The SMART4MD trial targeted PwMCI and their primary informal caregivers. SMART4MD is a tablet-based health app developed by Healthbit Ltd [21] specifically for people with MCI or mild dementia and their informal caregivers. The interface design prioritizes simplicity, familiarity, and customization to reduce cognitive load. It is a tablet-first app designed for tablet screens with large interactive zones and simple line charts rather than complex data visualizations for monitoring health progress. Key features of the SMART4MD app are medication reminders, reminders for appointments with health care professionals, and cognitive training through stimulating games and photographs. The app also includes an optional feature that allows users to share information with family members regarding their daily health condition, specific health issues, and QoL [20]. The project strictly followed a user-centered design approach involving testing with dyads (PwMCI and their caregivers) in Spain, Sweden, and other partner countries [22]. The interface was designed for simultaneous use by 2 types of users: PwMCI (the primary users) and caregivers (the secondary users). The design functions as an external memory aid by supporting reminders and daily schedules, thereby reducing the mental effort required to remember daily routines. Features such as calendars and photographs mimic real-world tools, making the app familiar and intuitive for users who are less familiar with technology. The app also allows users to customize features according to personal preferences, thereby enhancing accessibility and adaptability for PwMCI.

In this study, we used data collected from 3 sites in Sweden and Spain: (1) Blekinge Institute of Technology (BTH), Karlskrona, Sweden; (2) Consorci Sanitari de Terrassa (CST), Barcelona, Spain; and (3) Servicio Andaluz de Salud (SAS), Málaga, Spain. The site in Belgium was excluded because of insufficient participation (5 dyads) and incomplete data. Health outcomes were assessed at baseline and at 6, 12, and 18 months. Additional information on the intervention is available in the trial protocol [20]. The SMART4MD trial is reported in accordance with the CONSORT (Consolidated Standards of Reporting Trials) guidelines. The CONSORT checklist is provided in Multimedia Appendix 1.

The Czech Republic was not included in this sample because the Czech study withdrew from the project due to financial constraints. As no participants were recruited from this site, it was not included in the published protocol [20] or in this study.

Regarding the inclusion criteria, the originally specified Mini-Mental State Examination (MMSE) range of 20-26 points was later expanded to 20-28 points. This adjustment was informed by O’Bryant et al [23], who demonstrated that an MMSE score of 28 provides optimal sensitivity and specificity for detecting mild dementia in individuals with self-reported memory complaints. As the revised MMSE range of 20-28 points was incorporated into the SMART4MD trial protocol, this study is consistent with the published protocol [20].

The study was conducted in compliance with the ethical principles outlined in the Declaration of Helsinki [24]. The trial was approved by the regional ethical review board in Sweden (LU Numbers 650-00 and 744-00) and the local ethics committee in Spain (number 02-16-107-029). Dyads provided written informed consent before participation. No financial reimbursement was provided for participation in the trial. Due to the gradual deterioration in cognitive capacity among PwMCI, the study team consistently obtained renewed consent for data collection throughout the trial [20].

Participants were identified through primary care, secondary care services (memory clinics), outpatient clinics, day hospitals, specialist mental health care units, geriatric medicine units, and neurology service units. Clinical databases, such as the Swedish National Study on Aging and Care and the CST network in Spain, were screened for individuals meeting the basic age and cognitive criteria. Study staff or licensed medical providers identified individuals who had experienced cognitive impairment for more than 6 months. Potential participants were given or sent an information package containing a participant information sheet and a standard consent form. According to the standard operating protocol, there was a 24-hour cool-off period between the time a participant agreed to join the study and the meeting at which formal consent was obtained. Participant inclusion began in December 2017 and was expected to end in February 2019. Individuals aged 55 years or older who scored 20-28 on the MMSE were eligible for inclusion [25]. Additional inclusion criteria were that participants were able to manage their medications, were recipients of home care services, had no functional disability that could prevent them from using the SMART4MD app (eg, visual, hearing, or motor impairments), had an actively involved informal caregiver, and had experienced a noticeable decline in cognitive abilities for more than 6 months before inclusion.

Individuals were ineligible to participate if they scored higher than 11 on the Geriatric Depression Scale [26] or had a terminal illness with a life expectancy of less than 3 years, as determined by the research nurse during the initial visit. Furthermore, participants were excluded if they had any other known conditions that could affect cognitive impairment, such as substance abuse or psychiatric disorders including schizophrenia, bipolar disorder, and developmental disorders.

The sample size was determined based on the primary outcome measure, QoL, assessed using the Quality of Life in Alzheimer’s Disease (QoL-AD) scale for PwMCI. The protocol defined a recruitment target of 1200 PwMCI-caregiver dyads, split equally between the intervention and control groups. To detect a small effect size (0.2) between the control and intervention groups with 80% power using a 2-sided Student t test at a 5% significance level, a minimum of 394 participants per group (788 in total) was required. Assuming a 20% dropout rate, the required sample size increased to 493 participants per group (986 in total). The assumed minimum detectable effect size of 0.2 refers to a standardized mean difference, that is, 0.2 SD units on the QoL-AD outcome, corresponding to a small effect size. As the primary outcome for the sample size calculation was the PwMCI QoL-AD score, no separate adjustment was made for within-dyad correlation in the power calculation. Furthermore, the sample size calculation did not include an explicit inflation factor for site-level clustering or correlation; the estimate was based on a simple 2-group comparison rather than a cluster-adjusted design effect.

The SMART4MD trial successfully enrolled 1083 dyads consisting of PwMCI and their primary informal caregivers, which exceeded the minimum required sample size based on the power calculation. Twenty-six dyads were excluded from the analysis because the primary informal caregiver changed during the trial.

The duration of the trial was 18 months. Dyads were randomized, based on predetermined sequences, to either the intervention group (n=537) or the control group (n=541) in blocks using a 1:1 ratio through an internet-based randomization system (ALEA; FormsVision BV) established by the Anglia Ruskin Clinical Trials Unit. Dyads, research nurses, and researchers were not blinded to group assignments because of the nature of the intervention. Participants in the intervention group received standard care and the SMART4MD app on tablets for daily use by PwMCI. All dyads in the intervention group underwent a 1.5-hour training session on app use conducted by a research nurse. As data-enabled tablets and training were provided to trial participants, there were no specific technical prerequisites for inclusion regarding experience with or ownership of smartphones or tablets.

Dyads in the control group received standard care only, representing the routine health care services provided outside the trial. Care varied across participants depending on several factors, including general health, comorbidities, and health-seeking behavior. In Sweden, standard care for older adults generally includes an annual in-person physician visit for routine check-ups or a telephone consultation to discuss overall health and, if necessary, annual laboratory tests and prescription renewals. In Spain, standard care generally includes appointments at a health center, where diagnostic, therapeutic, and preventive services are provided, along with referrals to specialist services when needed [27].

The primary outcome of the trial was QoL, measured using the QoL-AD score for PwMCI at the 18-month follow-up. Other outcomes were treated as secondary outcome measures and categorized into confirmatory and exploratory end points to address multiplicity concerns and maintain statistical rigor. Confirmatory outcome measures for PwMCI included health-related QoL measured using the EQ-5D-3L, global cognitive function measured using MMSE scores [25], and medication adherence. Caregiver burden, measured using the Zarit Burden Interview (ZBI) score, was included as a confirmatory outcome for informal caregivers. All remaining outcome measures, subgroup analyses, and sensitivity analyses were considered exploratory.

Initially developed for persons with Alzheimer disease, the QoL-AD is a disease-specific instrument used to measure QoL, although its application has since expanded to other dementia disorders [28]. The scale consists of 13 items assessing physical health, energy, mood, living situation, memory, family, marriage, friends, self as a whole, ability to do chores, ability to engage in enjoyable activities, financial situation, and life as a whole. Responses are rated as 1 (poor), 2 (fair), 3 (good), or 4 (excellent), and summed to generate a total score, with higher scores indicating better QoL [29]. Ratings from PwMCI were obtained through interviews using standardized instructions, whereas caregiver ratings were obtained through a questionnaire. A weighted composite score was calculated by combining the PwMCI and informal caregiver ratings of the PwMCI’s QoL, with twice the weight assigned to the PwMCI’s self-assessment, following Logsdon et al [30].

The EQ-5D-3L questionnaire is a generic preference-based instrument used to measure quality-adjusted life years (QALYs). It consists of 5 domains: mobility, self-care, usual activities, pain/discomfort, and anxiety/depression. Each domain has 3 levels: no problems, some problems, and extreme problems, resulting in 243 possible health states [31]. These health states are converted into a score ranging from 0 (equivalent to death) to 1 (perfect health) using population preference values, commonly referred to as “tariffs” [32]. As tariffs differ across countries, the use of country-specific tariffs is recommended when calculating QALYs [33].

This study used the European EQ-5D Visual Analogue Scale (EVAS) tariff, which aggregates health state valuations from 6 EuroQol Group countries, including Spain and Sweden. This single European value set provides a standardized approach to measuring health utility across diverse populations. A key justification for this approach is the substantial agreement in EQ-5D VAS health state valuations across diverse European populations, supporting the use of a unified value set for cross-country comparisons [34]. Although the EVAS measures health-related QoL on a scale ranging from 0 (worst imaginable health) to 100 (best imaginable health), health utility scores for QALY calculation must be anchored between 0 (dead) and 1 (full health). To meet this requirement, we rescaled the EVAS scores into health utility weights for QALY calculation following the method described by Greiner et al [34]. The following formula was applied to convert the 100-point scale to a 0-1 utility index.

where VAS_i is the raw score (0-100) given by the respondent for a specific health state; VAS_{dead_median} is the raw score the respondent gave to the state “dead”; and VAS₁₁₁₁₁ is the raw score the respondent gave to “best imaginable health.”

Furthermore, we used the median value for death (2.0) rather than the mean value (5.3) for rescaling, based on the statistics reported in Table 4 of Greiner et al [34]. This choice was made to ensure that the anchors reflected the valuation of the typical respondent, as the median is more robust to skewed distributions and extreme outliers commonly observed in VAS data. In addition, using the median more accurately represents the consensus within the European sample that the worst health state is perceived as only marginally better than death. For example, if a patient’s VAS score for a specific health state was 78, then:

The resulting rescaled EVAS value is 0.7944 based on the aforementioned formula.

These scores were used to calculate accumulated QALYs over the trial period using the area-under-the-curve approach [35].

The MMSE is a standardized test of cognitive function in older adults that assesses memory, attention, orientation, visuospatial skills, and language [25]. The maximum MMSE score is 30, whereas a score lower than 24 indicates dementia; however, the cutoff score for MCI has not been consistently established [36]. The Instrumental Activities of Daily Living (IADL) is a validated tool used to assess the functional abilities of older adults in performing tasks necessary for independent living [37]. These tasks include shopping, housekeeping, meal preparation, laundry, independent travel, financial management, and medication management [37]. The summary score ranges from 0 (low function, dependent) to 8 (high function, independent) [37].

The outcome measures for informal caregivers were QALYs and caregiver burden assessed using the short-form 12-item ZBI (ZBI-12). The ZBI-12 is a 12-item self-report instrument that measures the impact of informal caregiving on the caregiver’s health and social life [38]. Each item is scored on a 5-point Likert scale ranging from “never” to “almost always,” with a total score ranging from 0 to 48, where scores of 17 or higher indicate a high caregiving burden [38].

At baseline, participating PwMCI registered up to 10 long-term prescribed medications, defined as medications intended to be taken for at least 6 months. Medication adherence was assessed for the 30 days preceding each follow-up visit for the 2 most frequently taken medications in tablet or capsule form (U Isaksson and J Frögen, unpublished data, 2018). A research nurse contacted participating PwMCI by telephone at 5, 11, and 17 months to confirm that they were taking the same 2 medications identified at baseline or to update the database if changes had occurred in their long-term prescribed medications. Dyads were asked to bring empty medication packaging used during the previous 30 days, remaining pills (or pill equivalents), and documents related to prescription history to the follow-up visits. Follow-up assessments at 6, 12, and 18 months were conducted for participants in both groups by research nurses using pill counts (number of pills taken). PwMCI who experienced changes in their medication regimen (eg, dose or frequency) were excluded from adherence calculations to prevent regimen modifications from biasing the results.

Overall medication adherence was measured by dividing the number of pills taken by the number of pills prescribed for each of the 2 drugs separately and expressing the result as a percentage. This percentage variable was then transformed into a binary variable representing overall adherence, where adherence rates between 80% and 110% were considered adherent, as described by Anderberg et al [20]. The composite measure of overall adherence for both drugs classified PwMCI as adherent only if they were adherent to both medications. PwMCI who were adherent to only 1 drug or who were prescribed only 1 drug were classified as nonadherent. Therefore, PwMCI prescribed only 1 drug were included only in the supplementary analysis of “change in medication adherence for at least one drug” and excluded from the analysis of “change in medication adherence for both drugs” to avoid measurement bias.

The primary analysis followed a modified intention-to-treat approach, including randomized dyads with available outcome data at the relevant follow-up. Missing outcome data were not imputed in the primary analysis and were addressed in sensitivity analyses. Baseline characteristics were compared between the intervention and control groups to confirm balance achieved through randomization. We also performed a dropout analysis at 18 months by comparing baseline characteristics between dropouts and nondropouts in the intervention and control groups.

Repeated measurements were collected at multiple time points (baseline, 6, 12, and 18 months), so differences between the intervention and control groups for each outcome were analyzed using mixed-effects models [39]. Results are presented as means with SEs and mean differences with 95% CIs. We fitted mixed-effects models including intervention group (SMART4MD vs control) and time (baseline, 6, 12, and 18 months) as fixed effects, with a random intercept for participants (clustered within group). The overall intervention effect and the interaction between intervention and time were used to estimate the differential impact of SMART4MD app use on outcomes over time. Mixed-effects logistic regression models were used to estimate medication adherence among PwMCI. In a supplementary analysis, we also fitted mixed-effects models to estimate adherence to at least one drug. All models were adjusted for study sites (BTH, CST, and SAS).

Statistical significance was set at a 2-sided P value <.05. Data analysis was performed using Stata/SE 15.1 software (StataCorp LP).

In this study, QALY was calculated for each follow-up assessment (6, 12, and 18 months). In the SMART4MD trial, an 180-day window with a tolerance of +18 or –18 days was allotted for conducting each visit. However, as is often the case in trials, some participant visits occurred outside the required 180-day window period. Hence, instead of relying on the predetermined 180-day interval, we used the actual number of days between 2 visits to calculate QALY at each follow-up assessment. To determine the total QALYs for the entire trial period (18 months), we summed the QALYs obtained at each follow-up assessment and excluded participants who missed any follow-up assessment.

For dyad-level QALYs, PwMCI and caregiver QALYs were summed to represent the total dyadic health benefit, and results are presented as mean (SD) or mean differences.

Results for exploratory analyses are reported with 95% CIs and nominal P values without formal adjustment for multiplicity and should therefore be interpreted as hypothesis-generating. Multiplicity adjustment for confirmatory outcomes was applied separately for PwMCI (medication adherence, MMSE, and EQ-5D-3L) and informal caregivers (ZBI scores) using the Holm-Bonferroni method [40].

The baseline demographic characteristics of PwMCI and informal caregivers in the intervention and control groups are presented in Table 1. Figure 1 presents the study flowchart.

There were no significant differences in medication adherence between groups at the 6-, 12-, or 18-month follow-ups (Table 4). At the 18-month follow-up, the estimated probability of adhering to both prescribed drugs was slightly higher in the SMART4MD group than in the control group; however, this difference was not statistically significant (absolute difference in probabilities=0.09; P=.12). The results for medication adherence to at least one drug also showed no differences between groups (see Table S1 in Multimedia Appendix 2).

Overall, the intervention showed benefits for PwMCI in terms of composite QoL-AD scores and nominally significant benefits for informal caregivers in terms of EQ-5D-3L index scores at 18 months (nominal P=.04)); however, no significant differences were found for the dyads at 18 months (nominal P=.21). There was no significant impact on medication adherence (6 months: P=.58; 12 months: P=.53; and 18 months: P=.12). All confirmatory outcomes for PwMCI (medication adherence, MMSE score, and EQ-5D-3L) and informal caregivers (ZBI scores) were nonsignificant before adjustment and remained so after Holm-Bonferroni correction (all P_adj>.99).

Attrition refers to the exclusion of participants from the final analysis, usually because of factors such as dropout from the RCT or missing data. The failure rate of dyads to respond to all health outcomes at the 18-month follow-up was 41.9% (225/537) in the intervention group and 37.7% (204/541) in the control group (P=.16). The stated reasons for exiting the trial were predominantly related to a general unwillingness to continue participation in the trial. PwMCI who dropped out from the intervention group were significantly more likely to be older (P<.001), female (P=.005), unmarried (P=.024), and less likely to live with a spouse (P=.02). They also tended to have worse MMSE and IADL scores at baseline. In the control group, PwMCI who dropped out were less likely to live with a spouse, less likely to have a high level of education, and had worse QoL-AD, MMSE, and IADL scores at baseline (Table S11 in Multimedia Appendix 2). The characteristics of informal caregivers among the dropouts are presented in Table S12 in Multimedia Appendix 2. These caregivers tended to be younger, less likely to live with a spouse, and to have a higher caregiver burden at baseline. At 18 months, multiple imputation under missing at random for imputed QoL-AD scores (mean difference 0.731; nominal P=.02), imputed QALY for PwMCI (mean difference 0.010; nominal P=.13), and imputed QALY for informal caregivers (mean difference 0.014; nominal P=.02) yielded similar estimates in tipping-point analyses applying δ shifts (Table S13 in Multimedia Appendix 2).

For PwMCI, the highest percentage of missing values (501/1078, 46.5%) was observed for QoL-AD scores at the 12-month follow-up (Table S14 in Multimedia Appendix 2). By contrast, for QALY, the highest percentages of missing values were nearly 40% for PwMCI (433/1078, 40.2%) and informal caregivers (436/1078, 40.4%) at the 18-month follow-up (Table S14 in Multimedia Appendix 2).

This study investigated whether the SMART4MD intervention improved QoL and medication adherence among PwMCI, as well as reduced caregiver burden during the 18-month trial period. Significant differences in QoL-AD scores were observed in the intervention group, and this finding remained robust in the sensitivity analysis using imputed composite QoL-AD scores. The model demonstrated a significant interaction between the intervention and time. The results fluctuated because no significant differences were observed at the 12-month follow-up; however, consistent differences in scores were found at the 6- and 18-month follow-ups. The SMART4MD intervention did not demonstrate a significant impact on the confirmatory outcomes for PwMCI (medication adherence, MMSE, and EQ-5D-3L) or informal caregivers (ZBI scores), with results remaining nonsignificant after controlling for multiplicity. Although exploratory results for EQ-5D-3L index scores were significantly different for informal caregivers in the intervention group, no significant improvement was found in total accumulated QALY at the 18-month follow-up. Sensitivity analyses based on imputed QoL-AD scores indicated significantly better QoL for PwMCI in the intervention group. The SMART4MD app includes reminders and other features intended to improve medication adherence; however, the study found that these features did not lead to a significant improvement in medication adherence. It is important to consider potential factors that might influence the outcomes, such as sex, age, cognitive status, and caregiver burden. However, the study did not find significant interaction effects on QoL-AD based on these factors that could explain the lack of improvement in QoL-AD and medication adherence.

QoL-AD scores were higher in the intervention group when imputed QoL-AD scores were used to account for missing data through multiple imputation. This suggests that the positive effect of the intervention on QoL-AD is robust and not substantially affected by missing data. In addition, QoL-AD scores remained consistent in sensitivity analyses using self-reported and proxy-reported scores. This implies that either perspective may be reliable for assessing QoL in PwMCI.

In site-specific analyses, significant effects on MMSE scores and QALY among PwMCI were observed at the SAS site in Spain, as well as improved medication adherence at the 6-month follow-up. These findings are exploratory and should be interpreted with caution, as they may reflect specific contextual factors, population characteristics, implementation nuances, or differences in local health care practices and caregiver support. Furthermore, the findings may be attributable to random chance. Future research should incorporate a priori effect-modifier analyses to formally test these associations and identify the specific factors, such as patient engagement, caregiver support, and local health care resources, that contributed to these positive outcomes.

The findings regarding PwMCI are consistent with existing evidence on the effectiveness of mobile apps for health promotion and disease management. A meta-analysis of 172 papers, primarily consisting of randomized trials, concluded that health-related apps provided a modest advantage in health outcomes compared with standard care [47]. A longitudinal study conducted in Ireland found no significant changes in the well-being of 12 PwD or their caregivers who used a smart health platform called Connected Health over a 12-month period. The Connected Health platform is a model of chronic care delivery that uses a portal to connect care providers and stakeholders, enabling a smooth and continuous flow of information. However, the results are not directly comparable because the Irish study lacked a control group, used a different QoL instrument (Dementia QoL), and had a smaller sample size consisting of patients with advanced dementia [48]. Another randomized clinical trial involving 390 persons aged 65 years and older evaluated the impact of an interactive website called ElderTree. The study found no significant improvements in QoL (Patient-Reported Outcomes Measurement Information System Global Health), independence (IADL), or related outcomes over a 12-month period [49].

We found no improvement in medication adherence with the SMART4MD app, which contrasts with a previous study by Li et al [50], who reported a significant improvement in medication adherence over 12 months using the Perx app. The Perx app included customized reminders and interactive games to incentivize medication adherence. However, the study population was younger and had health conditions other than cognitive decline. Additionally, users were required to upload a photo of their scheduled medication within 1 hour of receiving the reminder for verification [50]. Admittedly, SMART4MD did not include such interactive features. A scoping review also found that gamified apps may promote positive health behaviors, such as medication adherence [51], and may also drive app use [52]. Therefore, future mHealth apps may consider incorporating gamification strategies to support health behavior change, particularly medication adherence.

A direct comparison of our findings with previous studies of mHealth or eHealth interventions was not possible because of differences in the type of intervention, study design, outcome measures, and study duration. Another limiting factor is the limited evidence on the effectiveness of mHealth interventions designed for older adults with MCI or their informal caregivers. Therefore, it is difficult to draw definitive conclusions regarding the effectiveness of mHealth interventions for PwMCI and their informal caregivers based on the current evidence.

The high attrition rate in the current trial is unfortunately considered common in efficacy studies related to mHealth interventions [53]. However, to minimize attrition, SMART4MD trial investigators offered telephone interviews to participants who were unable to attend the 18-month follow-up in person. Consequently, 117 out of 650 (18%) follow-ups were conducted remotely by telephone. Most participants cited the COVID-19 pandemic as the reason for not participating in person at the 18-month follow-up. Although the pandemic may also have contributed to overall trial withdrawal, specific data on this reason for attrition were not collected at the time. Dropout was slightly higher in the intervention group (225/539, 41.7%) than in the control group (204/544, 37.5%), although the difference was not statistically significant. Assuming that dropout occurred randomly, the findings in Tables 2 and 3 provide unbiased estimates of health outcomes. However, we observed statistically significant differences in characteristics between dropouts and nondropouts (Tables S11 and S12 in Multimedia Appendix 2), indicating that missingness may not have been random, based on observed factors such as dyad age, decline in cognitive abilities, and increased caregiver burden. Previous studies have suggested using multiple imputation methods when missing data are assumed to be missing at random [54,55]. Sensitivity analyses based on multiple imputation of the primary outcome (QoL-AD) yielded significant effects (Table 5). Similarly, the exploratory results of the multiple imputation for QALY remained consistent with the base-case results, except that significant effects were observed at the 18-month follow-up for informal caregivers (nominal P=.02and dyads (nominal P=.01; see Tables S5 and S7 in Multimedia Appendix 2). Overall, we conclude that attrition is unlikely to have introduced substantial bias into the results. It is important to emphasize that significant differences in MMSE and ZBI scores between dropouts and nondropouts may suggest that the data were missing not at random. This implies that the likelihood of missing MMSE and ZBI data may have depended on baseline outcomes [56], although only 21 of 434 (4.8%) participants left the trial because of cognitive reasons.

As shown in the flowchart in Figure 1, most participants left the trial because they no longer wanted to continue participation. One possible reason for nonparticipation may have been difficulties in using the features of the SMART4MD app or dissatisfaction with the app relative to participants’ expectations [57]. However, a similar proportion of dyads withdrew from the control group, suggesting that the SMART4MD app itself may not have been the primary reason for attrition, but rather participation in the trial overall. Similar observations have been reported in other studies related to the SMART4MD project [22,58].

The small sample size in the medication adherence analysis reduced the likelihood of detecting statistically and clinically significant differences between groups. It is important to note that the number of participants recruited for the trial (1078 dyads) exceeded the required sample size for the primary outcome (QoL-AD), although it was lower than the required 1174 PwMCI needed to detect statistically and clinically significant differences in medication adherence [20]. At the 18-month follow-up, the sample included only 647 PwMCI for the main analysis and 418 PwMCI for the medication adherence analysis. The main reasons for further attrition in the medication adherence analysis were that participants reported not taking any medications or that the research nurse considered it impossible or inappropriate to count the medication.

In addition to the small sample size, the average number of times PwMCI launched the SMART4MD app was also low, indicating suboptimal utilization of the app. This may suggest that the SMART4MD app was not sufficiently user-friendly, making it challenging for PwMCI to use, as reported by Piculell et al [57]. A review of mHealth apps indicated a strong correlation between simple, easy-to-use, and interactive app features and their usability and acceptability [59]. If app functions are complex and PwMCI are unfamiliar with the app, this may even lead to distress [59]. A qualitative study based on individual semistructured interviews with 16 PwMCI from the SMART4MD trial found that participants did not perceive the app’s reminder feature as useful [57], which may have influenced their engagement with the app. This is further illustrated by the fact that only 195 of 312 (62.5%) PwMCI launched the app between the 6- and 12-month follow-ups, compared with 299 of 312 (95.8%) during the first 6 months. Participants may also have shown a lack of interest because they perceived themselves as too old to use a smartphone or tablet [57]. These findings suggest that participants may not have recognized the potential benefits of developing habits that could become useful as their condition progresses, underscoring the importance of educating participants about the benefits of mHealth interventions. As the nature of mHealth interventions differs from that of medications or surgical procedures, additional effort may be required to help participants appreciate the benefits associated with regular use of such interventions.

One of the inclusion criteria for the trial was that PwMCI had to have an MMSE score between 20 and 28. Although the use of an MMSE cutoff value of 28 is uncommon and associated with some limitations, it has been used in other studies [60]. Furthermore, previous research suggests that an MMSE cutoff score of 28 may be appropriate for identifying individuals with cognitive impairment, including those with mild dementia or MCI [23]. As PwMCI with an MMSE score of 28 had relatively higher cognitive abilities, we conducted sensitivity analyses using an MMSE cutoff of 26.

The main strength of this study is that it was based on a large multicenter RCT conducted across different cultural and health care settings and included both PwMCI and their informal caregivers. Another strength is the use of multiple health outcome measures, which increased comparability with other studies. The main limitation is the high attrition rate, and the dropout analysis revealed significant differences between dropouts and nondropouts, indicating that the results should be interpreted with caution.

Another limitation is the lack of data on user behavior at the 18-month follow-up. Furthermore, there are concerns regarding the quality of the 12-month user behavior data because they only indicated whether the app had been launched. Detailed data on the duration of interactions with the app and the specific functions used were not collected; therefore, objective measures of actual app use were unavailable. Consequently, inferences regarding a causal relationship between app use or reminder exposure and changes in QoL among PwMCI and their caregivers should be interpreted with caution. The main analysis of this study was limited to examining the presence of app use. Therefore, the study did not fully capture the effect of reminder features for medication adherence and appointments on behavioral change. Low app-launch frequency may partly explain the absence of an effect on medication adherence, although the available usage data do not permit conclusions regarding exposure to specific reminder functions or actual engagement with medication-related features. In addition, it is possible that the app itself imposed an additional burden on PwMCI and their caregivers. The reminders may also have had nuisance effects by causing distraction among PwMCI. This interpretation is consistent with findings from the SMART4MD feasibility study, in which some participants reported feeling too old to use the technology and perceived the reminder functions as not useful [22]. Previous research suggests that mHealth apps for dementia are more likely to be adopted when their features are simple [61]. It is therefore possible that the functions of the SMART4MD app were not perceived as sufficiently simple or intuitive by PwMCI. Moreover, the 18-month follow-up period may not have been long enough to identify minimal clinically important differences, as PwMCI consistently reported minimal changes in their condition during this period. For example, for the MMSE, a follow-up period of at least 36 months has been suggested to observe clinically meaningful changes [62]. This observation is consistent with the existing literature [63]. The trial design focused on 2 frequently used medications, which may have introduced an upward bias in adherence reporting. PwMCI may exhibit higher adherence to well-established and frequently repeated routines than to more varied treatment regimens; consequently, our results may overestimate overall patient adherence.

Neither the researchers nor the participants were blinded to the intervention and control groups, which may have introduced bias into the outcomes. However, blinding may be less relevant in pragmatic trials that aim to replicate real-world practice as closely as possible [64]. Future research in this area could consider providing the control group with an app containing limited features to address this concern. Another limitation of the trial was the lack of detailed characterization of informal caregivers. For example, information on the primary caregiver’s relationship to the PwMCI, caregiving hours per day or week, and employment status was not documented. Furthermore, the effect of the SMART4MD app was not stratified by caregiver type (eg, spouse, adult child, or other), which could have provided insight into which groups benefited most from the intervention.

The SMART4MD intervention was associated with a modest improvement in QoL among PwMCI, as measured by the composite QoL-AD score. Evidence of benefits among informal caregivers and dyads was less consistent and should be interpreted as exploratory. Further research is needed in this area, particularly studies with larger sample sizes, longer follow-up periods, and strategies to encourage app utilization. Future studies should also place greater emphasis on reducing dropout rates in mHealth programs.