This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Rehabilitation and Assistive Technology, is properly cited. The complete bibliographic information, a link to the original publication on https://rehab.jmir.org/, as well as this copyright and license information must be included.
Musculoskeletal care is now delivered via mobile apps as a health care benefit. Although preliminary evidence shows that the clinical outcomes of mobile musculoskeletal care are comparable with those of in-person care, no research has examined the features of app-based care that secure these outcomes.
Drawing on the literature around in-person physical therapy, this study examines how patient-provider relationships and program engagement in app-based physical therapy affect clinically meaningful improvements in pain, function, and patient satisfaction. It then evaluates the effects of patient-provider relationships forged through in-app messages or video visits and timely, direct access to care on patients’ engagement in their recovery.
We conducted an observational, retrospective study of 814 pre- and postsurveyed participants enrolled in a mobile app physical therapy program where physical therapists prescribed workouts, education, and therapeutic activities after a video evaluation from February 2019 to December 2020. We estimated generalized linear models with logit functions to evaluate the effect of program engagement on clinical outcomes, minimal clinically important differences (MCIDs) in pain (ΔVisual Analogue Scale ≤−1.5) and function (ΔPatient Specific Functional Scale ≥1.3), and the effects of patient-provider relationships and clinical outcomes on patient satisfaction—participant reported likelihood to recommend the program (Net Promoter Scores of 9-10). We estimated Poisson generalized linear models to evaluate the effects of stronger patient-provider relationships and timely access to physical therapy within 24 hours on engagement including the number of weekly workouts and weeks in the program.
The odds that participants (N=814) had a pain MCID increased by 13% (odds ratio [OR] 1.13, 95% CI 1.04-1.23;
Similar to in-person care, program engagement positively affects clinical outcomes, and strong patient-provider relationships positively affect satisfaction. In app-based physical therapy, clinical outcomes positively affect patient satisfaction. Timely access to care and strong patient-provider relationships, particularly those forged through video visits, affect engagement.
Physical therapists, providers with highly specialized knowledge in managing musculoskeletal conditions [
Increasing evidence supports that physical therapy via a mobile app delivers pain and functional outcomes comparable with those of in-person care [
In brick-and-mortar physical therapy clinics, “adherence” to a course of provider-prescribed care drives clinical outcomes [
Physical therapy delivered through a mobile app may not be structured similarly to in-person physical therapy with a specific number of weekly visits. In the program examined in this paper, care delivery focused on immediate access to care, ad hoc follow-up video visits, and direct, asynchronous communication between patients and their designated therapists. After an initial synchronous video evaluation, physical therapists designed recovery programs to accord with patients’ goals and altered these programs in response to synchronous and asynchronous feedback from patients. Physical therapists guided their patients through phases of their recovery in real time based on their activity levels, feedback to exercises, and changes in pain and function levels throughout an episode of care.
Owing to the real-time nature of physical therapy in this setting, we take a broader view of adherence and measure it as program engagement defined by 2 measures: the number of patient-recorded in-app–prescribed therapeutic weekly workouts and the number of weeks participants are active in the program. We first tested the hypothesis that clinical outcomes (clinically meaningful pain reduction and functional improvement) were positively associated with program engagement.
In concert with driving clinical outcomes by leveraging the best available evidence, evidence-based care is patient-centered, which is measured by patient satisfaction [
There is inconsistent evidence in the literature about how patients’ clinical outcomes affect patient satisfaction with physical therapy [
There are explicit trade-offs between care delivered through a mobile app versus in-person office visits. On the one hand, regular face-to-face visits may better strengthen patient–provider relationships than app-based video visits and chats. On the other hand, patients who arrive at in-person physical therapy only after referral, ineffective self-management, or alternative therapies (eg, acupuncture and massage), may be less motivated to engage in their treatment than those who can directly access care the same day via an app. Although we cannot interrogate these trade-offs in this paper, our secondary purpose is to understand if the strength of digital patient-provider relationships and immediate access to care via a mobile physical therapy program affects how readily participants engage in their own recovery.
In traditional clinical settings, provider communication with patients affects their adherence to treatment [
There is also evidence that early, direct access to physical therapy can affect clinical outcomes by treating conditions before they become more chronic and difficult to treat [
Our goal in this study is to examine the aspects of patient-provider relationships and program engagement that are associated with clinically important differences in pain and function along with patient satisfaction in physical therapy delivered via a mobile app. The secondary purpose of this study is to understand how 2 aspects of mobile app–based care delivery—relationships built on in-app interactions and immediate access to care—affect patient behaviors that are clinically meaningful: consistently working out and sticking with the program.
We conducted an observational, longitudinal, retrospective study using data collected from commercial users of a physical therapy program delivered via a mobile app offered as a health benefit with no cost or copay to privately insured employees by their employers [
We used established patient-reported outcome measures, including the Patient Specific Functional Scale (PSFS), Visual Analogue Scale (VAS), and Global Rate of Change (GROC), which were delivered asynchronously [
The baseline survey had an average of 4 questions across 5 screens. All the questions in the baseline survey were required to be answered. The final survey had an average of 4.5 questions across 4 screens, with responses to all but one open-ended question required. Participants could go back during their surveys and edit responses on previous pages, but they could not review their responses as a summary or alter their surveys after submission.
To enroll in the program, participants created an account in a mobile app and entered demographic information (age and gender), their chief complaint, and provided pain and function ratings in an in-app baseline survey. The participants were matched with a therapist licensed in their state to schedule an initial video evaluation visit. The program’s therapists were trained in evidence-based approaches to evaluate, diagnose, and treat patients on demand via a mobile app.
During the evaluation, physical therapists conducted an in-depth interview and performed a physical exam over secure in-app video to establish a functional baseline and arrive at a diagnosis. On the basis of the participant’s diagnosis and treatment goals, physical therapists then prescribed a course of care accessible through the app. Therapists also assigned educational content specific to patients’ conditions, therapeutic activities (eg, icing or going for a walk), and asynchronous digital physical assessments. Physical therapists modified their patients’ care plans in response to direct feedback from patients via in-app chat, regular pain and function surveys, or follow-up video visits.
All activities in the program were collected and quantified, including completion of prescribed in-app exercises and therapeutic activities, in-app chats with physical therapists, and subsequent video visits. At the end of the program, participants were asked to complete a final survey, which included final measures of pain and function.
We included participants in the study who enrolled after the launch of the program on February 15, 2019, and completed the program by December 31, 2020, if they were (1) aged ≥18 years; and (2) presented with a musculoskeletal condition such as low back pain, neck pain, arthritis, sprains, strains, or similar overuse injuries that would benefit from physical therapy or presented for postoperative rehabilitation; and (3) completed a participant survey of clinical outcomes at the end of their episode of care or reported reliable pain and function metrics toward the end of care in weekly surveys. We excluded participants if they (1) did not meet the inclusion criteria and (2) endorsed symptoms or multiple conditions during the initial video evaluation that physical therapists determined would preclude the use of app-based physical therapy as a first line of treatment and required referral for an in-person physical exam (eg, fractures, cervical central cord lesion, subarachnoid hemorrhage or ischemic stroke, unexplained weight gain or loss, fatigue and malaise, among other conditions).
Participants in our sample were not automatically excluded if they endorsed symptoms found on the Optimal Screening for Prediction of Referral and Outcome-Review of Systems (OSPRO-ROS) tool [
During the study period, 945 participants completed the program and a final outcome survey. Participants typically completed the voluntary final survey within 2 weeks of finishing the program and were neither incentivized nor reminded to do so. We carried forward 33 pain and function observations that participants reported in weekly in-app pain and function surveys if participants reported them less than 3 weeks before completing the program and more than 2 weeks after starting the program. Weekly pain and function surveys were not implemented until September 23, 2020, and participants responded more readily to these earlier in their recovery, resulting in few responses to carry forward. We also imputed 32 values for missing satisfaction scores using the modal responses of similar participants with similar earlier in-episode satisfaction scores. The average time between baseline and outcome responses collected during either the final survey or last weekly pain and function surveys was approximately 44 days.
To eliminate outliers, we calculated the standardized individual difference by dividing participant-level pre–post outcome differences by the SD of those differences and eliminating observations above and below 1% of the distribution for both clinical outcomes [
A total of 36 participants had too little activity to make reliable conclusions about the program’s outcomes (no workouts and <2 weeks in the program) and were excluded from the analysis. This left a total of 814 eligible participants included in the study (
Study participation flow diagram.
In the baseline and end-of-program surveys, participants rated their maximum pain levels over the last 24 hours using the VAS [
We created 2 binary variables for minimal clinically important differences (MCID) in pain (VAS) and function (PSFS): a value of 1 was assigned to participants’ episodes with changes in their pain ≤−1.5 points and ≥1.3 points in their functional ability [
We also created binary variables equal to 1 for large changes in pain (ΔVAS≤−3.5) and function (ΔPSFS≥2.7) and 0 otherwise based on thresholds identified in the literature [
Satisfaction with the program was measured by a final survey question that was used to calculate the Net Promoter Score (NPS) by asking participants to answer: “How likely is it that you will recommend the program to a friend or colleague?” on a scale from 0 “Not at all Likely” to 10 “Extremely Likely.” NPS defines categories of respondents as “Detractors” (0-6), “Passives” (7-8), and “Promoters” (9-10) [
Two variables measured program engagement: (1) the number of in-app workouts per week and (2) the duration of the program in weeks. The duration of the program was calculated as the difference between the time participants started the program after their initial evaluation to the end of the program, which was defined as the time when patients were either discharged directly by their provider or were inactive for 2 weeks, whichever came first.
During the program, physical therapists communicated asynchronously with participants through in-app chat to assess their progress and provide guidance. Physical therapists and participants also scheduled synchronous follow-up video visits. Patient-provider communications are used to measure the strength of these relationships and are captured by (1) the number of unique days a provider sends a message to participants through in-app chat per week and (2) a categorical variable for the number of video follow-up visits after participants’ initial video evaluations. The categories for follow-up visits included (1) no visits, (2) 1 to 2 visits, (3) 3 to 4 visits, and (4) 5 or more visits. The category for “no visits” was omitted from our models to serve as a comparator.
Prompt access to care was measured in days to the initial video evaluation after enrolling in the program. A binary variable was created with 1 assigned to those who accessed care within 24 hours, and 0 assigned to those who accessed care after 24 hours.
Chronicity, baseline pain and function levels, comorbid conditions, and adverse symptoms can affect participants’ recovery [
We controlled for baseline pain and function. Baseline pain was categorized as little to no pain (VAS≤1), mild pain (3.4≤VAS>1), moderate pain (7.4≤VAS>3.4), and severe pain (VAS>7.4) based on cut points identified in the literature [
To test our hypotheses, we estimated generalized linear models (GLMs). GLMs for MCIDs in pain, function, and satisfaction were estimated using the binomial family of exponential dispersion models and a logit link function, which is equivalent to a logistic regression model fit by maximum likelihood estimation. GLMs for the number of workouts per week and number of weeks in the program were estimated using the Poisson family of exponential dispersion models and a log link function. We interpreted our results by evaluating changes in the odds of an outcome, which were calculated by exponentiating the coefficients from the model, and by subtracting 1 from the odds to better interpret odds that were less than one (negative coefficients).
On average, providers frequently communicated with the participants. About one-third (257/814, 31.6%) of the participants completed 3 or more additional video visits beyond the initial evaluation. In between visits, physical therapists checked in with participants about 1.8 days per week via in-app chat. Provider chat messages consisted of single messages or in-depth live chat conversations with participants. Approximately 52.8% (430/814) of the participants completed their initial video consultation within 24 hours of registering for the program.
Baseline participant characteristics (N=814).
Characteristics | Values | |||
|
||||
|
Female, n (%) | 387 (47.5) | ||
|
|
|||
|
|
Value, mean (SD) | 40.85 (11.9) | |
|
|
≥50, n (%) | 214 (26.3) | |
|
||||
|
Low back pain | 172 (21.1) | ||
|
Shoulder | 132 (16.2) | ||
|
Knee | 118 (14.5) | ||
|
Neck | 104 (12.8) | ||
|
Upper body, elbow, wrist, hand, or arm | 84 (10.2) | ||
|
Lower body, ankle, foot or leg | 83 (10.3) | ||
|
Hip | 70 (8.6) | ||
|
Back or spine | 46 (5.7) | ||
|
Other | 5 (0.6) | ||
|
||||
|
Pain baseline (VASa) | 4.4 (2.2) | ||
|
Function baseline (PSFSb) | 5.2 (3.0) | ||
|
||||
|
Little to no pain (VAS≤1) | 61 (7.5) | ||
|
Mild pain (3.4≤VAS>1) | 218 (26.8) | ||
|
Moderate (7.4≤VAS>3.4) | 475 (58.4) | ||
|
Severe pain (VAS>7.4) | 60 (7.4) | ||
|
||||
|
Chronic (>3 months) | 497 (61.1) | ||
|
Subacute (1-3 months) | 128 (15.7) | ||
|
Acute (<1 month) | 189 (23.2) | ||
|
Reported comorbid conditions | 383 (47.1) | ||
|
Reported adverse symptoms | 281 (34.5) |
aVAS: Visual Analogue Scale.
bPSFS: Patient Specific Functional Scale.
Descriptive statistics for outcomes and predictors (N=814).
Variables | Values | |||
|
||||
|
Pain outcome (VASa), mean (SD) | 1.7 (1.9) | ||
|
|
|||
|
|
Pain MCIDb (ΔVAS≤−1.5) | 544 (66.8) | |
|
|
Large pain MCID (ΔVAS≤−3.5) | 289 (35.5) | |
|
Function Outcome (PSFSc), mean (SD) | 7.8 (2.365) | ||
|
|
|||
|
|
Function MCID (ΔPSFS≥1.3) | 519 (63.8) | |
|
|
Large function MCID (ΔPSFS≥2.7) | 421 (51.7) | |
|
||||
|
Likelihood to recommend, mean (SD) | 9.3 (1.5) | ||
|
Promoters, n (%) | 674 (82.8) | ||
|
||||
|
Number of workouts per week | 2.8 (2.2) | ||
|
Weeks in program | 9.1 (5.4) | ||
|
||||
|
Days messaged by physical therapist per week, mean (SD) | 1.8 (1.1) | ||
|
|
|||
|
|
None | 232 (28.5) | |
|
|
1-2 | 325 (39.9) | |
|
|
3-4 | 180 (22.1) | |
|
|
≥5 | 77 (9.5) | |
|
||||
|
24 hours to first visit | 430 (52.8) |
aVAS: Visual Analogue Scale.
bMCID: minimal clinically important difference.
cPSFS: Patient Specific Functional Scale.
Participants’ baseline chronicity and pain affected the odds of having an MCID in pain. We observed a 46% (
Participants’ age, chronicity, and baseline pain severity and function affected the odds that the participants saw an MCID in function. The odds of completing the program with an MCID in function were 53% (
Distribution of pain change by number of workouts. VAS: Visual Analogue Scale.
Odds ratios (ORs) for generalized linear models of program engagement and baseline controls on clinical outcomes (814 observations)a.
Variables | OR (95% CI) | |||||||
|
||||||||
|
Intercept | 0.13 (0.07-0.24) | <.001 | |||||
|
|
|||||||
|
|
Number of workouts per week | 1.13 (1.04-1.23) | .003 | ||||
|
|
|||||||
|
|
Age ≥50 years | 0.56 (0.38-0.83) | .003 | ||||
|
|
Chronic condition | 0.54 (0.38-0.77) | <.001 | ||||
|
|
Severe pain | 0.30 (0.12-0.74) | .01 | ||||
|
|
Baseline pain | 1.80 (1.63-2.00) | <.001 | ||||
|
||||||||
|
Intercept | 89.24 (43.52-182.98) | <.001 | |||||
|
|
|||||||
|
|
Number of weeks in program | 1.04 (1.00-1.08) | .03 | ||||
|
|
|||||||
|
|
Age ≥50 years | 0.47 (0.31-0.72) | <.001 | ||||
|
|
Chronic condition | 0.50 (0.35-0.74) | <.001 | ||||
|
|
Severe pain | 0.29 (0.14-0.57) | <.001 | ||||
|
|
Baseline function | 0.58 (0.54-0.63) | <.001 |
aComorbid conditions, adverse symptoms, and access were not significant, and there was no direct relationship between provider communication and outcomes.
bMCID: minimal clinically important difference.
Distribution of functional change by weeks in program. PSFS: Patient Specific Functional Scale.
Odds ratios (ORs) for generalized linear models of strength of patient–provider relationships, clinical outcomes, and baseline controls on satisfaction (814 observations)a.
Variables | OR (95% CI) | |||||||
|
||||||||
|
Intercept | 1.47 (0.56-3.84) | .43 | |||||
|
||||||||
|
|
1-2 follow-up visits | 2.06 (1.33-3.20) | <.001 | ||||
|
|
3-4 follow-up visits | 2.17 (1.27-3.70) | .01 | ||||
|
|
≥5 follow-up visits | 3.32 (1.42-7.79) | .01 | ||||
|
|
|||||||
|
|
Function MCIDb | 1.85 (1.17-2.93) | .01 | ||||
|
|
Large pain MCID | 2.84 (1.68-4.78) | <.001 | ||||
|
|
|||||||
|
|
Female | 2.23 (1.48-3.34) | <.001 | ||||
|
|
Baseline function | 1.09 (1.01-1.17) | .03 | ||||
|
|
Baseline pain | 0.85 (0.22-0.95) | .004 |
aA total of 32 imputed values (782 original).
bMCID: minimal clinically important difference.
The results in
Odds ratios (ORs) for generalized linear models of strength of patient–provider relationships, access, and baseline controls on program engagement (814 observations).
Variables | OR (95% CI) | |||||
|
||||||
|
Intercept | 2.04 (1.68-2.47) | <.001 | |||
|
|
|||||
|
|
1-2 follow-up visits | 1.16 (1.04-1.29) | .01 | ||
|
|
3-4 follow-up visits | 1.32 (1.17-1.49) | <.001 | ||
|
|
≥5 follow-up visits | 1.06 (0.90 to 1.25) | .49 | ||
|
|
Days messaged by physical therapist per week | 1.11 (1.07-1.16) | <.001 | ||
|
|
|||||
|
|
24 h to first visit | 1.14 (1.05-1.24) | .003 | ||
|
|
|||||
|
|
Age ≥50 years | 1.25 (1.14-1.37) | <.001 | ||
|
|
Adverse symptoms | 0.87 (0.79-0.95) | .002 | ||
|
|
Severe pain (VASa>7.4) | 0.76 (0.63-0.92) | .01 | ||
|
|
Baseline pain (VAS) | 1.02 (1.00-1.05) | .049 | ||
|
|
Baseline function (PSFSb) | 0.96 (0.95-0.98) | <.001 | ||
|
||||||
|
Intercept | 9.57 (8.80-10.40) | <.001 | |||
|
|
|||||
|
|
1-2 follow-up visits | 1.08 (1.01-1.14) | .02 | ||
|
|
3-4 follow-up visits | 1.28 (1.19-1.36) | <.001 | ||
|
|
≥5 follow-up visits | 1.91 (1.77-2.05) | <.001 | ||
|
|
Days messaged by physical therapist per week | 0.85 (0.83-0.87) | <.001 | ||
|
|
|||||
|
|
Age ≥50 years | 1.11 (1.05-1.16) | <.001 | ||
|
|
Chronic Condition (>3 months) | 1.13 (1.08-1.18) | <.001 | ||
|
|
Adverse symptoms | 1.11 (1.06-1.17) | <.001 | ||
|
|
Baseline function (PSFS) | 0.99 (0.98-0.99) | <.001 |
aVAS: Visual Analogue Scale.
bPSFS: Patient Specific Functional Scale.
Distribution of number of workouts per week by number of follow-up visits.
We included significant controls for age, adverse symptoms, pain severity, and baseline pain and function scores. Interestingly, participants aged ≥50 years had about 25% (
Participants who concurrently experienced adverse symptoms found on the OSPRO-ROS did approximately 13% (
Additional video follow-up visits were positively associated with program duration. Compared with participants who did not have follow-up visits,
Participants who were aged ≥50 years (
Mean physical therapist messages per week by distribution of program durations in weeks.
This study builds on prior studies that show that mobile app–based physical therapy delivers similar outcomes to in-person care [
However, the mechanism driving clinically meaningful changes in function requires a different form of engagement: time in the program. Exercise-induced analgesia is well documented in the literature, although the mechanism remains unclear [
Unlike traditional physical therapy, we observed that clinical outcomes were more closely associated with satisfaction. A minimal amount of functional change had a large effect on participants’ willingness to recommend the program under study. However, it took large changes in pain to influence participants to recommend the program to their friends and family.
Perceptions of functional changes may differ from perceptions of pain change. In our clinical practice, most patients during intake come to us “because they don’t want to hurt anymore” and expect pain to be eliminated. Patients may have more relative, vague expectations around functional recovery unless they cannot perform activities required for their livelihood. Patients often struggle to pinpoint goals for functional improvement. Pain alone may not be enough for patients to stop doing something altogether or they may not have a requisite daily task that they can no longer perform (eg, they must be able to lift 50 pounds for their job; they cannot pick up their child). This means that the elimination of pain (a hard outcome to achieve) must be met to be satisfied, but a lower level of functional improvement may yield satisfaction. Future research should unpack perceptions around changes in pain and function throughout recovery.
Physical therapy delivered via a mobile app resembles in-person physical therapy in that it depends on strong relationships between patients and providers to be successful. Frequent, albeit not weekly, video follow-up visits were positively associated with satisfaction, the completion of more weekly workouts, and persistence in the program, which were the key ingredients for recovery. Asynchronous messaging may also help strengthen patient–provider relationships because weekly workouts increased with each day per week that providers messaged participants. However, provider messaging may have a negative effect if used to chase unresponsive participants later in the program. Provider messages also did not have a significant effect on satisfaction, whereas video visits did.
Frequent, face-to-face interactions between providers and participants may keep participants motivated and remain active in the program until they see significant improvements. Future research should further explore how digital communication can build stronger therapeutic alliances between physical therapists and patients in a digital setting [
Unlike traditional in-person physical therapy, mobile physical therapy has the potential to reduce time to care [
We cannot eliminate the possibility that participants who access care sooner are more intrinsically motivated or have fewer barriers to exercising than those who delay their appointments. The delivery of care in a digital environment is a promising area for future research to understand how providers can optimize care to ensure better clinical outcomes and patient satisfaction.
We did not find any direct relationship between clinical outcomes and access to care or patient-provider communication that indicates strong ties. Rather, access and relationships between physical therapists and patients that were strengthened by digital communication were associated with patient behaviors that were then followed by significant recovery outcomes. Future work should aim to understand the causal relationships between the design of mobile app physical therapy programs in terms of access, indicators of different qualities of patient-provider relationships, and the recovery behavior of participants.
This study is inherently limited as an observational study of an employer-based population. The voluntary nature of, and lack of compensation for, completing the final survey meant that our sample size was reduced, potentially biasing our results. The results may not be generalizable to a broader population of employees, retirees, or children. Our study also lacked a control group. Future research should compare meaningful clinical outcomes, satisfaction, and program engagement of mobile app–based physical therapy to in-person physical therapy in a controlled clinical trial. Randomized control trials or other suitable experimental methods should be used to unpack causality around patient-provider communication and relational indicators, access to care, and program engagement.
Physical therapy delivered via a mobile app may be more likely to result in clinically important changes in pain and function if it engages patients by directly connecting them with physical therapists and by facilitating strong relationships with their providers. Synchronous communication, in particular video visits, may help physical therapists foster strong relationships that personalize app-based care and build in accountability and encouragement so that patients engage in recovery and, concomitantly, enjoy clinically important improvements in pain and function. In app-based physical therapy, clinical outcomes may be more closely associated with patient satisfaction, independent of patients’ relationships with their providers, than what is observed in studies evaluating in-person physical therapy.
Heatmap of significant Pearson correlations (
Models with main effects only.
generalized linear model
Global Rate of Change
minimal clinically important difference
Net Promoter Score
Optimal Screening for Prediction of Referral and Outcome-Review of Systems
Patient Specific Functional Scale
Visual Analogue Scale
The authors are indebted to Julie Mulcahy, Steve Bayer, Ryan Quan, Anna DeLaRosby, and Melissa Leebove for their insightful comments, edits, and support in crafting this paper. We are also grateful for Dan Rubinstein and Cameron Marlow, whose vision and desire to improve health care, led to the creation of Physera.
LB and TN are both employed shareholders of Omada Health Inc.