will be conducted as a single-blinded trial whereby participants will not be informed of their allocation to either the intervention or control group and will instead be informed that they will receive “health education material,” as stated in the patient information sheet. In addition, researchers responsible for data analysis will remain blinded to group assignment until all data have been cleaned and finalized to minimize analysis-related bias.

Participants assigned to the intervention hospitals will receive the DemensiaKITA mHealth app. The app will be installed directly by the research team to prevent unauthorized sharing. Participants may access the app freely during the study period.

The DemensiaKITA app is an offline mobile app prototype developed by a team of researchers from Universiti Teknologi MARA, Malaysia, and the Canadian University of Dubai (Figure 3). The app has been granted copyright protection under Universiti Teknologi MARA. It provides essential information on dementia and dementia care, facilitates easy access to nearby health services, optimizes the use of available local resources, and enhances caregiving skills. The app features appropriate and relevant content tailored for the Malaysian context, using the national language (Bahasa Malaysia) as the interface, specifically designed for dementia caregivers in Malaysia (Figure 3 and Textboxes 3 and 4) [12].

To promote engagement with the DemensiaKITA app, caregivers in the intervention group will receive a brief, standardized orientation on the app features at enrollment. Thereafter, the research team will send scheduled reminders via WhatsApp or SMS text messaging to encourage continued use of the app and completion of follow-up assessments. Reminders will be sent once weekly during the first month and every 2 weeks thereafter until the 3-month follow-up.

Caregivers in the control arm will receive standard Ministry of Health (MOH) dementia educational materials, which are the dementia leaflets and video. These materials will be explained in detail during the baseline session to ensure consistent educational exposure and engagement comparable to that provided in the intervention arm [42].

The dementia leaflets and video provided by the MOH Malaysia offer essential information on the nature of dementia, its common symptoms, and the groups at higher risk (Figure 4). They emphasize the importance of early detection and timely diagnosis, describe the available assessments conducted in clinics and hospitals, and outline practical steps to help maintain cognitive and physical well-being. The materials also highlight key measures that may reduce dementia risk, including mentally stimulating activities, regular exercise, healthy lifestyle practices, and effective management of chronic illnesses. In addition, they provide guidance on daily care, safety, behavioral management, and available support resources to assist people living with dementia and their caregivers [42].

Caregiver knowledge will be assessed using the DKAT-2, developed by Toye and colleagues [22] at Curtin University. The DKAT-2 consists of 21 items with Yes/No/Don’t Know response options and measures understanding of the dementia trajectory and evidence-based care approaches. Each item is scored as correct or incorrect, with “Don’t Know” categorized as incorrect. Total scores reflect overall knowledge, with higher scores indicating greater understanding of dementia care. The DKAT-2 demonstrates strong psychometric performance, including high internal consistency (Cronbach α=0.79), supporting its use in evaluating caregiver education and guiding patient-centered interventions [22].

Attitudes toward dementia will be measured using the DAS, developed by O’Connor and McFadden [23]. The DAS comprises 20 items rated on a 7-point Likert scale (1=strongly disagree to 7=strongly agree), capturing cognitive, emotional, and behavioral components of attitudes toward dementia. Higher scores denote more positive attitudes. The scale has demonstrated excellent reliability (Cronbach α=.83‐.85) and a stable 2-factor structure confirmed through exploratory and confirmatory factor analyses [23]. Its broad conceptual coverage makes it suitable for evaluating the impact of educational and psychosocial interventions.

Caregiving practices will be assessed using the CTI-25. The CTI-25 is a shortened and refined version of the original 45-item instrument developed by Clark and Rakowski in 1983 [24], with improvements introduced by Lee and Mok [25] to enhance clarity and reduce redundancy. The instrument contains 25 items across 5 domains: learning to cope with a new role; providing care based on the care recipient’s needs; managing one’s emotional needs; appraising supportive resources; and balancing personal and caregiving demands.

Items are rated on a 3-point Likert scale (0‐2), with higher scores reflecting greater caregiving difficulty. The CTI-25 demonstrates strong psychometric properties, including high internal consistency (α=.93) and construct validity supported by confirmatory factor analysis [25].

Caregiver burden will be measured using the ZBI-12, a validated abbreviated version of the original 22-item ZBI developed by Zarit in 1980 [26-28]. The ZBI-12, translated into Malay in 2023, evaluates key burden domains including privacy, social life impact, emotional strain, health effects, and time-related pressures. The shorter format enables rapid screening and ongoing monitoring while retaining strong psychometric performance. Total scores are categorized as: 0‐10 (no to mild burden); 11‐20 (mild to moderate burden); and 20 or more (high burden) [26-28].

This study received ethical approval from multiple institutional review boards. Approval was first obtained from the Faculty Ethics Research Committee, UiTM on November 7, 2024 (REC/11/2024 [PG/MR/549]). The study was subsequently approved by the Medical Research and Ethics Committee, Ministry of Health Malaysia, on December 16, 2024 (24‐03494-W6I [IIR]), in accordance with the ethical standards of the National Institutes of Health Malaysia. Additional approval was granted by the Medical Research Ethics Committee, UMMC, on May 21, 2025 (2025122‐14637).

Operational approvals were obtained from the Directors of Hospital Kuala Lumpur and UMMC for Phase 1 and from the Director of Selangor State Health Department and the Directors of the 4 participating hospitals for Phase 2. Permission to use all study instruments was formally granted by the respective copyright holders and instrument developers.

Before the study begins, all participants will be informed of the study objectives, procedures, and potential risks and benefits and will provide informed consent. Written informed consent will be obtained prior to enrollment. No financial or material compensation will be provided to participants for their participation in this study.

Participant data will be obtained through both hard-copy questionnaires and electronic forms, then entered into a password-protected database accessible only to authorized research staff. Double data entry and range checks will be performed for key variables to minimize errors. All electronic data will be stored on secure institutional servers with regular backup. Identifiers (names, phone numbers, and identification number) will be stored separately from research data and linked only through a unique study identification code. The DemensiaKITA app does not collect or transmit any identifiable health information; it functions as an information and support tool installed locally on participants’ smartphones.

All personal information obtained from participants will be treated as confidential and used solely for research purposes. Identifiable data will not be included in publications or presentations. Hard-copy documents will be stored in locked cabinets within restricted-access offices. Electronic data will be deidentified for analysis, and only the principal investigator and authorized team members will have access to the code key. Research procedures will comply with the Malaysian Personal Data Protection Act and institutional data protection policies.

Deidentified individual participant data underlying the main trial results may be made available upon reasonable request to the corresponding author after publication of the primary findings, subject to institutional data-sharing policies and approval by the ethics committees.

Content validation will be assessed using the CVI and content validity ratio. To ensure rigorous content validation, an item-level CVI of 0.79 or higher is required for item retention, while a scale-level CVI of 0.80 or higher is necessary to establish excellent content validity for the overall scale. A minimum content validity ratio of 0.99 is established based on expert consultation. This rigorous approach ensures that only the most relevant and important items are retained for subsequent phases of instrument development [29-31,34].

Face validation will be assessed by calculating the face validity index (FVI) to quantify the degree of agreement among respondents regarding the clarity and comprehensibility of the questionnaire items. A sample of 30 respondents was used for this process. Items with an item-level FVI and scale-level FVI of 0.80 or higher were retained. This threshold aligns with recommendations for establishing strong face validity [33-35].

Construct validation will be conducted using EFA to identify the underlying factor structure of the questionnaire. This will confirm that the items are effectively measuring the intended concepts. Additionally, correlational analysis will examine the relationships between different scales within the questionnaire to assess convergent and discriminant validity [34,36,45].

To ensure the questionnaire’s consistency, 2 key aspects of reliability will be evaluated. First, internal consistency will be assessed using Cronbach α, which measures how well the items within the questionnaire work together to measure the same concept. A Cronbach α of .70 or higher will be considered acceptable. Second, test-retest reliability will be assessed using the intraclass correlation coefficient. This determines the stability of the questionnaire’s results over time, with an acceptable value of 0.75 or higher [34,39,41]. This 2-pronged approach ensures the questionnaire produces reliable and consistent results.

All analyses will be conducted using a 2-sided significance level of α=.05. Descriptive statistics will summarize sociodemographic and caregiving characteristics by study arm [46]. Categorical variables will be presented as frequencies and percentages, and continuous variables as means and SDs or medians and IQRs, as appropriate. Baseline comparability between the intervention and control groups will be examined using chi-square tests for categorical variables and independent t tests (or nonparametric equivalents) for continuous variables [46,47].

The primary effectiveness analysis will use generalized estimating equations models to compare changes in knowledge, attitude, practice, and burden scores between the intervention and control groups over time (baseline, 1 mo, and 3 mo). For each outcome, a GEE model with an appropriate link function and working correlation structure will be fitted, including fixed effects for group (intervention vs control), time, and the group-by-time interaction. The group-by-time interaction term will be interpreted as the main estimate of the intervention effect. To account for potential confounding due to nonrandom allocation, the models will adjust for prespecified baseline covariates (age, sex, and relationship with the people living with dementia, caregiving duration, and baseline score of the corresponding outcome). Robust standard errors will be reported [21,48].

Effect sizes (Cohen d for between-group differences in mean change) will be calculated to quantify the magnitude of the intervention effect, in addition to P values and 95% CIs [21,48].

The primary analysis will follow the intention-to-treat principle, in which all participants will be analyzed in the groups to which their hospital was allocated, regardless of their level of app use, exposure to the leaflets, or subsequent adherence. Every effort will be made to obtain outcome data at 1 and 3 months for all enrolled caregivers, including those who discontinue or reduce use of the intervention [49].

Missing outcome data will initially be handled within the GEE framework, which uses all available repeated measurements under a missing-at-random assumption and provides robust standard errors. Patterns and reasons for missing data will be described. If more than a small proportion of outcome data (>20%) is missing, multiple imputation by chained equations will be considered as a sensitivity analysis, with covariates and outcome scores imputed under the missing-at-random assumption. Results from imputed and complete-case analyses will be compared to assess the robustness of the findings [21].

This study was conducted without any specific external financial support or grants from any public, commercial, or not-for-profit funding agencies. Recruitment and baseline data collection are currently underway at the time of manuscript submission (phase 1: 210 participants, phase 2 not yet started at the time of submission). The study findings are expected to be published in December 2026.