Testing Multiple Delivery Methods of the Virtual Exercise Games for Youth with T1D (ExerT1D) Peer Intervention: Protocol Development and Feasibility

Abstract

Introduction:

Adolescents with type 1 diabetes (T1D) face barriers to moderate-to-vigorous physical activity (MVPA) such as uncertainty with self-management, limited access to supportive environments, and stigma related to living with diabetes. Opportunities for peer activities with T1D role model support are limited. To address this need, we tested iterative refinements of pilot Virtual Exercise Games for Youth with T1D (ExerT1D) for feasibility and acceptability.

Methods:

The program included 6 versions: study 1 (1.1–1.4) included an active videogame, and study 2 (2.1–2.2) included a virtual reality active videogame. All versions included T1D exercise management education by clinicians and goal-setting guided by young adult coaches with T1D.

Results:

Seventeen adolescents (median age 15.4 [IQR 14.6–16.4] years, 7 non-Hispanic white, 8 male, median HbA1c 8.1% [IQR 7.4%−11.1%]) enrolled. Participants rated the program, comfort, clinicians, coaches, and group cohesion high/very high. Motivation for the videogame was high. Building T1D and MVPA self-management skills was rated excellent at most sessions, as were peer interactions and enriched communication after adding immersive virtual reality in study 2. Transitions between VR apps caused delays of 19±6 minutes per 60min-90min session. Compared to baseline, HbA1c or Glucose Management Indicator (GMI) decreased over time in an exploratory analysis (d=−1.12, 90% CI [−1.78,−0.48]).

Conclusions:

In a small cohort, the ExerT1D program facilitated a supportive environment for engaging diverse youth with T1D in an MVPA program led by T1D coaches. Larger studies are needed to assess the intervention’s impact on engagement with physical activity, glycemic outcomes, and quality of life.

INTRODUCTION

The adolescent years present unique challenges for youth with type 1 diabetes (T1D). Only ~20% of youth meet the recommended A1c level below 7% shown to correlate with long-term health.^1,2 Moderate-to-vigorous physical activity (MVPA) can be leveraged to improve glycemia, weight, and cardiovascular risk factors³ and has been demonstrated to improve overall body composition, muscle strength, cardiopulmonary fitness, mental health, diabetes-specific risks of glycemic outcomes, and insulin resistance.³ Sustained MVPA in adulthood is associated with fewer microvascular complications and mortality.⁴ Managing glycemia during MVPA requires more frequent glucose monitoring and adjustments to diet and insulin dosing, that are further complicated in adolescence by impaired executive function⁵ and puberty-specific hormones^3,6–8. Perceived barriers to physical activity reported by adolescents include the fear of glycemic destabilization⁹. Additionally, diabetes-specific stigma can be both internal (the negative thoughts and pressures put on oneself) and external (experienced when engaging with others without diabetes)¹⁰, particularly in public or social contexts such as exercise¹¹. Thus, interventions that create peer interactions with others who have shared lived experiences may be relevant when considering barriers to MVPA in this population.

Current pediatric T1D MVPA guidelines are available but not actively understood or prescribed by most clinicians¹². Additionally, there are limited data addressing feasibility and acceptability of studies supporting physical activity for youth with low baseline MVPA³. Active videogames and virtual reality (VR) are potential strategies to promote MVPA among youth^13,14. Group activities incorporating active videogame and VR technology with T1D mentors provide an exciting new potential avenue to deliver components of a successful T1D self-management intervention:¹⁵ 1) behavior modification (goal setting, diabetes self-management); 2) psychosocial support (engagement with T1D peers and role models); and 3) medical guidance (real-time diabetes self-management advice). Herein we present six cohorts of 3–5 participants evaluating the feasibility, acceptability, and preliminary safety of iterative refinements of a home-based virtual intervention to promote MVPA among adolescents with T1D.

METHODS

The two clinical trials were pre-registered on clinicaltrials.gov (study 1: NCT05163912, study 2: NCT05662826) and followed single-group longitudinal interventional designs. We observed feasibility and safety throughout the intervention. Participants completed a retrospective exit survey for acceptability. Reporting follows CONSORT guidelines for feasibility trials (Table S1).

Enrollment

Adolescents were recruited from the Yale Children’s Diabetes Center, clinicaltrials.gov, and diabetes social media groups between December 23, 2021 and June 15, 2022 (study 1; versions 1.1–1.4) and January 12^th – March 2^nd, 2023 (study 2; versions 2.1–2.2). Inclusion criteria were age (14–19yr old), T1D duration ≥6 months, not achieving recommended MVPA targets (i.e., <4days/wk with >60 minutes self-reported MVPA), literate in English, and under regular care by a healthcare provider (≥1 appointment in past year, 24-hour access to insulin dose adjustment advice), and willing to use the study videogames, a CGM, and a smartphone or computer with webcam (all provided by research team if not owned). Screening was conducted by telephone for these criteria, including the Prochaska screening question for MVPA levels¹⁶ and the Physical Activity Readiness Questionnaire¹⁷. Medical histories and glycemic data were verified against medical records (Epic: CareEverywhere). For any health conditions presenting special considerations for MVPA, additional documentation was requested from the primary care physician or sub-specialist for MVPA clearance (n=1, arthritis). Informed consent or parental permission/assent (<18yr) was completed and signed using REDCap eConsent over a HIPAA-compliant tele-video call (Zoom Video Communications, San Jose, CA). Participants were advised that they would share their first name with others in the group, and discussions would occur in HIPAA-compliant virtual environments of Zoom and Foretell Reality with confidentiality as a group rule.

Data on demographics, insulin therapy, anthropometrics, MVPA¹⁶, and sedentary time¹⁸ were collected by parental REDCap survey and verified against medical record. For participants that completed both study 1 and study 2, screening and baseline data for study 2 were used as follow-up data for study 1. Although no long-term follow-up was initially intended, we incorporated HbA1c and other variables with high potential to change over months (total daily insulin, insulin therapy modality, body mass index percentile) into an ancillary analysis of glycemic metrics and their possible covariates during and after study #1. The research team mailed supplies to the participant and held a second tele-video call to confirm proper setup.

Intervention Development

The intervention, based upon Social Cognitive Theory, integrated peer and role model interactions into home-based MVPA¹⁹: networked active videogame sessions and virtual discussions about experiences managing diabetes and MVPA. In study #2, we added VR as a delivery tool to enhance the drivers of adolescent development and motivation according to self-determination theory, which are: competence, autonomy, and relatedness^20–22. VR enhances competence by providing a safe space to learn competence in specific activities like MVPA and self-monitoring with a controlled degree of distractions and barriers. VR enhances autonomy by allowing users to embody a self-designed avatar. Finally, VR enhances relatedness by allowing users to experience new unique environments for the first time as a group.

Study 1 participants fitted their home television monitor with a Nintendo Switch Entertainment System (Nintendo of America, Inc., Redmond, WA) including Ring Fit Adventure which alternates low- (stationary walking or jogging) with 60+ higher-intensity exercises that are dynamic (e.g., knee-lifts) or resistance-based (e.g., abdominal press) tracked by a leg-strap accelerometer and handheld ring with gyroscopic/accelerometer sensors. Success with exercises unlocks more advanced adaptations (e.g., Russian twist). They were also provided a yoga mat, poster with Borg rating of perceived exertion scale, blood ketone meter and strips (Study #1 Abbott Diabetes Care Precision Xtra, Alameda, CA; Study 2 Ketomojo GK+, Amsterdam, Netherlands), and Fitbit Inspire 2 (Google, Menlo Park, CA). Participants were instructed to wear the Fitbit continuously, except when showering, bathing, charging the device, or briefly removing it for wrist comfort (1–2 hours every few days). They were introduced to the Fitbit mobile application but received no specific guidance on whether to retain or alter the manufacturer’s default step goal settings, except during the group goal-setting activities conducted at the conclusion of the diabetes management intervention sessions described below. Research staff also guided them to sign in using de-identified usernames and passwords created in advance by the study team and explained the privacy policies displayed by the app regarding how their anonymous data could be used (product improvement, personalization, safety). Sessions were livestreamed over Zoom (Figure 1).

Figure 1.

Schematic of a virtual exercise session livestreamed over Zoom in study 1. Participants set up their cameras so that they were visible during the exercise for purposes of instructors monitoring exercise safety, assessing and advising on exercise technique, and encouraging exercise. Sessions were guided by a coach, and a clinician with expertise in T1D exercise management was present to advise diabetes-specific adjustments for exercise management. An exercise physiologist was present as an observer for fidelity of the intervention and troubleshooting technical difficulties.

Study 2 participants engaged in physical activity sessions using a virtual reality headset (Meta Quest 2 headset, Menlo Park, CA). Participants completed headset tutorials to adjust the lenses for proper focus, draw a guardian boundary to demarcate an area without obstacles or tripping hazards, and learn how to use the controllers. Participants also designed a unique Avatar to represent themselves during activity using Meta’s features (clothing, accessories, general appearance [e.g., skin tone, eye color, body shape], hairstyle, and voice modifications) and use two applications within the headsets. The first application, FitXR (Meta, Menlo Park, CA) facilitated group exercise classes using the avatars side-by-side to other participants to perform exercises guided by verbal and visual instructions from an automated avatar coach. Classes were password-protected. Participants received individual and group scores within each activity. Following exercise activities, participants used VR headsets to meet in a virtual environment (Foretell Reality, Glimpse Group, New York, NY). Environments included a campfire with roasted marshmallow props, hot chocolate, and a guitar or a tropical beach with sea animals. These environments contain presence-enhancing features (spatial audio, hand and face tracking).

Four individuals living with T1D with experience managing their own MVPA with T1D were trained to be coaches that facilitated the sessions. Their demographic characteristics included an age range of 20 to 26 years (2 non-Hispanic White male, 1 non-Hispanic Black female, 1 non-Hispanic White female). The first coach was trained by reading a prototype manual and lesson guides prepared by the principal investigator with input from the multidisciplinary investigative team (pediatric endocrinologists, a pediatric psychologist, an exercise physiologist, an exercise telemedicine instructor, and behavioral interventionists). Then, the coach conducted two mock sessions with investigative team members as participants. The subsequent coaches were trained by reading the manual, observing recordings of prior sessions, and running mock sessions with the principal investigator and research assistants (RAs) as participants. All mock and live sessions were followed by feedback and refinements (c.f. Assessments below).

An RA facilitated intervention sessions and followed a source document to record the completion of each activity along with related diabetes-specific data (Table S2). The RA also helped participants troubleshoot technical difficulties with using the video games and equipment. A clinician (pediatric endocrinologist author LMN, certified diabetes care and education specialist author AC) provided guidance about diabetes-specific exercise management according to published guidelines^3,23 and an exercise physiologist monitored adherence to the intervention protocol (Table S2).

Participants were recruited in sequential cohorts for the 6-week (study #1) and 4-week (study #2) intervention with meetings twice weekly (Table S2) over Zoom. At the first session in each cohort, the clinician gave an overview of safe exercise practices related to T1D management based on national and international guidelines.^6,7,24 Prior to activity, coaches confirmed that the participants had water, fast-acting carbohydrates to treat hypoglycemia, proper exercise shoes and clothing, and a 2.4m-by-2.4m unobstructed space for exercise. Participants communicated to the coach if they needed to leave the session for any reason, and the RA maintained a list of each participant’s physical location in case emergency services were needed.

Diabetes Management

At each session, before and after exercise, teens reported their sensor glucose value, trend arrow, and insulin administration history (time and amount of last bolus and/or insulin-on-board from pump). Exercise was paused, and clinical guidance was provided if a glucose level was less than 70 mg/dL and/or symptoms of hypoglycemia were reported. Ketone levels were checked if a glucose level exceeded 250 mg/dL or there were clinical concerns for elevated ketones. Clinical guidance was provided when blood ketone levels were elevated ≥0.6 mmol/L or urine ketones were moderate or large. On a weekly basis, the clinician reviewed CGM and ketone data collected by questionnaire or passive upload (Keto-Mojo GK+, Napa, CA), identified any safety concerns (>4% time with hypoglycemia), and communicated these concerns with participants, and parents for participants <18 years old.

For the purposes of coaching exercise, we tracked the proportion of each session spent in active movement, repetitions of each Ring Fit exercise or total FitXR target hits, Borg 6–20 rating of perceived exertion of each Ring Fit exercise, Borg 1–10 rating of the total session, and average Fitbit heart rate.

Following the 20-to-40-minute exercise activity, the last 20 to 30 minutes of each session was spent on discussion over Zoom (study 1) or in virtual environments. Week 1 sessions were dedicated to creating individual fitness-related goals following the SMART (Specific, Measurable, Attainable, Relevant, Timely) principle. In addition, participants in each cohort formulated a group goal of cumulative achievements (e.g., group steps) to promote a cooperative rather than competitive approach.²⁵ Goal progress was checked and discussed each week, including the provision of MVPA and T1D management information that helped support the specific goals. During middle week sessions, coaches engaged teens to create role-playing skits on Zoom that involved educational points related to T1D management. During the final sessions, the coaches led a discussion and painting exercise allowing participants to express the relevance of the intervention to building and sustaining an overall active lifestyle (Figure 2).

Figure 2.

The study 2 virtual reality discussion environment included 3-dimensional painting where adolescents could express where their MVPA goals fit with their overall personal perspectives on MVPA, T1D, peers, family, other life topics, and their intersections. In the above example an adolescent represents themselves as a star that shines and dims with mood and life events, surrounded by diabetes (syringe), food managed with diabetes (hamburger), school, homework and projects (book), sports managed with diabetes (football), and being outside (runner). In study 1, similar discussions occurred verbally without visual expression.

Assessments

Session recordings were double-coded by two members of a rating team that included one exercise physiologist from the intervention support staff (author AH), a second exercise physiologist not involved with the intervention (JM, acknowledgements), and a psychology RA not involved with the intervention (RM, acknowledgements). The objective of coding was to identify missed opportunities for practicing each core competency of a role model and peer intervention^26–31 which were then discussed with coaches and clinicians weekly to iteratively improve the intervention. The competencies were:

1. Facilitating Peer Interactions. We encouraged peer support with group discussions led by coaches, inviting each participant to share general life updates, discuss diabetes management challenges and successes, and physical activity goals.

2. Practicing enriching communication. Instructors and clinicians modeled empathy, resilience, and realistic optimism, validating challenges, empowering youth to manage their condition with confidence and self-compassion. The team encouraged conversations that were supportive, meaningful, and growth-oriented.

3. Building T1D and MVPA Management Skills. Clinicians led T1D self-management skill-building exercises, reviewing participants’ glucose changes during exercise shared in the physical activity session to augment glucose pattern recognition and problem solving, emphasizing that glucose levels during physical activity can vary based on recent insulin doses (“insulin on board”), carbohydrate intake and time of day. We also reviewed strategies to mitigate hypoglycemia and hyperglycemia during activity (e.g., reducing insulin prior to activity or setting an “activity,” “exercise” mode or “temp target” prior to and/or after physical activity).

4. Sharing knowledge and experiences about MVPA and T1D between the teens occurred in some cases. Beginning with study 2, the competency to facilitate these occurrences was removed from the feedback provided to coaches. As peer interactions increased within the immersive VR environment, investigators concluded that focusing peer discussions specifically on T1D management was not a critical competency. They also recognized that such a focus might cause discomfort for adolescents who preferred to limit discussions about their T1D. Instead, the remaining three categories above effectively supported the creation of a peer intervention aimed at building skills in both T1D management and MVPA.

To quantify intervention fidelity, each competency was coded as “excellent” (competency evident with 0 missed opportunities), “fair” (competency evident with ≥1 missed opportunities), or “poor” (competency not evident). To generate rich discussion, each missed opportunity identified by either rater was included provided agreement by the other rater.

At the end of the study, participants completed a follow-up retrospective survey of acceptability including the motivation for exergame play inventory (α=0.79)³², perceived cohesion scale (α=0.78–0.83)³³, and a satisfaction survey from our past work²⁷ to evaluate program components/strategies (5 items, α=0.72), personal comfort level (6 items, α=0.81), and interactions with instructors (12 items, α=0.95). Participants also rated the top 3 factors by importance (Table S3 contains the survey).

Feasibility Outcomes and Analytic Methods

Study outcomes and pre-specified thresholds included recruitment uptake (≥35%)³⁴, attendance (≥75% attending ≥75% of sessions)³⁵, objective verification of MVPA by those in attendance (session average heart rate ≥55% age-predicted maximum³⁶ and session rating of perceived exertion ≥3 out of 10³⁷), biosensor wear-time (≥70%)³⁸, survey completion (≥85%), fidelity ratings (≥70% excellent, ≥90% at least fair), and psychosocial perceptions of acceptability (≥3.0 out of 5). Variable distributions were described using visual inspection, Shapiro-Wilks test of normality, Levene’s test of equal variances, and appropriate effect size formulas to explore temporal trends (SPSSv28, Chicago, IL). Inferential statistics such as ANOVA or comprehensive pairwise testing were not performed due to the small sample size. We refined the curriculum based on data after each cohort, ending the trial when no further refinements were found. This sample size (n=17) had 83% statistical power to detect any feasibility problems having ≥10% incidence according to Viecthbauer’s formula n = (ln[1-γ])/(ln[1-π]) where γ is power and π is minimum incidence of a feasibility problem to be detected.³⁹

The intervention included both short-term behavioral reinforcement (e.g., engagement of peers with T1D in group MVPA) and interactive education targeting long-term habit change (e.g., role-playing skits involving educational points). As such, we observed glycemic metrics not only during the 4–6-week interventions (e.g., CGM data) but also explored them up to 1 year later (e.g., medical record HbA1c) in accord with standard practices for brief psychoeducational interventions^40,41. CGM data were shared by participants using Dexcom Clarity and Abbott LibreView on dashboards exclusive to study staff, and CSV files were downloaded and de-identified. HbA1c data were point-of-care or laboratory measurements collected from the electronic health record.

RESULTS

Participants

Seventeen participants referred by local clinicians, Facebook posts, and T1D community organizations were enrolled and completed follow-up (Figure S1), representing diverse geographic (8 states) and demographic characteristics: 8 female sex, 6 female gender, 2 non-binary gender, 4 public insurance, 2 with household incomes below the federal poverty level, and 7 household income <$60,000 (Table 1). Most participants had A1c levels above American Diabetes Association targets: 12 had an A1c of greater than 7.0%, and 5 had an A1c level of greater than 10.0%.

Table 1.

Baseline characteristics of the study participants.

	Study #1	Study #2*
N	15	10
Age (yr)†	15.4 (14.5 – 16.4) Range:14.1 – 19.7	15.6 (15.0 – 16.0) Range: 14.8 – 17.5
Gender	6 female (40%) 2 non-binary (13%) 7 male (47%)	4 female (40%) 1 non-binary (10%) 5 male (50%)
Sex	8 female (53%) 7 male (47%)	5 female (50%) 5 male (50%)
Race	3 Black (20%) 10 White (67%) 1 More than 1 race (7%) 1 Other (7%)	3 Black (30%) 5 White (50%) 2 More than 1 race (20%) 0 Other (0%)
Ethnicity	4 Hispanic (27%) 11 Non-Hispanic (73%)	3 Hispanic (30%) 7 Non-Hispanic (70%)
A1c (%)†	7.8 (7.4 – 11.2)	8.4 (7.4 – 10.3)
Duration of type 1 diabetes (yr)	8.2±3.8	7.3±3.3
Household income annual	2 <$20,000 (13%) 0 $20,000–$39,999 (0%) 5 $40,000–$59,999 (33%) 0 $60,000–$79,999 (0%) 3 $80,000–$99,999 (20%) 4 >$100,000 (27%) 1 declined to respond (7%)	1 <$20,000 (10%) 0 $20,000–$39,999 (0%) 3 $40,000–$59,999 (30%) 0 $60,000–$79,999 (0%) 2 $80,000–$99,999 (20%) 4 >$100,000 (40%) 0 declined to respond (0%)
Insurance (N public, %)	4 (27%)	1 (13%)
Therapy	1 multiple daily injections (7%) 11 sensor-augmented pump (73%) 3 automated insulin delivery (20%)	0 multiple daily injections (0%) 4 sensor-augmented pump (40%) 6 automated insulin delivery (60%)
Prior CGM use (N, %)	15 (100%)	10 (100%)
Total daily insulin (U/kg)	0.9±0.3	0.9±0.2
Body mass index (%)†	67 (63 – 87)	71 (51 – 84)
Physical activity (days/wk with 60+ min MVPA)	1.5±1.1	1.4±1.1
Screen time (hours/day)	8.8±4.6	10.3±5.4

Data are presented as mean±SD for normally distributed data and median (interquartile range) otherwise.

^*Included 8 participants that participated in study #1.

^†Skewed distribution (age and HbA1c positive-skewed, body mass index negative-skewed).

Intervention Feasibility

Attendance and Adherence

Fifteen participants attended ≥80% of sessions, while the remaining two attended 50%−60% of the sessions. Absences were due to family activities (7), school activities (7), negative mood (6), oversleeping (6), acute illness (4), or no answer was given (2). Most of these absences (27/32, 84%) occurred on Saturdays, the few absences on Wednesdays being due to school activities (1), negative mood (3), or no answer was given (1).

Heart rate data were captured for 129/156 person-sessions (83%), with 15 out of 17 participants having at least one session. Among them, one averaged in the vigorous range of 70%−90% age-predicted maximum, 13 averaged in the moderate range of 55%−70%, and one averaged below the moderate range. The average participant heart rate across sessions was 61.4% ± 5.2% of maximum playing Ring Fit and 57.7% ± 6.0% playing FitXR. Ring Fit effort can be further corroborated since the game’s accelerometer and gyroscope display total active time at the end of each session. We screenshot these values over Zoom and found they averaged 15.6 (SD=7.0) minutes per session, or approximately 50% of the 20–40min MVPA windows we played for. These data are consistent with a circuit workout blending strenuous running and resistance exercises with recovery time, in a balance to sustain an average heart rate in the MVPA range.

The FitXR game inherently has less muscle tracking and feedback, so for further assurance of effort we asked participants after each session to rate their effort across the session (i.e., “session RPE”)³⁷. This metric was captured at 43/51 person-sessions in study 2 (84%) with all 10 participants having at least one session. Among them, 9 participants averaged in the region marked moderate-to-somewhat-hard on the scale (i.e., 3–4 out of 10) and 1 averaged in the region marked easy (i.e., 2 out of 10). Other metrics we have archived⁴² for future analyses describing the MVPA include repetitions of individual exercises, RPE of individual exercises, and hip accelerometry.

All participants were using a CGM in their clinical care before the study (14 Dexcom G6, 3 Freestyle Libre 2). Overall, CGM adherence was 73.6% in study 1 and 84.6% in study 2 across all study participants. Data completeness at the participant level was 79% (SD 25%) with 12 participants at ≥70%, 4 participants at 24%−54%, and 1 participant only sharing visual CGM data without a CSV file (excluded from data tabulations).

Fitbit wear time on average met literature standards (70.5%, or 16.9 hours per day suggesting an average of 10–15 hours during waketime)³⁸. Version 1.1 participants stopped wearing the Fitbit 2–4 weeks into the program, so we inserted a biweekly check of wear-time and troubleshooting after which such abandonment of the Fitbit did not repeat.

Technical Barriers

In study 1, one participant was unable to participate in one session due to “dead” Ring Fit controller batteries (rated as moderate-high disruption). This participant was also unable to communicate verbally in two other sessions due to phone audio interruption (though could communicate by text chat). Two other participants had brief Zoom connectivity interruptions on 3 occasions, respectively (reported as mild disruption).

In study 2, there were 15 malfunctions of the Oculus system (8 microphone, 3 speakers, 2 guardian boundary misplacement, 2 passcode), 9 with FitXR (5 times full class did not launch, 4 times individual participant ejected from class), and 5 with Foretell Reality Rooms (4 times individual participant locked out or ejected, 1 time no audio in room). FitXR was used in 12 of the 16 sessions and Foretell Reality was used in all 16 sessions. All issues resolved after restarting the app and/or system. FitXR classes were not started until all individuals could join the class without malfunctions, causing a delay of 19 (SD=6) minutes per session. When malfunctions occurred for individual participants mid-session, activities were paused unless rejoining was impossible. In these cases, a backup coach joined the participant in a separate FitXR class (4 instances) or group discussion (1 instance). Most malfunctions (80%) occurred during weeks 3–4 of study 2 before Meta updated its firmware (v50), when other users nationally reported similar issues.

Fidelity

In study 1, intervention evaluators rated coach competencies highest for building T1D self-management and MVPA management skills followed by enriching communication and facilitating peer interactions at the group sessions (Table 2, Figure 3). Most competencies trended upward for each successive cohort except for cohort 1.4. Cohort 1.3 was the largest size and most balanced with respect to race & ethnicity, income, HbA1c levels, and recruitment source (Table 3). Cohort 1.4 had lowest size plus lowest attendance among those enrolled. One coach covered the first two cohorts and another coach covered the last two, both achieved similar average scores on most competencies. In study 2, facilitating peer interactions increased and reached similar rating as building skills (Table 2, Figure 3). Enriching communications also increased, to a lesser extent. Building skills remained high but sharing knowledge and experiences decreased. We did not ask coaches to attempt correcting this trend, as building T1D and MVPA self-management skills persisted with input from the supervising clinician. Therefore, the research team opted to allow participant conversations to occur naturally. Two coaches alternated sessions in both cohorts, both achieved similar average scores on most competencies.

Table 2.

Satisfaction, perceived cohesion, and exergaming motivation by cohort and by study. Reported as Mean±SD with (% rated ≥4 out of 5)

Version	Interactions with Clinicians and Coaches	Group Cohesion (Perceived)	Program Components and Strategies	Comfort	Exergaming Motivation
#1.1 (n=3)	4.6±0.7 (67%)	4.2±0.9 (67%)	4.1±0.2 (100%)	3.9±0.9 (67%)	3.8±0.4 (33%)
#1.2 (n=4)	4.9±0.0 (100%)	4.6±0.4 (100%)	4.5±0.3 (100%)	4.5±0.1 (100%)	4.1±0.2 (75%)
#1.3 (n=5)	4.9±0.2 (100%)	4.2±0.4 (80%)	4.3±0.6 (60%)	4.2±0.4 (80%)	3.9±0.2 (60%)
#1.4 (n=3)	4.8±0.3 (67%)	4.7±0.2 (100%)	4.3±0.7 (33%)	4.3±0.6 (33%)	4.6±0.2 (100%)
Study #1 Total (n=15)	4.8±0.3 (93%)	4.4±0.5 (87%)	4.3±0.5 (73%)	4.2±0.5 (80%)	4.1±0.4 (60%)
#2.1 (n=5)	4.7±0.4 (100%)	4.3±0.5 (80%)	4.6±0.3 (100%)	4.7±0.4 (100%)	3.8±0.5 (40%)
#2.2 (n=5)	4.5±0.5 (100%)	4.4±0.8 (80%)	3.8±0.2 (40%)	3.7±0.6 (40%)	3.9±0.4 (60%)
Study #2 Total (n=10)	4.6±0.4 (100%)	4.3±0.6 (80%)	4.2±0.5 (70%)	4.1±0.7 (60%)	3.8±0.4 (50%)

Study #1 (versions 1.1–1.4) is 2-dimensional Ring Fit game and discussions by Zoom shown in methods Figure 1. Study #2 (versions 2.1–2.2) is virtual reality immersive environment shown in methods Figure 2. T1D, type 1 diabetes. MVPA, moderate-to-vigorous physical activity.

Figure 3.

Competency ratings. Numbered bars represent cohorts with color-coding by study (purple is study 1, blue is study 2) with brackets indicating 95% confidence intervals. The green shaded region indicates the a priori target for acceptability. The sessions not scored “excellent” were scored “fair” (i.e., none were scored “poor”).

Table 3.

Cohort Characteristics

Cohort (Symbol on Figure 3, n)	Race/ Ethnicity	Gender	Age	Household Income (annual)	A1c	Recruitment	Attendance (Mean %)1
#1.1 (X, n=3)	2 HispWhite 1 NHispWhite	1 NB 2 M	1, 14–15yr 2, 16–17yr	1 under $20k 2 under $60k	3 over 7.0% 1 over 10.0%	3 from clinic 0 from social media	75
#1.2 (+, n=4)	4 NHispWhite	4 M	3, 14–15yr 1, 16–17yr	0 under $20k 0 under $60k	2 over 7.0%	0 from clinic 4 from social media	92
#1.3 (▼, n=5)	3 NHispBlack 1 More than 1 race 1 NHispWhite	1 NB 3 F 1 M	5, 14–15yr	0 under $20k 3 under $60k 1 not reported	5 over 7.0% 2 over 10.0%	2 from clinic 3 from social media	95
#1.4 (●, n=3)	2 HispWhite 1 NHispWhite	3 F	1 14–15yr 1, 16–17yr 1, 18–19yr	1 under $20k 2 under $60k	2 over 10.0%	2 from clinic 1 from social media	72
#2.1 (X, n=5)	3 NHispBlack 2 More than 1 race	3 F 2 M	4, 14–15yr 1, 16–17yr	0 under $20k 3 under $60k	5 over 7.0% 2 over 10.0%	2 from clinic 3 from social media	90
#2.2 (+, n=5)	2 HispWhite 3 NHispWhite	1 F 3 M 1 NB	4, 14–15yr 1, 16–17yr	1 under $20k 1 under $60k	4 over 7.0% 1 over 10.0%	2 from clinic 3 from social media	83

HispWhite, Hispanic White. NHispWhite, Non-Hispanic White. NHispBlack, Non-Hispanic Black. NB, non-binary. M, male. F, Female.

¹Cohort #1.4 took one week longer than expected to recruit the minimum 3 participants. In fairness to the 2 participants who signed up on the preset schedule, we offered them 2 sessions during the week of delay which were credited as extra sessions.

Intervention Acceptability

Participant satisfaction ratings in both studies were highest for interactions with clinicians and coaches, followed by perceived group cohesion, program components and strategies, comfort, and lastly, active videogame motivation (Table 2). The clinicians and coaches component was also most frequently selected as a top-1 or top-3 (72%, 64%).

Intervention Safety and Monitoring

No participants experienced diabetic ketoacidosis, severe hypoglycemia, or other study-related adverse events during the study period. Two participants developed elevated ketone levels prior to exercise (1.1, 1.4 mmol/L) due to pump or infusion site malfunctions, that resolved after insulin administration. One participant deferred exercise due to feeling unwell as a result of having a high sensor glucose level (400 mg/dL) without elevated ketones. Three participants reported hypoglycemia symptoms during a session that resolved with carbohydrate treatment and three had sensor glucose levels <70mg/dL less than 1 hour after a session that participants resolved independently without staff guidance.

Weekly clinician reviews of CGM data resulted in 5 participants being referred to their diabetes healthcare providers for insulin dose adjustments to address excess time spent with hypoglycemia. Topics related to avoiding hypoglycemia after exercise were discussed in the group lesson and with participants directly.

Intervention Acceptability

Ratings of specific components were the highest for activities as follows: group 3-D painting, seeing personal and peers’ scores for hitting boxing targets during MVPA, freely exploring the virtual reality environments, and exercising with other teens with T1D. The lowest-rated activities were those either administered in the non-avatar Zoom audio-chat (icebreaker questions, blood sugar checks) or automated videogame characters not part of the peer group (FitXR automated coach) (Table 4).

Table 4.

Ranking of satisfaction of specific study #2 activities (n=10). Not done for study #1.

Component	Mean±SD (% rated ≥4 out of 5)
Group 3D painting	4.9±0.3 (100%)
Seeing individual score	4.6±0.5 (100%)
Seeing everyone’s scores	4.4±0.7 (90%)
Freely exploring virtual environments	4.4±0.7 (90%)
Exercising with other teens with type 1 diabetes	4.3±0.7 (90%)
Exercise Movements	4.2±0.6 (90%)
Exercising with coach that has type 1 diabetes	4.1±0.6 (90%)
Diabetes skit	4.1±1.0 (80%)
Talking about individual physical goals	4.1±1.0 (80%)
Discussing exercise management strategies with clinicians and coaches	4.1±0.7 (80%)
Talking about group physical activity goals	4.1±0.7 (80%)
Icebreaker questions	4.0±0.9 (80%)
Checking blood sugars and reporting them to clinicians and coaches	4.0±0.9 (67%)†
Automated Exercise Coach	3.7±0.7 (60%)

^†One participant declined to answer.

Participants set individual and group goals related to achieving Fitbit step counts, playing a sport, recreational exercise, or playing Ring Fit outside of group sessions, or managing diabetes (Table S4) with 28% success on individual goals and 50% success on group goals. Intervention evaluators noted that study 1 included a young adult instructor from an exercise physiology degree program and goal-setting discussions had more rigor in principles of exercise prescription (frequency, intensity, time, type, volume, progression).

Average steps in study 1 for those with Fitbit step goals were 8274±3734 per day and for the total sample were 7178±4013 per day. Average steps in study 2 for the one participant with a Fitbit step goal was 6,034 per day and for the total sample was 6467±2998 per day. Each skit successfully integrated 3–5 educational points (Table S5).

Exploratory Ancillary Analysis of Glycemic Outcomes

Between the two timepoints that HbA1c was available for the same participants - Study 1 baseline and follow-up (n=15) - it was on average lowered (8.9% vs 8.3%) but equally positive-skewed (Levene’s F_1,2=0.458) (Figure 4). Such effect is best standardized as the Wilcoxon Signed Rank test index W= −1.905, equal to biserial r= −0.49 or Cohen’s d= −1.12 (90% CI [−1.78, −0.48]) (i.e., probably a medium or large effect)⁴³. Total daily insulin showed was similar (0.88±0.27 to 0.83±0.24 U/kg/day). Some participants (n=7) who upgraded to a closed-loop system during follow-up had the same average HbA1c change as the full cohort. Body mass index percentile remained negative-skewed and showed no indication of change (median 67 vs 66, mean 75 vs 75, IQR 63–87 vs 53–86).

Figure 4.

Glycemic metrics over time. Top. Metrics indicating average glucose. Version #1 baseline is laboratory HbA1c taken an average of 7 weeks (range 1–14) before intervention start, extracted from medical record. Version #1 follow-up is laboratory HbA1c taken an average of 45 weeks (range 19–77) after intervention start, extracted from medical record. Missing values (1 participant at weeks 1–6 due to not sharing downloadable CGM data, 2 participants at follow-up due to not sharing long-term access to medical record) were imputed by intention-to-treat, last observation carried forward. Other boxes are glucose management indicator (estimation of HbA1c based on CGM). Dash through box indicates median, plus sign indicates mean, box indicates interquartile range, whiskers indicate minimum and maximum. Middle. HbA1c and glucose management indicator (GMI) metrics from top panel plotted by individual participants. Bottom. CGM metrics indicating time in glucose ranges. Data completeness at the participant level was 79% (SD 25%) with 12 participants at ≥70%, 4 participants at 24%−54%, and 1 participant only sharing visual CGM data without a csv file (excluded). The 1 participant who did not share exportable CGM data was excluded. Bars are omitted from timepoints where ≥5 participants had no shared data.

CGM clinical targets were met for time in hypoglycemia level 1 (1.8%−2.2% vs target ≤4%) and level 2 (0.5%−0.7% vs target ≤1%), but not level 1 (42%−51% vs target ≤25%) or level 2 hyperglycemia (21%−28% vs target ≤5%) or in target range (46%−55% vs target ≥70%) (Figure 4). Full CGM summary metrics are given in tables S6 and S7.

DISCUSSION

The primary finding of this study is that virtual delivery of an MVPA intervention was practical, safe, and well-received, providing meaningful opportunities for youth with T1D to engage with peers. Compared to our previous in-person clinical trial (Bright T1D Bodies) of MVPA in youth with T1D²⁶, the ExerT1D study had substantial improvements in recruitment uptake from 16% to 35%. This is comparable to recruitment uptake rates seen in personalized MVPA interventions (37%) that did not require coordinating activities to a specific group schedule³⁴. We also expanded from clinic to social media venues, which were well accepted by youth in the study. In addition, we improved the attendance from 56% to 85%−90%, while also doubling the frequency of sessions from once to twice per week. A glucose monitoring protocol was followed with no severe adverse events and occurrence of MVPA-induced hypoglycemia was similar to typical daily living. Recruiting for the program was also highly cost-effective. Social media advertisements utilized community postings rather than paid advertisements, and virtual technology was all within the scope of devices normally owned by diverse families of teens with T1D. Staffing of sessions was also cost-effective, and would have cost in the real-world US $100 per session for a certified diabetes care and education specialist trained in exercise physiology and US $25 per session for a college student health coach. In our previous program conducted in-person, each session cost US $294 in staffing since we had to prepare for widely variable attendance numbers, US $220 to rent a physical facility, and many participants needed taxi fare since not having transportation²⁶. Active videogame technology represents a promising modality for engaging youth with T1D from a variety of different socioeconomic backgrounds⁴⁴ in regular MVPA. These achievements were gained from an iterative process (Figure 5).

Figure 5.

Iterative design of the Virtual Exercise Games for Youth with Type 1 Diabetes (ExerT1D) Peer Intervention. Each blue arrow represents a set of interview themes that emerged from qualitative interviews and evaluations of session recordings in the study where the arrow starts and all subsequent studies. Box below each blue arrow details the specific themes. Row of red text indicates barriers that emerged in each study and informed us to change the venue for the next study. Bottom table indicates metrics of feasibility and estimation of efficacy. 2D, 2-dimensional (livestreamed on Zoom). VR, virtual reality. CGM, continuous glucose monitor. Hypo, hypoglycemia.

Engaging adolescents with T1D, the age group with the highest HbA1c levels, remains a persistent challenge⁴⁵. Notably, youth with the highest A1c levels (HbA1c > 9%−10%), who stand to benefit the most, have often been excluded from research. Diabetes-related stigma is a recognized barrier in social contexts, including exercise¹¹. ^46,47Virtually interacting with peers who have shared lived experiences with T1D may help reduce internal stigma by normalizing diabetes self-management behaviors, reducing feelings of isolation, and promoting positive identity formation, thus supporting engagement in MVPA.

Our program was found to be highly acceptable and comparable to our previously published in-person protocol²⁷. Participants rated group cohesion, peer interactions, and program comfort highly, reflecting strong social connectedness. Additionally, program structure and comfort received strong rates (4+/5) supporting the program’s acceptability. Perceived group cohesion has been associated with reduced internalized stigma and increased self-efficacy in other group interventions, particularly where shared lived experiences normalize health-related challenges and foster a sense of belonging^48–50. This suggests a plausible pathway by which the ExerT1D study provides a virtual environment to connect youth with T1D and mitigate stigma-related barriers to MVPA. Thus, future clinical trials are needed to directly investigate this connection.

A limited understanding of exercise management is commonly reported as a barrier for individuals with T1D. This knowledge gap can contribute to feelings of helplessness in diabetes self-management. Interventions that address these deficits may empower individuals to engage in physical activity while reducing the associated psychological burden. Participants in our study reported that their overall diabetes management knowledge and intentions were positively impacted by the intervention, in semi-structured exit interviews we have reported separately⁵¹. Prior studies supporting MVPA among adolescents with T1D provided self-management education and decision guidance^34,52–54, with only one including personal support from key family members such as a parent³⁴. In real-world contexts, however, 91% of adolescents with T1D report that their parents discourage MVPA⁹, and teachers and coaches have limited T1D knowledge⁵⁵. These findings underscore the need for interventions introduced during adolescence - a crucial developmental stage when they are developing the skills they will need for independent T1D management^26,27. While our program did not formally involve parents we sought to address parental concerns by ensuring parents were informed that the adolescents were receiving guidance to promote autonomous self-management. Parents were included for safety concerns or when insulin dose adjustments were needed. Parents were interviewed in our previous work, and they expressed positive sentiments including a sense of empowerment that their teen could manage T1D²⁷.

Our study builds on recent findings by Zaharieva et al⁵⁶, who demonstrated that quarterly visits incorporating T1D-specific exercise education management improved short-term glycemic outcomes among youth with type 1 diabetes. While their approach used structured, periodic education, our pilot feasibility study extends this model by delivering real-time, individualized guidance for exercise-related glycemic management over a shorter time frame. This more immediate and frequent support is designed to reinforce learning through repeated, real-world application and accelerate long-term learning, which is particularly valuable for adolescents developing foundational self-management skills. Support from peers and peer mentors is rated as important by adolescents with T1D^57,58 and associated with increased diabetes self-care⁵⁹. However, the incorporation of peers and role models is in nascent stages^28,60 and mostly focused on T1D-specific behaviors rather than those important for general health such as MVPA. Limiting peer support to T1D-specific behaviors has been negatively associated with diabetes self-care⁵⁹, similar to the “nagging” perceived from parents⁶¹. It may also be less engaging, as prior peer interventions have almost entirely been restricted to asynchronous communications without physical activities^29,30. When synchronous chats have been included just 39%−48% of participants attended one or more sessions^30,60. Overall, the present study shows the promise of peer group MVPA as a novel strategy both for supporting MVPA and generally engaging adolescents with T1D management.

The novel use of VR in clinical practice is expanding, with FDA-authorized indications for chronic pain⁶² and physical rehabilitation⁶³, and existing CPT codes (0770T, 0771T-0774T) supporting insurance coverage for select uses. Our institution has recently developed the XRPediatrics program, that focuses on virtual, augmented, and mixed reality game technology with the goal of providing effective and safe applications for youth. Extending VR interventions to adolescents with T1D represents a promising and innovative strategy, particularly given the development challenges in this population and the difficulties with in-person engagement. VR offers a novel platform to foster connection, motivation, and diabetes-specific skill development. VR-based interventions may complement clinical care by promoting greater engagement with MVPA and fostering social connections with peers.

A major strength of the study was the enrollment of a diverse group of adolescents in greatest need of support for MVPA and T1D self-management (under-resourced, people of color, elevated HbA1c)⁶⁴. This study had no exclusion criteria for elevated HbA1c levels, thus incorporating a group of youth with T1D at high risk of dysglycemia that has been largely excluded from research studies. Intervention attendance was excellent and overcame logistical barriers related to alignment of group schedules, availability of MVPA supplies, timing MVPA with insulin and diet, and provision of real-time guidance. Given the focus on feasible uptake of technology and inclusion of youth with all HbA1c levels, we enrolled participants even if they were wearing their CGM inconsistently at enrollment, which meant glycemic outcomes could only be assessed by combining HbA1c and GMI values in an exploratory ancillary analysis. HbA1c and GMI can be discordant due to the timeframe they reflect and non-glycemic factors that can impact HbA1c⁶⁵. Nonetheless, improvements in glycemia were seen during and after the intervention. These findings were encouraging, especially in light of the elevated HbA1c levels in the cohort at baseline. At the same time, the improvements were driven by participants with the highest baseline HbA1c levels (Figure 4, middle panel) which may reflect concurrent medical management tailored to elevated values - an effect we cannot isolate in the absence of a control group. We acknowledge some limitations of the intervention. Coaches were trained in group facilitation strategies derived from our expert planning group discussions and two pilot sessions, but for future studies we will utilize an accredited group facilitation course. Second, Fitbit measurements of heart rate may be less accurate for those with darker skin tone⁶⁶. However, the Fitbit was mainly used for step-counting which does not have this limitation, and heart rate was only used for an assessment of convergent validity with the validated videogame⁶⁷. That the study was limited to those who were able to afford Internet access at home and time to attend the sessions does limit some generalizability and implementation. Some post-intervention HbA1c values included in the exploratory analysis were more than 12 weeks after the intervention and may have been influenced by confounding variables. Enrolling participants from study 1 into study 2 may have led to bias in acceptability data, as returning participants may have been more engaged, motivated, and comfortable with the study procedures. Finally, as a small single-group study design, we had limited ability to robustly determine efficacy outcomes.

CONCLUSIONS

In summary, we report protocol development and feasibility of a virtual home intervention to promote MVPA among peer groups of adolescents with T1D. Our study successfully engaged a diverse group of youth with T1D, maintained high attendance rates at the sessions, included real-time individualized T1D self-management guidance that promoted safe MVPA, and created an environment for meaningful connections with shared experiences with T1D. The challenge was technical delays, especially transitions between virtual reality apps. Therefore, future trials testing efficacy and moderators warrant financial investment to integrate and interoperate the virtual reality apps. The advantages of our program were effectiveness at connecting teens to peers, clinicians, and young adult coaches in a way that was logistically feasible, time efficiency for the clinicians and coaches since they could interact with a group all at once, and sustainment of favorable MVPA and clinical metrics. Future clinical trials are needed to evaluate the efficacy of enhancing T1D exercise engagement, the impact on glycemia, and implementation of this intervention in broader, real-world settings.

Data Availability Statement

Individual de-identified participant data (including data dictionaries) are publicly available via Dataverse⁴². This includes individual participant data that underlies the results reported in any aspect of a published article (text, tables, figures, and appendices). Other documents that will be available include the study protocol, statistical analysis plan, informed consent form, and analysis code.