Level of participation in physical therapy or an internet-based exercise training program: associations with outcomes for patients with knee osteoarthritis

This study included participants from a study of Physical THerapy vs. INternet-Based Exercise Training for Patients with Knee Osteoarthritis (PATH-IN; NCT02312713); details of the study protocol and interventions have been previously reported [16]. PATH-IN participants were randomly assigned to one of three groups: PT, IBET, or wait list (WL) control. The PT intervention was modeled after standard care for patients with knee OA, with an emphasis on active interventions and a home exercise program; the intervention was delivered by physical therapists in multiple clinics. Based on a typical range of outpatient PT visits for knee OA, participants could receive up to 8 one-hour sessions during the initial 4 months of the study. Participants in the IBET group were given access to the website and encouraged to log on immediately after enrollment and to continue use of the program as often as possible throughout the study period. The IBET website provided an initial personalized exercise program, ongoing tailoring of exercise to facilitate appropriate progression and videos to demonstrate proper performance of stretching and strengthening exercises [17]. The website also sent reminders to participants via email after periods of not logging in.

Participants were identified through the University of North Carolina at Chapel Hill (UNC) and the Johnston County Osteoarthritis Project. All participants had a diagnosis of knee OA along with current joint symptoms. For the present study, only participants enrolled in either the PT (N = 140) or IBET (N = 142) groups were evaluated, albeit separately. Analyses were further restricted to participants who remained in the study at the 4-month assessment point, as these individuals had opportunity to fully participate in the PT intervention or had access to the IBET up to that time point. Five participants from the PT group were excluded because of developing health issues. In the IBET group, 12 participants withdrew and 6 participants were excluded due to health issues. Therefore, analyses included 135 participants from the PT group and 124 participants from the IBET group. This research is in compliance with the Helsinki Declaration and was approved by the Institutional Review Boards of UNC and Duke University Medical Center.

The mean age of participants was 64.9 years (standard deviation (SD) = 10.9) and 70.7% were female. Fewer than half of the participants (40.9%) were employed at the start of the study. Of the participants within the IBET group one-third (36.3%) were of non-white race, whereas only one-fifth (21.1%) of participants within the PT group identified with this classification. Additional participant characteristics are displayed in Table 1.

The mean number of PT visits was 5.7 (SD = 2.5), with a median of 7 visits. In the repeated measures models for WOMAC total and function scores, the interactions between the number of PT visits and time were not significant (p > 0.05), indicating that the association between PT visits and the respective outcome did not vary significantly between the two follow-up time points. For both WOMAC total and function scores, a greater number of PT visits was associated with greater improvement (decreased score) at follow-up (Table 2). Figure 1 shows the mean WOMAC total scores based on number of PT visits attended. Participants who attended 0–1 PT visits had increases in WOMAC total score, while those who attended 2–5 or 6–8 PT visits had decreases in WOMAC total score at both time points, with those in the 6–8 visit group experiencing the greatest improvement. For the 2-min Step Test, the interaction between the number of PT visits and time was not significant. There was a marginally significant (p = 0.05) association between the number of PT visits and the increase in 2-min Step Test score at follow-up, regardless of time point. A similar pattern was observed for 2-min Step Test score as compared to WOMAC total score: participants who attended 6–8 PT visits demonstrated the greatest improvements at the 4- and 12-month follow-ups (Fig. 1). In regard to WOMAC pain, the interaction between the number of PT visits and follow-up time was significant (p <  0.05). As shown in Fig. 2, the slope was steeper at 4 months compared to 12 months, indicating a stronger association between the number of PT visits and change in WOMAC pain score at 4 months when compared to 12 months.

During the first 4 months, the mean number of days participants logged onto the IBET website was 20.7 (SD = 24.6), median = 9.5. During the full 12-month period, the mean number of days participants logged on was 40.5 (SD = 59.8), median = 10.5. In the repeated measures model of WOMAC total scores, there was a nearly statistically significant interaction between number of days on the website and follow-up visit (estimate per day on website = − 0.008, 95% CI = − 0.02 – 0.00, p = 0.06; Table 3). Plots of this interaction showed that the association between number of days on the IBET website and change in WOMAC total score was steeper at 12 months than at 4 months (Fig. 2); this indicates a stronger association at 12-month follow-up. For WOMAC function and pain scores, the interaction between number of days on the website and follow-up time was not significant. However, there were trends for the number of days on the website being associated with improvement in WOMAC function and pain score over time (p = 0.07 and 0.10, respectively) (Table 3). For the 2-min Step Test, the interaction between number of days on the website and time was not significant, nor was there an association between number of days on the website and change in the outcome over time in general.

In bivariate analyses, the following characteristics were associated (p < 0.10) with greater number of PT visits attended: older age, non-white race, fair or poor self-rated health, and higher SEE; BMI and WOMAC pain were near the p < 0.10 threshold and were considered potentially important explanatory variables for number of PT visits (Table 4). In (Table 4). In multivariable analysis, only age was significantly associated (p < 0.05) with number of PT visits; specifically, older age was associated with greater number of PT visits attended (Table 4). When bivariate analyses were completed for the IBET intervention, there were no characteristics significantly associated with engagement to the IBET intervention. Since no variables met the criterion, a multivariable model was not completed.

In bivariate analyses of the PT group, the following characteristics were associated (p <  0.15) with greater activity on the PASE Leisure Activity subscale: male gender, white race, greater education, lower BMI, better performance on the 2-min Step Test, and higher SEE (Table 5). In multivariable models among the PT group, greater education (estimate = 0.42, 95% CI = 0.07, 0.77, p = 0.02) and greater SEE (estimate = 0.01, 95% CI = 0.003, 0.018, p <  0.01) were associated with more activity on the PASE Leisure Activity subcale. In bivariate analyses for the IBET group, the following characteristics were associated (p <  0.15) with greater activity on the PASE Leisure Activity subscale: lower BMI, better scores on the WOMAC function subscale, better performance on the 2-min Step Test, higher social support for exercise, and higher SEE. In multivariable models among the IBET group, better performance on the 2-min Step Test (estimate = 0.007, 95%CI = 0.002, 0.013, p = 0.012), greater social support for exercise (estimate = 0.016, 95%CI = 0.008, 0.024, p < 0.001) and greater SEE (estimate = 0.010, 95% CI = 0.001, 0.018, p = 0.024) were associated with higher scores on the PASE Leisure Activity subscale. In bivariate analyses of the WL group, the following characteristics were significantly associated (p < 0.15) with greater activity on the PASE Leisure Activity subscale: male gender, currently working, lower BMI, better performance on the 2-min Step Test, and higher SEE; in multivariable models among the WL group, none of these variables were significantly associated (p < 0.05) with the PASE Leisure Activity subscale.

In this study, participants who completed a greater number of PT sessions had overall better outcomes at follow-up. For one of our outcomes, WOMAC pain, greater number of PT visits had a stronger impact immediately following the completion of PT (4 months) than at a later time point. Importantly, however, for several outcomes (WOMAC total and function, 2-min Step Test), the association between number of PT visits and changes in outcomes did not differ significantly between the immediate post-treatment time point (4 months) and a later time point following an 8-month period of no study-delivered PT. The latter suggests that there may be a lasting, positive impact associated with receiving more PT visits on some outcomes. These results complement and expand on those of prior studies on exercise adherence and outcomes among patients with OA [8‐10]. van Gool et al. also found that greater exercise adherence (comprised of class attendance and home exercise) had an impact on longer term (18-month) outcomes [10]. However, patients in that intervention continued to receive some support (either in-person or via telephone) throughout the 18-month period. Our study additionally suggests that even after a formal exercise-based intervention (specifically PT) has ended, the degree of earlier use of an intervention may predict longer term outcomes. Pisters et al. (2010) found that following discharge from a PT intervention, patients with greater adherence to prescribed home exercise experienced better outcomes [9]. It is possible that participants in our study who had more PT visits also adhered more to recommended home exercise and exercise-based therapies after completing PT, though this was not directly measured.

To date there are no established guidelines for the optimal number of outpatient PT visits for participants with knee OA [23‐25]. Our results may provide information to help address this important question, which is highly relevant to both physical therapists and policy makers. In our analysis, participants who attended 0 or 1 PT visit experienced a worsening of their overall symptoms. In contrast, those with 2–5 and 6–8 visits showed clinically relevant improvement. Specifically, individuals with 2–5 visits had 18 and 11% improvement in WOMAC score at 4-month and 12-month follow-up, respectively. Individuals who attended 6–8 PT visits experienced even greater improvement: 28 and 21% at 4-month and 12-month follow-up, respectively. Prior research indicates that a 12% change in total WOMAC score represents a clinically relevant improvement in the context of this type of intervention [26]. These results suggest there may be added clinical benefit, even at 12-month follow-up, for providing 6–8 PT visits. It is likely that the optimal number of PT visits varies across patients, based on complexity, functional limitations and other factors. For some patients, it is possible that more than 8 visits would be beneficial.

The second main finding of this study was that although there was a trend for participants who logged into the website more frequently to have somewhat greater improvements in WOMAC total score, this association was small and not statistically significant. Therefore, associations of intervention participation level with outcomes differed between the supervised program (PT) and the self-directed program (IBET). One reason for the lack of association in the IBET group may be that use of the website overall was relatively low. Specifically, 28 of 142 participants never logged in, and the average (standard deviation) number of days participants logged in during the initial 4-month intervention period was 20.68 (24.62). Use of internet exercise interventions has varied across studies, with some showing higher rates than ours. Two studies that used similar web-based interventions for patients with arthritis had a greater number of logins, on average, as compared to our study [17, 27]. Although our study did not find greater use of the website to be associated with better outcomes, studies in other patient groups have found a positive association [28‐30]. For example, a retrospective study completed by Hwang et al., found that participants who logged into a weight-loss website at least 4 times over a thirty-day period had significantly more weight loss than those whom had fewer logins [31]. Given the increasing use of internet-based and mobile health platforms to deliver health-related interventions, additional research is needed to identify best strategies for maximizing use of these types of interventions.

A third main finding of this study was that only one participant characteristic, older age, predicted more PT visits in adjusted analyses. It is possible that older participants had more flexible schedules that allowed them to attend the PT visits more easily, though we did control for working status. No participant characteristics were associated with amount of use of the IBET program. Therefore, this is another difference between the supervised program (PT) and the self-directed program (IBET). This lack of association for the IBET group could be due to the relatively low levels of engagement with the website, as described above. However, these results suggest that participants with a variety of demographic and personal characteristics may be similarly likely to engage with the website.

Greater baseline exercise self-efficacy was associated with more improvement in PASE Leisure Time Activity subscale scores, for both PT and IBET groups. This aligns with prior research regarding the important association of self-efficacy with adherence to exercise [11, 21, 32]. There were some differences in other predictors of change in PASE Leisure Time Activity scores across study groups. Although this may indicate that different patient characteristics are important for predicting improvement in activity for supervised (PT) vs. self-directed (IBET) programs, results should be considered in light of the exploratory nature of these analyses. In the WL group, no participant characteristics were associated with change in PASE Leisure Time Activity in multivariable analyses.

There are several limitations to this study. Although this study focused on important metrics of participation in exercise-based interventions, other components such as adherence to home exercise were not measured; this is an important area for future study. The interventions provided in this study were freely available to participants. Therefore, the degree of participation or factors associated with particiaption apply to situations when financial costs and co-payments are not barriers; in real-world settings, these are likely to be considerations for some patients and may affect participation levels. We purposefully chose to include in these analyses only participants who remained in the study until the 4-month follow-up, since they had opportunity to fully participate in the interventions during this time period. We acknowledge that this strategy omits some participants who never engaged with the interventions for a variety of reasons. We did not systematically ascertain reasons participants discontinued PT visits before the maxium allowed number, which may have included discomfort or lack of perceived benefit. Although this study was longitudinal in nature, there are still limitations to inferring that a greater number of visits caused greater improvement in outcome. Specifically, it is possible that patients who experienced more improvement in symptoms during the course of therapy were motivated to continue attending PT visits. We did not obtain de novo radiographs on participants, though all had a previous physician diagnosis and / or prior radiographic verification of knee OA. For our analyses of participant characteristics associated with treatment use, we were limited in the numbers of participant characteristics that could be included in multivariable models due to sample sizes. However, since few participant characteristics met our criterion in the bivariate analyses for inclusion in the multivariable model, this was not considered a major limitation. Finally, these were exploratory analyses with a large number of comparisons, and results should be interpreted with this in mind.