Annals of Internal Medicine ORIGINAL RESEARCH

Effectiveness of an Unsupervised Online Yoga Program on Pain and

Function in People With Knee Osteoarthritis
A Randomized Clinical Trial
Kim L. Bennell, PhD; Sarah Schwartz, MScPT; Pek Ling Teo, PhD; Stephanie Hawkins, MPH; Dave Mackenzie, MScPT;
Fiona McManus, MPH, MBiostat; Karen E. Lamb, PhD; Alexander J. Kimp, DPT; Ben Metcalf, BSc; David J. Hunter, PhD; and
Rana S. Hinman, PhD

Background: Yoga is a mind–body exercise typically done quality of life, self-efficacy, fear of movement, and balance
in groups in person, but this delivery method can be inconven- confidence. Adverse events were also collected.
ient, inaccessible, and costly. Effective online programs may
increase access to exercise for knee osteoarthritis. Results: A total of 195 (92%) and 189 (89%) participants pro-
vided 12- and 24-week primary outcomes, respectively. Compared
Objective: To evaluate the effectiveness of an unsupervised with control at 12 weeks, yoga improved function (between-
12-week online yoga program. group mean difference in change, 4.0 [95% CI, 6.8 to 1.3])
but not knee pain during walking (between-group mean differ-
Design: Two-group superiority randomized trial. (Australian ence in change, 0.6 [CI, 1.2 to 0.1]), with more yoga partici-
New Zealand Clinical Trials Registry: ACTRN12620000012976) pants than control participants achieving the minimal clinically
Setting: Community. important difference (MCID) for both outcomes. At 12 weeks,
knee stiffness, quality of life, and arthritis self-efficacy improved
Participants: 212 adults with symptomatic knee osteoarthritis. more with yoga than the control intervention. Benefits were
not maintained at 24 weeks. Adverse events were minor.
Intervention: Both groups received online osteoarthritis in-
formation (control). The yoga group also received access to an Limitation: Participants were unblinded.
unsupervised online yoga program delivered via prerecorded
Conclusion: Compared with online education, an unsuper-
videos over 12 weeks (1 video per week, with each session to
vised online yoga program improved physical function but
be performed 3 times per week), with optional continuation
not knee pain at 12 weeks in people with knee osteoarthritis,
thereafter. although the improvement did not reach the MCID and was
Measurements: Primary outcomes were changes in knee pain not sustained at 24 weeks.
during walking (0 to 10 on a numerical rating scale) and physi- Primary Funding Source: National Health and Medical
cal function (0 to 68 on the Western Ontario and McMaster Research Council and Centres of Research Excellence.
Universities Osteoarthritis Index) at 12 weeks (primary time point)
and 24 weeks, analyzed using mixed-effects linear regression Ann Intern Med. 2022;175:1345-1355. doi:10.7326/M22-1761 Annals.org
models. Secondary outcomes were self-reported overall knee For author, article, and disclosure information, see end of text.
pain, stiffness, depression, anxiety, stress, global change, This article was published at Annals.org on 20 September 2022.

K nee osteoarthritis is a major contributor to global

disability (1). Pain and reduced function, together
with mood disorders, including depression and anxiety,
People typically access yoga in person by attending
a facility for supervised group classes (12). Limited
low-quality research in knee osteoarthritis shows some
are commonly reported by people with knee osteoarthri- benefits on pain, function, and knee stiffness from yoga
tis (2). Impaired muscle strength, mobility, and balance programs delivered in this manner (13). However, such
can also increase fall risk (3, 4), with about a quarter of programs can be inaccessible or inconvenient for peo-
people with knee osteoarthritis sustaining 1 or more falls ple to attend and often involve a cost to participate (14).
over 1 year (5). With an aging population, identifying Increasingly, people are accessing health and exercise
effective, conservative treatments that are widely accessi- services and resources online (15), driven particularly by
ble and can help people self-manage their knee osteoar- the global COVID-19 pandemic. Online interventions
thritis is an important priority (6). have the potential to reach a wide audience at little or no
Exercise can improve pain, function, muscle strength, cost regardless of a person's location, allowing exercise
and quality of life in people with knee osteoarthritis (7, 8) to be undertaken from the comfort and convenience of
and is universally recommended by clinical guidelines as
a core treatment (9, 10). Yoga is an increasingly popular
(11), low-impact mind–body exercise requiring minimal See also:
equipment. Its combination of static and dynamic pos-
Summary for Patients . . . . . . . . . . . . . . . . . . . . . . . . I-32
tures with mindfulness strategies, such as deep breathing
and relaxation, may help people living with knee osteoar- Web-Only
thritis by addressing both the physical and psychological Supplement
sequelae of their condition.
© 2022 American College of Physicians 1345
ORIGINAL RESEARCH Effectiveness of an Online Yoga Program in People With Knee Osteoarthritis

home and thereby overcoming some of the barriers to the affected side; 4) systemic or inflammatory arthritic
exercise reported in the osteoarthritis population (16). conditions; 5) regular (at least weekly) exercise (home-
Limited studies of online yoga programs have been done based leg strengthening exercises, yoga, or gym or
in other populations, such as caregivers, people with can- group exercise classes) for the previous 3 months; 6)
cer, and people with mood disorders, with results show- inability to commit to study requirements; or 7) inabil-
ing acceptability and feasibility (17) and generally positive ity to walk unaided.
outcomes (18–21). To date, an online yoga program Clearance to participate from a general practitioner
specifically for people with knee osteoarthritis has not was required for those who reported a fall in the previous
been investigated. The need for such evidence-based 12 months, those who were housebound due to immo-
packaged online exercise programs is highlighted bility, or those who indicated they had a health condition
in the 2020 U.S. National Public Health Agenda for that might compromise exercise safety on the Exercise &
Osteoarthritis (22). Sports Science Australia stage 1 preexercise screening
We aimed to evaluate the effectiveness of an unsuper- questionnaire (25).
vised 12-week online yoga program (“My Joint Yoga”)
plus online education in people with knee osteoarthritis Randomization and Interventions
compared with a control intervention of online education An independent biostatistician prepared a computer-
only. We hypothesized that improvements in the primary generated randomization schedule with permuted blocks
outcomes of knee pain during walking and physical func- of varying sizes. Password-protected software (REDCap)
tion would be greater with yoga than with the control. concealed allocation, which was revealed by a researcher
not involved in recruitment or outcome assessment.
Study interventions were disclosed to participants, but
METHODS not hypotheses. A statistical analysis plan (Supplement 1)
Design Overview was written and published on our website before data
This randomized controlled trial had a pragmatic analysis and while investigators were blinded to group
2-group superiority parallel design (participant-level ran- allocation. A biostatistician analyzed data while blinded to
domization, 1:1 allocation ratio) with unblinded participants group details.
(also deemed assessors given self-reported outcomes).
It was approved by the Institutional Human Research Control
Ethics Committee (#1852823) and registered prospec- Control participants were given access to a custom-
tively with the Australian New Zealand Clinical Trials ized trial website containing downloadable educational
Registry (ACTRN12620000012976). Participants signed material that covered understanding osteoarthritis, treat-
an electronic consent form in REDCap. Enrollment occurred ment options, exercise and physical activity, weight loss,
between 1 May 2020 and 28 June 2021, with follow-up understanding and managing pain, sleep, and patient
completed on 16 December 2021. There were no major stories. The website could be accessed as many times as
changes to trial methods (Supplement 1 and Table 1 of the participant wished over the 24-week trial.
Supplement 2, available at Annals.org). Findings are
reported according to the CLARIFY (CheckList stAndardising Yoga
the Reporting of Interventions For Yoga) guidelines (23). Yoga participants were given access to a different
customized trial website that contained online educa-
Setting and Participants tional material identical to that in the control intervention
The trial was coordinated from a single university but also included a self-directed, unsupervised 12-week
research center, with interventions accessed online. We progressive yoga program (https://myjointyoga.com.au).
recruited 212 participants from the Australian community The program was delivered from the website as a series
via advertisements, medical practitioners, social media of 12 different prerecorded 30-minute videos (1 video
(Facebook), print and radio media, our volunteer data- per week, with each session to be performed 3 times per
base, and Trialfacts (a patient recruitment agency). Eligibility week) of a yoga class with an instructor (who was a physi-
was confirmed via an online survey and a telephone otherapist and yoga teacher) and 3 consumers acting as
interview with study personnel. class members.
Inclusion criteria for participants were 1) meeting the The program (Table 2 of Supplement 2) was designed
National Institute for Health and Care Excellence osteoar- by the researchers and a panel comprising 5 yoga thera-
thritis clinical criteria (age ≥45 years, activity-related knee pists (registered with the International Association of
pain, and no knee morning stiffness lasting ≥30 minutes) Yoga Therapists or the Australasian Association of Yoga
(24), 2) knee pain on most days of the previous month, 3) Therapists), 2 people with knee osteoarthritis, and a phys-
knee pain for at least 3 months, 4) minimum average iotherapist with fitness leader qualifications and expertise
walking pain score of 4 on an 11-point numerical rating in teaching yoga. The program incorporated elements of
scale (NRS) over the previous week (0 = “no pain”, 10 = Hatha yoga, which involves slow-paced physical postures
“worst pain possible”), and 5) internet access. Exclusion and poses with a focus on breathing, controlled move-
criteria were 1) inability to speak English; 2) knee surgery ments, and stretching. In contrast, Vinyasa yoga is faster
or joint injection in the previous 6 months, scheduled and repeats specific sequences, and Iyengar yoga uses
appointment with an orthopedic surgeon, or planned props to emphasize alignment. Each session included a
surgery in the next 6 months; 3) knee replacement on warm-up consisting of an introduction by the instructor,
1346 Annals of Internal Medicine • Vol. 175 No. 10 • October 2022 Annals.org
Effectiveness of an Online Yoga Program in People With Knee Osteoarthritis ORIGINAL RESEARCH
breathing exercises, and yoga postures to warm up the the 12-week program. Upon completing the program,
core and lower-extremity muscles (5 minutes). The bulk participants were informed that continued participation
of each session featured static and dynamic yoga pos- with the yoga postures was recommended but optional.
tures intended to activate, strengthen, and stretch core
and lower-extremity muscles (20 minutes). Each session Outcomes and Follow-up
ended with a cooldown of stretches, breathing practices, Outcomes were self-reported using online question-
and relaxation exercises (5 minutes). The sessions con- naires at baseline and 12 weeks (primary time point) and
sisted of a core group of repeating postures, with new 24 weeks after randomization. For bilaterally eligible knees,
postures added throughout the 12-week progression. participants completed questionnaires about their more
The video class instructor demonstrated various modifi- symptomatic knee at baseline. Participants who completed
cations and levels for each posture so that it was safe and 24-week questionnaires were entered into a drawing to
feasible for all participants. The program incorporated win an Apple iPad.
behavior change techniques to encourage adherence Primary outcomes were 12-week changes in widely
(26, 27) (Table 3 of Supplement 2), including weekly re- used, reliable, and valid pain (28) and physical function (29)
minder and motivational e-mails to participants during measures that are recommended for knee osteoarthritis

Figure. Participant flow through the trial.

Assessed for eligibility via online Excluded (n = 1761)

questionnaire and/or telephone Met exclusion criteria (n = 974)
(n = 2004) Regular exercise for previous 3 mo: 540
Knee surgery/joint injection in previous 6 mo, current
appointment with surgeon, or planned surgery in next
6 mo: 202
Systemic/inflammatory arthritis: 147
Recent total knee replacement: 51
Unable to walk unaided: 18
Unable to commit to study: 15
Unable to speak English: 1
Did not meet inclusion criteria (n = 787)
Average pain during walking <4 on NRS*: 237
Morning knee stiffness for ≥30 min: 211
Not interested/not able to contact: 152
No knee pain on most days of previous month: 65
No difficulty in walking/climbing stairs: 25
No GP clearance to participate in exercise: 24†
Other condition affecting ability to exercise: 19
No activity-related knee pain: 19
Age <45 y: 15
Knee pain <3 mo: 15
No internet access: 5
Baseline assessment (n = 243) Excluded (n = 31)
Met exclusion criterion (did not complete consent/
baseline): 25
Did not meet inclusion criterion (average pain during
walking <4 on NRS*): 6
Randomly assigned (n = 212)

Control group (online education only) (n = 105) Yoga group (12-wk online yoga program plus
online education) (n = 107)

Provided both 12-wk primary outcomes (n = 97) Provided both 12-wk primary outcomes (n = 98)
Unable to contact: 5 Unable to contact: 8
Withdrew from study (1 lost interest, 2 reason Withdrew from study (lost interest): 1
unknown): 3

Provided both 24-wk primary outcomes (n = 93) Provided both 24-wk primary outcomes (n = 96)
Unable to contact: 9 Unable to contact: 10
Withdrew from study (1 lost interest, 2 reason Withdrew from study (lost interest): 1
unknown): 3

Included in the primary analysis (n = 105) Included in the primary analysis (n = 107)

GP = general practitioner; NRS = numerical rating scale.

* An 11-point NRS was used to report minimum average walking pain over the previous week (0 = “no pain”, 10 = “worst pain possible”).
† GP clearance was obtained for participants who reported a fall in the previous 12 months, were housebound due to immobility, or indicated they had
a health condition that might compromise exercise safety on the Exercise & Sports Science Australia stage 1 preexercise screening questionnaire.

Annals.org Annals of Internal Medicine • Vol. 175 No. 10 • October 2022 1347
ORIGINAL RESEARCH Effectiveness of an Online Yoga Program in People With Knee Osteoarthritis

Table 1. Baseline Characteristics of Participants Table 1–Continued

Characteristic Control Yoga Characteristic Control Yoga

(n = 105) (n = 107) (n = 105) (n = 107)

Mean age (SD), y 61.8 (7.2) 62.8 (8.2) No effect at all 3 (2.9) 1 (0.9)
Female, n (%) 78 (74.3) 70 (65.4) Minimal improvement 19 (18.1) 27 (25.2)
Mean height (SD), m 1.7 (0.1) 1.7 (0.1) Moderate improvement 67 (63.8) 61 (57.0)
Mean body mass (SD), kg 87.2 (17.3) 86.9 (18.4) Large improvement 16 (15.2) 17 (15.9)
Mean body mass index (SD), 30.7 (6.1) 29.9 (4.8) Complete recovery 0 (0.0) 1 (0.9)
kg/m2 Beliefs about effectiveness of
Place of birth, n (%) yoga, n (%)§
Australia/Oceania 69 (65.7) 80 (74.8) Not effective 2 (1.9) 2 (1.9)
Asia 7 (6.7) 4 (3.7) Somewhat effective 38 (36.2) 47 (43.9)
Europe 27 (25.7) 20 (18.7) Moderately effective 44 (41.9) 44 (41.1)
North America 2 (1.9) 3 (2.8) Highly effective 21 (20.0) 14 (13.1)
Currently employed, n (%) 57 (54.3) 46 (43.0) Median self-efficacy using com- 4.5 (3.8–4.8) 4.6 (4.0–4.8)
Right knee more painful, 69 (65.7) 57 (53.3) puter (IQR)||
n (%)
Unilateral symptoms, n (%) 33 (31.4) 33 (30.8) NRS = numerical rating scale (overall average pain of 0–10 in the pre-
Median knee symptom 5 (2–8) 4 (2–10) vious week, with higher scores indicating more pain).
duration (IQR), y * Defined as ≥1 time per week over the previous 3 months.
Problems in other joints, n (%) † From the incidental and planned exercise questionnaire (IPEQ-W),
Upper body 51 (48.6) 40 (37.4) which includes 10 questions about frequency and duration of inciden-
Back and neck 48 (45.7) 47 (43.9) tal and planned walking, sport, and recreational activities over the pre-
Hip 32 (30.5) 25 (23.4) vious 7 days. Scores are calculated as the product of the frequency
Foot 30 (28.6) 22 (20.6) and duration scores to create a total score for duration for the week,
Median spine pain severity 1 (0–3) 1 (0–3) with higher scores indicating higher levels of activity.
on NRS (IQR) ‡ Scored from a question asking about expectation of study treatment
Median hip pain severity on 2 (1–4) 1 (1–4) outcomes, with self-reported scores on a 5-point Likert scale ranging
NRS (IQR) from 0 ("no effect at all") to 4 ("complete recovery").
Comorbid conditions, n (%) § Participants were asked to rate their beliefs about the effectiveness
≥1 comorbid condition 59 (56.2) 59 (55.1) of yoga on a 4-point Likert scale ranging from 1 ("not effective") to 4
Heart disease 5 (4.8) 12 (11.2) ("highly effective").
High blood pressure 29 (27.6) 26 (24.3) || From a modified version of the computer self-efficacy scale, which
Depression 14 (13.3) 11 (10.3) involved taking the average of scored responses to 26 of the original
Diabetes 4 (3.8) 6 (5.6) 32 questions on a 5-point Likert scale indicating degree of confidence
Spine condition 9 (8.6) 17 (15.9) from 1 ("very little confidence") to 5 ("quite a lot of confidence").
Lung disease 12 (11.4) 17 (15.9)
Cancer 6 (5.7) 5 (4.7) clinical trials (30, 31). Overall average knee pain severity
Nervous system disease 0 (0.0) 1 (0.9) during walking in the previous week was assessed using an
Kidney disease 0 (0.0) 3 (2.8)
11-point NRS with terminal descriptors of 0 (“no pain”) and
Liver disease 0 (0.0) 2 (1.9)
Other 3 (2.9) 4 (3.7) 10 (“worst pain possible”) (28). The minimal clinically impor-
Treatment for knee in tant difference (MCID) is 1.8 units (32). Difficulty with phys-
previous 3 mo, n (%) ical function was assessed using the Western Ontario and
≥1 treatment 81 (77.1) 85 (79.4) McMaster Universities Osteoarthritis Index (WOMAC;
Massage/manual therapy 27 (25.7) 30 (28.0)
Gait aid 8 (7.6) 8 (7.5) Likert version 3.1) function subscale (33), with scores
Thermal therapy/ 40 (38.1) 43 (40.2) ranging from 0 (no dysfunction) to 68 (maximum dys-
electrotherapy function). The MCID is 6 nonnormalized units (34).
Orthotics, arch supports, 25 (23.8) 25 (23.4) Secondary outcomes included overall average knee
or wedging in shoes
Knee braces 28 (26.7) 29 (27.1)
pain in the previous week on an 11-point NRS (0 = “no
Land-based and water 29 (27.6) 34 (31.8) pain”, 10 = “worst pain possible”) (28); scores on the
exercises pain and stiffness subscales of the WOMAC (0 to 20 and
Joint injections 0 (0.0) 2 (1.9) 0 to 8, respectively, with higher scores indicating poorer
Acupuncture 5 (4.8) 4 (3.7)
Current pain medication
outcomes) (35); Depression, Anxiety and Stress Scale
use, n (%)* (DASS-21) score (0 to 42, with higher scores indicating
≥1 medication 75 (71.4) 80 (74.8) worse mental health) (36); Assessment of Quality of Life
Nonsteroidal anti-inflammatory 40 (38.1) 44 (41.1) (AQoL-6D) score ( 0.04 to 1.00, with higher scores indi-
drugs
cating better quality of life) (37); global change in condi-
Acetaminophen 59 (56.2) 65 (60.7)
Topical anti-inflammatory 37 (35.2) 39 (36.4) tion of knee via a 7-point Likert scale (“moderately better”
drugs or “much better” classified as “improved”); scores on the
Oral corticosteroids 1 (1.0) 1 (0.9) pain and other symptoms subscales of the Arthritis Self-
Oral opioids 3 (2.9) 0 (0.0) Efficacy Scale (1 to 10, with higher scores indicating
Median physical activity 22.8 (13.0–36.8) 22.8 (13.1–36.3)
(IQR), h/wk† greater self-efficacy) (38); Brief Fear of Movement Scale
Expectation of treatment for Osteoarthritis score (6 to 24, with higher scores indi-
outcome, n (%)‡ cating greater fear) (39); and Activities-Specific Balance
Continued Confidence Scale score (0% to 100%, with higher scores
indicating greater confidence in balance) (40).
1348 Annals of Internal Medicine • Vol. 175 No. 10 • October 2022 Annals.org
Effectiveness of an Online Yoga Program in People With Knee Osteoarthritis ORIGINAL RESEARCH
Baseline descriptive measures, adverse events, and Role of the Funding Source
co-intervention use (pain medications and nonpharma- The National Health and Medical Research Council
cologic treatments for knee pain over the prior 3 months) funded the study but played no role in the design or con-
were collected via an online questionnaire. Adherence duct of the study, statistical analysis, or the decision to
was assessed by the number of weekly yoga sessions publish the manuscript.
(0 to 3) during weeks 1 to 12, as recorded in a weekly
electronic logbook e-mailed to participants; the number
of yoga sessions (0 to 3) reported in the previous week at RESULTS
12 and 24 weeks; the number and duration (in minutes) We enrolled 212 participants from among 2004 peo-
of yoga website visits from data analytics; Exercise ple assessed for eligibility (Figure). Baseline participant
Adherence Rating Scale Section B score (0 to 24, with characteristics were generally similar between groups,
higher scores indicating better adherence) at 12 and 24 except more control participants were female, were
weeks (41); and the number and duration (in minutes) of employed, had the right knee affected, and had other
control website visits from data analytics. Process meas- joint problems (Table 1). Both primary outcomes were
ures at 12 weeks included willingness to pay for the yoga provided by 195 of 212 (92%) participants at 12 weeks
program, likelihood of recommending the program, and and 189 of 212 (89%) at 24 weeks, with some minor dif-
overall satisfaction with the program, all assessed using ferences between those who did and those who did
the NRS (0 = “not at all willing/likely/satisfied”, 10 = not provide these outcomes (Appendix Table 1, avail-
“extremely willing/likely/satisfied”), as well as the amount able at Annals.org).
participants were willing to pay (in Australian dollars). Adherence to the yoga program and website access
in both groups are shown in Table 4 of Supplement 2.
Statistical Analysis
Logbook data over the 12-week program showed a
We aimed to detect a small to moderate effect size
mean of 2.5 (SD, 0.7) yoga sessions per week. Participants
of 0.4 (consistent with exercise interventions in knee
reported a mean of 1.9 (SD, 1.3) yoga sessions in the final
osteoarthritis [42]), with a conservative correlation of 0.3
between baseline and follow-up measures (43), 80% week of the program, with 69% deemed to have “accept-
power, and a 2-sided significance level of 0.05 (unad- able” adherence (≥2 sessions). Yoga group participants
justed for two primary outcomes, with emphasis instead accessed their website a mean of 20.5 (SD, 19.5) times for
placed on the clinical meaning of differences and CIs as a mean of 25.4 (SD, 12.8) minutes per session over the 12
recommended [44]). Based on these assumptions and weeks. During the 12-week follow-up, the yoga website
using analysis of covariance adjusted for baseline score, was accessed a mean of 2.5 (SD, 6.4) times for a mean of
we needed 90 participants per group. Allowing for 15% 24.2 (SD, 18.0) minutes per session. Control participants
attrition, we recruited 106 people per group (212 accessed their online information website a mean of 1.6
overall). (SD, 1.5) times for a mean of 9.8 (SD, 14.3) minutes during
Comparative analyses were done using Stata, ver- the first 12 weeks. Yoga participants were highly satisfied
sion 16.1 (StataCorp), based on the intention-to-treat with the program and likely to recommend it to others
principle and using all randomly assigned participants. (Table 5 of Supplement 2).
Missing outcomes were multiply imputed using chained Primary outcomes are provided in Table 2 (mean
equations with predictive mean matching and 5 nearest outcome scores over time) and Table 3 (within-group
neighbors (Appendix, available at Annals.org). For pri- changes and between-group differences in change).
mary outcomes and continuous secondary outcomes, Compared with the control intervention at 12 weeks,
differences (follow-up minus baseline) were compared yoga led to greater improvements in function (between-
between groups by using separate mixed-effects linear group mean difference in change, 4.0 [95% CI, 6.8 to
regression models adjusted for the outcome at baseline 1.3]). Only small differences were detected between
and including fixed effects for week and an interaction groups for change in knee pain during walking at 12
between week and treatment group, with random effects
weeks (mean difference in change, 0.6 [CI, 1.2 to
for participants (45). The binary secondary outcome of
0.1]). More participants in the yoga group than the con-
global change and primary outcomes dichotomized based
trol group achieved the MCID for both function (60% vs.
on achievement of an MCID in improvement in pain (1.8
44%; relative risk, 1.34 [CI, 1.02 to 1.76]) and pain during
NRS units) and function (6 WOMAC units) were compared
between groups by using risk differences and risk ratios walking (60% vs. 41%; relative risk, 1.44 [CI, 1.09 to 1.91])
calculated from separate logistic regression models fit at 12 weeks (Table 6 of Supplement 2). Benefits of yoga
using generalized estimating equations, including a term on the primary outcomes were not maintained at 24
for week and an interaction between week and treatment weeks (Table 3). Analyses using complete-case data pro-
group as covariates. Standard diagnostic plots were used duced similar results (Tables 7 and 8 of Supplement 2),
to check model assumptions. as did adjustment for unbalanced baseline characteris-
The Appendix contains details of analyses for moder- tics (gender and problems in other joints) (Table 9 of
ation of yoga effects by baseline characteristics selected Supplement 2).
a priori and with adjustment for unbalanced baseline Continuous secondary outcomes are shown in Tables
participant characteristics (unplanned post hoc sensitivity 2 and 3. At 12 weeks, yoga led to greater improvements
analyses). in knee stiffness (mean difference in change, 0.4 [CI,
Annals.org Annals of Internal Medicine • Vol. 175 No. 10 • October 2022 1349
ORIGINAL RESEARCH Effectiveness of an Online Yoga Program in People With Knee Osteoarthritis

Table 2. Continuous Outcome Measures Over Time Across Treatment Groups*

Outcome Baseline 12 Weeks 24 Weeks
Control Yoga Control Yoga Control Yoga
(n = 105) (n = 107) (n = 98)† (n = 98)‡ (n = 93)§ (n = 96)||

Primary outcomes
Knee pain during walking (NRS)¶ 6.3 (1.3) 6.3 (1.3) 4.9 (2.4) 4.2 (2.4) 4.6 (2.3) 4.3 (2.4)
Physical function (WOMAC)** 24.0 (9.9) 25.5 (9.1) 19.3 (11.5) 15.8 (10.7) 18.2 (11.9) 16.5 (11.5)

Secondary outcomes
Overall knee pain (NRS)¶ 5.8 (1.5) 5.8 (1.5) 4.5 (2.3) 4.0 (2.3) 4.4 (2.2) 4.1 (2.3)
Knee pain (WOMAC)** 7.9 (2.6) 8.0 (2.5) 6.0 (3.6) 5.4 (3.4) 5.5 (3.5) 5.7 (3.6)
Knee stiffness (WOMAC)** 3.5 (1.6) 3.3 (1.7) 3.1 (1.5) 2.6 (1.5) 3.0 (1.7) 2.6 (1.5)
Anxiety (DASS-21)†† 2.7 (3.3) 2.5 (4.1) 3.7 (5.1) 2.5 (4.0) 3.5 (5.2) 2.9 (4.8)
Depression (DASS-21)†† 5.4 (6.5) 4.8 (7.2) 5.9 (7.1) 4.3 (6.0) 6.3 (7.5) 5.4 (7.0)
Stress (DASS-21)†† 7.0 (6.3) 6.1 (6.0) 7.4 (6.7) 5.9 (5.7) 7.3 (7.8) 6.6 (6.4)
Quality of life (AQoL-6D)‡‡ 0.72 (0.13) 0.74 (0.13) 0.73 (0.15) 0.78 (0.13) 0.73 (0.18) 0.76 (0.15)
Arthritis self-efficacy: pain§§ 6.0 (1.6) 6.1 (1.6) 6.1 (1.9) 6.8 (1.9) 6.2 (2.1) 6.5 (2.0)
Arthritis self-efficacy: other symptoms§§ 6.3 (1.8) 6.4 (1.6) 6.4 (1.8) 7.1 (1.6) 6.4 (1.9) 6.8 (2.0)
Fear of movement|||| 12.5 (3.4) 12.6 (3.1) 12.5 (3.7) 11.6 (3.3) 12.4 (3.6) 11.7 (3.4)
Balance confidence¶¶ 84.7 (15.6) 82.6 (13.0) 82.8 (15.7) 85.8 (12.7) 85.1 (15.4) 83.8 (14.1)

AQoL-6D = Assessment of Quality of Life; DASS-21 = Depression, Anxiety and Stress Scale; MCID = minimal clinically important difference; NRS =
numerical rating scale; WOMAC = Western Ontario and McMaster Universities Osteoarthritis Index.
* Data are for complete cases and are presented as means (SDs).
† Data are from 97 participants for WOMAC (physical function, pain, and stiffness) and AQoL-6D; 96 participants for DASS-21 (anxiety, depression,
and stress), arthritis self-efficacy (pain and other symptoms), and fear of movement; and 95 participants for balance confidence.
‡ Data are from 97 participants for NRS (overall knee pain) and WOMAC (pain and stiffness) and 96 participants for DASS-21 (anxiety, depression,
and stress), AQoL-6D, arthritis self-efficacy (pain and other symptoms), fear of movement, and balance confidence.
§ Data are from 92 participants for DASS-21 (anxiety, depression, and stress), arthritis self-efficacy (pain and other symptoms), and fear of movement
and 91 participants for balance confidence.
|| Data are from 95 participants for DASS-21 (anxiety, depression, and stress) and balance confidence.
¶ The NRS ranges from 0 to 10, with higher scores indicating greater pain. The MCID is 1.8 units.
** The WOMAC ranges from 0 to 68 for physical function (MCID, 6 nonnormalized units), from 0 to 20 for pain, and from 0 to 8 for stiffness, with
higher scores indicating more dysfunction, pain, or stiffness.
†† The DASS-21 ranges from 0 to 42, with higher scores indicating more anxiety, depression, or stress.
‡‡ The AQoL-6D ranges from 0.04 to 1.00, with higher scores indicating better quality of life.
§§ Based on the Arthritis Self-Efficacy Scale, which ranges from 1 to 10, with higher scores indicating greater self-efficacy.
|||| Based on the Brief Fear of Movement Scale for Osteoarthritis, which ranges from 6 to 24, with higher scores indicating greater fear of movement.
¶¶ Based on the Activities-Specific Balance Confidence Scale, which ranges from 0% to 100%, with higher scores indicating greater balance
confidence.

0.8 to 0.0]), quality of life (mean difference in change, DISCUSSION

0.04 [CI, 0.00 to 0.07]), arthritis self-efficacy for pain (mean Compared with an online education control interven-
difference in change, 0.6 [CI, 0.1 to 1.1]), and self-efficacy tion, a 12-week unsupervised home-based online yoga
for other symptoms (mean difference in change, 0.6 [CI, program improved physical function but not knee pain dur-
0.2 to 1.0]) than the control intervention. Confidence inter- ing walking in people with knee osteoarthritis. However,
vals of between-group differences in change in other the between-group difference in function of 4 units favor-
secondary outcomes at 12 weeks and in all secondary ing yoga was below the MCID of 6 units. Benefits of yoga
outcomes at 24 weeks included the null. Between-group on physical function were not sustained at 24 weeks. Some
differences in the proportions of participants reporting secondary outcomes (knee stiffness, quality of life, and ar-
global knee improvement were small (Table 6 of thritis self-efficacy) improved significantly more with yoga
than with the control intervention at 12 weeks but not 24
Supplement 2). Analyses using complete-case data
weeks. Participants performed about two thirds of the rec-
produced similar results (Tables 7 and 8 of Supplement 2).
ommended number of yoga sessions during the 12-week
Prespecified variables of age, gender, body mass
program, but adherence decreased over the subsequent
index, expectation of treatment effects, pain self-efficacy,
12-week follow-up when the program was optional.
and fear of movement did not moderate effects of yoga Mechanisms underpinning our observed improve-
on primary outcomes at 12 weeks (Appendix Tables 2 ments in physical function with yoga are unknown
and 3, available at Annals.org). because we did not evaluate treatment effect mediators,
There were no serious related adverse events and nor have they been formally examined in other yoga
few nonserious ones, with more reported by yoga partici- studies in osteoarthritis (46). Postulated mechanisms
pants (13 of 97 [13%]) than control participants (1 of 97 include physical, psychological, and lifestyle constructs,
[1%]) (Table 4). These were mostly musculoskeletal pain such as improved muscle strength, joint range, self-
episodes. Co-interventions appeared to be generally efficacy, and activity levels (46). Self-efficacy increased
similar between groups (Table 4). in our yoga group and has been identified as a mediator
1350 Annals of Internal Medicine • Vol. 175 No. 10 • October 2022 Annals.org
Effectiveness of an Online Yoga Program in People With Knee Osteoarthritis ORIGINAL RESEARCH
of benefits of other forms of exercise in osteoarthritis Our study is unique in investigating an unsupervised
(47). online yoga program in knee osteoarthritis. A recent
Several issues warrant consideration in explaining systematic review identified 5 trials in people with knee
our findings. First, because this was a pragmatic trial, we osteoarthritis that evaluated various in-person group
did not control for differences in contextual effects Hatha yoga programs, often with home practice and last-
between groups, and participants were unblinded. This ing 8 to 12 weeks (13). A meta-analysis of these trials
could have led to bias and overestimation of the true showed very-low-quality evidence for moderate to large
yoga effects, especially given our use of patient-reported beneficial effects of yoga on pain, function, and stiffness,
outcomes (48). However, because such bias might be but not quality of life or depression (13). However, CIs
expected to influence both primary outcomes, our incon- were wide, and there was potential risk of bias. Our results
sistent results showing that yoga benefits function but are generally similar, except our yoga effects were smaller
not pain raise uncertainty about how much these issues in magnitude and were not observed for pain. These may
influenced our findings. Furthermore, a recent large reflect differences in sample characteristics and methods,
meta-epidemiologic study showed no difference in esti- lack of supervision, or potentially lower adherence, although
mated treatment effects between trials with and without this cannot be confirmed because adherence data were
patient blinding (49). Second, not all participants were rarely reported in the other studies.
adherent, which may have attenuated detection of any Our study has limitations. As stated, participants were
true benefits of yoga. Third, there were modest differen- unblinded to interventions, which can inflate treatment
ces in baseline characteristics between groups, including effects. Not all of our sample may have had “structural”
greater prevalence of other joint problems in control par- osteoarthritis, given that we used clinical diagnostic crite-
ticipants, although post hoc analyses controlling for ria without radiographs, as recommended by guidelines
these did not alter our findings. Fourth, more yoga par- (50). However, there is little evidence to suggest that radi-
ticipants than control participants undertook other land- ographic changes moderate exercise outcomes in knee
based or water exercise during the intervention, which osteoarthritis (51). Because the yoga program was unsu-
may have improved function in the yoga group. pervised, we do not know whether the yoga elements

Table 3. Change in Continuous Outcome Measures Within and Between Groups Over Time Based on Multiply Imputed Data*
Outcome Mean (SD) Change Within Mean (95% CI) Mean (SD) Change Within Mean (95% CI)
Groups (12 Weeks Minus Difference in Change Groups (24 Weeks Minus Difference in Change
Baseline) Between Groups at Baseline) Between Groups at
12 Weeks for Yoga vs. 24 Weeks for Yoga vs.
Control Yoga Control Yoga
Control (n = 212) Control (n = 212)
(n = 105) (n = 107) (n = 105) (n = 107)

Primary outcomes
Knee pain during walking (NRS)†‡ 1.4 (2.4) 2.0 (2.3) 0.6 ( 1.2 to 0.1) 1.7 (2.5) 2.0 (2.4) 0.3 ( 1.0 to 0.4)
Physical function (WOMAC)ठ4.7 (10.8) 9.3 (9.6) 4.0 ( 6.8 to 1.3) 5.9 (11.9) 8.6 (10.8) 2.1 ( 5.1 to 0.9)

Secondary outcomes
Overall knee pain (NRS)†‡ 1.3 (2.3) 1.7 (2.2) 0.4 ( 1.0 to 0.2) 1.3 (2.5) 1.7 (2.3) 0.4 ( 1.0 to 0.3)
Knee pain (WOMAC)ठ2.0 (3.4) 2.5 (3.3) 0.4 ( 1.3 to 0.5) 2.3 (3.6) 2.3 (3.5) 0.1 ( 0.8 to 1.1)
Knee stiffness (WOMAC)ठ0.4 (1.7) 0.7 (1.6) 0.4 ( 0.8 to 0.0) 0.4 (1.8) 0.7 (1.8) 0.4 ( 0.8 to 0.0)
Anxiety (DASS-21)‡|| 1.0 (4.7) 0.2 (3.6) 0.9 ( 2.0 to 0.3) 0.9 (4.7) 0.7 (4.6) 0.3 ( 1.6 to 1.0)
Depression (DASS-21)‡|| 0.8 (5.8) 0.2 (6.2) 1.2 ( 2.7 to 0.4) 1.3 (6.9) 0.9 (6.0) 0.6 ( 2.5 to 1.3)
Stress (DASS-21)‡|| 0.4 (6.2) 0.2 (5.2) 1.0 ( 2.7 to 0.7) 0.6 (6.8) 0.9 (6.2) 0.1 ( 1.8 to 1.7)
Quality of life (AQoL-6D)¶** 0.00 (0.12) 0.03 (0.11) 0.04 (0.00 to 0.07) 0.00 (0.14) 0.01 (0.13) 0.02 ( 0.02 to 0.06)
Arthritis self-efficacy: pain**†† 0.1 (1.7) 0.7 (2.0) 0.6 (0.1 to 1.1) 0.2 (2.2) 0.5 (2.2) 0.3 ( 0.3 to 0.9)
Arthritis self-efficacy: other 0.1 (1.5) 0.6 (1.5) 0.6 (0.2 to 1.0) 0.1 (1.8) 0.4 (1.7) 0.3 ( 0.1 to 0.8)
symptoms**††
Fear of movement‡,‡‡ 0.1 (3.7) 0.8 (3.2) 0.7 ( 1.6 to 0.2) 0.1 (3.6) 1.0 (3.5) 0.9 ( 1.8 to 0.1)
Balance confidence**§§ 2.2 (17.5) 2.3 (12.4) 3.4 ( 0.4 to 7.2) 0.1 (16.8) 0.8 (12.4) 0.4 ( 4.0 to 3.2)

AQoL-6D = Assessment of Quality of Life; DASS-21 = Depression, Anxiety and Stress Scale; MCID = minimal clinically important difference; NRS =
numerical rating scale; WOMAC = Western Ontario and McMaster Universities Osteoarthritis Index.
* Results were adjusted for baseline values using multiply imputed data. Ten data sets were imputed and analyzed, with the results combined using
Rubin’s rules.
† The NRS ranges from 0 to 10, with higher scores indicating greater pain. The MCID is 1.8 units.
‡ For change within groups, negative changes indicate improvement. For difference in change between groups, negative differences favor yoga.
§ The WOMAC ranges from 0 to 68 for physical function (MCID, 6 nonnormalized units), from 0 to 20 for pain, and from 0 to 8 for stiffness, with
higher scores indicating more dysfunction, pain, or stiffness.
|| The DASS-21 ranges from 0 to 42, with higher scores indicating more anxiety, depression, or stress.
¶ The AQoL-6D ranges from 0.04 to 1.00, with higher scores indicating better quality of life.
** For change within groups, positive changes indicate improvement. For difference in change between groups, positive differences favor yoga.
†† Arthritis self-efficacy is based on the Arthritis Self-Efficacy Scale, which ranges from 1 to 10, with higher scores indicating greater self-efficacy.
‡‡ Fear of movement is based on the Brief Fear of Movement Scale for Osteoarthritis, which ranges from 6 to 24, with higher scores indicating
greater fear of movement.
§§ Balance confidence is based on the Activities-Specific Balance Confidence Scale, which ranges from 0% to 100%, with higher scores indicating
greater balance confidence.

Annals.org Annals of Internal Medicine • Vol. 175 No. 10 • October 2022 1351
ORIGINAL RESEARCH Effectiveness of an Online Yoga Program in People With Knee Osteoarthritis

Table 4. Adverse Events, Pain Medications, and Other Co-interventions

Event or Co-intervention Control, n/N (%) Yoga, n/N (%)
0 to 12 Weeks 13 to 24 Weeks 0 to 12 Weeks 13 to 24 Weeks

Adverse events*
Discontinued due to related adverse event 0/105 (0.0) 0/97 (0.0) 0/107 (0.0) 0/97 (0.0)
Any serious adverse event† 0/97 (0.0) 0/93 (0.0) 0/97 (0.0) 0/96 (0.0)
Nonserious related adverse event 1/97 (1.0) 1/93 (1.1) 13/97 (13.4) 11/96 (11.3)
Knee pain 1/97 (1.0) 1/93 (1.1) 5/97 (5.2) 3/96 (3.1)
Upper-limb pain 1/97 (1.0) 0/93 (0) 1/97 (1.0) 4/96 (4.1)
Back pain 1/97 (1.0) 1/93 (1.1) 6/97 (6.2) 2/96 (2.1)
Knee stiffness 0/97 (0.0) 0/93 (0) 1/97 (1.0) 0/96 (0.0)
Foot pain 1/97 (1.0) 1/93 (1.1) 0/97 (0.0) 2/96 (2.1)
Hip pain 0/97 (0.0) 0/93 (0) 1/97 (1.0) 2/96 (2.1)

Pain medication use‡

≥1 medication 70/97 (72.2) 68/93 (73.1) 61/97 (62.9) 67/96 (69.8)
Acetaminophen alone or in combined formulations 56/97 (57.7) 50/93 (53.8) 50/97 (51.5) 50/96 (52.1)
Topical anti-inflammatory drugs 35/97 (36.1) 30/93 (32.3) 34/97 (35.1) 32/96 (33.3)
Nonsteroidal anti-inflammatory drugs 37/97 (38.1) 36/93 (38.7) 33/97 (34.0) 35/96 (36.5)
Oral glucocorticoids 1/97 (1.0) 0/93 (0.0) 2/97 (2.1) 2/96 (2.1)
Oral opioids 3/97 (3.1) 5/93 (5.4) 2/97 (2.1) 3/96 (3.1)

Other co-interventions§
≥1 treatment 83/97 (85.6) 77/93 (82.8) 76/97 (78.4) 81/96 (84.4)
Massage/manual therapy 32/97 (33.0) 39/93 (41.9) 32/97 (33.0) 34/96 (35.4)
Gait aid 10/97 (10.3) 11/93 (11.8) 8/97 (8.2) 4/96 (4.2)
Thermal therapy/electrotherapy 42/97 (43.3) 33/93 (35.5) 31/97 (32.0) 33/96 (34.4)
Orthotics, arch supports, or wedging in shoes 37/97 (38.1) 31/93 (33.3) 30/97 (30.9) 31/96 (32.3)
Knee brace 28/97 (28.9) 25/93 (26.9) 16/97 (16.5) 17/96 (17.7)
Land-based and water exercise 58/97 (59.8) 52/93 (55.9) 64/97 (66.0) 64/96 (66.7)
Joint injection 1/97 (1.0) 1/93 (1.1) 0/97 (0.0) 1/96 (1.0)
Acupuncture 7/97 (7.2) 4/93 (4.3) 2/97 (2.1) 1/96 (1.0)
Knee surgery 0/97 (0.0) 1/93 (1.1)|| 1/97 (1.0)|| 0/96 (0.0)

* Adverse events were defined as any problem experienced as a result of the study, in the study knee or elsewhere in the body, lasting for ≥2 days,
and/or requiring cessation of the study intervention and/or treatment from a health care professional. Denominators depended on the number of
participants who completed the adverse events section of the questionnaire in each group.
† Serious adverse events were defined as any untoward medical occurrence that resulted in death, was life-threatening, required hospitalization,
resulted in significant disability, or required medical or surgical intervention.
‡ Defined as taken at least once per week over the prior month.
§ Defined as having tried the co-intervention specifically for pain in their study knee in the previous 12 weeks (but excluded study interventions).
|| Arthroscopy.

were performed correctly or completely. We also do not treatment. The program was safe, with no serious adverse
know whether control participants did yoga during the events and few nonserious adverse events in our sample
study or whether yoga participants engaged in more in- of older adults, most with comorbid conditions. This is
tensive nonyoga exercise (in addition to yoga) than con- important given that the program was unsupervised.
trol participants. Finally, our sample of English-speaking User acceptability was demonstrated, as participants
people who were highly confident using a computer may were highly satisfied with the program and likely to
limit generalizability. recommend it to others. Adherence was also good, partic-
Our results are promising, but further research is ularly compared with some other unsupervised online
needed to fully establish the effects of unsupervised online exercise programs for osteoarthritis (56, 57). Online yoga
yoga and to potentially improve on our program in order
has advantages of convenience, privacy, and flexibility (14)
to enhance treatment benefits. Given decreasing adher-
and is also accessible by a much larger audience than in-
ence over the follow-up period and the link between
person classes. Thus, even small benefits at a broader pop-
adherence and clinical outcomes (52, 53), longer-term
behavior change strategies to maximize engagement ulation level may be important for public health.
should be identified (54). There are also many different In summary, an unsupervised 12-week online yoga
types of yoga. Our program was low-intensity, emphasizing program for people with knee osteoarthritis improved
physical poses and movement, with less or no inclusion of function more than online education immediately after
elements such as deep relaxation, chanting, singing, and the program, although the improvement did not meet the
meditation. Thus, studies investigating different types, dos- MCID and was not sustained at 24 weeks. Benefits of the
ages, and formats of online yoga, together with exploration program for knee pain during walking were not apparent.
of moderators (55) and mediators of effects, will be useful.
This freely accessible, home-based online yoga From Centre for Health, Exercise and Sports Medicine, Department
program could be a scalable option to improve patient of Physiotherapy, The University of Melbourne, Melbourne, Victoria,
access to exercise, a core recommended osteoarthritis Australia (K.L.B., S.S., P.L.T., S.H., D.M., A.J.K., B.M., R.S.H.); Centre

1352 Annals of Internal Medicine • Vol. 175 No. 10 • October 2022 Annals.org
Effectiveness of an Online Yoga Program in People With Knee Osteoarthritis ORIGINAL RESEARCH
for Epidemiology and Biostatistics, Melbourne School of Population and years lived with disability for 354 diseases and injuries for 195
and Global Health, The University of Melbourne, Melbourne, countries and territories, 1990-2017: a systematic analysis for the Global
Victoria, Australia (F.M.); Centre for Epidemiology and Biostatistics, Burden of Disease Study 2017. Lancet. 2018;392:1789-1858. [PMID:
Melbourne School of Population and Global Health, and Methods 30496104] doi:10.1016/S0140-6736(18)32279-7
2. Stubbs B, Aluko Y, Myint PK, et al. Prevalence of depressive symp-
and Implementation Support for Clinical Health Research Hub,
toms and anxiety in osteoarthritis: a systematic review and meta-analysis.
Faculty of Medicine, Dentistry and Health Sciences, The University
Age Ageing. 2016;45:228-35. [PMID: 26795974] doi:10.1093/ageing/
of Melbourne, Melbourne, Victoria, Australia (K.E.L.); and Sydney afw001
Musculoskeletal Health, The University of Sydney, Sydney, New 3. Manlapaz DG, Sole G, Jayakaran P, et al. Risk factors for falls in adults
South Wales, Australia (D.J.H.). with knee osteoarthritis: a systematic review. PM R. 2019;11:745-57.
[PMID: 30609282] doi:10.1002/pmrj.12066
Acknowledgment: The authors thank Associate Professor Romy 4. Dor e AL, Golightly YM, Mercer VS, et al. Lower-extremity osteoarthri-
Lauche and the expert panel (Maria Kirsten, Lisa Westlake, tis and the risk of falls in a community-based longitudinal study
Therese Lew, Trish David, Margaret Clarke, and Elizabeth of adults with and without osteoarthritis. Arthritis Care Res
(Hoboken). 2015;67:633-9. [PMID: 25331686] doi:10.1002/acr.22499
Nelson) for their involvement in the development and design of
5. van Schoor NM, Dennison E, Castell MV, et al; EPOSA research
the yoga program; Lisa Westlake (instructor) and Shayne
group. Clinical osteoarthritis of the hip and knee and fall risk: the
Milsom, Margaret Clarke, and Bob Gardner (group actors) for role of low physical functioning and pain medication. Semin
starring in the yoga videos; and Dr. Jessica Kasza for formulat- Arthritis Rheum. 2020;50:380-6. [PMID: 32199610] doi:10.1016/j.
ing the sample size and randomization schedule. semarthrit.2020.02.006
6. Eyles JP, Hunter DJ, Briggs AM, et al. National Osteoarthritis
Grant Support: By a National Health and Medical Research Strategy brief report: living well with osteoarthritis. Aust J Gen
Council Program Grant (#1091302) and the Centres of Research Pract. 2020;49:438-42. [PMID: 32600001] doi:10.31128/AJGP-08-
Excellence (#1079078). 19-5051-02
7. Fransen M, McConnell S, Harmer AR, et al. Exercise for osteoar-
thritis of the knee. Cochrane Database Syst Rev. 2015;1:CD004376.
Disclosures: Disclosures can be viewed at www.acponline.org/ [PMID: 25569281] doi:10.1002/14651858.CD004376.pub3
authors/icmje/ConflictOfInterestForms.do?msNum=M22-1761. 8. Liao CD, Chen HC, Kuo YC, et al. Effects of muscle strength
training on muscle mass gain and hypertrophy in older adults with
Data Sharing Statement: The following data will be made avail- osteoarthritis: a systematic review and meta-analysis. Arthritis Care
able beginning 1 January 2023: deidentified participant data and Res (Hoboken). 2020;72:1703-18. [PMID: 31628720] doi:10.1002/
data dictionary. We will share all data that support published acr.24097
results of the trial. Additional data that have not been published 9. Bannuru RR, Osani MC, Vaysbrot EE, et al. OARSI guidelines for
the non-surgical management of knee, hip, and polyarticular osteoar-
will be withheld until at least 6 months after publication. Data will
thritis. Osteoarthritis Cartilage. 2019;27:1578-89. [PMID: 31278997]
be made available as required for specific, approved analyses and doi:10.1016/j.joca.2019.06.011
will be provided from a locked, cleaned, and deidentified study 10. Kolasinski SL, Neogi T, Hochberg MC, et al. 2019 American
database. Requests will be reviewed by the principal investigator College of Rheumatology/Arthritis Foundation guideline for the
before approval. The following supporting documents will be management of osteoarthritis of the hand, hip, and knee. Arthritis
made available beginning on 1 January 2023: statistical/analytic Rheumatol. 2020;72:220-33. [PMID: 31908163] doi:10.1002/art.41142
code and informed consent form (contact the principal investiga- 11. Zhang Y, Lauche R, Cramer H, et al. Increasing trend of yoga prac-
tor, Kim Bennell, at k.bennell@unimelb.edu.au). The investigators tice among U.S. adults from 2002 to 2017. J Altern Complement Med.
endorse the concept of data sharing to advance medical science. 2021;27:778-85. [PMID: 34076530] doi:10.1089/acm.2020.0506
12. Elwy AR, Groessl EJ, Eisen SV, et al. A systematic scoping review of
All requests for data sharing will be reviewed by the principal in-
yoga intervention components and study quality. Am J Prev Med. 2014;
vestigator to ensure no conflict with any planned subanalyses and
47:220-32. [PMID: 24996759] doi:10.1016/j.amepre.2014.03.012
to ensure that the data are shared in an ethical and protected 13. Lauche R, Hunter DJ, Adams J, et al. Yoga for osteoarthritis: a
manner. Analyses aimed at improving treatment of knee osteoar- systematic review and meta-analysis. Curr Rheumatol Rep. 2019;21:47.
thritis for noncommercial purposes are eligible. Data will be made [PMID: 31338685] doi:10.1007/s11926-019-0846-5
available after review and approval by the principal investigator. 14. Brinsley J, Smout M, Davison K . Satisfaction with online versus
Before any analysis, a signed confidentiality agreement and/or in-person yoga during COVID-19. J Altern Complement Med. 2021;
data sharing agreement is required (restrictions: none). 27:893-6. [PMID: 34339262] doi:10.1089/acm.2021.0062
15. Australian Bureau of Statistics. Household use of information
technology. 28 March 2018.
Corresponding Author: Kim L. Bennell, PhD, Centre for Health,
16. Dobson F, Bennell KL, French SD, et al. Barriers and facilitators
Exercise and Sports Medicine, Department of Physiotherapy, to exercise participation in people with hip and/or knee osteoarthritis:
School of Health Sciences, Faculty of Medicine, Dentistry and synthesis of the literature using behavior change theory. Am J Phys
Health Sciences, The University of Melbourne, Melbourne, Med Rehabil. 2016;95:372-89. [PMID: 26945211] doi:10.1097/
Victoria, Australia; e-mail, k.bennell@unimelb.edu.au. PHM.0000000000000448
17. Brosnan P, Nauphal M, Tompson MC . Acceptability and feasi-
Author contributions are available at Annals.org. bility of the online delivery of Hatha yoga: a systematic review of the
literature. Complement Ther Med. 2021;60:102742. [PMID: 34144493]
doi:10.1016/j.ctim.2021.102742
18. Martin AC, Candow D . Effects of online yoga and tai chi on
References physical health outcome measures of adult informal caregivers. Int
1. GBD 2017 Disease and Injury Incidence and Prevalence J Yoga. 2019;12:37-44. [PMID: 30692782] doi:10.4103/ijoy.IJOY_
Collaborators. Global, regional, and national incidence, prevalence, 5_18

Annals.org Annals of Internal Medicine • Vol. 175 No. 10 • October 2022 1353
ORIGINAL RESEARCH Effectiveness of an Online Yoga Program in People With Knee Osteoarthritis

19. Huberty J, Eckert R, Larkey L, et al. Perceptions of myeloproli- 35. Bellamy N . Osteoarthritis clinical trials: candidate variables and
ferative neoplasm patients participating in an online yoga interven- clinimetric properties. J Rheumatol. 1997;24:768-78. [PMID: 9101516]
tion: a qualitative study. Integr Cancer Ther. 2018;17:1150-62. 36. Delbaere K, Hauer K, Lord SR . Evaluation of the incidental and
[PMID: 30352518] doi:10.1177/1534735418808595 planned activity questionnaire (IPEQ) for older people. Br J Sports Med.
20. Uebelacker L, Dufour SC, Dinerman JG, et al. Examining the 2010;44:1029-34. [PMID: 19474003] doi:10.1136/bjsm.2009.060350
feasibility and acceptability of an online yoga class for mood disor- 37. Osborne RH, Hawthorne G, Lew EA, et al. Quality of life assess-
ders: a MoodNetwork study. J Psychiatr Pract. 2018;24:60-67. ment in the community-dwelling elderly: validation of the Assessment
[PMID: 29320386] doi:10.1097/PRA.0000000000000286 of Quality of Life (AQoL) Instrument and comparison with the SF-36. J
21. Zernicke KA, Campbell TS, Speca M, et al. A randomized wait- Clin Epidemiol. 2003;56:138-47. [PMID: 12654408]
list controlled trial of feasibility and efficacy of an online mindful- 38. Lorig K, Chastain RL, Ung E, et al. Development and evaluation
ness-based cancer recovery program: the eTherapy for Cancer of a scale to measure perceived self-efficacy in people with arthritis.
Applying Mindfulness trial. Psychosom Med. 2014;76:257-67. Arthritis Rheum. 1989;32:37-44. [PMID: 2912463]
[PMID: 24804884] doi:10.1097/PSY.0000000000000053 39. Shelby RA, Somers TJ, Keefe FJ, et al. Brief Fear of Movement
22. Osteoarthritis Action Alliance; Centers for Disease Control and Scale for Osteoarthritis. Arthritis Care Res (Hoboken). 2012;64:862-
Prevention; Arthritis Foundation. A National Public Health Agenda 71. [PMID: 22290689] doi:10.1002/acr.21626
for Osteoarthritis: 2020 Update. 2020. 40. Talley KM, Wyman JF, Gross CR. Psychometric properties of the
23. Moonaz S, Nault D, Cramer H, et al. CLARIFY 2021: explanation Activities-Specific Balance Confidence Scale and the Survey of Activities
and elaboration of the Delphi-based guidelines for the reporting of and Fear of Falling in older women. J Am Geriatr Soc. 2008;56:328-33.
yoga research. BMJ Open. 2021;11:e045812. [PMID: 34353794] [PMID: 18179492] doi:10.1111/j.1532-5415.2007.01550.x
doi:10.1136/bmjopen-2020-045812 41. Meade LB, Bearne LM, Godfrey EL. Comprehension and face
24. National Institute for Health and Care Excellence. Osteoarthritis: validity of the Exercise Adherence Rating Scale in patients with per-
care and management. Clinical guideline [CG177]. 12 February 2014. sistent musculoskeletal pain. Musculoskeletal Care. 2018;16:409-
25. Exercise & Sports Science Australia. Adult Pre-Exercise Screening 12. [PMID: 29575381] doi:10.1002/msc.1240
System (APSS). Accessed at www.essa.org.au/Public/ABOUT_ESSA/ 42. Fransen M, McConnell S, Harmer AR, et al. Exercise for osteo-
Pre-Exercise_Screening_Systems.aspx on 29 September 2020. arthritis of the knee: a Cochrane systematic review. Br J Sports
26. Michie S, Ashford S, Sniehotta FF, et al. A refined taxonomy of Med. 2015;49:1554-7. [PMID: 26405113] doi:10.1136/bjsports-
behaviour change techniques to help people change their physical
2015-095424
activity and healthy eating behaviours: the CALO-RE taxonomy.
43. Bennell KL, Campbell PK, Egerton T, et al. Telephone coaching
Psychol Health. 2011;26:1479-98. [PMID: 21678185] doi:10.1080/
to enhance a home-based physical activity program for knee osteo-
08870446.2010.540664
arthritis: a randomized clinical trial. Arthritis Care Res (Hoboken).
27. Nicolson PJA, Hinman RS, French SD, et al. Improving adher-
2017;69:84-94. [PMID: 27111441] doi:10.1002/acr.22915
ence to exercise: do people with knee osteoarthritis and physical
44. Wasserstein RL, Schirm AL, Lazar NA. Moving to a world
therapists agree on the behavioral approaches likely to succeed.
beyond “p < 0.05”. Am Stat. 2019;73(suppl 1):1-19. doi:10.1080/
Arthritis Care Res (Hoboken). 2018;70:388-97. [PMID: 28575544]
00031305.2019.1583913
doi:10.1002/acr.23297
45. Fitzmaurice G, Laird NM, Ware JH. Applied Longitudinal
28. Hawker GA, Mian S, Kendzerska T, et al. Measures of adult
Analysis. 2nd ed. J Wiley; 2011.
pain: visual analog scale for pain (VAS pain), numeric rating scale
46. Alaiti RK, Castro J, Lee H, et al. What are the mechanisms of
for pain (NRS pain), McGill Pain Questionnaire (MPQ), Short-Form
action of cognitive-behavioral, mind-body, and exercise-based
McGill Pain Questionnaire (SF-MPQ), Chronic Pain Grade Scale
(CPGS), Short Form-36 Bodily Pain Scale (SF-36 BPS), and measure interventions for pain and disability in people with chronic primary
of Intermittent and Constant Osteoarthritis Pain (ICOAP). Arthritis musculoskeletal pain? A systematic review of mediation studies
Care Res (Hoboken). 2011;63 Suppl 11:S240-52. [PMID: 22588748] from randomized controlled trials. Clin J Pain. 2022;38:502-9.
doi:10.1002/acr.20543 [PMID: 35686580] doi:10.1097/AJP.0000000000001047
29. McHugh M, Droy E, Muscatelli S, et al. Measures of adult knee 47. Lima YL, Lee H, Klyne DM, et al. How do non-surgical interven-
function. Arthritis Care Res (Hoboken). 2020;72 Suppl 10:219-49. tions improve pain and physical function in people with osteoarthri-
[PMID: 33091264] doi:10.1002/acr.24235 tis? A scoping review of mediation analysis studies. Arthritis Care
30. McAlindon TE, Driban JB, Henrotin Y, et al. OARSI Clinical Res (Hoboken). 2022. [PMID: 35866717] doi:10.1002/acr.24983
Trials Recommendations: design, conduct, and reporting of clinical 48. Zhang W, Robertson J, Jones AC, et al. The placebo effect and
trials for knee osteoarthritis. Osteoarthritis Cartilage. 2015;23:747- its determinants in osteoarthritis: meta-analysis of randomised con-
60. [PMID: 25952346] doi:10.1016/j.joca.2015.03.005 trolled trials. Ann Rheum Dis. 2008;67:1716-23. [PMID: 18541604]
31. Bellamy N, Kirwan J, Boers M, et al. Recommendations for a doi:10.1136/ard.2008.092015
core set of outcome measures for future phase III clinical trials in 49. Moustgaard H, Clayton GL, Jones HE, et al. Impact of blinding
knee, hip, and hand osteoarthritis. Consensus development at on estimated treatment effects in randomised clinical trials: meta-
OMERACT III. J Rheumatol. 1997;24:799-802. [PMID: 9101522] epidemiological study. BMJ. 2020;368:l6802. [PMID: 31964641]
32. Bellamy N, Carette S, Ford PM, et al. Osteoarthritis antirheumatic doi:10.1136/bmj.l6802
drug trials. III. Setting the delta for clinical trials—results of a consensus 50. Sakellariou G, Conaghan PG, Zhang W, et al. EULAR recom-
development (Delphi) exercise. J Rheumatol. 1992;19:451-7. [PMID: mendations for the use of imaging in the clinical management of
1578462] peripheral joint osteoarthritis. Ann Rheum Dis. 2017;76:1484-94.
33. Bellamy N, Buchanan WW, Goldsmith CH, et al. Validation [PMID: 28389554] doi:10.1136/annrheumdis-2016-210815
study of WOMAC: a health status instrument for measuring clinically 51. Quicke JG, Runhaar J, van der Windt DA, et al. Moderators of
important patient relevant outcomes to antirheumatic drug therapy the effects of therapeutic exercise for people with knee and hip
in patients with osteoarthritis of the hip or knee. J Rheumatol. osteoarthritis: a systematic review of sub-group analyses from
1988;15:1833-40. [PMID: 3068365] randomised controlled trials. Osteoarthr Cartil Open. 2020;2:100113.
34. Tubach F, Ravaud P, Baron G, et al. Evaluation of clinically rele- doi:10.1016/j.ocarto.2020.100113
vant changes in patient reported outcomes in knee and hip osteoar- 52. Cheung C, Wyman JF, Savik K . Adherence to a yoga program
thritis: the minimal clinically important improvement. Ann Rheum in older women with knee osteoarthritis. J Aging Phys Act. 2016;
Dis. 2005;64:29-33. [PMID: 15208174] 24:181-8. [PMID: 26214142] doi:10.1123/japa.2015-0048

1354 Annals of Internal Medicine • Vol. 175 No. 10 • October 2022 Annals.org
Effectiveness of an Online Yoga Program in People With Knee Osteoarthritis ORIGINAL RESEARCH
53. Pisters MF, Veenhof C, Schellevis FG, et al. Exercise adherence intervention trial. Exp Gerontol. 2020;134:110886. [PMID: 32088398]
improving long-term patient outcome in patients with osteoarthritis doi:10.1016/j.exger.2020.110886
of the hip and/or knee. Arthritis Care Res (Hoboken). 2010;62:1087- 56. Bossen D, Veenhof C, Van Beek KE, et al. Effectiveness of a
94. [PMID: 20235201] doi:10.1002/acr.20182 web-based physical activity intervention in patients with knee and/
54. Meade LB, Bearne LM, Sweeney LH, et al. Behaviour change tech- or hip osteoarthritis: randomized controlled trial. J Med Internet
niques associated with adherence to prescribed exercise in patients Res. 2013;15:e257. [PMID: 24269911] doi:10.2196/jmir.2662
with persistent musculoskeletal pain: systematic review. Br J Health 57. Allen KD, Arbeeva L, Callahan LF, et al. Physical therapy vs
Psychol. 2019;24:10-30. [PMID: 29911311] doi:10.1111/bjhp.12324 internet-based exercise training for patients with knee osteoar-
55. Park J, Sherman DG, Agogo G, et al. Frailty modifies the interven- thritis: results of a randomized controlled trial. Osteoarthritis
tion effect of chair yoga on pain among older adults with lower ex- Cartilage. 2018;26:383-96. [PMID: 29307722] doi:10.1016/j.
tremity osteoarthritis: secondary analysis of a nonpharmacological joca.2017.12.008

ACP JOURNALWISE
Personalized Literature Updating Service

Available exclusively to ACP members, ACP JournalWise is an e-mail

notification service that provides the literature alerts you want when you want
them. The alerts can be customized to meet your needs. Alerts are based on
your preferred areas of specialty and on clinical impact.

Benefits of ACP JournalWise include:

 Research alerts: Receive notifications of the latest research in clinical areas

you pick, filtered for quality and relevance
 Tables of contents: Browse the most recent tables of contents you select
from over 150 medical journals
 Folder sharing: Share article titles and abstracts with your colleagues and
follow those that others are reading
 Flexible format: Access it all on your desktop, tablet, or smartphone
Sign up today at journalwise.acponline.org.

Annals.org Annals of Internal Medicine • Vol. 175 No. 10 • October 2022 1355
Author Contributions: Conception and design: K.L. Bennell, R. scale). Imputed data sets were compared with complete
S. Hinman, D.J. Hunter, S. Schwartz, P.L. Teo. data using density plots for continuous outcomes and de-
Analysis and interpretation of the data: K.L. Bennell, R.S. Hinman, scriptive tabulations for the binary outcome.
D.J. Hunter, K.E. Lamb, F. McManus, B. Metcalf, P.L. Teo.
Drafting of the article: K.L. Bennell, A.J. Kimp, K.E. Lamb, B. Exploratory Analyses Prespecified in the
Metcalf, P.L. Teo. Statistical Analysis Plan: Assessing Moderation
Critical revision for important intellectual content: K.L. Bennell, R.S. of the Yoga Treatment Effect
Hinman, D.J. Hunter, K.E. Lamb, F. McManus, B. Metcalf, P.L. Teo. Exploratory analyses using multiply imputed data
Final approval of the article: K.L. Bennell, S. Hawkins, R.S. were conducted to assess the moderation of the effect of
Hinman, D.J. Hunter, A.J. Kimp, K.E. Lamb, D. Mackenzie, F. randomized treatment group on the primary outcomes
McManus, B. Metcalf, S. Schwartz, P.L. Teo. at 12 weeks by preidentified potential moderators (age,
Statistical expertise: K.E. Lamb, F. McManus.
pain self-efficacy, expectation of treatment effects, body
Obtaining of funding: K.L. Bennell.
mass index, and fear of movement). Expectation of treat-
Administrative, technical, or logistic support: K.L. Bennell, S.
ment effects was dichotomized into “benefit” (moderate
Hawkins, A.J. Kimp, D. Mackenzie, B. Metcalf, S. Schwartz, P.L. Teo.
improvement, large improvement, and complete recov-
Collection and assembly of data: A.J. Kimp, B. Metcalf, S.
Schwartz, P.L. Teo.
ery) and “no benefit” (no effect at all and minimal
improvement). Linear regression models were fit sepa-
rately for each outcome and each potential moderator
APPENDIX: EXPANDED DESCRIPTIONS OF adjusted for the outcome at baseline, with terms for
STATISTICAL ANALYSIS METHODS the potential moderator, the randomized group, and the
Comparative analyses between groups were per- interaction between the randomized group and the
formed using Stata, version 16.1, and using the inten- moderator included. Results for the continuous potential
tion-to-treat principle according to the group to which moderators were calculated as the estimated mean
the participant was randomly assigned, regardless of effect (and 95% CI) on each of the primary outcomes of a
whether they adhered to the intervention. 1-unit increase in each continuous potential moderator:
age, pain self-efficacy, body mass index, or fear of move-
Imputation of Missing Outcomes ment for each group at 12 weeks. Results for the binary
Missing continuous outcomes were imputed using potential moderator were calculated as the estimated
chained equations with predictive mean matching and 5 mean effect (and 95% CI) on each of the primary out-
nearest neighbors, using the “mi impute chained” com- comes of an expectation of “benefit” or “no benefit” for
mand with the “pmm” and “knn(5)” options. Binary out-
each group at 12 weeks. Mean differences (with 95% CIs)
comes were also imputed using the “mi impute chained”
between groups were also calculated.
command with logistic regression models. Data were
imputed for each treatment group separately. Estimates
Additional Exploratory Analyses Not Specified in
from 10 imputed data sets were combined using Rubin's
the Statistical Analysis Plan: Adjusting for
rules (58).
All continuous outcomes at 12 and 24 weeks were Participant Characteristics That Were
imputed together. In the imputation models for these con- Unbalanced at Baseline
tinuous outcomes, all outcomes at baseline were included Given differences observed at baseline between
in the model, as well as the following baseline characteris- groups for gender and the proportion of participants
tics: age, gender, body mass index, current employment with problems in other joints, we performed unplanned
status, laterality of symptoms, duration of knee symptoms, post hoc analyses of the 2 primary outcomes with adjust-
pain in other parts of the body, average spine pain, aver- ment for these baseline variables. This was done using
age hip pain, number and type of comorbid conditions, the same modeling approach as for continuous out-
pain medication use, physical activity level (assessed via comes, using multiply imputed data as described earlier.
the incidental and planned exercise questionnaire [IPEQ-
W]), expectation of treatment outcome, beliefs about Web Reference
effectiveness of yoga, and self-efficacy when using a com- 58. Carpenter JR, Kenward MG. Multiple Imputation and Its
puter (assessed via the modified computer self-efficacy Application. J Wiley; 2013. doi:10.1002/9781119942283

Annals.org Annals of Internal Medicine • Vol. 175 No. 10 • October 2022

Appendix Table 1. Baseline Characteristics of Participants Who Provided Both Primary Outcomes at 12 Weeks and Those Who
Did Not
Characteristic Incomplete Primary Outcomes (n = 17) Complete Primary Outcomes (n = 195)

Group, n (%)
Control 8 (47.1) 97 (49.7)
Yoga 9 (52.9) 98 (50.3)
Mean age (SD), y 60.1 (7.6) 62.5 (7.7)
Female, n (%) 12 (70.6) 136 (69.7)
Mean height (SD), m 1.7 (0.1) 1.7 (0.1)
Mean body mass (SD), kg 91.1 (20.7) 86.7 (17.6)
Mean body mass index (SD), kg/m2 30.7 (5.3) 30.3 (5.5)
Place of birth, n (%)
Australia/Oceania 11 (64.7) 138 (70.8)
Asia 0 (0.0) 11 (5.6)
Europe 6 (35.3) 41 (21.0)
North America 0 (0.0) 5 (2.6)
Currently employed, n (%) 8 (47.1) 95 (48.7)
Right knee more painful, n (%) 10 (58.8) 116 (59.5)
Unilateral symptoms, n (%) 5 (29.4) 61 (31.3)
Median knee symptom duration (IQR), y 6 (2–10) 4 (2–10)
Problems in other joints, n (%)
Upper body 9 (52.9) 82 (42.1)
Back and neck 7 (41.2) 88 (45.1)
Hip 6 (35.3) 51 (26.2)
Foot 2 (11.8) 50 (25.6)
Median spine pain severity on NRS (IQR) 2 (0–3) 1 (0–3)
Median hip pain severity on NRS (IQR) 3 (1–6) 1 (1–4)
Comorbid conditions, n (%)
≥1 comorbid condition 10 (58.8) 108 (55.4)
Heart disease 0 (0.0) 17 (8.7)
High blood pressure 4 (23.5) 51 (26.2)
Depression 2 (11.8) 23 (11.8)
Diabetes 2 (11.8) 8 (4.1)
Spine condition 4 (23.5) 22 (11.3)
Lung disease 4 (23.5) 25 (12.8)
Cancer 0 (0.0) 11 (5.6)
Nervous system disease 0 (0.0) 1 (0.5)
Kidney disease 0 (0.0) 3 (1.5)
Liver disease 0 (0.0) 2 (1.0)
Other 1 (5.9) 6 (3.1)
Treatment for knee in previous 3 mo, n (%)
≥1 treatment 10 (58.8) 156 (80.0)
Massage/manual therapy 4 (23.5) 53 (27.2)
Gait aid 3 (17.6) 13 (6.7)
Thermal therapy/electrotherapy 3 (17.6) 80 (41.0)
Orthotics, arch supports, or wedging in shoes 2 (11.8) 48 (24.6)
Knee braces 7 (41.2) 50 (25.6)
Land-based and water exercises 4 (23.5) 59 (30.3)
Joint injections 0 (0.0) 2 (1.0)
Acupuncture 0 (0.0) 9 (4.6)
Current pain medication use, n (%)*
≥1 medication 14 (82.4) 141 (72.3)
Nonsteroidal anti-inflammatory drugs 7 (41.2) 77 (39.5)
Acetaminophen 10 (58.8) 114 (58.5)
Topical anti-inflammatory drugs 5 (29.4) 71 (36.4)
Oral corticosteroids 0 (0.0) 2 (1.0)
Oral opioids 1 (5.9) 2 (1.0)
Median physical activity (IQR), h/wk† 30.1 (14.6–40.4) 22.5 (12.5–36.3)
Expectation of treatment outcome, n (%)‡
No effect at all 0 (0.0) 4 (2.1)
Minimal improvement 3 (17.6) 43 (22.1)
Moderate improvement 12 (70.6) 116 (59.5)
Large improvement 2 (11.8) 31 (15.9)
Complete recovery 0 (0.0) 1 (0.5)

Continued

Annals of Internal Medicine • Vol. 175 No. 10 • October 2022 Annals.org

 Appendix Table 1–Continued
Characteristic Incomplete Primary Outcomes (n = 17) Complete Primary Outcomes (n = 195)

Beliefs about effectiveness of yoga, n (%)§

Not effective 1 (5.9) 3 (1.5)
Somewhat effective 5 (29.4) 80 (41.0)
Moderately effective 9 (52.9) 79 (40.5)
Highly effective 2 (11.8) 33 (16.9)
Median self-efficacy using computer (IQR)|| 4.5 (3.5–4.7) 4.6 (3.9–4.9)

NRS = numerical rating scale (overall average pain of 0–10 in the previous week, with higher scores indicating more pain).
* Defined as ≥1 time per week over the previous 3 months.
† From the incidental and planned exercise questionnaire (IPEQ-W), which includes 10 questions about frequency and duration of incidental and
planned walking, sport, and recreational activities over the previous 7 days. Scores are calculated as the product of the frequency and duration
scores to create a total score for duration for the week, with higher scores indicating higher levels of activity.
‡ Scored from a question asking about expectation of study treatment outcomes, with self-reported scores on a 5-point Likert scale ranging from 0
(“no effect at all”) to 4 (“complete recovery”).
§ Participants were asked to rate their beliefs about the effectiveness of yoga on a 4-point Likert scale ranging from 1 (“not effective”) to 4 (“highly
effective”).
|| From a modified version of the computer self-efficacy scale, which involved taking the average of scored responses to 26 of the original 32 ques-
tions on a 5-point Likert scale indicating degree of confidence from 1 (“very little confidence”) to 5 (“quite a lot of confidence”).

Appendix Table 2. Potential Binary Moderator of the Effect of Yoga on Primary Outcomes at 12 Weeks Using Multiply Imputed
Data
Moderator, by Outcome (12 Weeks Mean (SD)* Mean Difference (95% CI) for P Value for
Minus Baseline) Yoga Minus Control (n = 212) Interaction
Control (n = 105) Yoga (n = 107)

Knee pain during walking (NRS)† 0.162

Expectation of treatment outcome‡
No benefit 0.15 (2.74) 1.79 (1.86) 1.44 ( 2.76 to 0.12)
Benefit 1.76 (2.18) 2.07 (2.43) 0.35 ( 1.08 to 0.38)

Physical function (WOMAC)§ 1.00

Expectation of treatment outcome‡
No benefit 3.50 (12.07) 7.74 (8.85) 4.16 ( 9.79 to 1.46)
Benefit 4.99 (10.36) 9.86 (9.72) 4.16 ( 7.37 to 0.95)
MCID = minimal clinically important difference; NRS = numerical rating scale; WOMAC = Western Ontario and McMaster Universities Osteoarthritis
Index.
* The effect on change in knee pain during walking or change in physical function at 12 weeks of an expectation of “benefit” or “no benefit” at base-
line in the control and yoga groups.
† The NRS ranges from 0 to 10, with higher scores indicating greater pain. The MCID is 1.8 units.
‡ Scored from a question asking about expectation of study treatment outcomes, with self-reported scores on a 5-point Likert scale ranging from 0
(“no effect at all”) to 4 (“complete recovery”) dichotomized as “benefit” (moderate improvement, large improvement, or complete recovery) and “no
benefit” (no effect at all or minimal improvement).
§ The WOMAC ranges from 0 to 68 for physical function, with higher scores indicating more physical dysfunction. The MCID is 6 nonnormalized
units.

Annals.org Annals of Internal Medicine • Vol. 175 No. 10 • October 2022

 Appendix Table 3. Potential Continuous Moderators of the Effect of Yoga on the Primary Outcomes at 12 Weeks Using
Multiply Imputed Data*
Outcome (12 Moderator Moderator Coefficient Moderator Coefficient Interaction Between
Weeks Minus for Control† for Yoga† Moderator and Group
Baseline)
Mean P Value Mean P Value Difference (95% CI) P Value for
(95% CI) (95% CI) in Coefficients Interaction
(n = 105) (n = 107) (Yoga Minus Control)
Knee pain during Age (years) 0.00 ( 0.06 to 0.07) 0.92 0.00 ( 0.06 to 0.05) 0.87 0.01 ( 0.09 to 0.08) 0.85
walking (NRS)‡
Physical function Age (years) 0.09 ( 0.17 to 0.36) 0.49 0.15 ( 0.09 to 0.39) 0.22 0.05 ( 0.30 to 0.41) 0.76
(WOMAC)§
Knee pain during Arthritis self-efficacy: 0.07 ( 0.35 to 0.21) 0.63 0.20 ( 0.47 to 0.08) 0.164 0.13 ( 0.52 to 0.27) 0.53
walking (NRS)‡ pain||
Physical function Arthritis self-efficacy: 1.05 ( 2.19 to 0.09) 0.072 0.61 ( 1.75 to 0.53) 0.29 0.43 ( 1.18 to 2.05) 0.60
(WOMAC)§ pain||
Knee pain during Body mass index 0.04 ( 0.04 to 0.11) 0.30 0.10 (0.00 to 0.19) 0.049 0.06 ( 0.07 to 0.18) 0.36
walking (NRS)‡ (kg/m2)
Physical function Body mass index 0.25 ( 0.07 to 0.57) 0.126 0.25 ( 0.14 to 0.65) 0.21 0.00 ( 0.50 to 0.51) 0.99
(WOMAC)§ (kg/m2)
Knee pain during Fear of movement¶ 0.03 ( 0.16 to 0.10) 0.69 0.12 ( 0.04 to 0.28) 0.147 0.14 ( 0.06 to 0.35) 0.170
walking (NRS)‡
Physical function Fear of movement¶ 0.07 ( 0.62 to 0.49) 0.81 0.37 ( 0.25 to 0.99) 0.24 0.44 ( 0.39 to 1.27) 0.30
(WOMAC)§
MCID = minimal clinically important difference; NRS = numerical rating scale; WOMAC = Western Ontario and McMaster Universities Osteoarthritis
Index.
* Results are presented in terms of a 1-unit increase in the potential continuous moderator in each group.
† The effect at 12 weeks on change in knee pain during walking or change in physical function of a 1-unit increase in the potential moderators in
each group.
‡ The NRS ranges from 0 to 10, with higher scores indicating greater pain. The MCID is 1.8 units.
§ The WOMAC ranges from 0 to 68 for physical function, with higher scores indicating more physical dysfunction. The MCID is 6 nonnormalized
units.
|| Based on the pain subscale of the Arthritis Self-Efficacy Scale, which ranges from 1 to 10, with higher scores indicating greater pain self-efficacy.
¶ Based on the Brief Fear of Movement Scale for Osteoarthritis, which ranges from 6 to 24, with higher scores indicating greater fear of movement.

Annals of Internal Medicine • Vol. 175 No. 10 • October 2022 Annals.org