A total of 29 adults aged ≥18 years and diagnosed with ASD were recruited through state and regional agencies that serve people with ASD across the United States and online autism support groups on social media such as Facebook and Reddit. Evidence of ASD diagnosed by a qualified medical professional such as a pediatrician or clinical psychologist (ie, when and where) was required for study participation and obtained via self-report. In addition, eligible study participants met the following inclusion criteria: (1) self-reported medical diagnosis of anxiety or self-identification of experiencing anxiety symptoms for the past 3 or more months, (2) access to a supported device (smartphones with Android 4.4 and higher or iOS 9.0 and higher operating system), (3) cognitive ability to understand the purpose of the study, and (4) no prior experience using the PA mobile apps used in the study. Individuals with low cognitive function, co-occurring intellectual disabilities, or mobility impairments were excluded from this study. We hereby use the term intellectually able adults with ASD to refer to those who can make their own decisions regarding health behaviors without much assistance. A participant with a self-reported mild learning disorder that did not significantly interfere with active daily living and study participation was included in the study. A formal screening interview was conducted with each participant through a phone or face-to-face video call to verify participant eligibility and identify potential barriers to study participation. All participants provided written or digital consent before data collection.

The Institutional Review Board of Indiana University approved this study (protocol number 1807483245).

On verification of the participant’s eligibility through the screening interview, participants completed a web-based demographic survey with an emphasis on BMI, waist circumference, medication use, and autism symptoms. Self-reported height and weight information were collected to calculate each participant’s BMI (ie, kg/m² or lbs/in² × 703) [51]. Participants were also asked to provide self-measured waist circumference, which can be an indicator of obesity-related disease risk [52]. The Autism Spectrum Quotient 10-item (AQ-10) questionnaire was included in the survey to examine the severity of autism symptoms (eg, social interaction deficits, sensory issues, and other autism-specific characteristics) [53]. The AQ-10 was designed for health care professionals to administer an informal ASD assessment of individuals with typical cognitive function [54]. The maximum possible score is 10, and individuals with a score of ≥6 are advised to consider referral for formal ASD assessment. The AQ-10 is one of the few autism screening tools available for adults [18]. After the completion of the demographic survey, study materials, including an accelerometer with an elastic belt, a USB cable charger, and study instruction sheets (ie, how to wear and charge the accelerometer and how to install and use the daily anxiety assessment [DAA] app for anxiety assessment) were either mailed or handed to remote and local participants, respectively.

A rubric was used to assess participant comfort and knowledge of study procedures before the start of the baseline and intervention periods, and case-by-case decisions were made by the research team when individuals were ready to start data collection. Before the start of the intervention period, participants were assigned to either the PuzzleWalk or Google Fit group according to age, sex, and BMI using a covariate-adaptive randomization process. Specifically, a minimization technique was applied to the randomization process by distributing the participants into 2 groups based on the aforementioned variables, which were identified before the start of data collection [55]. Covariate randomization aimed to minimize the imbalance in baseline characteristics across the 2 groups included in the study [56]. All participants received visualized step-by-step instructions (eg, search and download on Google Play or App Store, user registration, goal setting, and PA behavior tracking) on the assigned PA app (PuzzleWalk or Google Fit) and used it from the beginning of the intervention start (fourth week) until the end of the intervention (eighth week).

Participation required an approximate 2-month commitment: the first week for baseline and the fourth to eighth weeks for intervention. A 2-week interval was implemented between baseline and intervention onset to reduce the novelty effect in response to the accelerometer and PA app use [57] (see Figure 1 for the study timeline). Both the PuzzleWalk and Google Fit groups received reminders regarding the use of the PA app during the first week of the intervention period and autonomously continued to use the app until the intervention ended. Data collection occurred in the fall and early spring to avoid the impact of inclement weather on activity patterns. Participants who successfully completed the 2-month study received a US $100 e-gift card as a token of appreciation.

PA app use for the past 1 week was assessed through a self-report survey at 3 different data collection time points (ie, end of the fourth week, start of the eighth week, and end of the eighth week) (see Table 1). Participants were asked to report the frequency (days of app use during the past 7 days) and duration (hours and minutes usually spent for app use). The time spent using the PA app was calculated using a minutes per day format.

The Beck Anxiety Inventory (BAI) was used to assess participants’ prolonged state of anxiety. The BAI is a self-report scale comprising 21 items that measure the severity of anxiety symptoms during the past week, with high internal consistency (Cronbach α=.92) and test-retest reliability (r=0.75) [64]. Scores range from 0 to 63, with scores of ≥26 indicating potentially concerning levels of anxiety (ie, 0-7=minimal; 8-15=mild; 16-25=moderate, and ≥26=concerningly severe). The participants were asked to report the extent to which they had been bothered by each of the 21 symptoms in the past week. The participants completed the BAI at the start and end of each data collection period (a total of 6 times).

In addition to the BAI, time-specific and type of anxiety trigger questions were asked daily during each data collection period to better identify the contexts of potential anxiety triggers such as environmental, psychological, or sensory factors [65]. At the start of the study, participants downloaded the DAA mobile app developed by the research team. The participants received a prompt at 8 PM each day during the data collection period to answer specific questions. This time was deemed appropriate for assessing perceived emotional changes over the course of the day based on pilot work and considering typical work/school schedules and bedtimes. The questions included “When did you feel anxious today?” and “What caused your anxiety today?” with the possible choices of (1) change or disruption to routine, (2) sensory oversensitivity or overstimulation, (3) confusion and worries about social and communication situations, (4) specific fears or phobias, (5) too many demands or expectations, (6) being prevented from preferred behaviors or interests, and (7) not applicable. As the traditional retrospective assessment approach can be less reliable [66], the DAA adopted the ecologic momentary assessment method to improve response reliability [67]. Low compliance can be an issue with this type of daily report; therefore, we implemented routine strategies to address this issue, including establishing an efficient data/compliance-tracking system and providing regular reminders to complete the task and monetary incentives [68]. Reminders were sent to the participants via email or SMS text messages based on their preferences recorded in the demographic survey. Only participants with valid survey compliance (ie, participation for 6 times in the BAI and participation in at least four DAAs in each data collection period) and monitor wear (ie, at least 3 valid days in each data collection period) received a monetary incentive.

Daily walking steps, PA intensity, and sedentary time were measured using GT3X+ and ActiGraph triaxial accelerometers (ActiGraph). ActiGraph accelerometers have been extensively used to measure PA with moderate to high reliability in both laboratory and free-living conditions [69-71]. All participants were asked to wear an accelerometer on their right hip during waking hours for 7 consecutive days, including at least 2 weekdays and 1 weekend day over the 3 different data collection periods (baseline, start of the fourth-week intervention, and end of the eighth-week intervention). Accelerometers were programmed to calculate data in 60-second epochs [72]. Daily walking steps were measured using the ActiGraph pedometer function, and sedentary time and activity intensity were identified using the following activity counts per minute cutoffs: <100=sedentary, 100-2019=light, 2020-5999=moderate, >5999=vigorous, and ≥2020=MVPA [73]. These cutoff points were proposed by Troiano et al [73] and have been validated in large data sets of free-living adults [74-76]. The minimum wear time for a valid day is ≥10 hours per day for wake time [77]. Although there is no scientific consensus on the minimum number of valid days, 4 days have been widely accepted in previous studies to reliably estimate habitual PA [78,79]. However, this study required at least 3 valid days of monitor wear in each data collection period in an effort to minimize sample loss [78].

In addition, an intraclass correlation (ICC) analysis was performed to validate the use of at least 3 valid days of monitor wear by examining the relationship between 3 valid days and ≥4 valid days on all PA and sedentary time outcomes at baseline. ICC values of ≥0.75 are generally regarded as good or acceptable reliability [80]. The ICC results demonstrated that the reliability of 3 valid days for all PA and sedentary time variables was excellent (all ICC >0.90); therefore, participants with ≥3 valid days of monitor wear were included in the analyses. Overall, 2 participants’ accelerometry data at the intervention start (fourth week) and 1 participant’s accelerometry data at the intervention end (eighth week) did not satisfy the minimum number of 3 valid days; thus, their baseline accelerometry data were imputed according to the baseline observation carried forward method [81]. The criterion for nonwear time was 90 minutes of consecutive 0 counts [82], and the accelerometers were set to collect data at sampling rate of 30 Hz.

Independent t tests and chi-square tests were performed for continuous and ordinal variables, respectively, to compare the baseline differences between the 2 groups. Data are presented as mean (SD) or frequency (percentage), according to the variable type. According to the scoring guidelines, the BAI scores for all 21 items (ie, not at all=0, mild=1, moderate=2, and severe=3) were summed to yield a total score of anxiety severity [64]. The BAI scores collected twice during each data collection period were averaged to represent each time point (ie, baseline, intervention start, and intervention end). Data on common anxiety triggers and time-specific occurrences were screened to include only participants with at least four responses to the DAA at each data collection time point. All participants met the compliance criteria, and 134 counts (baseline, N=24), 124 counts (intervention start, 23/24, 96%), and 107 counts (intervention end, 20/24, 83%) were included in the frequency (percentage) analyses.

Objectively measured PA data were first systematically cleaned and verified by examining the ranges and missing values according to the established validation criteria. The collected raw accelerometry data were then converted into activity counts using the ActiLife 6 Data Analysis Software (ActiGraph). With the aforementioned activity intensity cutoffs, the PA data were processed and extracted into an editable spreadsheet. Manual data screening was sequentially performed to verify the minimum required hours and valid days of monitor wear for data analysis; invalid data were eliminated.

An intention-to-treat analysis was conducted to maximize external validity, and the baseline observation carried forward method was used to impute missing data after randomization [81,83,84]. According to the guidelines on missing data in clinical trials by the European Medicines Agency (2010), the baseline observation carried forward method can be appropriate in randomized trial design studies in which researchers reasonably assume that the outcomes of a participant would return to their baseline levels in the long term after dropout [85]. In light of the general span of a mobile health intervention’s effectiveness in promoting PA (eg, up to 3 months) [86], this method can also help maintain external validity and minimize sample loss.

The dependent variables were sedentary time (minutes per day), light PA (minutes per day), MVPA (minutes per day), steps per day, total activity counts (vector magnitude), average BAI anxiety score, and PA app use (minutes per day). To assess PA and anxiety changes over the 3 data collection periods; all measures collected at baseline, intervention start, and intervention end were compared between time points and groups using a repeated-measures analysis of covariance (ANCOVA). Repeated-measures ANCOVA models were adjusted for baseline characteristics, including age, sex, and BMI. The Mauchly test of sphericity was used for each outcome variable to examine the equality of variances of within-group differences across the 3 different data collection time points. In general, if the P value was <.05, the assumption of sphericity was violated. Owing to the violation of the sphericity assumption, the Greenhouse-Geisser correction was applied to MVPA, steps, and total activity count variables to interpret the results of the within-group effects.

The effect size (partial η²) was calculated and defined as >0.02=small, >0.13=medium, or >0.26=large [87]. Owing to the violation of the normality assumption, Spearman rank correlation analyses were performed to determine the baseline correlations between the outcome variables and the impact of increased PA or decreased sedentary time on anxiety change. The degree of change in the outcome variables following PA app use was calculated by subtracting the baseline value from the average of intervention outcomes (eg, MVPA change = [MVPA at intervention start + intervention end]/2 – MVPA at baseline). Data analyses were performed using SPSS 26.0, and significance was declared at P<.05 (2-tailed).

The findings from this feasibility study indicate that gamified PA-promoting mobile apps can be an effective tool for decreasing sedentary time and increasing MVPA in intellectually able adults with ASD. Notably, the PuzzleWalk mobile app, developed using BCTs and autism-specific design elements, was comparable with the commercially popular Google Fit in inducing changes in PA and sedentary time. However, the positive improvements in PA and sedentary time did not significantly reduce anxiety levels, although overall anxiety severity for participants with ASD was positively changed from moderate to mild at the end of the intervention.

It appears that intellectually able adults with ASD can benefit similarly from both commercially available and best practice–guided PA mobile apps. Although there were no significant differences in PA and sedentary time between the 2 groups, there were varying trends in the data according to groups, which are worthy of mention. In the PuzzleWalk group, there was a downward trend in sedentary time and a short-term upward trend in MVPA, step count, and total activity count. Conversely, there was a slight intervention-induced increase in light PA and a clear upward trend in steps and total activity counts in the Google Fit group. These observations, in conjunction with the significantly greater time spent on app use in the PuzzleWalk group compared with the Google Fit group, suggest that the specific research-based design elements of PuzzleWalk, including BCTs and competitive gamification, may pose an advantage over commercially popular apps that do not incorporate these elements. As shown in previous gamified PA interventions for neurotypical samples, adding competition elements to the intervention may be more effective in inducing PA or sedentary behavior changes than collaboration when there are no social connections among users [88,89]. It is of interest to see whether these trends are enhanced in a larger sample of adults with ASD.

Increasing evidence suggests that mobile app interventions can be effective in promoting PA mostly in the short term (eg, up to 3 months) [86]. In a recent study, the short- and long-term effects of a mobile phone–based PA app, together with brief in-person counseling, were evaluated for PA behavior change among inactive middle-aged women (n=210). The findings of this randomized clinical trial indicated that the intervention group’s MVPA and total steps were significantly increased during the first 3-month intervention period, although the intervention effect was not maintained in the following 6 months [88]. Similar results of regression fallacy have been reported in earlier randomized controlled trials that examined the efficacy of smartphone sensor–based interventions on PA promotion in healthy adults, which found a substantial PA decline during long-term follow-up [83,89]. Given that adherence and engagement in mobile app interventions generally decline before 3 months [33,86], the attenuation of intervention effects observed in this study (ie, regression of PA outcomes at the intervention end) may correspond with the overall decline in time spent on PA app use during the intervention period. As such, it is critical for mobile app interventions to maximize participant engagement with the app and exposure to the intervention through additional health behavior management strategies such as motivational short message services and telephone coaching [86,90].

This study aimed to identify an anxiety treatment adjunct to, and not in replacement of, conventional therapies. The overall anxiety level decreased from moderate to mild, with decreases in sedentary time and increases in PA over time, in both groups. Although this change was not statistically significant, it may be of clinical or personal value to adults with ASD. A recent review and meta-analysis reported that PA may have a protective effect against anxiety symptoms and disorders; however, the authors cautioned against drawing firm conclusions because of the limited and heterogenic available research [1]. Kim et al [91] recently reported that the optimal range of PA for decreasing anxiety symptoms was 600 to 9000 metabolic equivalents (METs) minutes per week and 1200 to 3000 METs minutes per week for neurotypical men and women, respectively [91]. Individuals with ASD in this study had mean PA levels of 569.9 (SD 45.6) METs minutes per week at the intervention end, which did not meet this criterion [92]; therefore, the intervention-induced changes in PA may have been insufficient to further affect anxiety levels. In addition, many adults with ASD have difficulty articulating their emotions and feelings, and this could have affected the ability of participants in this study to report the full extent of their anxiety symptoms [93]. Despite the lack of robust outcomes regarding intervention effectiveness, additional research on the long-term effects of PA interventions as a simple and economical way of alleviating the health burden of anxiety in adults with ASD is encouraged.

Interestingly, small intervention-induced changes in PA, sedentary time, and anxiety were unfavorably associated. Increases in light PA, steps, and total activity counts and decreases in sedentary time were associated with increased anxiety in adults with ASD. This is contrary to the preponderance of research and a general understanding of the relationships among PA, sedentary time, and anxiety [24,94]. Previous studies have shown that anxiety symptoms are associated with low PA levels and increased sedentary time [95]. Furthermore, Lee et al [96] found that adults with ASD reporting higher sedentary time had increased odds of developing physical and mental health conditions, including anxiety, associated with cardiovascular risk. The underlying mechanisms that explain the relationship between PA, sedentary time, and anxiety are understudied [1]; however, a potential explanation for the findings in this study is that PA-related social situations and environmental changes may have negatively contributed to anxiety symptoms in the sample [7,97]. Confusion and worries about social and communication situations were common anxiety triggers in adults with ASD in this study, as has been previously reported among those with ASD who are often self-aware of their incompetence and difficulty in social situations [7,65,97,98]. As the mobile apps used in this study prompted outdoor walking, which is difficult to perform in social isolation, it is reasonable to conclude that the prospect of entering the public to walk could have elevated anxiety in study participants. Specific types of anxiety were not assessed in this study; however, Hollocks et al [4] found that social phobia was a highly prevalent type of anxiety disorder in adults with ASD. Therefore, future efforts to study PA and sedentary time interventions for this population must consider the social expectations of the prescribed activity.

The major strengths of this study include (1) the use of a covariate-adaptive randomized controlled trial design to control the influence of covariates in the results [99], (2) objective assessment of PA and sedentary time, and (3) relatively good adherence to the study protocol and low attrition. Limitations include a small sample size; the use of self-report for the assessment of anxiety symptoms and app use time; and potential underestimation of PA, as accelerometry cannot accurately detect bicycling, swimming, and other upper-body movements [100]. Regardless, this feasibility, proof-of-concept study demonstrated that intellectually able adults with ASD could favor a gamified PA-promoting mobile app, and this use can induce small improvements in PA, sedentary time, and anxiety, which are worthy of further investigation.

This study demonstrates that a gamified BCT-based mobile app, PuzzleWalk, may be able to decrease the level of sedentary time and create a short-term impact on increasing MVPA, daily steps, and total activity counts among adults with ASD. However, the findings also suggest that anxiety in adults with ASD was unfavorably related to increased light PA, steps, and total activity counts and decreased sedentary time after the intervention. Further longitudinal research is warranted to evaluate the long-term efficacy of PuzzleWalk in improving physical and mental health and to elucidate the underlying mechanisms that explain the roles of PA and sedentary time in changing anxiety symptoms in adults with ASD. Given the promising usability of a gamified app as a supplementary behavior change tactic, it is recommended that the design elements of mobile health interventions be user centered to meet the unique needs and leverage the strengths of the target population (eg, visual support for users with autism). Furthermore, as supported by many previous findings, mobile health interventions should focus on increasing sustainability and long-term behavior change that can continuously promote regular PA participation. Finally, there is a need for more experimental studies conducted in real-world settings to verify the evidence for gamification and other BCTs in underserved populations [62].