This is a prospective, longitudinal cohort study aimed to assess the clinical outcomes of digital rehabilitation programs provided by SWORD Health.

Individuals eligible for workers’ compensation under health plans, which have entered into a commercial agreement with SWORD Health, acting as an in-network provider of physical therapy services, were recruited. Patients were initially assessed by their orthopedic surgeon and referred for physical therapy after confirmation of an ankle sprain (based on clinical and imaging findings). Referral to in-network providers of physical therapy was managed administratively, with the possibility of explicit referral to the digital program by the orthopedic surgeon.

The plan of care was based on therapeutic exercise sessions to be performed independently by the patient at home using the biofeedback device provided by SWORD Health, in accordance with the protocol presented in Multimedia Appendix 1. Patients were instructed to perform 1 exercise session per day. The plan was adapted by the treating physician in articulation with the physical therapist as needed. Program duration and patient discharge were determined by the treating physician.

The primary outcome was the change in the self-reported Numerical Pain Rating Scale (NPRS) score (0-10), as well as the change in the Foot and Ankle Ability Measure (FAAM) [46]—for activities of daily living (FAAM–ADL) and sports (FAAM–Sports) between baseline and the 6-month follow-up. Patients were assessed for these outcomes at baseline, end of the rehabilitation program, and at 6 months, through an electronic survey. The NPRS was self-reported at each assessment survey with the question, “How would you rate your pain over the last 7 days – from 0 (no pain at all) to 10 (worst pain imaginable)?”

Secondary outcomes included the following user experience–related outcomes, collected along the program by the digital therapist:

To assess differences in primary and secondary outcomes among the 3 time points, a Bonferroni multiple comparison test was performed with time as a categorical variable. Both unadjusted and adjusted differences for covariates with 95% CIs were estimated. Included covariates were age, gender, BMI, days to start treatment, grade of sprain, exercise level, and previous injury.

Linear mixed-effects models (LMM) were also utilized to assess participant change across NPRS, FAAM–ADL, and FAAM–Sports metrics from baseline to the end of the study. This type of model was chosen over a repeated-measures analysis of variance (ANOVA) since the former allows for a relaxation of model assumptions (ie, it does not assume that variances and covariances among groups are equal), a more flexible treatment of time (which is treated as a continuous variable and not a category), and makes it easier to include covariates (as additional fixed-effects) [47].

For each outcome, a model was created without including covariates (unadjusted) and including covariates (adjusted). Multiple imputation using 50 imputed data sets was used to account for attrition in each variable across time [48].

A bivariate correlation analysis was also performed to investigate covariates’ association with outcomes.

A repeated-measures ANOVA was also performed for the primary outcomes with time as the within-subjects factor and grade of sprain (grades I-III per the guidelines of Lynch [49]) as the between-subjects factor. The same approach was used with program duration categories (<4 weeks and >4 weeks) as the between-subjects factor.

LMM analysis was performed using R. All the other statistical analyses were performed using SPSS (version 17.0; SPSS Inc).

Deidentified individual participant data are provided in Multimedia Appendix 2.

In total, 104 patients from 4 different recruitment sites and 26 different orthopedic surgeons were consecutively enrolled in SWORD Health’s fully remote physical therapy program for ankle sprain between February and November 2020 (Figure 1). The dropout rate was 10.6% (11/104); 1 (1.0%) patient subsequently refused all types of care proposed by the physician, 5 (4.8%) dropped out owing to unknown reasons (missing exercise sessions and medical appointments), and 5 (4.8%) did not adhere to the digital program. In total, 93 (89.4%) patients completed the program, and 79 (76.0%) were available for follow-up.

Participant’s baseline characteristics are presented in Table 1. Baseline assessment of the outcome variables is summarized in Table 2.

Table 2 presents the primary and secondary clinical outcomes. All patients showed significant improvement in NPRS, FAAM–ADL, FAAM–Sports, and VAS pain scores (P<.001) from baseline to 6-month follow-up (Table 3).

Essentially, considering the minimal clinically important difference values established for FAAM subscales [50] (ie, 8 points for FAAM–ADL and 9 points for FAAM–Sports), the registered changes from baseline were also clinically meaningful. Overall, patients experienced reductions of 50% and 33% in the NPRS and VAS pain scores, respectively, which can also be considered clinically significant.

LMM analysis with 50 imputed data sets revealed that for all unadjusted models, significant effects of time were observed in the expected directions. This indicates that over the larger study timeline, participants reported significant improvements in NPRS, FAAM–ADL, or FAAM–Sports scores (Multimedia Appendix 3). Longitudinal changes in ankle function and pain perception are depicted in Figure 2.

Similar effects of time were found for all adjusted models (Multimedia Appendix 3). Regarding NPRS scores, participants with a higher average BMI (estimate=0.08; P=.04) and those who took days to start treatment (estimate=0.01; P=.02) had significantly higher scores. A significant negative linear effect of sprain grade was also observed (estimate=–1.00; P=.02). No covariates significantly affected FAAM–ADL scores. Regarding FAAM–Sports scores, older participants showed significantly lower scores (estimate=0.58; P=.03). A significant negative quadratic effect of sprain grade was also observed (estimate=0.01; P=.02).

Among those who finished the program, 83.9% (78/93) of patients were classified by the treating physician as having obtained maximum medical improvement and no residual disability, while 12 (13%) patients were rendered with permanent partial disability, 2 (2%) with temporary total disability, and 1 (1%) with permanent total disability.

In total, 48.4% (45/93) of completers returned to work before clinical discharge. Within those discharged with no residual disability, the majority (53.8%, 42/78) did not return to work before clinical discharge, likely because of the shorter treatment period (median 2.8 weeks, IQR 4.04 weeks; range 0.86-9.57 weeks).

The reinjury rate was 2.5%; 2 of the 79 available patients for follow-up reported ankle sprain recurrence as a result of a fall. One was a patient who had had 4 previous ankle sprains, and the other did not have a history of previous injury. Both had been discharged with no residual disability.

The adverse event rate was 3.2% (Multimedia Appendix 4). One patient reported exacerbated ankle pain caused by long walks; one could not perform the sessions due to a suspected plantar fasciitis; and another reported intense lower back pain, limiting compliance with the exercise protocol. None were related to the digital intervention.

This study shows that a fully remote, home-based, digital rehabilitation program for acute ankle sprains delivered at patients’ homes allowed patients to attain clinically meaningful improvement in pain (evident from their VAS and NPRS scores), activities of daily living (FAAM–ADL scores), and sports activities (FAAM–Sports scores). Furthermore, these programs led to a full recovery without residual disability in 83.9% of patients, which compares favorably with the published literature showing that at least one-third of individuals will experience residual symptoms [13-15].

There is a dearth of studies on digital programs for acute ankle sprains, as supported by recently published systematic [51] and literature [52] reviews on the subject. We therefore broadened the search to include exercise-based approaches in general [23,53-55]. Overall, the results obtained in this study are similar to those reported for other supervised exercise programs, and the first detailed positive outcomes with a fully digital program.

One RCT (n=90) [55] assessed the effectiveness of exercise training using the Nintendo Wii Fit balance board in comparison to physical therapy and to a control receiving no therapy. Investigators found this tool was not more effective than PT only or no exercise. Of note, patients enrolled in this study had little room left for improvement, with near-normal scores at baseline on the FAAM–ADL (mean 71-83) and FAAM–Sports (mean 37-52), and low VAS pain (approximately mean 1 point), which may have been the reason behind no difference between physical therapy only or no exercise.

In another RCT (n=74) [54] comparing an manual therapy and exercise (MTEX) program with a home exercise program (HEP), the improvement in the MTEX program at 4 weeks was similar to what we observed in this study: FAAM–ADL score, mean 21.3 (95% CI 18.2-24.5) points; FAAM–Sports score, mean 27.1 (95% CI 22.7-31.6) points; and NPRS score, mean –2.7 (95% CI –2.9 to –2.5) points. When compared to the HEP group, our intervention also provided superior outcomes in terms of functional recovery and pain.

Both NPRS baseline values and its magnitude of change from baseline to the end of the program were similar to the ones reported for other exercise interventions after ankle sprain [23,54,56].

This study corroborates previous findings of high recurrence rates both among nonathlete (24%-54%) [57,58] and athlete (12%-47%) populations, with 29% of all enrolled patients having had previous injury.

Also consistent with our findings, the group from Verhagen found that a home-based proprioceptive 8-week training program, delivered through a mobile app after usual care, was successful in reducing recurrences of ankle sprains in a 12-month period as against conventional care alone (22% versus 33%, as revealed through an RCT with 522 athletes from the Netherlands) [59]. Although the rate of reinjury was still much higher than that reported here at 6 months (2.5%), this further supports the effectiveness of remote interventions in preventing ankle reinjuries.

Previous findings indicate an association between the rate of resprain and incomplete recovery [57]. Therefore, the high percentage of complete recovery attained in this study may explain the lower rate of recurrence, even if the 2 patients who experienced recurrence had been discharged with no residual disability. In fact, by the 6-month follow-up, 45.6% (36/79) and 35.5% (11/31) of patients in this study, respectively, achieved scores compatible with the normative values for FAAM–ADL and FAAM–Sports reported for the adult population (92.3 and 85.1 points, respectively) [60].

In a systematic review and meta-analysis, Bleakley et al [61] found no clear consensus on an optimal training volume, with rehabilitation times ranging from 3.5 to 21 hours (median 12 hours). The highest total rehabilitation time was 21 hours, equivalent to 1.75 hours per week over 12 weeks. In our study, the mean total exercising time was 12.5 (SD 10.50, range 1.3-47.3) hours, equivalent to 2.4 (SD 0.87, range 0.4-4.6) hours per week. Hence, the total training volume was similar to that of other interventions, but dosage per week was much higher.

Even though overall changes from baseline to follow-up were not significantly different between patients discharged before or after 4 weeks (no interaction found between time and program duration), the latter patients had worse FAAM–ADL scores both prior to participating in the program and at discharge, and worse NPRS and FAAM–ADL scores at 6 months.

We hypothesize this could be a consequence of the particularly long period between the injury date and treatment initiation—mean 53.2 (SD 48.26, range 4-281) days—mainly in relation to disruptions in health care delivery in the wake of the COVID-19 pandemic. (ie, a delay between injury and the physician appointment) (Table 1). Indeed, we found a correlation between longer waiting periods and extended program duration (Pearson r₉₃=0.48; P<.001), with mean waiting times of 40.9 (SD 38.79) days for patients who were discharged within 4 weeks versus 68.1 (SD 54.57) days for those discharged after 4 weeks (P=.01).

Recent reviews have not found sufficient evidence regarding independent predictors of clinical outcomes [62,63]. Only 1 study so far gathered proof that a low injury grade is a predictor for better outcomes [64].

In this study, no differences were found in terms of program duration between injury grades (P=.11, 1-way ANOVA; grade I: 4.8 weeks, 95% CI 3.8-5.8 weeks; grade II: 4.7 weeks, 95% CI 3.2-6.2 weeks; grade III: 7.5 weeks, 95% CI 3.7-11.2 weeks). Nonetheless, we found differences between injury grade and clinical improvement. Patients with grade II injuries experienced the greatest improvement during the program, followed by those with grade I and then grade III injuries. This could be explained by the fact that grade I sprains are only directed to physical therapy in case of aggravation, at a point where they actually become slower respondents. In the long term, however, patients with grade III sprain reported the greatest improvement in NPRS and FAAM–Sports scores, followed by those with grade II and grade I sprain, as confirmed by LMM analysis. This is most likely related to the lower FAAM scores and higher pain levels at baseline in patients with grade III sprain, consequently with a higher margin of progression. These aspects, along with the convergence of clinical outcomes over time for the 3 groups, do not support the notion of a low injury grade being a predictor for better outcomes.

The limitations of this study are mainly related to the study design and the referral process. This was a prospective cohort study and, as such, did not include a control group. However, as shown above, our results are comparable to those reported previously for supervised exercise programs in this same context. Regarding the referral process, while patient assignment to an in-network provider was largely performed administratively, explicit referral to digital programs was possible. This may have introduced a selection bias toward patients more likely to engage in digital care.

This was the first study presenting the outcomes from a fully remote exercise-based rehabilitation program for acute ankle sprains, demonstrating clinically meaningful change in both pain and function, as well as complete recovery in 81.7% of patients, with sustained results over time. As such, this study demonstrates not only the feasibility of fully digital programs in this context, but also that these programs can achieve clinical outcomes comparable to face-to-face interventions.