Introducing the Adherence Promotion with Person-Centered Technology (APPT) Trial: Rationale, Methods, and Baseline Characteristics

Abstract

Home-based cognitive training programs delivered via computers and tablets hold promise as cost-effective, population-level interventions to prevent or mitigate age-related cognitive decline. However, adherence to such programs is often low. Using message-tailoring techniques and an adaptive algorithm, we developed a person-centered reminder smart system that delivers motivational messages at times when participants are predicted to be available for training activities. This paper presents the background, study design, methodology, and baseline data for a randomized controlled trial examining the system’s efficacy in supporting adherence to cognitive training. A total of 199 cognitively normal, community-dwelling older adults aged 62 to 88 were randomly assigned (1:1) to either the smart reminder or the control condition. Participants were instructed to engage in training activities for 30 minutes per day, five days per week, for 18 consecutive weeks. Those in the smart reminder condition received personalized messages, delivered at optimized times, and with targeted content, while those in the control condition received generic messages at a fixed time. Adherence rate will be the primary outcome measure, calculated and compared across conditions. Findings from this study will have implications not only for adherence support in cognitive training but also for broader applications of technology-mediated smart reminder systems, including physical exercise, nutrition, medication management, telehealth, and social connectivity. By enhancing intervention engagement, these systems have the potential to improve the health and well-being of older adults on a large scale.

1. Introduction

Home-based cognitive training, delivered via computers and tablets, can be a cost-effective, large-scale strategy for preventing or slowing cognitive decline (see, e.g., [1], [2] for meta-analytical findings supporting its benefits; but also see [3], [4] for contrary evidence). However, prior research indicates that adherence to cognitive training is often low [5], [6], [7]. This pattern has also been observed in our laboratory, where older adults completed only 59% of required sessions in a 12-week intervention [8] and 65% in an 8-week intervention [9]. Low adherence poses a significant challenge, both in evaluating the effectiveness of cognitive training programs [10] and in implementing them successfully on a large scale.

Automated electronic reminders have shown promise in promoting adherence. These strategies, which use telecommunication-based messaging, often rely on tailored messages [11] that are personalized based on pre-assessed individual differences [12]. Research has consistently demonstrated that tailoring increases the personal relevance of messages, which in turn enhances adherence to behavioral interventions (see, e.g., [13], [14],[15], [16] for reviews). Beyond message content, the timing of message delivery may also influence its effectiveness in promoting adherence. Studies suggest that behavioral interventions are more effective when delivered at moments when individuals are most receptive and able to take action. These approaches, known as just-in-time adaptive interventions [17], have shown greater success compared to interventions that do not adjust timing based on user availability (see [18] for a review).

Effective adherence support may also require an understanding of facilitators and barriers to adherence. Although it remains unclear which factors predict adherence to cognitive training and other health behaviors that support cognitive health, extensive research has explored the adoption and maintenance of health behaviors more broadly [19]. The Health Belief Model (HBM) identifies several key predictors of behavior, including perceived susceptibility to and severity of a health condition, perceived benefits and barriers related to the behavior, and self-efficacy in carrying out the behavior [19]. In addition to attitudes and beliefs, HBM also highlights internal and external cues—such as changes in symptoms and external reminders—as triggers for action. The Theory of Reasoned Action (TRA) and the Theory of Planned Behavior (TPB) propose that behavioral intention is the strongest predictor of actual behavior [20]. While sharing some core elements with HBM, these theories also emphasize the role of social influences on both behavioral intentions and actions. Additionally, demographic factors and personality traits have been suggested as modifying influences that shape perceptions and indirectly impact health-related behaviors [19][20].

Empirical research supports the explanatory power of HBM, TRA, and TPB in predicting a range of health behaviors, including participation in cervical screening [21], mammography screening [22], and COVID-19 vaccination [23]. Within adherence research, systematic reviews have identified additional barriers and challenges. Individuals with more depressive symptoms, lower self-care, less social support, poorer self-rated health, busier schedules, greater medication burden, and lower motivation are more likely to be non-adherent to medication regimens [24], [25], [26], [27], [28]. Furthermore, a history of poor adherence and non-adherence to other therapies are strong predictors of medication non-adherence [28].

Although awareness of aging is not typically examined as a predictor of adherence, it may influence adherence through its effects on perceptions and play a unique role in cognitive health interventions. Awareness of aging is theorized to shape age-related outcomes by influencing social contexts, expectations, goals, behaviors, experiences, and available resources [29]. Perceptions of susceptibility to and severity of cognitive decline, as well as expectations regarding the effectiveness of cognitive training in preventing decline, may be among the many beliefs shaped by awareness of aging. Specifically, older adults with more negative attitudes toward aging may be more pessimistic about the aging process and view cognitive decline as inevitable. These perceptions and beliefs can influence goals and behaviors, potentially leading to earlier disengagement from cognitively challenging activities when faced with obstacles [30] or a higher likelihood of non-adherence to interventions and healthy behaviors when encountering the same barriers [31].

The goal of the Adherence Promotion with Person-Centered Technology (APPT) trial is to evaluate the benefits of an adaptive, tailored smart reminder system in improving engagement with a cognitive training program and to examine individual differences that may influence adherence. Based on prior research, we expect the adaptive, tailored smart reminder system to be more effective in promoting adherence compared to a standard support system that delivers generic messages at a fixed time. We also expect adherence to be predicted by factors identified in the preceding section.

Older adults were recruited to participate in a cognitive training study and asked to engage with the intervention program (Mind Frontiers, Aptima Inc.) for 30 minutes per day, five days per week, over a four-month (18-week) period. Participants had the flexibility to complete training sessions at any time and location of their choosing to better reflect real-world conditions for large-scale interventions. They were randomly assigned to one of two conditions: the smart reminder condition, in which they received tailored messages with adaptive timing, or the control condition, in which they received generic messages at a fixed time.

This paper presents the trial design, recruitment activities, and participant characteristics.

2. Methods

2.1. Overview of study design

This trial is registered on ClinicalTrials.gov (Identifier: NCT05016856). It is a single-site randomized controlled trial (RCT) conducted at Florida State University (FSU) in Tallahassee, FL. Following a telephone screening and baseline assessments, eligible participants were randomly assigned to either the smart reminder or control condition (Figure 1).

Figure 1.

CONSORT diagram for the trial

After randomization, participants completed pre-test assessments and received a touch-screen tablet with the intervention program installed. They were also provided with training videos and take-home manuals to guide them through the program. Participants were instructed to engage with the intervention for 30 minutes per day, five days per week, over four months. At the end of the study, all participants completed post-test assessments.

To gain further insight into adherence patterns, a subset of participants representing high, medium, and low adherence levels from both groups was invited to participate in interviews about their study experience. The trial followed highly standardized protocols for recruitment, screening, assessments, intervention administration (including take-home manuals), and data transfer from intervention devices. The study protocol was approved by the Institutional Review Board (IRB) at Florida State University.

2.2. Eligibility criteria

Participants were eligible for the study if they were 65 years of age or older, planned to remain in the Tallahassee, Florida area for the next six months, had normal or corrected-to-normal vision and hearing, and scored above 35 on the Telephone Interview for Cognitive Status–Modified (TICS-M, [32]).

Participants were ineligible if they had significant visual or hearing impairments that cannot be corrected, had cognitive impairment (TICS-M score of 35 or below [33]), had arthritis in their hands, or self-reported a medical condition that could limit participation, such as a terminal illness, severe motor impairment, Alzheimer’s disease, Parkinson’s disease, or another neurodegenerative disorder. Additional exclusion criteria included an inability to read at a sixth-grade level or higher and prior training with Mind Frontiers (Aptima Inc.) in previous interventions.

2.3. Intervention conditions

All participants were provided with tablets preinstalled with Mind Frontiers software package along with training materials on how to play the games. The Mind Frontiers software package consists of seven different gamified cognitive tasks (detailed in Materials section) dressed in Western (cowboy) themes. Game play data were uploaded to a server at FSU in real time. Participants were instructed to play 30 minutes at any time of their choice within a day, five days of their choice out of the seven days within a week, for 18 consecutive weeks. They were encouraged to play all the available games. No specific instructions were given on the order in which they should play the games or what games to play on which day.

The number of game sessions opened and played within a day were actively monitored to determine whether to send a message the next day. A message would be sent if a participant had not played any of the games for a total number of two, four, and six days within the week. Messages were sent to participants’ cellphone numbers via short message service (SMS). A scheduled message would be cancelled if the participant played any of the games on the day of the scheduled message before the scheduled time. Overall, a participant would receive at most three messages per week. They would not receive any messages if they played on more days than required.

2.3.1. The smart reminder condition

When reminder messages were sent, a motivational message that aligned with the participant’s reason for participating was randomly selected from the message library and delivered at the hour the participant was most likely to engage in training, as predicted by a machine learning algorithm.

The algorithm dynamically adjusted hourly weights (0–23) across days of the week (Monday–Sunday, 0–6), increasing or decreasing them based on user activity within those time slots. To model temporal engagement patterns, we implemented a week-by-week weighting algorithm in which each day-hour time slot initially starts as empty (i.e., no assigned weight). To seed the system during the first week, before any behavioral data was available, participants were asked to indicate the time of day they might expect to use the program on each day of the week. These self-reported preferences were treated as initial activations and were assigned a baseline weight of 1.0. This ensured that reminder messages could be delivered even in the absence of prior engagement behavior on every possible day of the week from the very start of the trial. From that point forward, these seeded activations were treated identically to behaviorally observed activity in the weighting algorithm.

In each subsequent week, the algorithm updated all previously activated time slots based on the participant’s behavior that week. If a participant engaged with the program during a time slot that had already been activated, the associated weight increased by 50% (i.e., was multiplied by 1.5). If a previously used time slot was not activated during that week, the weight was reduced by 50% (i.e., multiplied by 0.5). However, weights were not permitted to fall below 1.0, ensuring that once a time slot had been used, it retained a minimal level of influence in future scheduling decisions. Time slots that had never been used remained empty until first-time activity was detected, at which point they were initialized with a weight of 1.0 and included in subsequent weekly updates. This approach allowed the system to dynamically prioritize time slots with consistent engagement, remain sensitive to changing patterns of behavior, and accommodate new time slots as they emerged.

In terms of readable pseudo code, after the initial seeding, this can be approximated by:

FUNCTION updateWeightsForWeek(player, currentWeekActivity)

FOR EACH timeSlot IN player.timingData:

IF player.timingData[timeSlot] IS NOT null:

IF timeSlot IS IN currentWeekActivity:

SET player.timingData[timeSlot] = player.timingData[timeSlot] * 1.5 ELSE:

SET player.timingData[timeSlot] = player.timingData[timeSlot] * 0.5

IF player.timingData[timeSlot] < 1.0:

SET player.timingData[timeSlot] = 1.0

FOR EACH timeSlot IN currentWeekActivity:

IF player.timingData[timeSlot] IS null:

SET player.timingData[timeSlot] = 1.0

SAVE player.timingData

END FUNCTION

Weights were rounded to the nearest whole integer. When a reminder message was needed, it was delivered in advance of the hour corresponding to the time slot with the highest weight for that day of the week. If multiple time slots had the same weight, the system prioritized the one that had been most recently updated.

As a concrete example, during intake, a participant might predict that they would engage with the intervention on Saturday at 1 p.m., resulting in a weight of 1 prior to the first week of engagement. During the first week, if they used the intervention on Saturday at 1 p.m., the weight would be updated to 2 (1 × 1.5, rounded). If they again used the intervention on Saturday at 1 p.m. during the second week, the weight would be updated to 3 (2 × 1.5, rounded). In the third week, if the pattern continued, the weight would be updated to 5 (3 × 1.5, rounded). However, if no engagement occurred during the fourth week, the weight would be reduced to 3 (5 × 0.5, rounded). Note that this results in a steeper decay than growth function (e.g., four steps to go from 1 to 8, but only three steps to go from 8 to 1). This was intentional to allow the system to more quickly adapt to participants’ changing schedules from week to week if necessary.

Tailored messages were designed to reflect three of the most commonly endorsed reasons for participating in research studies: maintaining cognitive health, contributing to scientific research, and enjoying games [34]. Participants self-report their reason to participate by ranking the three reasons and received messages that aligned with their highest-ranking reason for participation. Messages were developed and pilot-tested before the trial [35]. The research team iteratively created 25 messages to address each of the three reasons, resulting in a total of 75 unique messages.

Messages related to cognitive health reinforced the idea that playing the training games could help maintain cognitive function (e.g., Keep your mind active with some brain games). Messages in the scientific contribution category emphasized that participation provided valuable data for advancing research (e.g., Each game you play provides valuable data about cognition). Finally, messages in the enjoyment category highlighted the fun and engaging nature of the games (e.g., Take a moment to enjoy your favorite brain game). The full message library is provided in the appendix.

2.3.2. The control condition

When reminder messages were sent, a generic message was scheduled for delivery at 9:00 AM (EST). The message read: “Please don’t forget to play Mind Frontiers. Thank you.”

2.4. Contact Schedule

Participants were recruited through various methods, including participant registries, advertisements in local newspapers, and referrals from other participants. Interested participants contacted researchers and completed a telephone screening to determine eligibility. Those who qualified received electronic links to baseline assessments via email after providing informed consent. Participants were then randomly assigned to study conditions.

After completing the baseline assessments, all participants received a tablet preloaded with Mind Frontiers games, electronic links to training videos on how to play the games, and user manuals for both the tablet and the games. Researchers did not initiate contact during the four-month intervention but remained available via phone, email, and video conferencing to assist with troubleshooting, technical issues, or other inquiries.

At the end of the four-month intervention, participants received electronic links to post-test assessments via email. Once a participant completed the post-test, a researcher calculated their adherence rate (total number of days played divided by the total number of required days). A subset of participants representing low, moderate, and high adherence levels was invited via email to participate in a follow-up interview. Those who agreed took part in a 30-minute Zoom interview about their experience.

A researcher later visited each participant’s home to collect the tablet, and participants received $400 as compensation for their participation. All researchers interacting with participants were blinded to treatment conditions.

2.5. Measures

Table 1 provides a description of each measure along with reliability statistics, where applicable.

Table 1.

Description of assessment batteries

Construct	Name of Measure	Description
Screening
Cognitive Screening	Modified Telephone Interview for Cognitive Status (TICS-M [32])	A 13-item test of cognitive functioning. Scores range from 0–50.
Potential Predictors of Adherence Derived from Health Behavior Literature
Perceived Susceptibility	Perceived Susceptibility of Cognitive Decline [49]	A 4-item scale measuring one’s beliefs about the chances of experiencing cognitive decline, Alzheimer’s disease or dementia. Scores range from 1–5. A higher score indicates higher perceived susceptibility (Cronbach’s α = .829).
Perceived Severity	Perceived Severity of Cognitive Decline [49]	A 9-item scale measuring one’s belief about how serious cognitive decline, Alzheimer’s disease or dementia and their sequelae are. Scores range from 1–5. A higher score indicates higher perceived severity (Cronbach’s α = .791).
Self-efficacy	Self-Efficacy of Cognitive Training [49]	A 12-item scale measuring confidence in one’s ability to adhere to cognitive training programs. Scores range from 1–7. A higher score indicates lower self-efficacy to adhere (Cronbach’s α = .960).
	The General Self-Efficacy Scale [50]	A 10-item scale measuring perceived self-efficacy. Scores range from 10–40. A higher score indicates higher self-efficacy (Cronbach’s α = .892).
Perceived Benefit	Brain Training Expectation Scale [51]	A 7-item scale measuring perceived effectiveness of brain training. Scores range from 1–7. A higher score indicates higher perceived effectiveness (Cronbach’s α = .910).
Barrier - Busyness	Martin and Park Environmental Demands Questionnaire [52]	An 11-item scale measuring the level of self-reported environmental demands of day-to-day events. The scale consists of 2 dimensions (Busyness, Routine). A higher score indicates a higher level of busyness and routine. Cronbach’s α for Busyness and Routine are .835 and .780, respectively.
Barrier - Depression	Center for Epidemiologic Studies -Depression Scale (CES-D [53])	A 20-item scale assessing depressive symptoms. Scores range from 0–60. A higher score indicates a higher level of depressive symptoms (Cronbach’s α = .882).
Barrier - Health	36-Item Short Form Survey (SF-36 [54])	A 36-item scale that assesses 8 dimensions of health and well-being. Cronbach’s αs for Physical Functioning, Role Limitations, Role Limitations due to Emotional Problems, Energy/Fatigue, Emotional Well-Being, Social Functioning, Pain and General Health are .906, .857, .700, .903, .831, .849, .817, and .790 respectively.
Barrier – Technology Use	Mobile Device Proficiency Questionnaire (MDPQ-16 [55])	A 16-item scale assessing proficiency in using mobile devices in 8 domains (Mobile Device Basics, Communication, Data and File Storage, Internet, Calendar, Entertainment, Privacy, Troubleshooting and Software Management). Total score ranges from 5–40. A higher score indicates a higher level of proficiency. Cronbach’s α for the scale is .912 and Cronbach’s αs for the subscales range from .616 to .976.
	Technology Readiness Index (TRI 2.0 [56])	A 16-item scale measuring people’s propensity to embrace and use new technologies. The scale consists of 4 subscales. A higher score indicates a higher level of readiness to embrace new technologies. Cronbach’s α for the scale is .870. Cronbach’s αs for Optimism, Innovativeness, Discomfort, Insecurity are .791, .902, .760, and .715 respectively.
Barrier – Effort and Workload	NASA-TLX [57]	A 10-item scale measuring perceived workload of a task. The scale is modified to reflect expected workload for baseline assessments. A higher score indicates higher expected workload.
Barrier - Social Support	Interpersonal Support Evaluation List [58]	A 12-item scale that assesses overall social support and 3 dimensions of social support (appraisal, belonging and tangible). A higher score indicates more support. Cronbach’s αs for Overall Support, Appraisal, Belonging, and Tangible Support are .896, .844, .810, and .763 respectively.
	Processes of Exercise Behavior Change Scale [59]	3 items assessing the amount of social support available for behavioral change, modified to reflect expected social support for a generic cognitive training program. A higher score indicates more social support for change (Cronbach’s α = .825).
Cue to Action – Dementia Concerns	Dementia Worry Scale [60]	A 12-item scale measuring dementia worry. Scores range from 12–60. A higher score indicates greater worry about dementia (Cronbach’s α = .922).
Cue to Action – Subjective Cognitive Decline	Perceived Deficit Scale [61]	A 5-item scale measuring subjective cognitive declines. Scores range from 0–20. A higher score indicates a higher level of perceived cognitive dysfunction (Cronbach’s α = .704).
Motivation	Brain Training and Independence Survey [62]	7 items assessing amount of time one is willing to invest in daily cognitive training to extend their functional independence by different amount of time (ranging from 1 week to 3 years).
Social Influences	Descriptive Norm	1 item measuring one’s belief about whether their partner or best friend would adhere to a cognitive training program. Scores range from 1–7. A higher score indicates higher likelihood for the referent to adhere.
	Injunctive Norm	1 item measuring one’s belief about whether their partner or best friend would approve of adhering to a cognitive training program. Scores range from 1–7. A higher score indicates higher likelihood for the referent to approve.
Behavioral Intention	Behavioral Intention	1 question measuring one’s intention to perform the behavior. Scores range from 1–7. A higher score indicates higher perceived likelihood.
Adherence Habits	Health Promoting Lifestyle Profile II (HPLP II [63])	The 8 items in Physical Activity subscale and the 9 items in the Nutrition subscale were administered in this study. The Physical Activity subscale assesses one’s involvement in regular light, moderate, and vigorous activities. The Nutrition subscale assesses one’s consumption of foods essential for sustenance, health, and well-being. Scores range from 1–4. A higher score indicates a healthier lifestyle. Cronbach’s αs for Physical Activity and Nutrition are .840 and .717 respectively.
	Adherence Starts with Knowledge (ASK-20 [64])	A 20-item scale assessing barriers to medication adherence. Scores range from 10–100. A higher score indicates greater barriers.
Self-care	Appraisal of Self-care Agency Scale Revised (ASAS-R [65])	A 15-item scale assessing self-care agency. The scale has 3 dimensions (Having power for self-care, Developing power for self-care, Lacking power for self-care). A higher score indicates having more power over self-care (Cronbach’s α = .840). Cronbach’s αs for Having power for self-care, Developing power for self-care, Lacking power for self-care are .846, .672, and .683 respectively.
Demographics	Demographics Characteristics	Questions include, age, education, occupational status, income, race, ethnicity, and living arrangements.
Personality	Big Five Inventory (BFI-2 [66])	A 60-item scale assessing the Big Five personality domains and facet traits nested with in each domain. Scores range from 1–5. A higher score indicates stronger endorsement to the corresponding trait. Cronbach’s αs for Openness, Conscientiousness, Extraversion, Agreeableness, Neuroticism are .795, .862, .841, .754, and .891 respectively. Cronbach’s αs for the 15 facets range from .461 to .865.
	Need for Cognition [67]	An 18-item scale assessing an individual’s tendency to engage in and enjoy effortful cognitive endeavors. Scores range from 18–90. A higher score indicates a stronger desire to engage in cognitively challenging activities (Cronbach’s α = .909).
Aging Attitudes	Attitude Toward Own Aging (ATOA [68])	A 5-item scale assessing one’s attitudes and views on their own aging. Scores range from 1–5. A higher score indicates more positive aging attitudes (Cronbach’s α = .754).
	Subjective Age [69]	1 item assessing how old one feels currently in general.

2.5.1. Primary Outcome: Adherence Differences between Groups

The primary outcome of the trial is the overall adherence rate to the cognitive training program over four months. Adherence will be calculated as the number of days played, as recorded on the server, divided by the total number of required play days. G*Power (http://www.gpower.hhu.de/) was used to estimate statistical power. The analysis indicated that 86 participants per group would be required to detect a medium effect (Cohen’s f = .25) using ANOVA with alpha set to .05 and power set to .90. To account for possible attrition or data loss prior to intervention delivery, we increased the sample size to 95 per group.

In two previous Mind Frontiers studies of shorter duration that did not involve reminders, adherence was poor, supporting a reasonable expectation of adherence challenges in the current study. These studies reported adherence rates of 59% and 65% over 12-week and 8-week periods, respectively [8, 9].

2.5.2. Secondary Outcome: Predictors of Adherence

Predictors of adherence were assessed before the trial began, and the choice of predictors was either theory based or based on previous empirical findings. These included perceived susceptibility to and severity of cognitive decline, self-efficacy (both general and specific to cognitive training), and potential barriers to adherence, such as busyness, depression, physical health, technology proficiency, technology readiness, effort and workload, and social support. Additional factors measured included cues to action (dementia concerns, subjective cognitive decline), motivation, social influences (descriptive and injunctive norms), behavioral intentions, adherence to activities in other domains (medication adherence, healthy lifestyle behaviors), self-care, personality traits (Big Five Inventory, Need for Cognition), and awareness of aging (subjective age, attitudes toward one’s own aging). Name and brief descriptions of each measure are presented in Table 1. We plan to conduct factor analyses to reduce the number of predictors as we expect several measures will be correlated. For regression-based analyses involving predictors of adherence, each group is sufficiently powered (90 percent) to detect a medium effect (Cohen’s f² = .15, two-tailed, alpha = .05). The full sample, with group included as a dummy-coded variable, is powered to detect a small-to-medium effect (Cohen’s f² = .085, alpha = .05, power = .90).

To supplement these quantitative measures, a semi-structured interview was conducted with a subset of participants representing low, medium, and high adherence levels. These interviews aimed to identify additional factors influencing adherence that were not captured by the predictor measures. Questions were tailored based on adherence levels: for high adherers, the focus was on strategies used to maintain engagement, whereas for medium and low adherers, questions explored reasons for disengagement and barriers to adherence.

2.5.3. Rationale for choosing Predictors

The selection of predictor variables was informed by existing literature on adherence in other domains, research on health behaviors, and behavioral change models. Concepts from the Health Belief Model (HBM), the Theory of Reasoned Action (TRA), and the Theory of Planned Behavior (TPB) were adapted, where appropriate, to align with the cognitive training activities in this trial.

2.5.4. Participant Interviews

A subset of participants was interviewed after the intervention to better understand barriers and facilitators to adherence. We aimed to sample participants across the adherence spectrum, using cutoffs of 0–25%, 50–80%, and 95–100% to represent low, medium, and high adherence, respectively, with the option to adjust these cutoffs as needed based on the evolving distribution of adherence observed during the study. In total, 77 particpants were interviewed.

2.6. Materials

2.6.1. Cognitive Training Delivery Apparatus

All participants received a Samsung Galaxy Tab A tablet (10.1” screen) along with a charger. They were instructed to keep the tablet connected to the Internet while playing.

2.6.2. Cognitive Training Activities

The Mind Frontiers software package was used for cognitive training activities. This package includes gamified cognitive tasks similar to commercially available brain training games. It features seven different games: Ante Up, Irrigator, Pen ‘Em Up, Riding Shotgun, Trader Jack’s, Sentry Duty, and Supply Run. Additionally, the package offers a bonus feature that allows players to use points earned from the games to build a virtual town. All games are designed with adaptive difficulty, ensuring they remain challenging for players throughout the intervention. The Mind Frontiers package has been used in previous studies examining adherence (e.g., [8], [9]).

In Ante Up, players are shown a set of cards arranged in a specific pattern and must replicate this pattern within a given number of moves using the cards provided. This game challenges reasoning ability and is based on the Tower of London test described by Shallice (1982) [36].

In Irrigator, players construct a water pipeline from a well to various targets before time runs out, using pipe pieces that change with each turn. As players progress, the number of targets increases, along with obstacles they must navigate. This game challenges visuospatial processing and is based on a training task described by Mackey et al. (2011) [37].

In Pen ‘Em Up, players sort objects dropped from a UFO into two pens by swiping left or right based on specific criteria provided at the start of each round. This game is based on task-switching paradigms described by Karbach and Kray (2009) [38].

In Riding Shotgun, players ride in a horse-drawn wagon and observe a grid of tiles that briefly light up. They must remember the sequence and replicate the pattern in the correct order. This game targets visuospatial memory and is based on the training task described by Klingberg et al. (2002) [39].

In Trader Jack’s, players select an item or set of items that balances the weight of objects placed on a scale. This game exercises inductive reasoning skills and is based on the training described by Willis and Schaie (1986) [40].

In Sentry Duty, players memorize the sequence in which sentries lift a lantern and say a word. They must determine whether the current sentry’s location and word match those of the sentry N turns prior. This game challenges working memory and is based on the dual n-back training task described by Jaeggi et al. (2008) [41].

In Supply Run, players take on the role of a traveling merchant. Along the way, townspeople request items from different categories. Players must remember the last requested item from each category and purchase them at a town store at the end of the trip. This game challenges working memory and is based on the training task described by Dahlin et al. (2008) [42].

2.6.3. Intervention Delivery Apparatus

All messages were sent automatically using services provided by Twilio. Gameplay data were transmitted to a server at Florida State University (FSU) whenever the tablets were connected to the internet. The adaptive algorithm, written in Python, was stored on the same server. It ran every hour, on the hour, to check gameplay data, update time slot weights, and communicate with Twilio to send reminder messages.

2.7. Treatment Fidelity

The trial followed a highly manualized approach, with a detailed operations manual developed for all study protocols. Study measures and the smart reminder system were pilot tested among the research team, while training videos and user manuals were tested with older adults in a prior study [35]. To ensure consistency, a single researcher handled all participant inquiries related to technical difficulties throughout the intervention, and the same researcher conducted all follow-up interviews.

Issues encountered during the intervention, whether related to a task or the device, were logged in a spreadsheet. This log included detailed descriptions of technical issues (e.g., inability to connect to WiFi, power loss, or server breakdown).

2.8. Data and Safety Monitoring

This study is a single-site, minimal-risk trial overseen by the Principal Investigator (PI), Dr. Boot, who is responsible for ensuring participant safety on a daily basis.

2.9. Sample

Participants were recruited through various methods, including participant registries, advertisements in local newspapers, and referrals from other participants.

A total of 288 individuals were prescreened by the study team. Of these, 71 were excluded due to ineligibility, with the most common reasons being failure to meet the cognitive criteria (n = 29, 40.85%) and previous participation in cognitive training studies using the Mind Frontiers software (n = 25, 35.21%). Among the 205 eligible participants, five declined to participate, and one was lost to follow-up. Ultimately, 199 participants were recruited and randomized, with 99 assigned to the smart reminder condition and 100 to the control condition.

The final sample consists primarily of women (67.34%) and ranges in age from 62 to 88 years (M = 72.69, SD = 4.97). The sample is predominantly white, highly educated, and has a high income. There were no significant differences in age, gender, race, education, or income between the smart reminder and control conditions (Table 2). Note that despite passing the prescreening, which required participants to be 65 years of age or older, two participants later reported birthdates indicating they were younger (one was 62, and the other was 63).

Table 2.

Baseline demographics for participants in the APPT study

	Total group	Smart reminder group	Control group	t or χ2
Age	72.67 (SD=4.97)	72.34 (SD=5.21)	73.00 (SD=4.73)	t (185) = 0.90
Gender	Female n=134 (67.34%)	Female n=70 (70.70%)	Female n=64 (64.00%)	χ2 (1) = 0.99
Education	High school or less n=6 (3.02%)	High school or less n=2 (2.02%)	High school or less n=4 (4.00%)	χ2 (4) = 6.99
	Some college n=31 (15.58%)	Some college n=13 (13.13%)	Some college n=18 (18.00%)
	Bachelor n=48 (24.12%)	Bachelor n=24 (24.24%)	Bachelor n=24 (24.00%)
	Master n=79 (39.70%)	Master n=47 (47.47%)	Master n=32 (32.00%)
	Doctoral n=24 (12.06%)	Doctoral n=8 (8.08%)	Doctoral n=16 (16.00%)
Race	White n=163 (81.91%)	White n=79 (79.80%)	White n=84 (84.00%)	χ2 (2) = 1.42
	Black n=15 (7.54%)	Black n=9 (9.09%)	Black n=6 (6.00%)
	Others n=6 (3.02%)	Others n=4 (4.04%)	Others n=2 (2.00%)
Income	Less than 20k n=3 (1.51%)	Less than 20k n=2 (2.02%)	Less than 20k n=1 (1.00%)	χ2 (4) = 2.34
	20k-39,999 n=17 (8.54%)	20k-39,999 n=10 (10.10%)	20k-39,999 n=7 (7.00%)
	40k-59,999 n=27 (13.57%)	40k-59,999 n=13 (13.13%)	40k-59,999 n=14 (14.00%)
	60k-79,999 n=42 (21.11%)	60k-79,999 n=17 (17.17%)	60k-79,999 n=25 (25.00%)
	80k and above n=83 (41.71%)	80k and above n=42 (42.42%)	80k and above n=41 (41.00%)

Note:

^*p<.05

^**p<.01

^***p<.001; Education: 1=no formal education, 2=less than high school, 3=high school or GED, 4=some college, 5=Bachelor, 6=Master, 7=Doctoral; Income: 1=less than 10k, 2=10k-19,999, 3=20k-39,999, 4=40k-59,999, 5=60k-79,999, 6=80k and above.

3. Discussion

This study examines the benefits of an adaptive, tailored smart reminder system designed to improve adherence to a cognitive training program. In addition, we collected data on individual characteristics that may influence adherence, as well as qualitative data from participants with low, medium, and high adherence. The trial’s outcomes will provide valuable insights into the effectiveness of personalized adherence support strategies. Baseline data on individual differences will further our understanding of facilitators and barriers to adherence, while in-depth interviews with selected participants will help identify additional factors not captured through surveys or pre-test measures. Finally, these findings may have broader implications for adherence support across other domains, including medication adherence, diet, exercise, and technology-based interventions.

Several challenges arose during the trial. First, the study was originally planned to last six months (24 weeks) but was reduced to four months (18 weeks) due to COVID-19-related delays and resultant difficulties in recruitment and administration. Second, internet connectivity issues may have affected participants’ experiences. Since the algorithm was stored on a server and required real-time access to participants’ gameplay data, instances of poor connectivity could result in unrecorded gameplay sessions. Participants with inconsistent internet connections may have also received unintended reminder messages. Additionally, server downtime caused by hurricanes, and in one instance, a tornado disrupted data collection. During these periods, participants received error messages indicating they could not connect to the server, and gameplay data from those days were not recognized. Similarly, reminder messages scheduled for delivery during server outages were not sent. Because the system relies on past gameplay data to adjust future reminders, disruptions in data collection affected the timing of subsequent messages in the smart reminder condition. These challenges are common in technology-based intervention studies (e.g., [43], [44]).

This trial also has some limitations. The smart reminder system relied on a relatively simpler adaptive algorithm to determine when to send messages. While previous research has shown that more complex machine learning algorithms can predict adherence to cognitive training programs with moderate to high accuracy (e.g., [45], [46]), more advanced, person-centered deep neural networks designed to predict daily engagement (e.g., [47], [48]) tend to be computationally expensive and require large datasets. Future research could explore pre-trained models using data from participants with similar adherence patterns or individual characteristics to improve real-time adherence prediction. We also relied on predefined categories to target motivations, and it is possible that important motivations for some individuals were not included in this list. Additionally, the efficacy of cognitive training programs remains a topic of debate (e.g., [1], [2], [3], [4]), and the optimal regimen for effective cognitive training is not well established. Given that the regiments used in research examining the efficacy of cognitive training programs could be program-specific and different from each other, only the knowledge about the principles of what makes adherence support systems effective, and the methodologies involved in designing such systems would be generalizable.