A single-arm pilot study of a 12-week online intervention combining PA prescription and individualized coaching. The current study measured the efficacy of the intervention for helping participants to meet age-appropriate PA guidelines [12,14], with secondary outcomes exploring acceptability, feasibility, and safety, and whether the intervention modified dementia risk, overall PA levels, mental health symptoms, and stages of change. We examined pre-post changes in measures of cognitive mechanisms of behavior change.

This study received ethical approval from the University of Melbourne Human Research Ethics Committee (2021-20479-16470-3). All published data is deidentified. Participants provided written informed consent to take part in this study before screening. Participants kept all programs and equipment provided during this study but did not receive other compensation for participating.

Participants were eligible for inclusion if they were aged between 45 and 80 years; English-speaking; living in the community; had access to a phone and/or internet; reported concerns about their cognition (with or without objective cognitive decline) but not dementia; and experienced mild to moderate symptoms of depression, anxiety, or stress (measured using the short-form Depression Anxiety Stress Scales [DASS-21] [29]). Concerns about cognition were reported by a “yes” response to either of the questions: “Do you think your memory or thinking has changed?” or “Do you think your memory or thinking is worse than for other people your age?” Exclusion criteria were lack of capacity to provide informed consent; inability to read or write in English; diagnosis of dementia or a score indicative of dementia (27 or lower) on the Telephone Interview for Cognitive Status [30]; scores in the severe category or higher on either the depression, anxiety, or stress dimensions of the DASS-21 [29]; acute or past history of significant mental illness; current suicidality or history of suicidal attempts within the past 2 years; significant uncorrected sensory impairment; severe mobility impairment (either an inability to walk unaided or a score of 12 or more, indicative of medium to high risk, on the Falls Risk Assessment Tool [31]); any other significant medical condition that would prevent safely undertaking moderate-intensity PA (gauged using the Adult Pre-Exercise Screening System [32] and the Physical Activity Readiness Questionnaire [33]); or a BMI of less than 18.5 or 38 or higher (indicating that they were underweight or very obese). Where applicable, ethnically adjusted BMI indicators were used [34]. Participants were recruited from various sources across Australia, including state-wide seniors’ newsletters, university media, general practitioners, online self-registration services (StepUp platform) [35], and a database of volunteers who previously consented to be approached for research. Our target sample size was informed by a 1-sample proportion test [36] based on the AIBL Active Study [25]. To detect a 15% increase in participants meeting moderate-intensity PA guidelines with α=.05 and an expected attrition rate of 30% [37], we calculated a required sample size of 70, which would yield power of 0.80 to detect this effect.

EXCEL was a 12-week home-based PA intervention comprising the following resource and support components.

We developed a PA guide incorporating aerobic, strength, and balance programs tailored to 2 age streams: 45‐59 years (based on the standard national PA guidelines [14]) and 60‐80 years (based on national guidelines for older adults with MCI or subjective cognitive decline [12]). While warm-up and cool-down programs were identical for both streams, aerobic and strength programs were customized. Within each stream, participants received either a starter or intermediate program based on their health, abilities, and current activity level. Strength programs differed slightly between males and females. All older age stream participants received a balance program with options for progression. Core program content was developed from successful past studies, including the INDIGO (Individual Goal-Setting and Volunteer Mentors) study [38] and work by Yang et al [39]. The PA guide (Figure 1) provided structured guidance on increasing duration, intensity, number of sessions, and resistance or weights to meet the PA guidelines in 12 weeks. The program incorporated flexibility, with progression guides suggesting rather than prescribing exercises. Each warm-up and cool-down, strength, and balance program included summary pages for easy reference (Figure 2), plus exercise-specific pages detailing performance instructions, pictures, and links to YouTube (Google LLC) demonstration videos (Figure 3). Along with the PA guide and program, participants received a wearable wrist activity tracker, dumbbells, and resistance bands via post if they did not already own this equipment. Participants had access to an online PA diary hosted on the REDCap (Research Electronic Data Capture; Vanderbilt University) platform [40], which could be printed in hard copy or PDF format upon request.

The online support structure consisted of 3 types of sessions as outlined below.

The introduction session was an approximately 70-minute online videoconference session that familiarized participants with key sections of the PA guide, including guidelines and safety information. Participants received personalized directions from an exercise specialist (JS) and were able to ask questions.

Of fortnightly catch-up sessions, there were online sessions (20 min or less) with EXCEL coaches, who checked participants’ diaries, discussed difficulties, and used behavior change strategies tailored to individual needs [22]. These strategies included education, motivation, incentivization, reinforcement, training, environmental restructuring, modeling, and enablement. The 4 EXCEL coaches came from various backgrounds, including psychology, PA, social work, and occupational therapy.

Of exercise specialist sessions, there were as-needed online sessions with the team exercise science specialist (JS) provided extra assistance on exercise elements of the program for participants who requested or required additional support.

Figure 4 summarizes the participant procedure. Participants completed a consent form and prescreening survey and proceeded to a screening call (50 min), completing additional measures to confirm eligibility. Eligible participants were assigned to a PA program and scheduled for an introductory session before completing self-report baseline measures (≈40 min). Their general practitioner was notified of their involvement in the PA intervention. Participants completed the DASS-21 again at weeks 4 and 8 (each ≈5 min). Safety was prioritized, with participants scoring in the severe range or higher on any DASS-21 subscale contacted using a safety protocol script. Fortnightly consultation sessions were held between EXCEL coaches and the research team, which consisted of psychiatrists (NTL, TWHC, TR, and EC), a clinical psychologist (KAE), and exercise physiologists (KLC and JS), to monitor participant safety. Following the 12-week program, participants completed self-report postintervention measures (≈40 min) and an evaluation survey (≈8 min) via Zoom.

Outcome measures are summarized in Table 1. The primary outcomes reflected successful uptake of age-appropriate PA guidelines, specifically meeting the aerobic recommendation of at least 150 minutes per week of moderate or vigorous intensity activity at follow-up and achieving the full set of PA guidelines encompassing aerobic, strength, and balance activities. Secondary outcomes included acceptability, feasibility, and safety, changes in general dementia risk levels, changes in total moderate-to-vigorous PA that incorporates everyday activities, changes in mental health symptoms, and progression through stage of change. Finally, 2 categories of mechanism-focused outcomes were evaluated. The first relates to potential cognitive mechanisms underpinning exercise behavior change, including shifts in knowledge, motivation, and self-efficacy from pre- to postintervention. The second examined potential mechanisms associated with general dementia risk-reduction behavior change over the same period.

All quantitative data analyses were conducted using SPSS (version 29; IBM Corp) and R (version 4.1; R Foundation). The lme4 package was used for the mixed-effects modeling [50], and the emmeans package was used to calculate the effect sizes [51], while the coin package was used for the asymptotic marginal homogeneity test [52]. Standard assumption checks were undertaken for the linear mixed-effects analyses covering residuals, random effects, variance homogeneity, multicollinearity, influence, and missing data patterns and found no departures that would likely affect the findings. Given the single-group pre-post design, key assumptions required for valid causal mediation analysis, including the ability to rule out unmeasured confounders and separate intervention effects from natural temporal change, could not be met. Therefore, we examined pre-post changes in hypothesized mediators descriptively rather than conducting formal mediation analysis. Descriptive statistics (means, SDs, frequencies, and proportions) were used to describe demographic and outcome measures. The Marginal Homogeneity test, an extension of the McNemar test, was used to compare the proportion of participants who met PA guidelines at baseline to those who met the PA guidelines at postintervention, with the sample restricted to those participants with both baseline and follow-up data. For continuous variables, linear mixed-effects models were used to compare changes in outcome scores from baseline to follow-up, accounting for missing data at follow-up. The models included a fixed effect of time and a random intercept for participants. The Kenward-Roger degrees of freedom approximation was used to calculate P values, CIs, and effect sizes. For these models, Cohen d effect size was calculated as the mean difference between baseline and follow-up scores divided by the square root of the sum of the intercept (participant) and residual random effect variances [53]. Cohen d effect sizes were interpreted as small (d=0.2), medium (d=0.5), and large (d≥0.8). The Benjamini and Yekutieli [54] approach was used to correct for multiple testing of the primary outcomes and secondary outcomes, respectively. For these analyses, an α of .05 was used for statistical inference.

Participants’ self-reported diaries collected over the 12-week intervention period were used to assess adherence to PA guidelines. No consensus exists on defining PA adherence; similar studies commonly use meeting 2/3 or more of the prescribed PA [55]. We will apply two thresholds: (1) full adherence, defined as meeting the complete guideline each week (>150 min/wk moderate-plus aerobic PA, >2 strength sessions/wk, and for older adults >1 balance session/wk); and (2) 2/3 adherence, defined as >100 minutes per week of aerobic PA, ≥1 strength session per week, and ≥1 balance session per week for older adults. Aggregate total adherence over the 12-week intervention period will use a realistic expectation of 8 or more out of a possible 12 weeks’ meeting in full or with 2/3 guidelines.

Figure 5 summarizes this study recruitment and screening process; 55 participants were enrolled and completed baseline assessment (46 female/9 male; mean age at screening 62.2, SD 7.6 y). Demographic characteristics are presented in Table 2.

We performed sensitivity analyses adding age, gender, baseline depression score, and baseline total minutes of PA as covariates; their inclusion had a negligible effect on the estimated mean difference and effect size and did not meaningfully change the CIs, indicating the main results are robust to these potential confounders, and the full results are reported in Multimedia Appendix 1.

These findings support our hypothesis that the EXCEL program, a 12-week digitally delivered, tailored PA intervention, could effectively facilitate middle-aged and older Australians experiencing mild to moderate mental health symptoms alongside cognitive concerns to meet PA guidelines. The program was acceptable, feasible, and safe and was associated with reductions in both dementia risk and mental health symptoms and progression through stages of change. Changes to action planning and positive outcome expectancies were observed and present potential mechanisms of action, reflecting components of the behavior change model that informed the intervention’s design [14]. Given the well-documented challenges in engaging this population in sustained PA and the critical importance of PA for DRR, these results are particularly encouraging.

EXCEL was highly efficacious; nearly half of the cohort (24/55, 44% participants) met PA guidelines at the end of the program, up from 6% (3/55) of participants at baseline. The increases in PA are particularly encouraging given the low proportion meeting PA guidelines at baseline (<6%), compared to the Australian population levels of 22.4% of adults and 33.4% of older adults [56]. Low baseline PA levels in our cohort fit with evidence that people with mental health disorders are significantly less likely to meet PA guidelines compared to controls [57]. This further underlines the critical importance of interventions that are tailored to support this cohort at high risk for dementia. Retention rates (95%) and positive feedback from participants (ie, 98% found it useful and 100% would recommend it to others) demonstrate that the program was well-received. High levels of acceptability might be attributed to the individually tailored nature of the intervention, which was designed based on comprehensive formative research incorporating the preferences and needs of the target demographic [15]. It is also relevant to note that EXCEL was conducted in part during the COVID-19 lockdowns, which may have increased participants’ involvement due to a reduction in alternative activities and distractions. Evidence suggests that while COVID-19 had a mixed impact on research participation, research that used PA was associated with increased involvement [58].

Safety was a central focus of our design, particularly given the online delivery. While the safety, scalability, and effectiveness of digital PA interventions have been well documented in younger adults (see umbrella review by Russell et al [59]), there remains a gap in the literature regarding older adults. We used a robust system of support and monitoring that included email alerts and professional consultations that managed physical and mental health risk successfully. Our current results, taken together with data that show the majority of middle-aged and older adults are online (98% of Australians aged 65‐74 y and 94% of those aged 75 y + [60]), suggest that age should not be seen as a barrier to accessing safe and effective online PA interventions. Another advantage suggesting future scalability was that coaches were not required to have mental health credentials or exercise physiology expertise themselves to be highly effective. The safety of our trial and the training and support of our coaches did, however, require qualified mental health and exercise physiology clinicians in an oversight capacity, which has implications for real-world implementation.

A significant reduction in dementia risk was observed over the 12 weeks, as measured by the CogDRisk, a validated online measure that assesses individualized risk factors of dementia in various population settings [42,61]. These results support a developing literature showing that PA reduces dementia risk in other high-risk groups, including those with new-onset type II diabetes [62]. Baseline dementia knowledge in our cohort was low and was not modified by our intervention, which is to be expected given that this was not an aim of the current study. This does demonstrate, however, that it is not essential to increase dementia knowledge to reduce dementia risk, which may be particularly pertinent for individuals with cognitive difficulties. Nonetheless, integrating additional psychological or social support structure components could further enhance the intervention’s impact on self-efficacy and sustained behavior change.

We evaluated whether the intervention impacted mental health symptoms and found a significant and clinically meaningful improvement across all DASS-21 dimensions. After 12 weeks, our cohort’s anxiety levels had returned to within the normal range, while depression and stress symptoms were now mild. This is a meaningful change from the extremely severe (stress), severe (depression), and moderate (anxiety) average symptom scores observed 12 weeks earlier, at baseline. Our findings add to the established literature showing the important positive effect that exercise can play in improving mood; a finding reflected in clinical practice guidelines in the United States, United Kingdom, and Australia [63-65]. A recent meta-analysis by Noetel et al [66] showed exercise to be equally effective for people with and without comorbidities, and with different baseline levels of depression. Their findings reflected over 200 studies [66] and suggest that walking and strength training are among the most effective exercises for depression reduction, in line with PA guidelines that underlie EXCEL. While evidence for the anxiety-reducing effect of PA is currently less consistent (see Chong et al [67] for a meta-analysis), we were encouraged to see that PA reduced anxiety levels in our cohort.

EXCEL was designed to specifically target proposed mechanisms of action for optimizing engagement in this cohort, including emotional regulation, self-efficacy in existing skills, and the capacities to act on intentions despite barriers, and we measured these potential cognitive mechanisms pre- and postintervention. Our findings showed a small to moderate, but significant, increase in positive outcome expectations for exercise. Positive outcome expectancy refers to an individual’s belief that PA will lead to certain (positive) outcomes and is 1 of 2 key constructs in social cognitive theory, alongside self-efficacy. Our findings are consistent with previous research showing that (1) positive, rather than negative, outcome expectancy is associated with PA levels, and (2) there is an apparent age-related effect whereby outcome expectancy is more predictive of PA in older adults [68]. Positive outcome expectancies have previously been observed to facilitate the formation of PA intentions [68] and are associated with progression through the stages of change, as seen in this study. Considered within the COM-B framework, positive outcome expectancies may enhance motivation and support the development of strong intentions to engage in PA.

However, transitioning from intention to behavior often relies on effective action planning—that is, setting concrete, detailed plans that outline the when, where, and how of intended PAs [69]. Our findings support this notion. The mental stimulations involved in planning have also been shown to have a self-regulatory effect that may help individuals connect their behavioral responses to future scenarios and account for potential barriers [70]. In this context, our intervention may have supported participants in mentally rehearsing the steps needed to translate PA intentions into action, reducing the intention-behavior gap. This aligns with our findings on the stages of change measure: at baseline, participants primarily scored in the “preparation” stage, characterized by intentions to begin regular exercise or making sporadic attempts to be active. Following the intervention, averages shifted to the “action” stage, reflecting active and consistent engagement in structured exercise. These results suggest that aging adults with cognitive difficulties related to subjective or MCI and co-occurring mental health symptoms may particularly benefit from pragmatic, structured action planning that bridges the gap between intention and behavior.

While our results are promising, they should be interpreted with caution due to the pilot nature of this study, modest sample size, and pre-post design. This study’s design prevented a valid causal mediation analysis, meaning that changes in hypothesized mediators can only be described rather than inferred as causal. Further research using a larger sample and a randomized controlled trial design with mediation analyses is needed to determine whether the proposed mechanisms contribute causally to the observed outcomes. Sensitivity analyses indicated that age, gender, baseline depression score, and overall minutes of PA did not confound the findings. However, technological or digital literacy was not measured, and engagement may have been higher during COVID-19 lockdowns; these unmeasured confounders may limit generalizability to nonpandemic conditions. Despite these limitations, the findings exemplify the utility of tailored interventions and the value of a behavior-change framework. It is recommended that future studies include larger cohorts and longer follow-up periods to replicate findings and evaluate the durability of effects. Given the complexity of factors influencing PA adherence and mental health among aging adults, multicomponent interventions that address both physical and cognitive health holistically are recommended.

This pilot study demonstrated EXCEL to be effective, acceptable, feasible, and safe. The tailored, PA-delivered PA intervention supported middle-aged and older adults experiencing mild to moderate mental health symptoms alongside cognitive concerns to meet PA guidelines. Important secondary gains were a reduction of dementia risk and clinically meaningful improvements in mental health symptoms, emphasizing the advantages of boosting PA in high-risk groups. Improvements in action planning and outcome expectancy may have contributed to bridging the PA intention-behavior gap, in line with the COM-B model underlying the intervention design. While findings require confirmation in larger controlled trials, the results suggest that EXCEL offers a scalable, evidence-informed model for supporting dementia-risk reduction in adults who face well-documented barriers to engaging in sustained PA.