A multicomponent behavior change intervention to promote walking in adults after traumatic brain injury: A Pilot Randomized Control Trial

Abstract

Background

Sedentary behavior is highly prevalent following traumatic brain injury (TBI) and compounds existing risks for cardiovascular, neurodegenerative, and affective disorders. The cognitive and behavioral sequelae of TBI, including impaired decision-making, blunted reward processing, and cognitive fatigue, create particular barriers to adopting and maintaining an active lifestyle. Effective behavior change interventions targeting physical activity in community-dwelling TBI survivors remain scarce. Here, we evaluated the feasibility, compliance, and preliminary efficacy of a 12-week remotely delivered walking intervention combining planning, reminders, and monetary micro-incentives.

Methods

Fifty-six community-dwelling adults aged 40–80 years with a mild-to-moderate TBI diagnosed between 3 months and 15 years prior were randomized to either a planning, reminders, and micro-incentives intervention (n=23) or a health advice control condition (n=25). Participants wore a Fitbit Inspire 3 continuously throughout the study. Intervention participants completed weekly phone calls to plan five 30-minute walks for the following week, received daily text message or email reminders on planned walk days, and earned small incentives (redeemed for money at the end of the study) upon walk completion. Control participants received weekly health education calls. Feasibility was operationalized as recruitment, retention, and adverse event rates. Compliance was assessed via phone call completion rates and Fitbit wear time. Efficacy outcomes included weekly walk counts, walking duration, and step counts. Modifiers and exploratory outcome measures included environmental and psychosocial variables.

Results

Forty-eight participants completed the study (retention rate: 84.2%), with high phone call compliance in both groups (intervention: 98.4%; control: 98.1%). Intervention participants completed more walks than controls from week 1 onward (aIRR = 5.33, 95% CI: 2.27–12.5), with the group difference growing over time (interaction aIRR = 1.09 per week, 95% CI: 1.01–1.17). The intervention group also walked significantly longer at week 1 (b = 62.14 min, 95% CI: 1.05–123.23), with the difference growing over time. Similarly, the intervention group accumulated significantly more steps during walks at week 1 (b = 4,779 steps, 95% CI: 45.50–9,513.00). Baseline intrinsic motivation increased the probability of walking on any given day and those who on average walked for >150 minutes per week saw higher improvement in their perceived cognitive abilities (b = 4.21, 95% CI: 0.15, 8.27).

Conclusions

Remotely delivered, multicomponent walking interventions consisting of planning, reminders, and micro-incentives are possible candidate programs to fill a gap in community-based healthcare for individuals living with lifelong consequences of TBI. With high retention and compliance, and strong preliminary efficacy of the intervention these findings provide compelling support for a larger, fully powered trial prior to implementation in a clinical setting.

Introduction

Once re-integrated back into the community, individuals who have sustained a traumatic brain injury (TBI) are susceptible to long-term cognitive, behavioral, emotional and social deficits^1,2. Adoption of unhealthy habits, like sedentary behaviors, places these individuals at greater risk for long-term health issues, including increased risk of cardiovascular disease, neurodegenerative disorders and affective disorders^3–5. A critical gap in access to community-based healthcare and programs can leave survivors of TBI without meaningful support to build and maintain healthy habits once reintegrated back into their communities.

Physical activity (PA) and exercise have been shown to be effective lifestyle strategies to improve cardiovascular health, cognitive and brain health⁶, favorable disease outcomes⁷ as well as overall morbidity and mortality^8–10. The beneficial health effects of PA engagement appear even larger in those with pre-existing conditions^11,12. However, maintaining a physically active lifestyle is particularly difficult for those with long-term injuries. Damage to neural networks that underpin decision-making abilities, self-regulation and reward pathways^13–15 shift the balance of favor towards automatic processes (e.g., being sedentary), making it harder to initiate effortful behavior (e.g., PA). Engaging in PA requires substantial executive cognitive resources to plan and make decisions about when, where, and how to be physically active¹⁶. This cognitive load becomes particularly challenging when competing priorities in one’s environment demand attention, which is particularly difficult with mental fatigue after TBI¹⁷. Moreover, one needs to actively interrupt the automatic habitual behavior of being sedentary to initiate the planned behavior. Interruption of habitual sedentary behavior requires conscious effort of moment-to-moment awareness as well as self-regulatory control¹⁸. Since the health benefits of PA are often delayed rather than immediately rewarding, individuals must rely on higher-order cognitive processes to overcome the natural inclination toward effort minimization (while maximizing rewards)^19,20. This reward deficit creates additional cognitive burden, as an individual must actively convince themself to engage in effortful behavior despite often an absence of immediate, tangible benefits. Together, multicomponent interventions to scaffold complex physical activity behaviors may be particularly effective in individuals with TBI. For example, planning can reduce the cognitive burden of behavior initiation by creating automatic if-then behavioral scripts²¹. Behavioral reminders, (for example in the form of push notifications or text messages) can interrupt habitual sedentary behaviors and trigger pre-planned activites²². Small but timely incentives may provide proximal rewards that compete with the immediately rewarding sedentary behavior²³. Around ~70% of PA research with TBI and stroke patients have tested the associations between physical activity and health outcomes, whereas only 6% have tested the effectiveness of interventions to promote independent PA behaviors, according to one review²⁴. Indeed, the only systematic review of physical activity behavior change interventions in TBI report that the overall efficacy of specific self-management interventions remains inconclusive²⁴. As such, evidence of interventions to promote PA adoption in TBI survivors is needed.

Here, we conducted a 12-week, two-arm, pilot randomized control trial of a virtually delivered, planning, reminders, and micro-incentives intervention to promote walking in adults with TBI. The objectives of this pilot trial were to assess the feasibility, compliance and preliminary efficacy of this intervention, compared to a health advice only control group. In an exploratory step, we aimed to examine potential modifiers of walking behaviors in adults with TBI and ask whether walking habits contribute to improved health outcomes.

Methods

Details of all study protocols are described below. This study received ethical approval from the Northeastern University Institutional Review Board (#23-06-18) and was registered on clincialtrials.gov before enrollment of the first participant (NCT06222502).

Study Design

This was a 12-week, 2-arm pilot randomized control trial. Participants were randomized (using block randomization of 4–6 per block) into either an active or control intervention. Randomization allocation was concealed using REDCap. A series of cognitive function and psychosocial questionnaires were collected at baseline (before randomization) and after the 12-week intervention, either in-person (if the participant was local to the greater Boston area) or over a HIPPA-compliant zoom video call. Local participants were invited to participate in magnetic resonance imaging session however these data are not presented here due to the small N (n=10). Participants were blinded to study hypotheses and told that the study intended to assess walking habits in adults with a TBI. Investigators were not blinded to allocation. Outcome measures of PA were collected continuously over 12 weeks using a provided Fitbit Aspire 3.

Participants

Adults aged 40–80 years with a diagnosis of mild or moderate TBI between 3 months and 15 years ago were recruited via advertisements in the United States. Participants were required to receive a formal mild or moderate TBI diagnosis by a physician (AS) using OSU-TBI-ID questionnaire ²⁵, requiring a direct blow or impact to the head with either loss of consciousness of 0–30 minutes (mild) or >30 minutes and <24 hours (moderate) or post-traumatic amnesia of 0–1 day (mild) or >1 day and <7days (moderate). Additional inclusion criteria include ability to walk unassisted. Exclusion criteria included a history of severe TBI (loss of consciousness >24 hours, post-traumatic amnesia > 7 days of a Glasgow Coma Scale of <9) current speech or physical therapy, mild cognitive impairment (as assessed by a mini-mental status exam score of ≤30) or neurological, neurodegenerative or cardiovascular disease.

Sample size justification

The intended sample size for this pilot study was n=60 and derived pragmatically based on the feasibility of recruiting initially within a 1-year time frame. As a pilot study we expected that N=60 would provide precise estimates of feasibility and to quantify the precision and variance of the mean effect, which aids in the planning of larger, sufficiently powered efficacy trial. We anticipated a 20% attrition rate and expected to have a final sample of 48, 24 in each group. A posthoc power analysis using a Wald-based approximation on the primary group-by-time interaction term was run to calculate the achieved power with the final sample size (see Results).

Planning, Reminders and Micro-incentives Intervention

Those randomized to the intervention arm received weekly phone calls to plan their walk days for the following seven days. During weekly phone calls with an interventionist, participants planned five 30-minute walks for the following seven days. When, where, and how was not intentionally part of the script, but interventionists often reported participants shared voluntarily when and how they were going to walk. Participants received a daily text message on planned walk days at 8am with a reminder of their daily plan and incentive notifications. Points were awarded when Fitbit data, connected to our propriety platform, confirmed walk completion, typically within 20 minutes to 1 hour (dependent upon the Fitbit API syncing). 500 points were awarded for each walk, with 1000 points converting to $1 at the end of the study. Participants were encouraged (during weekly phone calls) to track their progress on the study website, view completed walks and point totals. Points were only awarded on planned walk days.

Control Intervention

Those randomized to the Control condition received similar weekly phone calls with information focusing on health education. This information was delivered with weekly phone calls from staff (separate from intervention staff). Controls were provided with information in the form of “health tips” about the benefits of physical activity and other healthy behaviors. For example: “Walking 10,000 steps daily benefits cardiovascular health”. Participants were reminded to wear their Fitbit consistently and check the study platform for weekly health tips. Consistent scripts and staff training aimed to apply the control condition content uniformly and match the participant-staff contact time with that if the intervention group.

Intervention Monitoring

Two password-protected websites were used to deliver and monitor the intervention: a staff website for inputting and modifying participant schedules and pulling Fitbit data automatically in near-real time, and a participant website displaying a walk calendar, point tallies, and color-coded completion status (see supplementary figure 1). The system connected to Fitbit’s Web API for near-real-time data collection and used Twilio to send automated reminders. Data synced typically within 20 minutes after a walk, points were awarded automatically upon walk completion, and individual data were downloaded for offline analysis. Participants were monitored for adverse events via weekly phone calls, with any health or safety concerns documented, reviewed by the principal investigator, and classified according to National Institute on Aging guidelines. An Inspire 3 Fitbit was mailed after baseline, with setup scheduled 7–14 days later via Zoom (~30 minutes). Study staff created a de-identified Google account for each participant prior to setup, which covered platform familiarization and device and application configuration. Daily data were extracted from participants’ Fitbit Google Accounts, including minute-level heart rate, daily step counts, and walk activity logs (start time, duration, steps, distance, heart rate, caloric expenditure), filtered to each participant’s active intervention period. Study weeks were defined as 7-day periods beginning from each participant’s first weekly phone call. Device wear time was calculated using heart rate data as the primary indicator of wear, following Baroudi and colleagues²⁶, as the number of unique minutes with heart rate data divided by 1,440, multiplied by 100. Daytime (7:00AM–11:00PM) versus nighttime (11:00PM–7:00AM) wear percentages were also calculated to characterize wearing patterns. Walk data were extracted from Fitbit activity logs (labeled “Walk” only) and validated as complete if they comprised either a single walk ≥27 minutes or two walks of 15–27 minutes on the same day. The ≥27-minute threshold was chosen to reduce participant frustration for walks of 27–29.99 minutes. All walks >15 minutes were tallied for total walk counts in both groups. For intervention participants, the first completed walk on a planned day was classified as “planned,” with additional walks on that day or any unscheduled day classified as “unplanned,” ensuring planned walk counts could not exceed scheduled walks while capturing activity beyond the protocol. For control participants, all completed walks contributed to their total walk count.

Cognitive Function and Psychosocial Assessments

Verbal Fluency²⁷ and Hopkins Verbal Learning Test²⁸ were administered at baseline and endpoint. A paper version of the Trail Making Test was administered to participants who completed their baseline session in person but not presented here due to the small N (N=10). These were chosen to provide measures of domains with known deficits after a TBI and are commonly used in clinical practice. All cognitive testing was administered by two trained research staff (EMT and LH). Patient-Reported Outcomes Measurement Information System (PROMIS)^29–31 questionnaires, including Global Physical Health (GPH) subscale and the Global Mental Health (GMH) subscale from the PROMIS Global Health Scale v1.2 (PROMIS-10), and the PROMIS Cognitive Function – Abilities Subset v2.0, Short Form 8a were collected at baseline and endpoint. Additionally, a demographic survey, the Pittsburgh Sleep Quality Index (PSQI)³², International Physical Activity Questionnaire (IPAQ)³³, Behavioral Regulation in Exercise Questionnaire 3 (BREQ-3)^34,35, Mediterranean Diet Adherence Screener ³⁶, Self-Efficacy for Exercise Questionnaire³⁷, Patient Health Questionnaire (PHQ-9)³⁸, and Generalized Anxiety Disorder Questionnaire (GAD-7)³⁹. Only the PROMIS, BREQ-3 and Self-Efficacy for Exercise questionnaires were analyzed in this study. Detailed methods of each test and questionnaire are found in supplementary material 2.

Environmental variables

Recreational Area:

Geographic data were obtained from the PAD-US-AR-V1 dataset, an accessible and recreationally focused curation of the United States Geologic Survey’s (USGS) Protected Areas Database-United States (PAD-US) dataset, available on an Open Science Framework (OSF) repository⁴⁰. For each participant, the total accessible recreational area existing within a one-mile radius of participants’ mailing address was quantified.

Walkability Index:

National Walkability Indices (NWI) were obtained from the United States Environmental Protection Agency (USEPA)⁴¹. NWI was associated with each participant using mailing address spatial coordinates⁴² and NWI associated census block group spatial data⁴³. For interpretation of NWI, participant addresses were classified as either urban or rural. If the 1-mile radius buffer area around a participant’s address fell entirely within an “Urban Area”, a Census classification for densely populated areas of at least 50,000 people⁴⁴, the address was considered urban.

Weather Variables:

Daily weather data were obtained from the National Oceanic and Atmospheric Administration Climate Data Online API (Global Historical Climatology Network Daily dataset)⁴⁵. For each participant, the nearest weather station was identified using a ZIP code lookup, and a coordinate-based search within a 50 km radius served as a backup. Daily summaries included temperature, apparent temperature, and total precipitation.

Endpoint Satisfaction Survey

Participants were asked to complete an optional satisfaction survey, answering Likert scale questions about scheduling, communication, difficulty, and open-ended responses about satisfaction with the study. Result of this survey are presented in supplementary material 8.

Statistical analysis

All analyses were conducted in R (version 4.5.2; R Core Team, 2025). Descriptive statistics were used to characterize sample demographics and feasibility outcomes. Continuous variables are reported as means and standard deviations (SD), and categorical variables as frequencies and percentages. Group differences at baseline were assessed using chi-square tests for categorical variables and one-way ANOVAs for continuous variables.

Feasibility and Compliance:

Feasibility outcomes (recruitment, retention, and adverse events) are reported descriptively. Baseline group comparability was assessed using chi-square tests for categorical variables and one-way ANOVAs for continuous variables. Phone call compliance was expressed as the percentage of scheduled calls completed per participant, with group differences in average call length examined using a Mann-Whitney U test. Fitbit wear time compliance metrics are reported descriptively by group.

Efficacy:

Group differences in weekly walk frequency over time were examined using Poisson generalized linear mixed models (GLMMs) with a log link function, fit using the lme4 package with the bobyqa optimizer. The primary model included group (intervention vs. control), week (continuous, 1–12), their interaction, and covariates: age, biological sex, years of education, and time since injury (scaled to unit variance). A random intercept and random slope for week (1 + week | id) were included to account for individual variability in baseline walk counts and rates of change. Race was excluded as a covariate due to limited sample diversity. Results are reported as adjusted incidence rate ratios (aIRR) with 95% confidence intervals and z-statistics. Estimated marginal means (EMMs) at weeks 1, 6, and 12 were derived using the emmeans package, with pairwise comparisons corrected using the false discovery rate (FDR).

Walking duration (total minutes per week) and step count (total steps per week accumulated during walks) were each examined using linear mixed-effects models (LMMs) fit with restricted maximum likelihood (REML) via the lme4 package. Both models included the same fixed effects structure as the walk count models (group × week interaction, age, biological sex, years of education, and scaled time since injury). For walking duration, one influential outlier was removed prior to analysis based on Cook’s distance of 0.9, and a random intercept only (1 | id) was specified due to convergence issues with a random slope. The steps model retained both a random intercept and random slope for week (1 + week | id). Fixed effects are reported as unstandardized coefficients (b) with standard errors, t-statistics, degrees of freedom estimated via the Satterthwaite approximation, and 95% confidence intervals. Estimated marginal means at weeks 1, 6, and 12 are reported in the supplementary tables 2 and 3.

Modifiers of walking:

To examine potential modifiers of daily walking behavior, we specified three multilevel logistic regression models corresponding to the Opportunity, Capability, and Motivation domains of the COM-B framework. The binary outcome was whether a participant completed a walk of at least 27 minutes on a given day. Because participants contributed multiple daily observations across up to 12 weeks, a random intercept for participant was included in all models. Prior to fitting the final models, a demographics-only model was built containing all candidate demographic covariates: age, sex, years of education, days since injury and randomization (intervention vs. control). Only demographic variables reaching statistical significance (p < .05) in this preliminary model were carried forward as covariates. Three separate models were then estimated, each combining the significant demographic covariates identified above with predictors specific to one theoretical domain which included: Opportunity: daily weather conditions (temperature, precipitation, percentage cloud cover), neighborhood walkability, percentage local accessible for recreational greenspace, urban classification, and season. Capability: self-reported baseline physical health, mental health, and perceived cognitive function. Motivation: baseline intrinsic motivation, amotivation, and exercise self-efficacy. Continuous predictors measured at the daily level (temperature, precipitation, cloud cover) were standardized within-person prior to analysis, such that estimated effects reflect within-individual variation in exposure rather than between-person differences in average conditions. Person-level continuous predictors were standardized across participants. All models were estimated using the lme4 package in R. Fixed effects are reported as odds ratios (ORs) with 95% Wald confidence intervals.

Psychosocial and cognitive outcomes:

To test if changes in cognitive function and perceived cognitive, mental and physical health occurred over the 12-week intervention, we classified participants based on if their total walking duration exceeded 150 minutes in at least 6 of the 12 intervention weeks, regardless of randomization assignment. This was chosen to mimic World Health Organization recommendations for physical activity and as a measure on consistency across intervention weeks, rather than testing the effect of intervention assignment given the heterogeneity in walking adherence within the intervention group. We fit separate models for each outcome (PROMIS mental health, PROMIS physical health, PROMIS cognitive abilities, verbal fluency, verbal learning and memory). We used separate (for each perceived health outcome) LMMs fit with a random intercept, main fixed effects of group (<150, >151 minutes), time (Pre, Post) and a group by time interaction. Age, sex, years of education, days since injury, and randomization group (intervention/control) were included as covariates. Similar model assumptions, outlier detection and removal were used as previously described. As these analyses were secondary exploratory to the main aims, multiple comparison corrections were not performed, planned comparisons were made on non-significant interaction terms and only 95% confidence intervals are presented.

Results

Feasibility

Between June 2024 and September 2025, 341 inquiries were received, 118 individuals passed initial screening and were prescreened for eligibility. Fifty-six participants were consented and randomized into either the intervention (N=28) or control arm (N=28). Eight participants (15.8%) dropped out of the study, five in the intervention group and three from the control group. Five of these dropouts were for motivation reason, two for musculoskeletal injuries not due to study procedures and one due to issues associated with a major hurricane. See the consort diagram in Figure 2 for details.

Figure 2.

Consort diagram of participant enrollment

No serious adverse events related to study procedures were recorded throughout the study. Seven adverse events not related to study procedures were recorded. Four reported musculoskeletal injuries sustained in the home, two reported head injuries, also sustained in the home and one reported a major illness.

Groups were well-matched at baseline across all demographic and injury-related characteristics, with no significant differences observed between intervention, control, and dropout participants (see Table 1).

Table 1: Baseline Characteristics by Group.

	level	Overall	Intervention	Control	Dropout	p
n		57	23	25	9
Age (years) (mean (SD))		52.8 (8.9)	53.0 (9.3)	52.7 (8.5)	52.6 (10.1)	0.989
Biological Sex (%)	Female	48 (85.7)	18 (78.3)	23 (92.0)	7 (87.5)	0.392
	Male	8 (14.3)	5 (21.7)	2 (8.0)	1 (12.5)
Gender Identity (%)	Man	8 (14.3)	5 (21.7)	2 (8.0)	1 (12.5)	0.392
	Woman	48 (85.7)	18 (78.3)	23 (92.0)	7 (87.5)
Race (%)	Asian	1 (1.8)	1 (4.3)	0 (0.0)	0 (0.0)	0.721
	Black or African American	3 (5.4)	2(8.7)	1 (4.0)	0 (0.0)
	More Than One Race	1 (1.8)	1 (4.3)	0 (0.0)	0 (0.0)
	Unknown	1 (1.8)	0(0.0)	1 (4.0)	0 (0.0)
	White	50 (89.3)	19 (82.6)	23 (92.0)	8 (100.0)
Ethnicity (%)	Hispanic or Latino	2 (3.6)	0 (0.0)	1 (4.0)	1 (12.5)	0.257
	NOT Hispanic or Latino	54 (96.4)	23 (100.0)	24 (96.0)	7 (87.5)
Time Since Injury (days) (mean (SD))		1246.5 (976.2)	1219.7 (1099.2)	1241.7 (896.5)	1351.0 (943.7)	0.954
Loss of Consciousness (%)	No	35 (62.5)	14 (60.9)	16 (64.0)	5 (62.5)	0.975
	Yes	21 (37.5)	9 (39.1)	9 (36.0)	3 (37.5)
Memory Gap (%)	No	31 (55.4)	12 (52.2)	13 (52.0)	6 (75.0)	0.483
	Yes	25 (44.6)	11 (47.8)	12 (48.0)	2 (25.0)

Compliance

98.1% of weekly phone calls were made in the control group and 98.4% in the intervention group (supplementary figure 1C), showing high compliance with the weekly phone calls. No significant differences in phone call length between groups was found (U = 275.00, p = 0.80, CI_95% [−0.36, 0.28] supplementary figure 1A). See supplementary table 4 for expanded details.

Table 2 presents valid wear time analysis for Fitbit wear compliance. Participants in the intervention group had higher % of days with data 89.7% compared to 87.5% in the control group. Participants in the control group on average wore their Fitbit device for longer on any given day and had more valid days with >10 hours of wear time. However, participants in both groups wore their device during the day (7.00am to 11.00pm) for similar amount of time ~70% of total wear time. Together, suggesting that participants in the intervention group likely wore their device around their walks and removed it mor readily after engaging in activity. See supplementary figure 1B for a comparison of mean daily wear time between group.

Table 2. Activity Tracker Wear Time Compliance.

group	N	Days with Data Mean (SD)	% of Expected Days	Mean Daily Wear Time (hours)	Valid Days ≥10hr Mean% (SD)	Daytime Wear % Mean (SD)	Nighttime Wear % Mean (SD)
Control	25	75.3 (22.4)	87.5%	18.9 (5.1)	81.7 (25.5)	67.4 (8.3)	32.6 (8.3)
Intervention	23	75.5 (11.7)	89.7%	16.4 (6.4)	68.6 (30.1)	67.1 (8.9)	32.9 (8.9)
Overall	48	75.4 (17.9)	88.6%	17.7 (5.8)	75.4 (28.3)	67.3 (8.5)	32.7 (8.5)

Efficacy

Walk counts

A numerical majority of completed walks were >27 minutes in length (Supplementary Figure 1D). Less total walks were completed on a weekend and most walks happened between 2pm and 6pm in the evenings (Supplementary Figure 1E and 1F). Figure 3A displays the estimated trajectories of weekly walking activity across 12 weeks (raw walk counts per group is illustrated in Figure 3B). The intervention showed an immediate effect, where at week 1, those in the intervention group completed more walks than controls (adjusted incidence rate ratio (aIRR) = 5.33, 95% CI: 2.27–12.5, z = 3.85). The control group (orange line) showed a declining trend, with predicted walks decreasing from 0.71 at week 1 to 0.31 at week 12 (aIRR per week = 0.927, 95% CI: 0.862–0.996, z = −2.07), representing a 7.3% weekly decrease. In contrast, the intervention group maintained relatively stable walk counts across the study period. For total walks (teal line), intervention participants’ estimated walks remained between 4.1 (95%CI; 2.27,7.4) at week 1 and 4.4 (95%CI; 2.2, 9.0) at week 12. Planned walks (Figure 1, red line) accounted for (60%) of total intervention walks. The difference between intervention and control walks grew over the study period (interaction aIRR = 1.09 per week, 95% CI: 1.01–1.17, z = 2.18, 72% power achieved using Wald-Based posthoc approximation), where intervention participants completed 5.8x more walks at week 1 and 14.5x more walks at week 12 (estimated marginal means are found in Table 3). A statistical comparison of planned walks only is found in supplementary material 2.

Figure 3.

Multipaneled plot of intervention efficacy. Panel A illustrates model predicted walks over time for the control group, planned walks only and total walks (planned + unplanned) in the intervention group. Significance asterisks show significant marginal effects between control group and intervention (significant for both total and planned only comparisons) at reference weeks one, six and twelve. Panel B illustrates individual (dots) raw walk counts for the control (teal bars) and intervention participants. Intervention participants shown using stacked bar plot with unplanned walks (orange) stacked on panned walks (rust). Panel C and D illustrate mean plus standard error (shaded) trajectories for duration and step counts within walks per week.

Figure 1.

Overall study design. Virtual or in-person consent was followed by randomization to one of two arms (planning, reminders and micro-incentives or health tips control). Intervention delivery for both groups consisted of weekly phone calls and then reminder text messages for the intervention group.

Table 3. Estimated Marginal Means of Weekly Walks.

Estimated marginal means of walking count for reference weeks 1,6 and 12. Values represent the model-predicted weekly walk count for each group. Ratios calculated by dividing marginal means for the intervention group by the control group with a value >1.0 representing higher walk counts in the intervention group. P values corrected for multiple comparisons using FDR.

week	Control	Intervention	Ratio	Ratio_Label
1	0.74	4.08	5.50	5.5× more
6	0.49	4.30	8.82	8.8× more
12	0.29	4.58	15.52	15.5× more

Duration

Figure 3C illustrates total duration of walks per week both groups. The intervention group walked on average 63.14 minutes more than the control group in the first intervention week (b = 62.14, SE = 30.56, t(62.04) = 2.03, 95% CI [1.05, 123.23]). The control group declined in walking duration over time (b = −3.93, SE = 1.44, t(460.26) = −2.74, 95% CI [−6.75, −1.11]). For every additional week, the intervention group walked an average of 7.54 minutes more than the control group during walk activities (b = 7.54, SE = 2.17, t(462.58) = 3.47, 95% CI [3.27, 11.80]). Estimated marginal means (supplementary table 2) show that during the final week, intervention participants (187.70 minutes) walked for an average 5.3x longer than control participants (35.12 minutes).

Steps

Figure 3D illustrates total number of steps taken during walks each week for both groups. The intervention group on average took 4,779.20 more steps than the control group during walks at week one (b = 4,779.20, SE = 2,359.71, t(52.64) = 2.03, 95% CI [45.50, 9,513.00]). A negative effect of week was observed in the control group, indicating a decline in step count over time (b = −187.46, SE = 90.03, t(454.06) = −2.08, 95% CI [−364.00, −10.50]). For every additional week, the intervention group accumulated an average of 394 more steps during walks compared to the control group (b = 394.00, SE = 136.29, t(455.47) = 2.89, 95% CI [126.00, 662.00]). Estimated marginal means (supplementary table 3) shows that at week 12, intervention participants took 3.1x more step during walks than the control group.

Modifiers

Figure 4 illustrates the percentage of days with completed walks of greater than 27 minutes by participant. No opportunity variable significantly increased or decreased the probability of walking on any given day. Although daily temperature, daily precipitation and walkability index, rural areas, spring and winter (compared to fall) were associated with lower probability of walking and more cloud cover, more access to recreational greenspace and summer (vs fall) were associated with increased odds of walking (supplementary table 5). Higher baseline perceived cognitive abilities was associated with lower probability of walking (OR = 0.88, 95% CI [0.81, 0.96]), SE = 0.037, z = −3.03). Additionally, higher intrinsic motivation at baseline was associated with increased odds of walking at least 27 minutes on any given day (OR =2.55, 95% CI [1.54, 4.24]), SE = 0.661, z = 3.62) (Figure 4C).

Figure 4.

Multipaneled plot of modifiers of walking. Panel A illustrates the proportion of days that a walk of >27 minutes was completed broken down by participants and intervention assignment. Panel B provides Odds ratios and 95% confidence intervals for all predictors in each of the three hierarchical logistic regression models, predicting the probability of walking on any given day. Each model was categorized based on the COM-B framework. Panels C and D show significant effects of baseline intrinsic motivation and perceived cognitive abilities on increasing and decreasing, respectively, the probability of walking, broken down by intervention assignment.

Outcome measures

Promis self-report scales:

Figure 5 shows the proportion of individuals who engaged in at least 150 minutes of walking in at least 6 of the 12 intervention weeks, stacked by intervention assignment. A detailed breakdown of meeting this threshold by week is found in Supplementary material 9. Participants who met the 150-minute walking threshold showed greater improvement in PROMIS Cognitive Function from pre to post compared to those who did not meet the threshold (b = 4.21, SE = 2.07, t(44.70) = 2.03, 95% CI [0.15, 8.27]). Those who met this threshold increased the perceived cognitive function by ~3.9 T-score points (b = 3.85, SE = 1.49, t(44.69) = 2.58, 95% CI [0.85, 6.86]) whereas those who did not meet this criteria declined slightly in their perceived cognitive function (b = −0.36, SE = 1.44, t(44.04) = −0.25, 95% CI [−3.25, 2.54]). No notable difference in change over time were seen for PROMIS mental or physical health (supplementary table 6).

Figure 5.

Changes in perceived cognitive abilities. Left illustrates the proportion of participants who walked or did not walk for at least 150 minutes per week for at least 6 of the 12 weeks, broken down by intervention assignment. A detailed illustration of the number of weeks that each participant completed >150 minutes of walking is found in supplementary figure 4. Right: pre-post line plot showing the significant change in perceived cognitive abilities in the group who met the 150 minute threshold versus those who did not meet this threshold.

Cognitive function:

No changes between groups were seen for any of the cognitive tests pre-to-post intervention when dichotomizing the participants into those who averaged more than 150 minutes of weekly walking.

Discussion

In this twelve-week, two-arm pilot randomized control trial, we show the feasibility and preliminary efficacy of a remotely delivered planning, reminders and micro-incentives intervention to increase walking in community-dwelling adults with TBI. We successfully recruited participants across the continental USA, with just a 15% attrition rate and high compliance to the weekly intervention and activity tracking. Preliminary efficacy suggests that a multicomponent intervention, compared to health advice only, can substantially increase walking over twelve weeks. An analysis of possible modifiers of walking revealed that baseline intrinsic motivation increased the probability of walking, whereas higher perceived cognitive abilities pre-intervention decreased the odds of walking. Consistently engaging in at least 150 minutes of walking per week over the course of the intervention was associated with increased perceived cognitive abilities.

Individuals with TBI face a lack of meaningful care and programs to maintain and increase healthy lifestyles after re-integration into their communities. Engaging in a bout of PA in absence of habit or external cues requires decision making (should I walk? Where should I walk? When? For how long?), moment-to-moment awareness of sedentary behaviors and an effort-reward trade-off (if I become active, what will I gain?). Fatigue, reward deficits and poor situational awareness can make sustained engagement in a physically active lifestyle particularly difficult after TBI^17,46. Here we show that a multicomponent scaffolding intervention consisting of weekly planning, daily reminders and monetary micro-incentives can together target barriers to physical activity by reducing the burden of momentary decision making and providing a proximal reward to competing rewarding sedentary behaviors. This multicomponent intervention, and small variations to those components (like additional points for returning after a missed session) has been shown to increase gym attendance in a very large multi-site study in healthy adult members of a large gym chain⁴⁷. The current study shows that this intervention is also on average, potentially effective at increasing walking, time spent walking and steps during walks, in adults with TBI who face specific injury-related barriers to PA⁴⁸. Habituation to behavioral interventions can lead to steady declines in the intervention efficacy over time⁴⁹, however we saw that the effect of the intervention continued to increase compared to the control group over time. Given the 12-week timeframe, it could be that the intervention habituation period had not peaked and had the intervention continued beyond 12 weeks, a steady decline in its effect might have been seen. Or, given little data exists on the effect of behavioral interventions to increase physical activity in TBI, it is possible that the scaffolding effect of this intervention could have longer lasting effects than it does in a generally healthy population.

Beyond the overall pattern of group differences, a closer examination of the composition of walking activity within the intervention group reveals interesting insights relevant for future intervention development. Specifically, 60% of total walks completed by intervention participants were planned walks (defined as the first walk of >27 minutes on a planned walk day or the first two walks of >15 minutes each). The other 40% of the walks were spontaneous and unplanned. We found that around half of the participants completed these unplanned walks on the same day as their planned walks (i.e., they did extra walks) and the other half completed extra unplanned walks on unplanned days (see supplementary material 7). On one hand, this could reflect a lack of adherence to the intervention components or efficacy of the intervention components (e.g., not completing planned walks but spontaneously walking on unplanned day). Another possible explanation for this though can be found in our analysis of modifiers of walking. Higher baseline intrinsic motivation increased the probability of engaging in a walking bout at least 27 minutes on any given day. Given 40% if walks were unplanned, it could suggest that those with higher intrinsic motivation are less likely to require the scaffolding effects of a multicomponent intervention. However, given the associations between intrinsic motivation and probability of walking was minimal in the control group (Figure 4C), it is more likely that walking bouts themselves, aided by the multicomponent intervention, boosted the effect of intrinsic motivation on walking behaviors as a positive feedback loop, where more walking led to more motivation to walk, reflected in the sustained increase in the intervention effect over time (Figure 3A). This is particularly important as maintenance of PA long-term is likely key to long-term health in TBI survivors. However, whether this intervention can be used long-term or promote sustained engagement is yet unclear.

The relationships of perceived cognitive abilities in the present study are complex and worth further study. We found that on one hand, consistently meeting at least 150 minutes of walking per week led to improvements in perceived cognitive abilities, but on the other, higher baseline perceived cognitive abilities decreased the probability of walking on any given day. Higher self-reported PA has been associated with better scores on the same perceived cognitive abilities questionnaire as in the current study¹¹ and interventional studies of aerobic exercise in TBI populations have reported consistent improvements in executive function. Therefore, our study contributes dose-related evidence to the growing body of literature on the role PA can play in cognitive outcomes in individuals living with TBI. However, individuals perceived health is a complex system where persistent reductions in self-awareness can often make participants unaware of their cognitive disabilities⁴⁶. In non-injured individuals, higher executive function has been associated with greater adherence to PA¹⁶, in contrast to the current study. Several possible scenarios may explain this discrepancy. For one, those with low self-awareness may overestimate their perceived cognitive functioning but given an actual cognitive deficit, may find consistent engagement in PA particularly difficult⁵⁰. Given PA itself can improve actual cognitive function in TBI, those who could engage (via scaffolding effects of the intervention itself) in consistent walking could have both re-calibrated their own perceptions of their cognitive function at the same time as seeing real gains in certain cognitive domains (not measured in this study), reflected as genuine improvements in perceived cognitive abilities. Alternatively, a self-efficacy paradox may lead those with higher perceptions of their own cognitive abilities to devalue the participation in PA as a rehabilitative or healthy habit tool, leading to less engagement across the intervention. This may be particularly true for PA, where greater perceived benefits of engagement are related to higher levels of engagement⁵¹. This is strengthened by the fact that the majority of participants who did engage consistently in at least 150 minutes of walking each week were receiving the active intervention (Figure 5).

Despite walking significantly more than the control group, intervention participants only completed on average 2.6 planned walks per week, just over half of the intended five planned walk. Optimizing the intervention components could lead to improvements in this effect. For example, participants were only asked to plan which days they would walk. A future intervention could reduce decision fatigue further by having participants plan the where (e.g., around the park), the when (e.g., before breakfast) and the how (e.g., by going before breakfast I have time to walk before going to work). Further optimization of the reminder messages could impact the interruption of automatic sedentary behaviors. In this study, a generalized reminder message was sent in the morning at the same time on every planned day. Finding the optimal time of day for a given individual to send a prompt that is then subsequently responded to has been shown to increase step count in another study⁵². Although non-significant, higher precipitation and higher temperature were associated with lower probability of walking (supplementary figure 2). Contextual information is important when prompting individuals to engage in a behavior⁵² and so development of just-in-time adaptive interventions in this population may lead to improved efficacy through personalization of this intervention.

All primary efficacy outcomes were derived from device-based Fitbit data, providing objective, continuous measurement of walking behavior and reducing reliance on self-report. Interestingly, despite intervention participants wearing their device for fewer hours per day and having fewer valid wear days than controls, they still demonstrated substantially greater walking activity, suggesting the observed effects are likely conservative estimates of the true behavioral difference between groups. From a practical and clinical standpoint, the intervention was delivered entirely remotely to participants distributed across the continental United States, without requiring in-person attendance. This is particularly relevant given that TBI survivors face numerous barriers to physical activity participation, including fatigue, transportation difficulties, limited access to adapted community programs, and lack of professional support^48,53. Remote delivery circumvents some of these barriers and, combined with the low cost of the micro-incentives, suggests this intervention could offer a scalable and accessible program once individuals have left in-person rehabilitation.

Several limitations to this study should be considered when interpreting the results. First, this was a pilot study and therefore not powered to test overall efficacy of the intervention; effect sizes are reported with 95% confidence intervals and standard errors to inform the planning of a larger trial and the study achieved only 72% power to detect the achieved effect size of the difference in walking bouts over time between groups (primary efficacy analysis). Second, participants varied considerably in time since injury, though this heterogeneity likely reflects the true diversity of the community-dwelling TBI population rather than a methodological limitation. Third, as a 12-week pilot trial, the long-term maintenance of the observed walking increases remains to be established, and future research should examine whether effects are sustained beyond the active intervention period. Fourth, the sample was not balanced by sex, with a predominance of female participants, and lacked racial and ethnic diversity, which limits the generalizability of the findings to more diverse TBI populations. Fifth, while Fitbit provided objective and continuous measurement of walking behavior, we only captured activities explicitly logged as walks, and incidental ambulatory activity or other activities (such as cycling or rowing for example) not recorded as a discrete walk event may have been missed. Finally, the intervention protocol has several modifiable features that may have constrained its efficacy, including the absence of detailed implementation intention planning, the use of generalized and fixed-time reminder messages, and uniform incentive structures applied across all participants regardless of individual reward sensitivity.

Conclusions

Physical inactivity after TBI is common, consequential, and largely unaddressed. This trial demonstrates that a remotely delivered, multicomponent behavior change intervention combining planning, reminders, and monetary micro-incentives is feasible, well-tolerated, and associated with sustained increases in walking activity over 12 weeks in community dwelling adults with TBI. A larger, fully powered efficacy trial is now warranted.