Long-Term Clinical Outcomes of a Remote Digital Musculoskeletal Program: An Ad Hoc Analysis from a Longitudinal Study with a Non-Participant Comparison Group

Digital solutions have emerged as an alternative to conventional physiotherapy, particularly for chronic musculoskeletal pain (CMP) management; however, its long-term effects remain largely unexplored. This study focuses on patients reporting CMP, assessing 1-year clinical outcomes after a multimodal digital care program (DCP) versus non-participants, who enrolled in the program but never engaged in a single exercise session or partook of the educational content made available to them. In this longitudinal study ad-hoc analysis, pain and functionality outcomes at 1-year reassessment were studied, focusing on the odds of reaching minimal clinically important difference (MCID) and, overall average reduction in both outcomes. Healthcare utilization was additionally studied within the same period. From 867 individuals allocated to the study, 460 completed the 1-year reassessment (intervention group: 310; comparison group: 150). The intervention group presented sustained and greater pain reduction until 1-year reassessment than the comparison group, reflecting greater odds ratio of achieving the MCID both in pain and functionality (1.90 95% CI: 1.27–2.86, p = 0.002 and 2.02 95% CI: 1.31–3.12, p = 0.001, respectively). A lower healthcare utilization during the 1-year follow-up was observed in the intervention group than in the comparison group. This study suggests that a digital CMP program may have a lasting impact on improved pain and functionality in patients with CMP.


Introduction
Musculoskeletal (MSK) disorders are the main cause of disability worldwide, being collectively responsible for nearly 150 million years lived with disability (YLD), a figure that increased by 59% since 1990 [1].
The burden of MSK disorders is mainly driven by chronic musculoskeletal pain (CMP), i.e., pain that lasts beyond 3 months [2,3]. The anatomical regions most frequently affected by CMP are back, shoulder and knee [4].
CMP is associated with tremendous economic impact, with total costs estimated at approximately $560-635 billions in the US, exceeding those for heart diseases and cancer [5]. Patients suffering from CMP have diminished quality of life, impaired job performance (often resulting in loss of employment, early retirement, and disability), and informed consent. All participants who started the program a year prior were encouraged to submit 1-year follow-up reassessment surveys, through an email campaign offering $20 gift cards, with friendly reminders in the following 2-weeks. Baseline, 12-weeks and 1-year data was collected via email. The intervention group consisted of participants who engaged in the DCP program. Early discharge was possible depending on the physical therapist's (PT) clinical judgment. The comparison group corresponded to participants who registered for the program but did not start the intervention (never engaged in a single exercise session or partook of the educational content made available to them).

Intervention
The DCP, previously described elsewhere [53,54,60], consisted of an 8 or 12-week program comprising exercise, education and cognitive behavioral therapy (CBT). Upon enrollment each participant was assigned to a PT. A total of 56 PT were involved in the study. Each patient received a Digital Therapist (DT), an FDA-listed class II medical device consisting of a tablet with a pre-installed app (with instructional videos), that uses motion tracking technology to digitize motion and to provide real-time biofeedback during exercise. It was recommended that participants perform at least 3 exercise sessions per week. The DT connects with a cloud-based platform, which enables the PT to asynchronously monitor progress and adjust the protocol as per clinical judgment. The exercise program followed pre-determined phases, which were adapted by the PT to each participant's particular progress. Exercises included gradual exposure to movements through mobility/stretching, strength, and balance. Phase progress would include an increase of: (i) number of repetitions and sets, (ii) external load, (iii) range of motion, and (iv) multi-articular and multidirectional exercises. The education and CBT components were developed according to current clinical guidelines and research [12,15,[61][62][63], focusing on topics related to anatomy and physiology, evidence-based treatments, prognosis of symptoms, importance of staying active and fear-avoidance. The CBT program, based on third generation techniques (mindfulness, acceptance and commitment therapy and empathy-focused therapy) consisted of self-guided interactive modules specifically designed to address pain catastrophizing, active coping methods, and fear-avoidance. Bi-directional communication was ensured through a built-in secure chat within a smartphone app and video calls. The education content provided was condition-specific, while the CBT was generally MSK pain-oriented.

Outcome Measures
The following outcomes were collected at baseline, 12-weeks, and 1-year for the intervention group, and at baseline and 1-year for the comparison group: (i) Pain using the Numerical Pain Rating Scale (NPRS), through the question: "Please rate your average pain over the last 7 days" from 0 (no pain at all) to 10 (worst pain imaginable)" [64]. Following the IMMPACT recommendations [65], a 30% pain improvement was considered as a minimal clinically important difference (MCID).
(iii) Health care utilization through the question "Specify additional care if you seek additional care for your condition." (Conservative care: face-to-face physical therapy and chiropractor/osteopath; Invasive or emergency care: injections, emergency room and surgery).

Safety and Adverse Events
Patients were advised to report any adverse events [75] to the dedicated PT through the available communication channels for further assessment. Additionally, pain and fatigue levels during the exercise sessions (assessed by Numeric Rating Scale, range 0-10) were collected at the end of each session.

Data Availability
All relevant data is provided in the article or as supplementary material. The protocol, de-identified data and analysis codes may be provided upon request to the corresponding author.

Sample Size
A power analysis by simulation was conducted to estimate the proper sample size necessary to detect significant changes over time in pain above and beyond the minimal clinically important difference (see outcome section) after 12-weeks. Simulated data was generated using the lavaan package within the R statistical computing environment. Sample sizes in this simulation ranged from 25 to 300 participants. For each sample size, 10,000 simulations were conducted to assess statistical power at the p < 0.050 level of significance. Each simulated data set was obtained with parameters from previous studies of longitudinal change in treatment outcomes [52][53][54][55]. Additionally, a 30% attrition rate was assumed across the simulated data set. The results of this analysis showed that an initial sample of 130 participants in each group would be sufficient to achieve at least 80% statistical power to find a significant effect, greater than or equal to the MCID of 30% for pain [65].

Statistical Analyses
A descriptive analysis of the study population demographics (age, body mass index (BMI), gender and employment status) and clinical data at baseline (pain, functionality, anxiety by Generalized Anxiety Disorder 7-item questionnaire (GAD-7, range 0-21 [76]), and depression by Patient Health 9-item scale (PHQ-9, range 0-27) [77]) was performed. BMI categories were defined following Weir et al. [78]. Statistical analysis between intervention and comparison groups was performed using independent samples t-test or Mann-Whitney U test for quantitative variables, or Chi-squared test or one-sided two-proportion Z-test for qualitative variables.
To ensure comparability between groups, inverse probability weighting (IPW) was applied for controlling for confounding effects of differing baseline characteristics, as these reduce the ability of researchers to derive causal inference from clinical studies [79]. IPW differentially weights each participant's data based upon their baseline characteristics, while keeping all data, being one of the methods recommended by ROBINS-I tool to address bias due to confounding [80]. These weights are then included during data analysis to strengthen causal inference. IPW was performed on participants in both the intervention and comparison groups based upon baseline characteristics of age, BMI, gender, pain level, depression score, and anxiety score. IPW R package was used in order to estimate all weights [81]. These weights were then used in the estimation of logistic regression models predicting if participants met MCID status based upon the study group.
To address the potential bias brought by missing data, the multiple imputation of chained equations (MICE) algorithm was applied to enable data analysis for the entire cohort at 1 year follow-up [82]. Data imputation via the MICE algorithm has been shown to be a statistically reliable means of recovering information from missing data, thereby reducing uncertainty and bias of model parameters [83]. By coupling missing data imputation via MICE and IPW, the statistical models can give more robust and reliable estimates of treatment effects while controlling for confounding. Odds ratios (OR) were calculated to compare the likelihood that an outcome will occur given a particular exposure-intervention, compared to the odds of the outcome occurring in the absence of that exposure.
All statistical analyses were conducted using commercially available software (SPSS v22, IBM, Armonk, NY, USA) or coded using R (version 1.4.1717, R Foundation for Statistical Computing). The level of significance was set at p < 0.05 for all tests.

Participants
A total of 867 individuals were allocated to the study, of which 460 individuals delivered 1-year follow-up assessment, 310 belonged to the intervention group (IG) and 150 to the comparison group (CG). The study flowchart is presented in Figure 1. During the intervention, no serious adverse events [75] were reported. Patients from the intervention group performed on average 3.33 (SD 0.98) sessions per week, including 22% who performed below the 3 sessions.  Table 1 shows the demographic characteristics of the study sample. Mean age of the total cohort was 49.2 years (SD 11.4) with the majority being women (53.4%), middle age (40-60 years: 55.4%), overweight or with obesity (33.6% and 41.5%, respectively), and employed (96.3%). The most prevalent MSK condition was low back pain, which was reported by 36.6% of participants, followed by shoulder (16.0%) and knee (15.6%) conditions.   Table 1 shows the demographic characteristics of the study sample. Mean age of the total cohort was 49.2 years (SD 11.4) with the majority being women (53.4%), middle age (40-60 years: 55.4%), overweight or with obesity (33.6% and 41.5%, respectively), and employed (96.3%). The most prevalent MSK condition was low back pain, which was reported by 36.6% of participants, followed by shoulder (16.0%) and knee (15.6%) conditions.

Clinical Outcomes
The intervention group had a pain change from 4.40 (95% CI 4.19-4.61) at baseline to 2.19 (95% CI 1.99-2.39) at the end of the 12-week program, which did not clinically differ from the obtained data for 1-year outcomes.
Greater pain reduction was observed in the intervention group when compared to the comparison group (CC), who reported a mean change of −1.79 (95% CI 1.43-2.14 points, p < 0.001) in the completers cohort and of −1.78 (95% CI 1.57-2.00 points, p < 0.001) in the entire cohort.
Considering the recommended MCID of 30% for pain [65], a 71.6% response rate was found for the intervention group and 57.3% for the comparison group. Within completers, this translated into an odds ratio (OR) of being a responder for pain of 1.90, 95% CI 1.27-2.86, p = 0.002 for the intervention group when compared to the comparison group (OR 1.86 (95% CI 1.40-2.47, p < 0.001) for the entire cohort), Table 3. When analyzing the effect of each covariate in the odds for being a responder, only pain at baseline had a significant effect in the result (OR: 1.12, 95% CI 1.00-1.26, p = 0.045 Supplementary Table S1). Regarding functionality, a 47.4% response rate for reaching functional MCID was observed for the intervention group versus 36.9% in the comparison group, respectively. Among the completers, a significant OR of 2.02 (95% CI 1.31-3.12, p = 0.001) for being a responder was obtained favoring the intervention group (OR 2.25 (95% CI 1.72-2.96, p < 0.001), entire cohort), Table 3. None of the covariates influenced the likelihood of being a responder for functionality, in the completers cohort. When analyzing the entire cohort, low OR was observed for both body mass index (BMI) (OR of 0.97 95% CI 0.95-1.00, p = 0.037) and invasive care (OR of 0.36 95% CI 0.16-0.80, p = 0.013), Supplementary Table S1. These results suggest that patients with higher BMI or those seeking invasive and emergency care had difficulty reaching MCID for functionality.
When analyzing conservative care utilization during the year, we found that the intervention group had a lower number of consultations with physical therapists and osteopaths/chiropractors than the comparison group (CG: 21.3% vs. IG: 14.8%, p = 0.041). A similar tendency was observed for invasive and emergency care, corresponding to usage of injections, emergency rooms and surgery, where the intervention group reported a utilization of 6.1%, versus 10.0% on the comparison group (p = 0.068).

Main Findings
A significant pain reduction was observed in the intervention group after the digital care program (DCP), which remained stable until the 1-year reassessment, and was higher than the pain reduction observed in the comparison group. The response rate for pain was superior in the intervention group (71.6% versus 57.3%) to that which was observed in the comparison group. Similarly, the response rate for functionality was higher in the intervention group (47.4%) than in the comparison group (36.7%) at 1-year. These results, translated in a significant odds ratio (OR) for being a responder both for pain and functionality, favor the intervention group. Additionally, the intervention group reported lower conversative care utilization during the 1-year follow-up.

Comparison with Literature
Chronic MSK pain (CMP) affects one in every five people, representing a clinical and social problem worldwide [84]. Herein, the most common reported affected areas by both intervention and comparison groups were low back, followed by shoulder and knee, in line with the prevalence of MSK conditions found in the US general population [4]. In addition, the cohort follows a demographic pattern (mainly women, and patients who are overweight or obese) that has been previously associated to a higher likelihood of developing CMP [85][86][87][88], with a similar trend in both groups.
In the present study, pain levels reported by both groups at baseline (4.6-4.8 points) were within the range of those described in the literature for CMP both in telerehabilitation (5.8-4.5 points) [31][32][33][34]52,53,89] and in conventional physical therapy (4.6-5.8 points) studies [90][91][92]. When analyzing the impact of telerehabilitation interventions on pain reduction at 1-year, the results varied, possibly depending on a number of factors, including baseline pain severity, type of pain management strategy, length of the intervention and particularly the affected joint (e.g., several anatomic regions (50.9%) [31], low back (44.8%) [93], knee (43.9%) [32], hip (34.5%) [32], or hip and knee (17.2%) [94]). In the present study, a 2.43-points change was observed, with the comparison group reporting significantly lower improvement. Accordingly, higher response rates were observed in the intervention group (71.6% versus 57.3%) than in the comparison group. Even though we accounted for baseline differences using inverse probability weighting, we found that the likelihood of being a responder for pain at 1-year was influenced by initial pain levels, which is in accordance with previous literature [95]. Similarly, the comparison group presented higher BMI levels at baseline, which did not impact the likelihood of being a responder, despite high BMI being a known factor for poor prognosis of MSK pain management [96,97]. Direct comparison of response rates with other studies is difficult owing to different pain assessment tools and minimal clinical important change applied thresholds. Nevertheless, these results are consistent with those reported in studies evaluating the effect of digital programs on CMP in several anatomic regions [31] or in a single affected area (low back, knee and hip) [32,93,94]. For example, Kent et al. conducted a cluster pilot randomized controlled trial (RCT) to assess the effect of a multidisciplinary digital intervention (consisting of exercise, education and biofeedback) in a mixed acuity cohort (80% chronic) of patients with low back pain, reporting a response rate of 68% at 12 months in comparison to 21% in the standard care group [93]. Wang et al., in a similar intervention for chronic MSK for several anatomic regions reported a response rate of 72.2% in the intervention group, versus 56.2% in a control group composed of participants who enrolled but did not engage in any session [31]. In another longitudinal observational study, focusing on assessing the long-term outcomes (48-weeks) of a digital self-management program (composed of exercise and educational components) for hip and knee osteoarthritis, a response rate of 72% was reported [32], considering a lower minimal clinically important change of 20% for pain.
Functional long-term outcomes reported across the literature diverge, following either in-person or digital physiotherapy. In a systematic review assessing non-invasive nonpharmacological treatment, such as exercise for chronic pain, no long-term functional improvement for chronic low back pain and osteoarthritis pain was reported [98]. However, for chronic neck pain, exercise seemed to have a short-and long-term effect in improved function [98]. On the other hand, telerehabilitation studies failed to report evidence of sustained functional improvement [31,89]. In the present study, alongside pain reduction, significant long-term functional improvements were observed in the intervention group (47.4% response rate vs. 36.9% in the comparison group), as suggested by a significant OR both in completers and the entire cohort.
Significantly fewer participants sought physical therapists and osteopaths/chiropractors than non-participants (IG: 14.8% vs. CG: 21.3%, p = 0.041), with a similar trend regarding invasive and emergency care (injections, emergency rooms and surgery), where the intervention group reported a utilization of 6.1%, versus 10.0% of the comparison group (p = 0.068). One possible explanation might be that the improvements in pain and functionality experienced by patients in the intervention group may have prevented the need for other healthcare services. That is, this digital MSK program may have acted as a substitute for usual in-person care, particularly because this program was delivered during the COVID-19 pandemic. It may be reasonable to assume that the percent of additional healthcare usage was likely lower than usual in both groups, although we cannot conclude if the pandemic equally influenced healthcare usage in both groups.
Multidisciplinary rehabilitation may offer the possibility of patient-centered care, with condition management being holistically adjusted to patients' needs over time. A digital multimodal MSK program (as the herein reported) may address MSK pain through exercise and psychoeducation helping participants to change their beliefs about the causes and threat value of pain [99]. This may help them experience less fear of movement and more self-efficacy about their condition, thus reducing both anxiety and depressive symptoms exacerbated by MSK pain [100]. The exact weight that each DCP component has on the overall outcome change cannot be clarified in the current study design and should be further pursued in future trials.

Strenghts and Limitations
The major limitation of this study is the lack of randomization between groups, which impacts the establishment of causality of the intervention on the outcomes. Stronger evidence may be retrieved from future research using a randomized clinical trial considering conventional in-person therapy standard care as a control group. Nevertheless, the included cohorts have similar demographic patterns to the general US population [101], fostering generalization, and baseline differences were adjusted using inverse probability weighting and imputation for missing data to decrease bias. Additionally, despite weighting for levels of anxiety and depression at baseline, we cannot rule out the possibility that other psychological and cognitive traits (including motivation to perform the program) may have influenced the acceptance of the program and corresponding outcomes. Other limitations may include that the study design does not allow determination of the weight of each component of the DCP on the observed changes. In addition, while some people reported to have sought additional conservative and invasive and emergency care throughout the year, this was self-reported data (versus claims-based data). Additionally, we did not collect data about pharmacological usage, or other types of additional care, and therefore the influence of these on the clinical outcomes was not controlled. Nevertheless, the present study contributes to minimizing the gap in the literature regarding the long-term effect of a completely remote multimodal digital intervention.

Conclusions
This study was aimed at gaining further insight into the impact of a completely remote multimodal DCP in a real-world setting on long-term clinical outcomes of CMP. Participants going through the program reported sustained benefit both in pain and functionality at 1-year follow-up, as well as significantly higher improvements in pain and function in comparison to a control group of non-starters. This study suggests that a multimodal digital MSK program may have a lasting impact on improved pain and functionality in patients with chronic MSK pain.
Supplementary Materials: The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/healthcare10122349/s1, Table S1: Association of baseline variables with odds of being a responder for pain and disability.

Institutional Review Board Statement:
The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the New England IRB (protocol number 120190313) and prospectively registered in ClinicalTrials.gov, NCT04092946, on 17 September 2019.
Informed Consent Statement: All subjects gave their informed consent for inclusion before they participated in the study.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author. The data are not publicly available due to privacy restrictions.