Timing in smartphone-guided home stroke rehabilitation with functional electrical stimulation: feasibility of distributed supervision and retrospective comparison with a front-loaded model

Abstract

[Purpose] This study reports: (Aim 1) a feasibility study of a distributed supervision model (DSG) within a smartphone-guided home stroke rehabilitation program incorporating functional electrical stimulation (FES); and (Aim 2) a retrospective cohort comparison of DSG outcomes against a previously published front-loaded supervision group (FSG). [Participants and Methods] Thirty-one adults with chronic hemiparetic stroke (onset >6 months) completed a 6-week multicomponent home exercise program (3 sessions/week, 1 hour/session) incorporating gait, strength, and balance training with FES delivered via smartphone app. FSG (n=12) received 2 weeks of supervised onboarding followed by 4 weeks of independent home training. DSG (n=19) received one supervised and two independent sessions weekly. Pre- and post-assessments included the 10-Meter Walk Test, Mini-BESTest, Berg Balance Scale, Timed Up and Go, and 30-Second Sit-to-Stand Test. [Results] No adverse events were reported. Both groups demonstrated high adherence (DSG: 90.3 ± 10.2%; FSG: 87.3 ± 8.7%) and significant within-group improvements across all outcomes, exceeding minimal clinically important differences. Between-group comparisons showed significantly greater improvements in the DSG on the 30STS (Welch’s t, p<0.001, d=1.57, achieved power=0.98) and TUG (p=0.020, d=0.74, achieved power=0.48); contrasts for BBS, Mini-BESTest, and 10MWT were non-significant with low achieved power (0.05–0.21) and are reported as inconclusive. All findings should be interpreted as preliminary given the retrospective, non-randomized design. [Conclusion] Both supervision models demonstrated high feasibility, safety, and clinically meaningful functional gains. Supervision timing may influence the magnitude of functional gains and warrants investigation in future randomized trials.

INTRODUCTION

Stroke remains a leading cause of long-term disability worldwide, with over 12 million new cases reported annually¹⁾. Despite early rehabilitation efforts, >50% of stroke survivors continue to experience persistent gait and balance dysfunction due to residual motor system impairments^{2, 3)}. These deficits are associated with limitations such as reduced functional mobility⁴⁾, decreased physical activity levels⁵⁾, and restrictions in community and social participation⁶⁾, all of which negatively impact quality of life⁷⁾. Physical therapy (PT) remains a cornerstone of post-stroke rehabilitation, playing a critical role in promoting neuroplasticity, restoring motor function, and improving independence in daily activities^{8, 9)}. Core components of PT programs typically include task-specific gait training, balance exercises, strength conditioning, and mobility retraining, all of which have been associated with measurable improvements in postural control, walking speed, and functional mobility^{2, 3)}. Despite its efficacy, the magnitude of functional gains is often modest, and there is considerable variability in outcomes based on intensity and individual response to treatment¹⁰⁾.

Recent meta-analyses and clinical trials suggest the importance of increasing therapy dosage to optimize outcomes. Studies demonstrated a positive dose-response relationship, wherein higher intensities and greater volumes of therapy are linked to better improvements in motor performance and activities of daily living¹¹⁾. Moreover, while the American Heart Association guidelines emphasize intensive and repetitive task practice to enhance the magnitude of functional gains⁹⁾, most individuals in the chronic phase do not sustain adequate therapy volumes after outpatient discharge, making home-based and technology-supported models particularly relevant for this population¹²⁾. Among neuromodulatory techniques, functional electrical stimulation (FES) has demonstrated efficacy in enhancing motor relearning and reducing motor impairment when used adjunctively with conventional therapy^13,14,15). However, most individuals in the chronic phase of stroke recovery do not receive high training doses after outpatient discharge^{16, 17)}. Barriers such as insurance restrictions, travel limitations, insufficient prescription, or limited access to regional care contribute to this gap^{9, 18)}. Even among those receiving regular PT, carryover into functional settings could be suboptimal due to limited adherence and lack of task-specific practice in real-life environments¹⁹⁾. Home-based training has emerged as a strategy to bridge this gap and enhance training volume. Systematic reviews indicate that home-based interventions provide comparable short-term outcomes to clinic-based programs in upper and lower limb function^20,21,22,23). Moreover, participants and caregivers often prefer home-based care due to its convenience, lower costs, and reduced caregiver burden^{24, 25)}. Nonetheless, existing home-based protocols predominantly target isolated functions rather than complex, functional activities such as gait or dynamic balance²¹⁾. Additionally, most home-based programs require significant therapist or caregiver involvement via telerehabilitation or home visits²⁶⁾, limiting scalability. Incorporating neuromodulatory tools such as FES could enhance the efficacy of home programs²⁷⁾. FES has been shown to facilitate motor relearning by normalizing muscle activation patterns²⁸⁾. Several meta-analyses have confirmed their benefits in improving gait speed, endurance, functional mobility, and balance when used in conjunction with clinical PT²⁹⁾. However, most FES interventions are clinic-based, rely on wired systems, and have not been widely implemented in home settings, partly due to equipment complexity, cost, and limited reimbursement³⁰⁾.

Mobile health (mHealth) technologies represent a rapidly growing frontier in stroke rehabilitation, offering scalable, cost-effective, and user-centered solutions to extend care beyond traditional clinical settings. With the increasing global penetration of smartphones and mobile internet, mHealth interventions have become more accessible, particularly for stroke survivors who face barriers to regular in-person therapy³¹⁾. Smartphone-based applications have demonstrated high levels of feasibility, usability, and participant engagement across diverse stroke populations³²⁾. These applications can deliver structured rehabilitation protocols, offer real-time reminders, and provide participants with educational content and performance metrics, contributing to enhanced self-management and autonomy. Additionally, mHealth platforms could foster greater adherence to prescribed rehabilitation regimens by incorporating behavioral prompts, gamification elements, and personalized goal-setting features³³⁾. When combined with neuromodulatory tools such as FES, smartphone-based platforms offer even greater potential by enabling real-time biofeedback³⁴⁾, remote monitoring³⁵⁾, and integrated motor training³⁶⁾. These hybrid platforms not only increase engagement but also support individualized progression³⁷⁾ and allow therapists to supervise remotely³⁸⁾, making them particularly advantageous for stroke survivors in underserved or rural regions.

Despite these perceived benefits, the optimal structure and timing of therapist supervision within mHealth-FES home programs to maximize outcomes, adherence, and resource efficiency remains unclear. Our recent study showed the feasibility and efficacy of combining a front-loaded home-based exercise program delivered via smart technology and integrated with FES in individuals with chronic stroke³⁹⁾. Building on this prior work, the present study addresses two aims. Aim 1 assessed the feasibility (safety and adherence) and preliminary functional effects of a distributed supervision model (DSG) delivered via the same smartphone-guided FES platform in a new cohort of individuals with chronic stroke in Chile. The rationale for testing distributed supervision stems from evidence that spaced therapist contact aligns with principles of self-determination theory supporting autonomy, competence, and relatedness and motor learning frameworks emphasizing distributed feedback for long-term skill retention^{40, 41)}. Aim 2 performed a retrospective cohort comparison of the DSG outcomes against the previously published FSG data³⁹⁾ to generate preliminary signals regarding whether supervision timing influences functional gains. We hypothesized that both groups would demonstrate feasibility and clinically meaningful functional gains. For Aim 2, given the non-randomized and retrospective nature of the cohort comparison, between-group results are interpreted as preliminary and hypothesis-generating to inform future randomized trial design, rather than to establish causal superiority of one supervision model over the other.

PARTICIPANTS AND METHODS

We included 31 adults (age 60.4 ± 11.3 years, 14 women) with chronic hemiparetic stroke (onset >6 months), and a median of 46 months after stroke. Participants were recruited from two locations: Santiago, Metropolitan Region, Chile; and Chicago, Illinois, USA, via flyers, advertisements, public screenings, and direct physician referral. All interested participants were first telephone screened and included if their age was between 18 and 90 years, had experienced hemiparesis due to a stroke with an onset of more than 6 months, used a smartphone regularly, and were able to understand and communicate in English or Spanish. Participants were excluded if they had the presence of any neurological condition other than stroke, cardiopulmonary, musculoskeletal, or systemic diagnosis, recent major surgery or hospitalization within the last 6 months, deep venous thrombosis, active cancer, peripheral nerve injury, or neuropathy in the affected limb. Demographic characteristics are included in Table 1.

Table 1. Demographic characteristics of research participants with their respective means and standard deviations.

Variables		Front-loaded supervision (n=12)	Distributed supervision (n=19)
Age, years (means and SD)		67.6 ± 5.6	55.9 ± 11.8
Sex (% male)		6 (50)	11 (57.9)
Stroke type, n (%)
	Ischemic	5 (41.7)	12 (63.2)
	Hemorrhagic	7 (58.3)	7 (36.8)
Side of lesion, n (%) †
	Dominant hemisphere	7 (58.3)	8 (42.1)
	Non-dominant hemisphere	5 (41.7)	11 (57.9)
Chronicity, years (means and SD)		9.2 ± 3.1	6.5 ± 7.3
Mini-Mental State Examination (/30)		26.2 ± 1.2	26.1 ± 1.1

^†Hemispheric dominance was assigned from participant-reported pre-stroke handedness; the lesioned hemisphere was classified as dominant if it was contralateral to the participant’s pre-stroke dominant hand. SD: standard deviation; %: percentage.

Participants who passed telephone screening were scheduled for an in-person screening session within one week. Inclusion criteria at this stage required the ability to walk independently with or without an assistive device for 300 feet. Participants were excluded if their body weight exceeded 250 lbs, if they had cognitive impairments (Mini Mental State Exam score <25), speech impairments (aphasia score of >71/100 on Mississippi Aphasia Screening Test), poor bone density (T score <−2 on heel ultrasound), or loss of protective sensations (on 5.07/10 g monofilament test). We also tested for possible cardiovascular risk and excluded participants with a heart rate >85% of age-predicted maximal heart rate post Six-minute walk test completion, resting systolic blood pressure >165 or diastolic blood pressure >110 mmHg, shortness of breath, or uncontrolled pain (3/10). Participants were excluded if they had Botox treatment in the past 3 months or a skin condition not tolerant with FES therapy, history of uncontrolled/controlled epilepsy or other seizure disorders, spasticity (Ashworth scale >2), or uncontrolled hypertension or diabetes. The study was approved by the Institutional Review Board at the University of Illinois Chicago (#IRB:2022-0524) and registered as a clinical trial (NCT05849532). Written informed consent was obtained from all participants.

This study employed a two-aim design. Aim 1 was a prospective single-group feasibility study examining the safety, adherence, and preliminary functional effects of the DSG delivered through a smartphone-guided FES platform. Aim 2 was a retrospective cohort comparison in which DSG outcomes were compared against those of the previously published FSG³⁹⁾, which was conducted at a different site (Chicago, IL, USA) and timeframe. Because the groups were assembled at different sites and time points with partially overlapping inclusion criteria, no causal or randomization-based inferences are drawn from between-group comparisons. Rather, Aim 2 is intended to generate preliminary signals to inform the design of a future randomized controlled trial. The FSG (Chicago, IL, USA) and DSG (Santiago, Chile) cohorts differ not only in supervision schedule but also in country, healthcare-system context, post-discharge rehabilitation access, and recruitment timeframe. These factors cannot be disentangled from the supervision-model contrast in a non-randomized retrospective design, and the Aim 2 between-group results are therefore interpreted as preliminary and hypothesis-generating, even where they reach statistical significance.

Participants were conveniently assigned to one of two home-based supervision models. The Front-Loaded Supervision Group (FSG; n=12) received 2 weeks of supervised home onboarding (3 sessions per week) followed by 4 weeks of independent home-based training (3 sessions per week), totaling 18 sessions. The Distributed Supervision Group (DSG; n=19) received one supervised and two independent sessions per week for 6 weeks, also totaling 18 sessions. Both intervention arms received equal total therapy dosage. Because the groups were recruited at different sites and time points and were not randomized, between-group comparisons are exploratory and intended to generate hypotheses for a future randomized controlled trial, even where they reach statistical significance.

The intervention sessions for both groups were structured around four main components: dynamic balance, functional strength, gait, and stretching (Fig. 1). Each 1-hour training session incorporated approximately 10 minutes for each of the four main components. Dynamic balance exercises began with mini-lunges, mini-squats, or standing with feet together and progressed to forward or skater lunges. Functional strength exercises began from sit-to-stand or standing weight shifts and progressed to stool touch or alternate stepping on a stool. The gait component began from walking on level ground and progressed to walking on ramps, heel-toe walking, and backward walking or climbing stairs. Stretching exercises included trunk flexion, side-bends, active upper limb movements, and lower limb stretches, progressing in terms of dosage.

Fig. 1.

Scheme of intervention design. After recruitment, participants were allocated to the Front-Loaded Supervision Group (FSG) or the Distributed Supervision Group (DSG). The medical devices and the mobile app user interface are associated with the TRAINFES platform.

All clinical training sessions were supervised by a physical therapist and delivered via the TRAINFES ADVANCED^® system (Biomedical Devices SpA, Santiago, Chile) mobile application. Participants wore a gait belt while performing exercises. Participants first logged into the mobile app on their smartphone, completing safety precautions, contraindications, and an overview of exercise components. Participants received electrical stimulation to paretic lower limb muscles (quadriceps, gluteus medius, tibialis anterior, triceps surae) during each exercise. FES was delivered at 25–45 pulses per second. Amplitude was titrated at the first supervised session to produce a visible, comfortable contraction of the target muscle without pain or unwanted joint motion, within the device’s 20–100 mA tolerable range. Mean delivered intensity, extracted from the TRAINFES Cloud logs, is reported in the Results. Both sites used the same written protocol, the same TRAINFES exercise library, and the same progression rule: advance when rate of perceived exertion (RPE) was ≤13/20 for two consecutive sessions; hold or regress when RPE was ≥16/20 or when the in-app form checklist failed. The supervising therapist completed a fidelity checklist at every in-clinic session. The TRAINFES ADVANCED system (Biomedical Devices SpA, Santiago, Chile) consisted of an FES stimulator unit, an inertial measurement unit (IMU) sensing unit, and a mobile user interface communicating via Bluetooth 4.1. Three stimulation algorithms were used: Sequential, Sensor-triggered, and WalkFES (closed-loop gait facilitation). Participants were provided with the equipment package and received training regarding precautions, equipment setup, home safety procedures, and safe FES use at home. The RPE scale was used for self-assessment of exertion during home sessions. Therapists at both sites completed the same training prior to data collection and met regularly throughout the study to align on screening, FES titration, RPE-based progression, and use of the in-app session structure. The fixed time allocation embedded in the app (~10 minutes per component) constrained therapist discretion to over-emphasize any single component.

Safety and adverse events were monitored across both supervised and unsupervised sessions. A serious adverse event was defined as an anticipated or unanticipated physical or psychological occurrence resulting in a life-threatening injury or disease, inpatient hospitalization, outpatient admission, persistent or significant disability, or death. Adherence was measured as the percentage of completed sessions out of prescribed sessions and tracked automatically through the TRAINFES Cloud platform. Adherence was classified as low (<50%), moderate (51–80%), or high (80–100%)^{42, 43)}. Clinical outcomes were collected within one week before and after training (Table 2), including the Timed Up and Go test (TUG; MCID=3.4 s), Berg Balance Scale (BBS; MCID=4 points), Mini Balance Evaluation Systems Test (Mini-BESTest; MCID=4 points), Ten-Meter Walk Test (10MWT; MCID=0.16 m/s), and 30-Second Sit-to-Stand Test (30STS; MCID=2 repetitions).

Table 2. Outcome measures.

Stage in protocol	Outcome measures and variables assessed		MCID
Screening tests		1. Montreal Cognitive Assessment (MoCA)	-
		2. Mini-Mental State Examination (MMSE)	-
		3. Ability to walk independently with or without an assistive device for at least 300 ft (100 m).	-
Pre- and post-intervention assessment	Clinical and Biomechanical
		1. Berg Balance Scale (BBS)	4 points
		2. Mini-BESTest	3 points
		3. Gait Speed - 10 meters walking test (10MWT)	0.16 m/s
		4. Timed Up and Go (TUG) test	3.4 s
		5. 30 seconds Sit to Stand (30STS)	2 reps

Divided into three main stages of the Experimental Procedure: Screening, Pre, and Post assessments. MCID: minimal clinically important difference.

Because this study was a feasibility investigation, sample size was based on methodological guidance for early-phase rehabilitation research rather than powered hypothesis testing. Feasibility trials often include 12–30 participants per arm. We pre-specified a target enrollment of approximately 30 participants to ensure adequate representation across both supervision models while remaining practical for an early-stage digital intervention. Data were analyzed using a mixed statistical approach. Normality was first assessed using the Shapiro–Wilk test. For within-group comparisons, paired t-tests were used for normally distributed variables and Wilcoxon signed-rank tests for non-normally distributed data. Effect sizes were calculated using Cohen’s d for parametric analyses and rank-biserial correlation (r) for non-parametric comparisons. For Aim 2, between-group change-score contrasts were planned as secondary analyses. Because the study was not prospectively powered for between-group testing, we conducted a post-hoc achieved-power analysis (G*Power 3.1.9.7) on the observed effect for each outcome to allow the reader to weight each between-group result. The choice of parametric versus non-parametric test was determined per outcome by the Shapiro–Wilk normality of change scores in each group, following the rule specified above for within-group analyses: Welch’s t-test was used for TUG, BBS, 30STS, and 10MWT; Mann–Whitney U was used for the Mini-BESTest, where the DSG change scores were non-normal. Effect sizes were Cohen’s d for parametric contrasts and rank-biserial r for the non-parametric contrast; the rank-biserial r was converted to an equivalent d for the power calculation. Achieved power was computed using α=0.05, two-tailed, with the observed effect and the actual group sizes (n₁=12, n₂=19). We acknowledge that post-hoc power calculated from an observed effect has well-known interpretive limitations and is most informative when achieved power is high; we therefore report it transparently for every contrast rather than only for the contrasts that reach significance, and we frame null findings on low-powered contrasts as inconclusive rather than as evidence of no between-group difference. No adjustments for multiple comparisons were performed given the exploratory nature of the between-group analyses. A significance threshold of α≤ 0.05 was used for all hypothesis tests. All analyses were conducted using GraphPad Prism v8.0.2 and PAST 4.12b.

RESULTS

All participants completed the intervention without adverse events, demonstrating high adherence (FSG: 87.3 ± 8.7%; DSG: 90.3 ± 10.2%). Beyond binary session completion, mean session duration (FSG: 56.2 ± 7.4 min; DSG: 58.1 ± 6.8 min) and mean post-session RPE (FSG: 13.4 ± 1.5; DSG: 13.7 ± 1.4) were comparable between groups. Mean delivered FES intensity, pooled across the four stimulated muscles, was 38.4 ± 9.7 mA in the FSG and 40.1 ± 11.2 mA in the DSG. Both groups demonstrated significant pre-to-post improvements across all outcomes (Table 3).

Table 3. Pre- and post-intervention outcomes for the front-loaded (FSG) and distributed (DSG) supervision groups, with within-group change scores, effect sizes, and between-group inferential statistics.

		Front-loaded group (n=12)				Distributed group (n=19)				Between-group comparison
	outcome	Pre	Post	Δ (mean ± SD)	ES	Pre	Post	Δ (mean ± SD)	ES	Test	p	ES	Mean diff (95% CI)	Power
Balance
	BBS	40.92 ± 4.9	46.25 ± 4.4	5.27 ± 1.95	1.14**	38.16 ± 11.18	43.37 ± 10.79	5.21 ± 1.96	0.88**	Welch’s t	0.93 (ns)	d=0.03	−0.06 [−1.60, +1.48]	0.05
	Mini-BESTest	17.83 ± 4.3	21.40 ± 1.2	3.36 ± 1.21	0.85**	14.74 ± 5.69	19.84 ± 5.70	5.11 ± 3.35	0.88**	Mann–Whitney U	0.32 (ns)	r=0.22	+1.0 [−3.0, +9.0] §	0.21
Gait
	10MWT (m/s)	0.57 ± 0.15	0.75 ± 0.19	0.16 ± 0.06	1.01**	0.52 ± 0.23	0.72 ± 0.30	0.19 ± 0.11	0.70**	Welch’s t	0.30 (ns)	d=0.34	+0.03 [−0.03, +0.10]	0.14
Functional strength
	30STS (reps)	10.00 ± 2.7	12.58 ± 3.05	2.55 ± 0.93	0.89**	7.79 ± 3.4	12.84 ± 3.9	5.05 ± 1.87	1.37**	Welch’s t	< 0.001 **	d=1.57	+2.51 [+1.46, +3.56]	0.98
Functional mobility
	TUG (s)	28.02 ± 7.2	22.02 ± 7.4	−5.40 ± 1.49	0.82**	25.87 ± 12.46	17.17 ± 9.65	−8.71 ± 5.43	0.88**	Welch’s t	0.020 *	d=0.74	−3.31 [−6.05, −0.56]	0.48

Within-group p-values: ns: non-significant; *p<0.05; **p<0.005. Within-group effect sizes: Cohen’s d for paired t-tests; rank-biserial r for Wilcoxon signed-rank tests. Between-group comparison (shaded columns, added in this revision): Test choice was determined per outcome by Shapiro–Wilk normality of change scores; Welch’s t-test was used for normally distributed contrasts and Mann–Whitney U for the Mini-BESTest (DSG change scores non-normal, Shapiro p=0.005). Cohen’s d is reported for parametric contrasts; rank-biserial r for the non-parametric contrast. The mean difference for the Mini-BESTest (§) is the Hodges–Lehmann estimator of the median difference (DSG − FSG). Post-hoc achieved power was computed in G*Power 3.1.9.7 using α=0.05, two-tailed, the observed effect, and group sizes n₁=12, n₂=19; for the Mann–Whitney U contrast, rank-biserial r was converted to an equivalent d for the power calculation. BBS: Berg Balance Scale; Mini-BESTest: Mini Balance Evaluation Systems Test; 10MWT: 10-Meter Walk Test (m/s, meters per second); 30STS: 30-Second Sit-to-Stand Test; TUG: Timed Up and Go Test (s, seconds); Δ: change score (post − pre); ES: effect size; SD: standard deviation; CI: confidence interval.

Both groups significantly improved in TUG performance (p<0.001 for both). The FSG reduced their mean TUG time from 28.02 seconds (SD: 7.2) to 22.02 seconds (SD: 7.4; mean Δ=−5.40 ± 1.49 s, effect size=0.82), with 100% of participants exceeding the MCID. The DSG showed a larger median reduction from 22.0 seconds (IQR: 17.6–34.5) to 13.0 seconds (IQR: 10.2–24.0; effect size=0.88), with a median improvement of 9.0 seconds and 89.5% (17/19) meeting the MCID. The DSG showed a significantly greater reduction in TUG time than the FSG (FSG Δ=−5.40 ± 1.49 s; DSG Δ=−8.71 ± 5.43 s; mean difference −3.31 s, 95% CI [−6.05, −0.56]; Welch’s t-test p=0.020, Cohen’s d=0.74, post-hoc achieved power=0.48). Because achieved power for this contrast was modest, this finding should be regarded as a directional signal consistent with the 30STS result rather than as a definitive between-group effect on its own (Fig. 2).

Fig. 2.

Changes in functional mobility measured by the Timed Up and Go (TUG) test across supervision groups. Left panel: individual participant trajectories (light per-participant lines) with group-mean lines overlaid in bold for FSG (front-loaded supervision) and DSG (distributed supervision) before (PRE) and after (POST) the 6-week intervention. Both groups demonstrated statistically significant within-group improvements (FSG: p<0.001, Cohen’s d=0.82; DSG: p<0.001, rank-biserial r=0.88). Right panel: between-group comparison of mean change scores (ΔTUG), with greater negative values indicating larger improvement. The dashed line represents the minimal clinically important difference (MCID). The DSG showed significantly greater improvement than the FSG (Welch’s t-test p=0.020, Cohen’s d=0.74, post-hoc achieved power=0.48). The shaded region denotes clinically meaningful improvement beyond the MCID threshold.

Berg Balance Scale scores increased significantly in both groups (p<0.001). The FSG improved from 40.92 (SD: 4.9) to 46.25 (SD: 4.4; mean Δ = +5.27 ± 1.95, effect size = 1.14), with 66.7% (8/12) achieving clinically meaningful gains exceeding the MCID of 4 points. The DSG improved from a median of 40 (IQR: 33–46) to 47 (IQR: 38–50; median Δ=+5, effect size=0.9), with 63.2% (12/19) meeting the MCID. The between-group BBS change-score contrast was non-significant (FSG Δ=5.27 ± 1.95; DSG Δ=5.21 ± 1.96; mean difference −0.06, 95% CI [−1.60, +1.48]; Welch’s t-test p=0.93, Cohen’s d=0.03). The observed effect was negligible and post-hoc achieved power was correspondingly low (0.05); this null result is therefore inconclusive rather than evidence of no between-group difference (Fig. 3). Mini-BESTest scores rose significantly in both groups (p<0.001). The FSG increased from 17.83 (SD: 4.3) to 21.4 (SD: 1.2; mean Δ=+3.36 ± 1.21, effect size=0.85), with 50% (6/12) reaching the MCID of 4 points. The DSG improved from a median of 14 (IQR: 9–19) to 21 (IQR: 17–25; median Δ=+4, effect size=0.9), with 52.6% (10/19) achieving the MCID. The between-group Mini-BESTest change-score contrast was non-significant (FSG median Δ=3, IQR 2.5–4; DSG median Δ=4, IQR 2.5–7.5; Mann–Whitney U test p=0.32, rank-biserial r=0.22). Because the DSG change scores were non-normal (Shapiro–Wilk p=0.005), the non-parametric test was used in accordance with the rule specified in the Methods. Post-hoc achieved power for this contrast was 0.21, and the result is therefore reported as inconclusive rather than as evidence of no between-group difference (Fig. 4).

Fig. 3.

Improvements in static and dynamic balance assessed by the Berg Balance Scale (BBS). Left panel: individual participant trajectories (light per-participant lines) with group-mean lines overlaid in bold for FSG and DSG at baseline (PRE) and following the 6-week intervention (POST). Both groups exhibited statistically significant within-group improvements (FSG: p<0.001, Cohen’s d=1.14; DSG: p<0.001, rank-biserial r=0.88). Right panel: comparison of mean BBS change scores (ΔBBS) for each group. The dashed line indicates the MCID=4 points. The between-group contrast was non-significant (Welch’s t-test p=0.93, Cohen’s d=0.03, post-hoc achieved power=0.05); this null result is reported as inconclusive rather than as evidence of no between-group difference.

Fig. 4.

Changes in dynamic balance performance measured by the Mini-BESTest. Left panel: individual participant trajectories (light per-participant lines) with group-mean lines overlaid in bold for FSG and DSG at baseline (PRE) and after the 6-week intervention (POST). Both groups demonstrated statistically significant within-group improvements (FSG: p<0.001, Cohen’s d=0.85; DSG: p<0.001, rank-biserial r=0.88). Right panel: box plots of the change scores (ΔMini-BESTest) for each group. The dotted line represents the MCID=4 points. The between-group contrast was non-significant (Mann–Whitney U test p=0.32, rank-biserial r=0.22, post-hoc achieved power=0.21); this null result is reported as inconclusive rather than as evidence of no between-group difference.

Results from the 30-Second Sit-to-Stand Test revealed significant improvements in both groups (p<0.001). The FSG improved from 10.0 (SD: 2.7) to 12.58 (SD: 3.1; mean Δ=+2.55 ± 0.93, effect size=0.9), with 50% (6/12) exceeding the MCID of 2 repetitions. The DSG demonstrated greater gains, improving from 7.79 (SD: 3.4) to 12.84 (SD: 3.9; mean Δ=+5.1, effect size=1.4), with 95% (18/19) meeting the MCID. The DSG mean change exceeded the FSG mean change by approximately 2.5 repetitions (FSG Δ=2.55 ± 0.93; DSG Δ=5.05 ± 1.87; mean difference 2.51, 95% CI [1.46, 3.56]). A Welch’s t-test for this contrast yielded p<0.001 (Cohen’s d=1.57; post-hoc achieved power 0.98). Because this is a non-randomized, retrospective between-site comparison, the result is interpreted as a preliminary signal consistent with a between-group effect of the magnitude reported, not as evidence of a causal effect of supervision timing (Fig. 5). Gait speed (10MWT) improved significantly in both groups (p<0.001). The FSG increased from 0.57 m/s (SD: 0.15) to 0.75 m/s (SD: 0.19; mean Δ=+0.16, effect size=1.0), with 50% (6/12) achieving the MCID of 0.16 m/s. The DSG improved from 0.52 m/s (SD: 0.23) to 0.72 m/s (SD: 0.30; mean Δ=+0.19, effect size=0.7), with similar MCID attainment. The between-group 10MWT change-score contrast was non-significant (FSG Δ=0.16 ± 0.06 m/s; DSG Δ=0.19 ± 0.11 m/s; mean difference 0.03, 95% CI [−0.03, +0.10]; Welch’s t-test p=0.30, Cohen’s d=0.34). Post-hoc achieved power for this contrast was 0.14, and the result is therefore reported as inconclusive rather than as evidence of no between-group difference (Fig. 6).

Fig. 5.

Improvements in lower extremity functional strength assessed by the 30-Second Sit-to-Stand Test (30STS). Left panel: individual participant trajectories (light per-participant lines) with group-mean lines overlaid in bold for FSG and DSG before (PRE) and after (POST) the intervention. Both groups showed statistically significant within-group improvements (FSG: p<0.001, Cohen’s d=0.89; DSG: p<0.001, Cohen’s d=1.37). Right panel: box plot comparison of change scores (Δ30STS) between groups. The dotted line represents the MCID=2 repetitions. DSG showed significantly greater improvement than FSG (Welch’s t-test p<0.001, Cohen’s d=1.57, post-hoc achieved power=0.98).

Fig. 6.

Increases in gait speed measured by the 10-Meter Walk Test (10MWT). Left panel: individual participant trajectories (light per-participant lines) with group-mean lines overlaid in bold for FSG and DSG before (PRE) and after (POST) the 6-week intervention. Both groups showed statistically significant within-group improvements (FSG: p<0.001, Cohen’s d=1.01; DSG: p<0.001, Cohen’s d=0.70). Right panel: box plots of change in gait speed (Δ10MWT) for each group. The dotted line indicates the MCID=0.16 m/s. The between-group contrast was non-significant (Welch’s t-test p=0.30, Cohen’s d=0.34, post-hoc achieved power=0.14); this null result is reported as inconclusive rather than as evidence of no between-group difference.

DISCUSSION

This study evaluated: (Aim 1) the feasibility and preliminary functional effects of a distributed supervision model (DSG) within a smartphone-guided FES home rehabilitation program for chronic stroke; and (Aim 2) an inferential, retrospective cohort comparison of DSG outcomes against those of a previously published front-loaded supervision group (FSG,³⁹⁾). The DSG demonstrated high feasibility, safety, and clinically meaningful improvements across key outcomes including functional mobility, balance, gait speed, and lower extremity strength. Between-group comparisons identified two statistically significant findings favoring the DSG (30STS and TUG), with the 30STS contrast adequately powered (0.98) and the TUG contrast carrying modest achieved power (0.48); contrasts for BBS, Mini-BESTest, and 10MWT were non-significant with low achieved power (0.05–0.21) and are reported as inconclusive. Given the non-randomized and retrospective nature of the Aim 2 comparison, all between-group findings should be interpreted as preliminary and hypothesis-generating rather than as definitive evidence of differential effectiveness.

Both intervention models yielded high adherence (FSG: 87.3%; DSG: 90.3%) and no reported adverse events, reinforcing the feasibility, safety, and acceptability of integrating FES with mHealth platforms for unsupervised, home-based rehabilitation in individuals with chronic stroke. These adherence rates were notably higher than those reported in traditional home exercise programs for post-stroke populations, which often fall below 70% due to limited supervision, lack of feedback, and motivational decline^{44, 45)}. The absence of adverse events further supports previous findings that home-based FES systems, when embedded within structured digital platforms and supported by initial therapist training, are safe and well-tolerated in neurologic populations^{46, 47)}. The smartphone-based platform used in this study included interactive features such as video demonstrations, sensor-based stimulation control, automated reminders, and safety lockout mechanisms, which likely contributed to both high adherence and reduced risk. These behavioral and technical features are grounded in self-determination and behavioral reinforcement theories^{41, 48)}.

The DSG demonstrated greater improvements in functional mobility and strength, particularly in the 30-Second Sit-to-Stand and TUG tests. Notably, 95% of DSG participants exceeded the MCID for 30STS, compared to approximately 50% in the FSG, indicating a more robust functional response in lower extremity strength and transitional movements. These tasks are known to be particularly sensitive to improvements in sit-to-stand biomechanics, trunk control, and quadriceps activation, all of which are amenable to neuromuscular facilitation with sustained therapist involvement⁴⁹⁾. TUG improvements in the DSG were also more pronounced, reflecting meaningful gains in gait initiation, turning ability, and postural transitions. The significantly greater strength and mobility gains observed in the DSG may reflect a synergy of neurophysiological reinforcement, motor learning facilitation, and behavioral support associated with sustained therapist interaction; however, causal attribution is not possible from this retrospective comparison.

Between-group differences on the BBS, Mini-BESTest, and 10MWT did not reach statistical significance, although achieved post-hoc power for these three contrasts was low (0.05–0.21), so these null results are inconclusive rather than evidence of equivalent response. One possible explanation, if the absence of a between-group effect on these measures were to be replicated in an adequately powered trial, is that clinical functional scales may be less sensitive to subtle differences in motor learning or neural recovery⁵⁰⁾, especially when participants receive equal training dosage and content across groups. In contrast, the 30STS may better capture volitional control, neuromuscular recruitment, and strength gains elicited through active engagement. The DSG’s spaced supervision model could have enhanced opportunities for self-regulated learning, problem-solving, and deeper internalization of movement^51,52,53), aligning with principles of the challenge point framework and self-determination theory. However, this is only one of several candidate explanations, and several alternatives that this retrospective design cannot rule out are equally plausible. These include baseline imbalance in stroke etiology, lesion laterality, age (the DSG was on average 11.7 years younger), and chronicity; higher expectancy or motivation in a newer cohort enrolled into a novel intervention compared with the historical FSG cohort; site-specific differences in referral source, self-selection, and therapist style; and country-level differences in baseline physical activity and post-discharge rehabilitation access. The between-group difference is therefore not attributed to supervision timing alone, and these mechanistic possibilities are framed as hypotheses for a future randomized trial.

This study has several limitations. Group allocation was non-randomized, introducing potential selection bias. The small sample size limits generalizability and power to detect between-group differences. Adherence was monitored at the session level (binary completion, mean session duration, mean post-session RPE, and per-muscle FES amplitude setting). Granular within-session metrics—actual time-on-task per component, real repetition counts, active FES-on time per muscle, and peak-versus-tonic stimulation amplitudes—were not exportable from the TRAINFES Cloud platform in the version available during data collection. This limits the strength of any claim that supervision timing, rather than delivered dose, drives the between-group differences observed on the 30STS and TUG. Future randomized trials should treat these granular metrics as primary fidelity outcomes. Fidelity to the standardized progression rules was tracked via in-session checklists but was not independently audited by a blinded third reviewer. Baseline habitual physical activity was not quantified with a validated questionnaire such as the IPAQ or PASIPD; we cannot rule out that the DSG cohort was more physically active at baseline in ways that would have favored response to lower-extremity strength training such as the 30STS. Long-term follow-up was not conducted, leaving the durability of gains unknown. Three of the five between-group contrasts (BBS, Mini-BESTest, 10MWT) had achieved post-hoc power below 0.30 with the observed effect sizes; the corresponding non-significant results are therefore inconclusive rather than evidence of no between-group difference, and they motivate a future adequately powered randomized trial rather than a definitive conclusion that supervision timing has no effect on these outcomes. The TUG contrast, although reaching nominal significance, also had modest achieved power (0.48) and should be regarded as a directional signal consistent with the 30STS finding rather than as a definitive between-group result on its own. Therapist supervision was not blinded and may have varied in style or responsiveness between sites. Future trials should prioritize randomized, adequately powered designs with real-time adherence monitoring, long-term follow-up, and strategies to minimize therapist-related bias. Incorporating adaptive supervision, qualitative assessments, and economic analyses will further support scalability.

The Aim 2 comparison is materially confounded by site. The two cohorts were recruited in different countries, the United States and Chile, with different healthcare-system structures, reimbursement environments, post-discharge outpatient access patterns, cultural norms around therapist–patient interaction, and baseline digital literacy. Any between-group difference in functional gain, including the larger DSG improvement on the 30STS, must be read as jointly attributable to these geographic and socio-economic factors and to the supervision schedule. We do not claim supervision timing is the causal driver of the observed difference. We report the comparison to motivate a future randomized trial in which supervision timing can be isolated within a single, contemporaneous cohort. Further, the two cohorts were also not balanced on stroke etiology or lesion laterality. The FSG had a higher proportion of hemorrhagic strokes (58.3% vs. 36.8%) and dominant-hemisphere lesions (58.3% vs. 42.1%) than the DSG, although neither imbalance reached statistical significance. Because hemorrhagic stroke and dominant-hemisphere involvement carry distinct recovery patterns for lower-extremity strength and trunk control, the larger 30STS gain in the DSG may reflect these baseline differences rather than the supervision schedule. We treat this as one of several alternative explanations, and as part of the rationale for a future randomized trial.

The between-group differences observed in this study suggest that supervision timing may be relevant to tailoring strategies based on stroke chronicity, baseline function, digital literacy, and individual preferences. Distributed supervision may be particularly suitable in the chronic phase, where ongoing input may reinforce correct motor patterns and sustain adherence. Conversely, front-loaded supervision may benefit those in the subacute phase by supporting rapid skill acquisition during heightened neuroplasticity; however, these are hypotheses to be tested in future randomized trials^{54, 55)}. Although both models delivered equivalent total therapy dose, the between-group differences observed on the 30STS and TUG suggest that the timing of therapist contact may be associated with a different magnitude of functional gains, though causal conclusions cannot be drawn. A hybrid model combining higher-frequency early contact with consistent check-ins may combine the strengths of both approaches. These findings are consistent with prior work emphasizing the role of stage-specific therapy intensity in optimizing neurorehabilitation outcomes^{56, 57)}.

The DSG demonstrated high feasibility, safety, and adherence, with clinically meaningful pre-to-post improvements across functional mobility, balance, gait speed, and lower extremity strength (Aim 1). A retrospective cohort comparison with the previously published FSG (Aim 2) identified two statistically significant between-group findings favoring the DSG (30STS, adequately powered at 0.98; TUG, modestly powered at 0.48); contrasts for BBS, Mini-BESTest, and 10MWT were non-significant with low achieved power and are reported as inconclusive. These findings must be interpreted with caution given the non-randomized design, differences in recruitment site and timeframe, and the exploratory nature of the comparison. These findings collectively support the feasibility of both supervision models and highlight supervision timing as a clinically relevant variable for the magnitude of functional gains, warranting investigation in a future randomized controlled trial. Future research should examine long-term retention, neuroplastic mechanisms, cost-effectiveness, and scalability within adequately powered and rigorously controlled designs.

Data availability statement

Data may be made available upon request.

Funding

This work was supported by a predoctoral pilot grant awarded to Rudri Purohit by the Midwest Roybal Center for Health Promotion and Translation and by departmental funds from the University of Illinois Chicago Department of Physical Therapy.

Conflict of interest

The authors declare that they have no conflicts of interest.

Funding Statement

This work was supported by a predoctoral pilot grant awarded to Rudri Purohit by the Midwest Roybal Center for Health Promotion and Translation and by departmental funds from the University of Illinois Chicago Department of Physical Therapy.