Background:

Most adults with knee osteoarthritis (OA) are inactive, highlighting the need for continued efforts to promote regular engagement in exercise. Few studies have directly compared different strategies, ranging in intensity of resources required, for improving exercise and related outcomes among patients with OA.

Objective:

The objective of this 3-arm study was to compare the effectiveness of physical therapy (PT; with an emphasis on a home exercise program) and internet-based exercise training (IBET), both compared with a wait list (WL) control arm, among individuals with knee OA.

Methods:

This study was a randomized controlled trial of 350 participants with symptomatic knee OA, allocated to standard PT, IBET, and WL control group in a 2:2:1 ratio, respectively. The PT group received up to 8 individual visits within 4 months. The IBET program provided tailored exercises, video demonstrations, and guidance on exercise progression. The primary outcome was the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC, range 0-96), assessed at baseline, 4 months (primary time point), and 12 months. General linear mixed effects modeling compared changes in WOMAC among study groups, with superiority hypotheses testing differences between intervention groups and WL and noninferiority hypotheses comparing IBET with PT (noninferiority margin = 5 points on WOMAC). Higher WOMAC scores indicate worse symptoms; therefore, negative scores for comparisons at follow-up indicate more improvement for PT and IBET at groups compared with the WL group.

Results:

Baseline mean WOMAC scores for PT, IBET, and WL groups were 32.0, 31.3, and 33.6, respectively. At 4 months, improvements in WOMAC score did not differ significantly for either the PT group or IBET group compared with WL (PT: −3.36, 95% CI, −6.84 to 0.12, P = .06; IBET: – 2.70, 95% CI, −6.24 to 0.85, P = .14). Similarly, at 12 months, mean differences compared with WL were not statistically significant for either group (PT: −1.59, 95% CI, −5.26 to 2.08, P = .39; IBET: −2.63, 95% CI, −6.37 to 1.11, P = .17). IBET was noninferior to PT at both time points.

Conclusions:

Improvements in WOMAC score following IBET and PT did not differ significantly or clinically from the WL group. This contrasts with prior studies showing that PT improves pain and function for people with knee OA. Additional research is needed regarding strategies to maximize the effects of exercise-based interventions for individuals with knee OA.

Limitations and Subpopulation Considerations:

This study was conducted in 1 geographic region and included only participants who had regular internet access. More than half of participants had a college education. These factors could affect generalizability to other subpopulations.

Aim 1

Compare the effects of IBET and standard PT for knee OA on short-term (4-month) patient-centered outcomes, vs a WL control group.

Aim 2

Compare the effects of IBET and standard PT for knee OA on longer-term (12-month) patient-centered outcomes, vs a WL control group.

Aim 3

Examine whether individual patient characteristics (particularly age and baseline functional status) are associated with differential improvement with the IBET and/or standard PT interventions.

Types and Number of Stakeholders Involved

Our stakeholder panel comprised 5 patients with knee OA, 2 physicians who treat patients with knee OA (1 primary care, 1 orthopedic specialist), 2 physical therapists (including 1 representative from the American Physical Therapy Association), an exercise physiologist with experience in delivering scalable exercise-based interventions, and 2 representatives of national organizations seeking to improve outcomes for people with OA (Arthritis Foundation, Centers for Disease Control and Prevention Arthritis Program).

How the Balance of Stakeholder Perspectives Was Conceived and Achieved

We (the research team) developed the stakeholder panel to reflect different “key voices” related to OA, PT, and exercise-based interventions. Patients with OA are obviously crucial to the conversation because they have direct experience in managing OA and engaging in exercise-based interventions in the context of chronic pain and functional limitations. Patients were the largest group of stakeholders on the panel, based on the importance of their perspectives. In addition, inclusion of multiple patients allowed diversity in demographic characteristics (age, gender, race, rural/urban residence, severity of OA systems) and prior/current experience with exercise-based interventions. Clinicians and public health representatives are also key voices for this research, so we sought at least 2 participants from each of these categories. The perspective of the physical therapy profession was particularly important. The study team sought to achieve balance in perspective through both regular group phone calls that emphasized the importance of all team members' input and individual conversations to ensure stakeholders were comfortable in sharing their perspectives. In particular, the investigator team met individually with each patient stakeholder to answer questions and receive input on various aspects of the study during the start-up phase.

Methods Used to Identify and Recruit Stakeholder Partners

We identified patient stakeholders in 2 ways: recommendations from health care providers at the University of North Carolina at Chapel Hill (UNC) who treat patients with knee OA and participants in the Johnston County Osteoarthritis Project (JoCo OA, a long-standing cohort study in a rural county about an hour from UNC). Potential patient stakeholders were contacted by the principal investigator, who provided details of the study and stakeholder panel role.

Methods, Modes, and Intensity of Engagement

Throughout the study period, we had monthly contact with the stakeholder panel. Engagement primarily involved the whole panel, but, as noted above, we met with patient stakeholders individually to allow them to voice their input on the study privately. The study team and stakeholder panel mutually agreed that phone calls would be held every other month, with the study team providing updates via email on alternate months. Additional group calls could be scheduled at any time, and study team members were available for individual calls or email-based discussions on specific issues.

Impact of Engagement on the Study

The following are key contributions of our stakeholder panel across various phases of the study:

Study Design

The study's design was a randomized controlled trial with participants assigned to 3 groups: standard PT, IBET, and WL control, with allocation of 2:2:1, respectively (Figure 1).²⁹ Randomization was stratified by enrollment source. During the full study period, all participants could receive other usual clinical care for OA outside the study setting.

Figure 1

Overview of Trial Design.

Forming the Study Cohort

Study inclusion criteria consisted of the following: (1) diagnosis of knee OA verified by a radiograph or self-report of a physician diagnosis, along with items based on the American College of Rheumatology clinical criteria for knee OA³⁰; and (2) self-report of pain, aching, or stiffness in 1 or both knees on most days of the week. Table 1 lists exclusion criteria. We used 2 recruitment methods. First, participants could self-refer in response to posted advertisements or brochures given by a health care provider. Second, we actively recruited patients were actively recruited based on UNC medical records and from the JoCo OA; in both cases, potentially eligible participants were identified from existing data (enrolled JoCo OA patients and patients with codes for OA in UNC medical records), sent an introductory letter, and then called to assess interest. All potential participants were screened for eligibility via telephone and then asked to meet a study team member to complete consent, HIPAA authorization, and baseline assessments. After completion of consent and all baseline assessments, participants were given their randomization assignment by the project coordinator via telephone. Reasons for nonparticipation are shown in the Results section and in Figure 2.

Table 1

Exclusion Criteria.

Figure 2

CONSORT Diagram.

Study Setting

Recruitment was conducted at UNC and in the surrounding communities, as well as in Johnston County, a more rural community. Study PT visits took place at 4 different clinics in 4 different towns (3 in and around UNC Chapel Hill and 1 in Johnston County); 2 clinics were affiliated with UNC and 2 were private/community based.

Interventions

This study compared 2 interventions that have a common focus on encouraging home-based exercise appropriate for knee OA, based on treatment guidelines. We designed the IBET program was designed to mirror aspects of PT visits for knee OA.²⁶ However, our pragmatic study did not seek to align the exact content between the 2 interventions. Rather, we compared the IBET tool to usual PT care for knee OA, as these 2 types of care are available options for individuals with knee OA.

PT Intervention

The PT intervention was modeled after recommended elements of care provided to patients with knee OA,³² including (1) evaluation of strength, flexibility, mobility, balance, function, knee alignment, and limb length inequality; (2) evaluation of the need for assistive devices, knee braces, patellar taping, heel lifts, shoe wedges, and other footwear modifications; (3) instruction in a home exercise program; (4) instruction in pacing daily activities and protecting joints; (5) manual therapy (manual force, ie, soft tissue massage, repeated passive joint glides, applied by the therapist to improve mobility of the tibiofemoral or patellofemoral joints and soft tissues around the knee), if appropriate; and (6) modalities for pain management, if appropriate. Emphasis was placed on the home exercise program, which was initiated at the first visit and addressed at all subsequent visits. Physical therapists often provided participants with handouts describing their assigned home exercise, but this was not required for every visit. Based on a typical range of outpatient PT visits for knee OA, study participants could receive up to 8 one-hour sessions. To mirror standard clinical practice, physical therapists were permitted to tailor visits to patients' needs and functional limitations. However, physical therapists were encouraged to utilize the full 8 visits whenever possible based on patient willingness to continue. Physical therapists at multiple clinics administered the intervention following training by PT co-investigators. Table 2 shows the general guidance given to study physical therapists for delivering the intervention. Physical therapist co-investigators observed visits for each study therapist to ensure fidelity and adherence to the study intervention, as outlined in Table 2.

Table 2

Guidance for Structure and Content of Physical Therapy Visits.

Follow-up

Research assistants blinded to randomization (via concealment in the database) conducted follow-up assessments 4 months (primary outcome time point) and 12 months after baseline. Four-month follow-up occurred immediately after the eligible window for all PT visits. The 12-month follow-up was considered a “maintenance” time point (because PT visits ended by 4 months), though the IBET group-maintained website access through 12 months.

Study Outcomes

Baseline study measures were conducted in person. Follow-up assessments were conducted in person, with allowance for phone-based assessment if participants could not attend in person. For all self-report measures, a trained interviewer read questions and response options to participants and then entered their responses into a database.

Data Collection and Sources

Participants were contacted by phone to schedule follow-up assessments; multiple attempts were made when needed, including calling at different times of day. We used several processes to maximize retention. First, the WL group was offered treatment after completing the final follow-up. Second, we provided compensation adequate to support participants' time and travel for study assessments. Third, we worked with 4 different clinics to deliver the PT intervention arm, minimizing participants' driving distances. Fourth, we made reasonable accommodations for participants to complete their baseline and follow-up assessments. If participants were unable to come to a study location for a follow-up visit, we allowed for telephone-based assessments to minimize loss to follow-up. Information on withdrawals, loss to follow-up, and exclusion was entered into the study database, which included automated functions to minimize and monitor for missing data.

Analytical and Statistical Approaches

For superiority hypotheses (H₁, H₃), we based our primary conclusions on intention-to-treat (ITT) analyses.⁵² For noninferiority hypotheses (H₂, H₄), the ITT analysis would not necessarily be the conservative approach, as this is thought to attenuate toward no between-group difference (which, in turn, could bias in favor noninferiority).⁵³ We therefore performed analyses on both an ITT and per-protocol basis^54,55; for the latter, we excluded individuals who did not adhere to their assigned study group, including those in the PT group (n = 9) who attended no visits, those in the IBET group who did not log in to the website (n = 28), and those in the IBET (n = 5) and WL (n = 4) groups who received PT outside the study during the initial 4-month period. We fitted a general linear mixed effects with changes from baseline in WOMAC scores as the dependent variables. We applied an unstructured covariance matrix to account for within-patient correlation between follow-up repeated measures. We included fixed effects for follow-up time (2 levels), intervention group (3 levels), their interaction, baseline WOMAC score, and enrollment source (3-level stratification variable). We used the SAS MIXED procedure to fit these models and test linear contrasts corresponding to each hypothesis. To test the null hypothesis of noninferiority of IBET vs standard PT in management of OA symptoms, we constructed the 95% CI of the appropriate linear contrast; we concluded noninferiority of IBET to PT at 4 months if the upper limit of the interval was less than the noninferiority margin of 5 points for WOMAC total.⁵⁵ We selected this margin because it is on the border of what would be considered a clinically important effect (approximately 12% difference) in this context.^56,57 Specifically, we expected a baseline WOMAC score of approximately 40,^58,59 so a 12% difference would be 4.8 points. Superiority hypotheses involved 2 comparisons vs WL control, so we conducted each at the 2-sided .025 significance level. The noninferiority hypotheses involve only 1 comparison, so we tested these at the full 1-sided .025 significance level that is appropriate for a noninferiority comparison. We used corresponding analytic strategies for most secondary outcomes, though for these outcomes we found insufficient information in the literature to define a noninferiority margin. Because the Global Assessment of Change variables do not have baseline values, we managed the actual values (rather than change from baseline) as the response variable, with no baseline score as a covariate.

For Aim 3 involving heterogeneity of treatment effects (HTE), we focused on 2 patient characteristics: age (categorized as <65 years, 64-74 years, and ≥75 years) and baseline functional status (categorized by quartile); we used ordinal variables for interpretability. We chose these characteristics based on the study team's experience with likely contributors to patient response to these different interventions, because we had little prior research to guide the selection of variables for theses analyses. We evaluated 3 objective function tests: 30-Second Chair Stand, 2-Minute March Test, and Timed Up and Go. Because of the distributions of scores for these function variables, we created categories based on the lowest quartile, a pooled category of the middle 2 quartiles, and the upper quartile. We adhered each patient characteristic to separate linear mixed effects models (described above), along with all 2-way and 3-way interactions with intervention group and time. A hypothesis test of the 3-way interaction of intervention group × time × patient characteristic determined whether we concluded evidence of HTE for that characteristic.

We also explored associations of number of PT visits attended or number of days the IBET website was accessed during the initial 4-month study period with changes in WOMAC scores and the 2-Minute March Test. For each outcome variable, a linear mixed model included the baseline level of the respective outcome variable, time in months, the engagement variable (number of PT visits or days on the website), and the interaction between time and the engagement variable. We also fitted a main effects model (no interaction term); the engagement variable term in these models reflected the homogeneous association between the applicable engagement variable and the respective outcome across follow-up time points.

We had several strategies for handling missing data. When individual items were missing from self-report scales, we followed guidelines regarding when to impute scores vs leave them missing⁶⁰; when guidelines were not available, we treated the scale as missing if >1 item was missing. When participants declined to complete or could not complete function tests, we assigned them the lowest value for that test; when participants ran out of time to complete function tests or when assessments were completed via telephone, we treated data as missing. For sensitivity analysis for the ITT approach, we performed a sensitivity analysis using a multiple-imputation approach for all primary and secondary outcomes that included additional exploratory variables beyond the explanatory variables included in our primary model (ie, follow-up time, intervention group, the corresponding baseline score, and enrollment source).

As a first step, we applied 2-sample independent-groups t tests for continuous baseline variables (ie, age at baseline, BMI, number of joints with arthritis, duration of arthritis symptoms, WOMAC total, and KOOS [Knee Injury and Osteoarthritis Outcome Score] Pain, Self-efficacy for Exercise, PHQ-8, and PASE scores) and chi-square tests for binary baseline variables (ie, gender, nonwhite race, working status, fair or poor health, married or living with partner, some education above high school, adequate income) to identify those baseline variables that differed between completers and noncompleters. We included any variable with a P value of .25 or less as an explanatory variable in the imputation model. We found the following variables to be associated with the missingness status at the P ≤ .25 level: gender, working status, adequate income, PASE, KOOS Pain, BMI, WOMAC total score, and duration of arthritis symptoms.

The imputation model additionally included variables for intervention group and the stratification variable nonwhite race. We used the fully conditional specification in SAS PROC MI; to impute missing values in continuous and binary variables, we used regression and logistic regression approaches, respectively. We generated imputations sequentially by specifying the imputation model for each variable given the set of other variables. Because race, income, and baseline WOMAC total score were missing for some participants, we imputed each of these characteristics first. Then, for all outcomes, we used all characteristics identified in the first step to impute the corresponding missing baseline score. Next, we imputed missing 4-month scores as a function of the baseline characteristics plus its baseline score. Finally, imputation of 12-month scores included the 4-months score, the baseline score, and the baseline characteristics.

Using this approach, we created 30 imputed data sets, with the number of burn-in iterations before each imputation set at 20. We conducted the analytic model on each of the 30 imputed data sets, and we synthesized these 30 sets of results using SAS PROC MIANALYZE to produce the multiply imputed version of the results.

Conduct of the Study

The full study protocol has been published previously.²⁹ Only minor modifications were made to the protocol, shown in Table 3. The IRB approvals are current as of this report's date of submission.

Table 3

Summary of Modifications Submitted to the IRBs.

Participants and Retention

We identified 11 274 potential participants (Figure 2). Of the 683 who completed the telephone screening, 350 (51%) were eligible, enrolled, and randomized. Among participants who were eligible but declined (n = 65), the following reasons were given: not interested (15), too busy (14), moving (2), planning joint replacement (2), transportation issues (2), poor health (14), unable to schedule (6), other (1), and no reason given (9). Because randomization was stratified by enrollment source, allocation across groups was slightly different than the 2:2:1 ratio: IBET = 142, PT = 140, and WL = 68. At 4-month follow-up, 87% of participants overall completed the primary outcome: 93% of PT group, 80% of IBET group, and 90% of WL group. At 12-month follow-up, 87% of participants overall completed the primary outcome: 92% of PT group, 79% of IBET group, and 93% of WL group (Figure 2). Compared with participants who completed follow-up assessments at 12 months, noncompleters had a higher (worse) baseline mean WOMAC total score (31.2, SD = 17.6 vs 37.6, SD = 19.1, respectively). Participant characteristics are shown in Table 4.

Table 4

Participant Characteristics at Baseline.

Adverse Events

The PT group experienced 4 nonserious study-related events (1 fall, 3 increased knee pain) and the IBET group experienced 4 (2 increased knee pain, 1 shoulder pain, 1 ankle pain). Table 5 shows serious adverse events; the investigators did not deem any to be study related.

Table 5

Adverse Events.

Intervention Delivery

Participants' use of the IBET website is summarized in Table 6. Between baseline and 4-month follow-up, 114 (80%) participants in the IBET group logged in to the website; the mean (SD) number of days logged in was 20.68 (24.62), median = 9.5; the difference between mean and median was primarily because of the skewed distribution and substantial number of 0 days. Between baseline and 12-month follow-up, 115 (81%) participants in the IBET group logged in to the website; the mean (SD) number of days logged in was 40.46 (59.81), median = 10.5. The mean number of days logged in to the website between 4-month and 12-month follow-up was 19.77 (37.72), median = 0. Seven physical therapists contributed to intervention delivery, with numbers of participants treated by each PT ranging from 2 to 40; this wide range was primarily because of participants' geographic proximity to the different study PT clinics. Among participants in the PT group, 131 (94%) attended at least 1 visit; 51% attended 6 to 8 visits. The mean (SD) number of visits was 5.72 (2.51), median = 7.0 visits. Between baseline and 4-month follow-up, 5 patients in IBET, 1 patient in PT, and 4 patients in WL reported that they received PT for knee OA outside the study. Corresponding numbers between baseline and 12 months were 8, 8, and 3, respectively.

Table 6

Descriptive Statistics of Use of IBET Website.

Primary Outcome: WOMAC Total Score

Because of the large number of analyses and results described regarding both primary and secondary outcomes, we have included Table 7, which summarizes the significant comparisons.

Table 7

Summary of Significant Between-Group Differences (P < .025) by Analysis Strategy and Time Point.

Intention-to-Treat Analyses

Superiority Hypotheses

Neither IBET nor PT were superior to WL at 4 months or 12 months at the specified P < .025 (Table 8; Figure 3). Multiple imputation analyses showed similar but slightly stronger effects for both interventions (Table 9), with the greatest difference between PT and WL at 4 months (estimate = −3.56; 95% CI, −7.16 to 0.04; P = .05).

Table 8

Within- and Between-Group Mean Changes in Outcomes and 95% CIs: Results of Intention-to-Treat Analyses.

Figure 3

Estimated Mean WOMAC Total Scores and 95% CIs by Group and Time Point.

Table 9

Within- and Between-Group Mean Changes in Outcomes and 95% CIs: Results of Intention-to-Treat Analyses With Multiple Imputation (N = 350).

Noninferiority Hypotheses

The upper limit of the 95% CI for IBET was within the prespecified noninferiority limit of 5 points on the WOMAC total score at both 4 months (estimate = 0.67; 95% CI, −2.23 to 3.56; P = .65) and 12 months (estimate = −1.04; 95% CI, −5.26 to 2.08; P = .39). See Figure 4.

Figure 4

Comparison of Changes in WOMAC Total Scores Between IBET and PT Group (Noninferiority Hypotheses): Results of ITT Analyses.

Per-Protocol Analyses

Per-protocol analyses yielded similar results (Table 10): At 4 months, the greatest difference was between PT and WL (−3.65; 95% CI, −7.34 to 0.03; P = .05). Differences between IBET and PT were within the prespecified noninferiority limit at 4 months (estimate = 1.3; 95% CI, −1.9 to 4.5; P = .43) and 12 months (estimate = −1.13; 95% CI, −4.49 to 2.23; P = .51).

Table 10

Within- and Between-Group Mean Changes in Outcomes and 95% CIs: Results of Per-Protocol Analyses.

Secondary Outcomes

Functional Tests

For both the Unilateral Stand Test and the 30-Second Chair Stand Test, no between-group differences occurred when using ITT (Table 8), multiple-imputation, or per-protocol analyses (Table 9 and Table 10). For the 2-Minute March Test, the largest difference was between the PT and WL groups at 4 months (ITT estimate = 7.75; 95% CI, 0.43-15.07; P = .04), favoring the PT group; using multiple imputation, this difference was 8.97 (95% CI, 1.68-16.26; P = .02). No between-group differences occurred in the Timed Up and Go (Table 11).

Table 11

Differences in Mean Changes Between IBET and PT and 95% CIs.

Process Measures

Heterogeneity of Treatment Effects (Aim 3)

In linear mixed effects models examining HTE based on age and physical function tests, no significant interactions (P > .05) occurred based on quartiles of age, 2-Minute March Test, or Timed Up and Go test; this indicates that treatment effects did not vary for these outcomes based on these participant characteristics at baseline. For analyses including interactions with baseline chair stand test, 1 significant 3-way interaction occurred; at 4-month follow-up, a significant interaction of intervention group with baseline chair stand score occurred (F = 5.59; P = .011). Specifically, for participants in the lowest (worst) of chair stand scores, the PT group had a substantially greater improvement than the other 2 groups; in contrast, changes in chair stand scores were similar across study groups for participants in the top 3 quartiles of chair stand scores at baseline.

Associations of Treatment Engagement With Outcomes

In the repeated-measures models for WOMAC total and function scores, interactions between the number of PT visits and time were not significant (P > .05; Table 12), indicating that associations between PT visits and these outcomes did not vary between follow-up time points; a greater number of PT visits was associated with greater improvement at follow-up (regardless of time point). Participants who attended 0 to 1 PT visits had increases in WOMAC total score, while those who attended 2 to 5 or 6 to 8 PT visits had decreases in WOMAC total score at both time points (Figure 5). For the 2-Minute March, the interaction between the number of PT visits and time was also not significant. A marginally significant (P = .05) association occurred between the number of PT visits and the increase in 2-Minute March Test score at follow-up, regardless of time point. Participants who attended 6 to 8 PT visits demonstrated the greatest improvements at follow-up (Figure 5). For WOMAC pain, the interaction between the number of PT visits and follow-up time was significant (P < .05). The slope was steeper at 4 months compared with 12 months (Figure 6), indicating a stronger association between number of PT visits and change in WOMAC pain score at 4 months than at 12 months.

Table 12

Results From Repeated-Measures Models for Outcomes With Number of PT Visits.

Figure 5

Model-Predicted Mean Changes In Outcome by Number of PT Visits Attended: WOMAC Total.

Figure 6

Associations Between Engagement With PT or IBET and Change in WOMAC Pain and Total Scores at 4- and 12-Month Follow-ups.

For the IBET group, in the repeated-measures model of WOMAC total scores, a nearly statistically significant interaction occurred between number of days on the website and follow-up visit (Table 13). The association between number of days on the IBET website and change in WOMAC total was steeper at 12 months than at 4 months (Figure 6). For WOMAC function and pain scores and the 2-Minute March, the interaction between number of days on the website and follow-up time was not significant (Table 13). Also, no significant associations occurred between number of days on the website and change in the outcomes over time in general.

Table 13

Results From Repeated-Measures Models for Outcomes With Number of Days on IBET Website.

Context for Study Results

This project aimed to provide evidence to support decision making for both patients with knee OA and health care providers by comparing the effectiveness of 2 exercise-based interventions—PT and IBET—both compared with a WL control group. In particular, because access to PT (an evidence-based treatment) is limited for some patients, this study aimed to assess whether IBET may be a similarly effective alternative that could be disseminated at relatively low cost. Results of this study showed that for the primary outcome, a self-report measure of pain, stiffness, and function (WOMAC), improvements were comparable for the IBET and PT group. However, results also showed that neither the PT nor IBET group improved significantly or clinically more in WOMAC score than the WL group.³⁵ However, for several secondary outcomes that are likely important from the patient perspective (including sleep, fatigue, satisfaction with function, and global assessment of knee symptom change), the PT group did show significantly greater improvement than the WL group at 1 or both follow-up time points. The mechanism(s) underlying these changes in secondary outcomes for the PT groups are unclear, given the lack of significant change in WOMAC scores.

Given prior studies on the effectiveness of exercise and PT care for knee OA,^61-63 it is unclear why the PT intervention was not statistically superior to WL for most outcomes. However, prior studies have included a wide range of intervention types and have been heterogeneous in terms of in dose (eg, number and duration of sessions), duration.⁶² This makes comparison across studies challenging. However, the results of this study can be considered in the context of the range of prior interventions in this area. A meta-analysis of exercise- and PT-related interventions for knee OA found that, with respect to pain, standardized mean differences (SMDs) were −0.21 (−0.35, −0.08) and −0.69 (−1.24, −0.14) for programs that focus on aerobic and strengthening exercise, respectively.⁶² The SMD for pain immediately following our PT intervention was smaller than these, −0.14 (−0.39, 0.11), and declined at 12 months, −0.02 (−0.31, 0.27). The meta-analysis found that for disability/function, SMDs were −0.21 (−0.37, – 0.04) and −0.16 (−0.48, −0.16), for programs that focus on aerobic and strengthening exercise, respectively. The SMD for function immediately after our PT intervention was somewhat larger than these, −0.27 (−0.52, −0.01), but declined to −0.19 (−0.45, 0.08) at 12 months. Therefore, our PT intervention was comparable to prior PT-related studies regarding function but less effective for pain. Overall, though, pooled effect sizes from the meta-analysis, as well as our study, are considered relatively small. One recent study compared PT with Tai Chi among individuals with knee OA.⁶⁴ The PT intervention was similar to the PT intervention in our study but involved more planned visits. This study used a different version of the WOMAC than our study (0-100 scale for each item, with subscale scores summing across all items); therefore, direct comparison between the studies should be interpreted with caution. However, we converted our WOMAC subscale scores to the same metric and found that changes in the PT arm immediately following the intervention were smaller for our study, for both WOMAC pain and function scales. Specifically, WOMAC pain scores decreased 143 points in the other study and 28 points in our study; WOMAC function sores decreased 456 points in the other study and 119 points in our study. We hypothesize that the greater effects observed in the other study may result from greater dose or intensity, because it involved more planned visits, and phone-based follow-up following the in-person visits.

Other recent meta-analyses of OA studies indicate that exercise-based interventions adhering to American College of Sports Medicine (ACSM) dose recommendations (eg, specific duration, intensity and frequency of cardiorespiratory, strengthening and stretching exercises) resulted in larger improvements in pain, function, and strength than those not adhering to those recommendations.^63,65 The PT intervention emphasized stretching and strengthening exercise, with less focus on cardiorespiratory exercise. In addition, we did not specify a minimum intensity for strengthening exercises; ACSM guidelines recommend 1 to 3 sets of 8 to 12 repetitions at 60% to 70% of 1 repetition maximum of a given exercise (for increasing muscular strength). Effects of the PT intervention may have been stronger if therapists had been given a more specific protocol regarding intensity of the home exercise program. Future studies should consider identifying strategies to help patients achieve ACSM exercise dose recommendations in the context of a routine course of PT for knee OA.

Exploratory analyses in this study showed that participants who attended more PT visits showed a greater improvement in study outcomes, including the WOMAC total score and 2-Minute March Test (Figure 5). Further, these changes persisted at 12-month follow-up, 8 months after PT sessions ended. These findings lend support to the notion that a greater dose of PT may be important and signal that further rigorous research is needed in this area. Unfortunately, we did not collect standardized information on reasons patients did not complete the maximum number of PT sessions allowed in this study. In real-world clinical settings, patients may not complete a full “course” of PT for various reasons, including perceived lack of benefit, difficulty of affording copayments, or time and travel requirements to attend PT sessions. Future studies should aim to determine both the optimal dose of PT for knee OA and how best to support patients in engaging maximally during a course of PT.

Little research exists on internet-based exercise programs for OA. However, a Netherlands-based study found that individuals with hip and knee OA who participated in an internet-based exercise program improved from 58.8 to 67.8 on the Knee Injury and Osteoarthritis Outcome Scale after 3 months, which was significantly better than a control group.⁶⁶ Participants in that study self-referred in response to an advertisement, which also may have resulted in a more activated sample than in our trial. Taken together, results of these studies suggest that internet-based exercise programs could yield clinically meaningful improvement for individuals with knee OA, but results may vary based on the patient sample and degree of self-motivation to participate. Also, as noted above, effects of these types of interventions may be greater if the interventions adhere to ACSM guidelines. The IBET intervention did encourage daily stretching and strengthening 2 to 3 times per week, which is in accordance with those guidelines. The IBET program also recommends progressively challenging exercises for participants. However, the intensity of strengthening exercises could have fallen below ACSM guidelines for improvements in muscular strength.

Implementation of Study Results

The PT intervention examined in this study was modeled after standard care; therefore, this intervention is already implemented. However, exploratory results regarding association of numbers of PT visits and outcomes provide practical guidance for implementation. No specific recommendations exist regarding the optimal number of outpatient PT visits for individuals with knee OA. Results of this study suggest that patients and clinicians may want to consider providing at least 6 PT visits (because patients in this study who had 6 or more visits showed the most improvement [Figure 5]), although additional research is needed to support clinical decisions in this area. A member of our stakeholder panel is on the leadership of the American Physical Therapy Association and will guide the study team on appropriate ways to disseminate these study results to professionals in that clinical area. We did not experience substantial barriers in participant attendance at PT visits. However, some participants attended no visits, despite repeated attempts at phone contact and scheduling. In real-world clinical settings, patient nonattendance at PT visits tends to be an even greater challenge, partly because patients often have copayments for PT visits, which may be a challenge for some to afford (visit costs were covered for our study participants). Research is needed on strategies to best engage patients in PT visits, identify and solve barriers, and encourage attendance throughout the treatment course.

The specific IBET program studied in this project is proprietary, and widespread implementation/availability of the program is in that company's purview. However, we have worked closely with the company during this project and will discuss implementation possibilities with their leadership. We will also work with stakeholder panel members from the Centers for Disease Control and Prevention Arthritis Program and the Arthritis Foundation regarding key messages to disseminate to patients and clinicians about the potential effectiveness of internet-based exercise programs for people with knee OA in general. A main challenge we experienced with the IBET program was that 20% of participants never logged on to the website, and among those who did, the number of days logged on was low. The extent to which participants performed their home exercises even on days they did not complete their Patient Health Questionnaire on the website is not known. However, this challenge highlights a need for more work to optimize internet-based exercise to best foster engagement among patients with OA. Strategies such as daily diaries may be a simple approach to enhance use.

Generalizability

We chose study inclusion and exclusion criteria to reflect as representative of a sample of individuals with knee OA as possible. We limited exclusion criteria to health conditions that would make home exercise unsafe or that would confound study outcomes. The study sample was about 72% female, which may limit generalizability somewhat for males. However, knee OA is more common in women, particularly after age 50. Although about 60% of participants had at least a bachelor's degree, a substantial proportion of individuals had less education, which supports the generalizability of the findings. About 27% of the study sample was nonwhite, but racial minority participants were almost all African American, so generalizability for other racial and ethnic groups is limited. This study included only individuals who had regular internet access, which also may affect generalizability.

Subpopulation Considerations

Analyses completed for Aim 3 of this project showed very little difference in treatment effects based on participant age group or performance on physical function tests at baseline. One exception was that for participants with the poorest performance on the chair stand test at baseline, those in the PT group showed greater improvement than those in the IBET or WL groups; this same pattern did not emerge for those with better chair stand performance at baseline. These results suggest that participants with poorer function may benefit most from a PT intervention. However, these results should be viewed as exploratory, particularly because other measures of function did not exhibit this same pattern in this study. These results illustrate the need for more focused, rigorous research to identify the most appropriate exercise-based interventions for specific patients with knee OA.

Study Limitations

This study had several limitations. First, we did not confirm an OA diagnosis with standardized de novo radiographs or independent physician assessments. However, all participants had either a prior radiographic or physician diagnosis of OA, so it is unlikely that participants had neither radiographic or symptomatic OA. Second, self-reported physical activity is often overreported. However, it is unlikely that this differed among study groups. Third, home exercise was not assessed, and we included no objective measure of physical activity (eg, accelerometry). Fourth, global assessment of change (secondary outcome) has limitations in accuracy and may be biased when asking participants to consider change over a long period of time; this may be a particular limitation for our 12-month time point. Fifth, exploratory HTE analyses were limited to 2 variables because of limitations in power. Sixth, we did not ascertain whether patients were currently seeking care for OA. Seventh, we had slightly higher attrition at 4-month follow-up than was anticipated in our sample size/power analysis (13% vs 10%). However, we calculated that, even with this attrition rate, we still had 79% power (in the context of other original assumptions) to detect the specified between-group differences for stated hypotheses. We also acknowledge that the per-protocol analyses had less power because of the smaller sample size; we calculated a 73% power for these comparisons.

Future Research

A key area for future research is understanding whether and how the effects of specific exercise-based treatments can be augmented. For PT, effects tend to wane over time following a course of treatment, and strategies including booster visits or mobile health interventions may help patients to maintain effects over time.⁶⁷ For the IBET intervention, engagement with the website was low in this study; engagement has varied across other studies of web-based interventions for OA.^26,68 A key area of future research is examining whether features such as text reminders or peer support components may bolster use.