Predictors of Exercise Adherence in Patients with Meniscal Tear and Osteoarthritis

Abstract

Objective

To identify predictors of poor exercise adherence in patients with osteoarthritis (OA) and meniscal tear.

Design

Secondary analysis of data gathered over the first 12 weeks in the Meniscal Tear in Osteoarthritis Research (MeTeOR) Trial, a multicenter, randomized controlled trial.

Setting

Seven referral centers in the US.

Participants

This analysis was conducted in 325 of the 351 MeTeOR patients, each of whom was ≥ 45 years old and had meniscal tear and osteoarthritic changes on imaging studies; 26 were excluded due to missing data from which to derive the primary outcome variable.

Interventions

All patients received a structured supervised exercise program focused on strengthening, along with prescribed home exercises; half were randomized to also receive arthroscopic partial meniscectomy.

Main Outcome Measure

Poor exercise adherence through 12 weeks, defined as performing <50% of prescribed exercise.

Results

38% of the MeTeOR cohort showed poor exercise adherence. In the multivariate model, adjusting for treatment group, those who earned ≤ $29,000/year had 1.64 times the risk of non-adherence (95% CI: 1.10, 2.43) than those who earned >$100,000 /year; and, those without baseline pain with pivoting and twisting had 1.60 times greater risk of non-adherence than those with these symptoms (95% CI: 1.14, 2.25).

Conclusion

Low income was associated with poor exercise adherence among patients ≥ 45 with osteoarthritis and meniscal tear, as was absence of pain with pivoting and twisting. Our findings highlight the need for further research into exercise adherence and for interventions to enhance adherence among those with low incomes.

Introduction

Osteoarthritis (OA) compromises the mobility and quality of life of over 250 million people around the world [1–3] and knee OA is the most frequent cause of musculoskeletal disability in industrialized nations [4–9]. Degenerative meniscal tear is a common cause of morbidity in adults, and arthroscopic partial meniscectomy (APM) is among the most commonly performed orthopedic surgeries [10]. Exercise is a subset of physical activity that is planned, structured and purposeful [11]. Aerobic and strengthening exercises have been shown to be effective treatments for reducing pain and improving function in patients with knee OA [12–20]. Postoperative exercises are associated with improvements in self-reported measures of knee function and range of motion among patients who have undergone arthroscopic partial meniscectomy (APM) for degenerative meniscal tear [21]. In patients with OA and symptomatic meniscal tear, the treatment effect of exercise is comparable to that of APM [15–17].

Increasingly, scientists are moving beyond the known benefits of exercise among adults with degenerative knee disease and are studying risk factors that influence patient adherence to exercise [13, 14, 18]. In this clinical population, it has been shown that the benefits of exercise cannot be realized unless patients adhere to their programs [12, 20]. Furthermore, as a 2005 multidisciplinary expert panel suggested, long-term adherence to a recommended exercise program is the key predictor of the long-term clinical success of that program [22].

Although a positive dose-response relationship between exercise adherence and health benefits among adults with OA has been described [14, 18], adults with knee OA often become non-adherent to their exercise program just weeks after exercise is prescribed [13, 19, 20]. No exercise adherence study has been conducted among patients with knee OA and concomitant meniscal tear.

Literature on exercise adherence often describes two broad categories of factors that motivate exercise: intrinsic (i.e. patient-level clinical and psycho-social factors) and extrinsic (i.e. physical and environmental factors outside of the patient) [23–26]. Intrinsic factors include mood/affect, pain, attitude towards exercise, and expectation of outcome, among others, while extrinsic factors include social support, socioeconomic status, accessibility of exercise facility and transportation, among others [13, 27–33]. A recent review noted the factors that most significantly influence exercise adherence in the OA population are extrinsic, and fall into the category of ‘Environmental Context and Resources’ [34]. The factors associated with poor exercise adherence in patients with OA and concomitant symptomatic meniscal tear have not been examined explicitly. The aim of the study was to help fill this research gap.

Methods

Design and Setting

We performed a prospective, observational secondary analysis of data from the Meniscal Tear in Osteoarthritis Research Trial (MeTeOR; Clinical Trials.gov NCT00597012), a multicenter randomized controlled trial (RCT) involving adult patients with symptomatic mild-to-moderate osteoarthritic changes and concomitant meniscal tear [17, 35]. Between June 2008 and July 2011, eligible patients were randomized to receive APM followed by physical therapy or a physical therapy regimen alone. Inclusion criteria for the MeTeOR trial included clinical symptoms consistent with a torn meniscus, availability of knee radiograph and MRI, evidence of cartilage defects on knee MRI or evidence of osteophytes or joint space narrowing on plain radiograph, and evidence of meniscal tear on knee MRI. MRI readings were standardized by using the Boston Leeds Osteoarthritis Knee Score [36] and a quantitative 3D assessment of articular cartilage volume.

Before randomization, all MeTeOR patients received a baseline questionnaire examining clinical data such as perceived pain and demographic data such as socioeconomic status. After randomization, all MeTeOR patients were called every other week to assess adherence to the prescribed exercise program. Both operative and non-operative study arms received an exercise program consisting of supervised physical therapy sessions and an unsupervised home exercise program. The operative arm commenced exercise after surgery and the non-operative arm commenced exercise within two weeks of randomization. This analysis used MeTeOR data from the baseline questionnaires and the biweekly phone interviews conducted through the first 12 weeks of follow-up.

Exercise Program

The MeTeOR physical therapy program was a standardized, land-based protocol administered to patients in both arms. The research coordinator distributed the protocol to each patient at the randomization visit, along with a set of 2.5-pound therapy weights. All patients progressed through three stages addressing range of motion, proprioception, balance, eccentric strength, concentric strength, and aerobic conditioning. Patients worked with physical therapists up to twice weekly and undertook a written home exercise program in addition to their outpatient therapy sessions. Readiness to advance through stages of the exercise program was determined by the physical therapist using a protocol that included self-reported pain, observed strength, knee range of motion, functional mobility, and knee effusion. Over-the-counter pain medications were allowed as needed during the trial period, as were prescribed pain medications and intra-articular corticosteroid injections as determined by patient and physician.

Underlying Theoretical Model

Our framework was informed by elements of an exercise behavior model described by Bennell, et al. [33]. We anchored our analysis by first distinguishing intrinsic and extrinsic factors. Intrinsic factors are clinical and psycho-social factors within a patient such as pain, disease severity (i.e. Kellgren-Lawrence OA grade), attitude towards health, and health expectations. Extrinsic factors are physical and environmental factors outside of a patient such as socioeconomic status and social support. We present the intrinsic and extrinsic factors considered in this analysis in Table 1. The variables we used to represent these intrinsic and extrinsic factors included validated measures such as the SF-36 [37], Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) function scale [38], Knee injury and Osteoarthritis Outcome Scale (KOOS) [39], and the five-item mental health index (MHI-5) contained within the SF-36 [40].

Table 1.

Intrinsic and Extrinsic Factors Examined

Intrinsic factors	Source	Scoring system
Demographic Data
Age, race, gender, height, weight	Self-reported
Living status	Who do you currently live with?	Alone versus Not alone
Subjective Clinical Data
General affect	MHI-5 [46] (a subscale within SF- 36[43])	Index score: 5–25 (5 items each score 1–5)
Symptoms of anxiety/depression	Anxiety/Depression items in EQ- 5D[46]	Item score: 0–2
Expectations of intervention	How much do you expect your treatment (surgery or physical therapy) to improve your quality of life over the NEXT COUPLE YEARS?	Item score: 0–4
Perception of personal health	Current health status subscale of EQ-5D [55]	Item score: 0–100
Clinical symptoms of knee OA	KOOS symptoms subscale[45]	Index score: 0–20
Clinical symptoms of meniscal tear	In the LAST WEEK, have you had any of the following symptoms in your [INDEX] KNEE? Clicking, catching, popping, giving way, locking, swelling, pain with pivoting/twisting?	Yes versus No for each item
Knee pain with specific activities	KOOS pain subscale[45]	Index score: 0–36
General pain	Pain/Discomfort subscale of EQ- 5D[55]	Item score: 0–2
Index knee-related function during athletic activities or specific postures and activities of daily living	SF-36 activities subscale[43]	Index score: 9–27 (9 items scores 1–3)
General function during ADLs and mild physical activity	KOOS function/daily living subscale[45]	Index score: 0–68 (17 items scores 014)
Objective Clinical Data
Need to take oral medication for knee pain	How often do you take each of the following medications for your knee?	Index score: 0–6
Need for intra-articular knee injection(s)	Have you had injections in your knee in the LAST THREE MONTHS?	Yes versus No
Kellgren and Lawrence OA grade on plain films[47]	AP standing plain radiograph reviewed and scored during pre-trial screening by participating physician	Grade 0 – 4
Score on Timed Up and Go test	Timed Up and Go test supervised by physical therapist during pre- trial musculoskeletal exam[48]	Seconds
Score on Quadriceps and Hamstrings manual muscle testing	Manual muscle testing[49] performed and recorded by physical therapist during pre-trial musculoskeletal exam	Average of 3 trials using the manual muscle test hand held dynamometer, pounds
Extrinsic Factors	Source	Scoring system
Annual income	What is your annual household income from all sources before taxes?	Item score: 1–5
Randomization arm	Randomization arm assigned during pre-trial screening by research coordinator (Surgical / Non- Surgical)	Surgery versus Physical Therapy
Highest level of education achieved	What is the highest level of education you have achieved?	Item score: 1–5

Outcome

The primary outcome was poor exercise adherence over the course of 12 weeks. Based on prior literature, we defined those who adhered to < 50% of their program elements over the course of 12 weeks as non-adherent [41].

Research staff called patients bi-weekly during first 12 weeks post-randomization and asked about the number of PT sessions patients were scheduled to attend in the prior 2 weeks and the number they actually attended. Similarly, the researcher asked patients the number of days on which they did all, part, or none of their prescribed home exercise program.

Thus, exercise adherence for each study participant was calculated as:

$$\begin{array}{l}[(\#\mathit{attended}\mathit{physical}\mathit{therapy}\mathit{days})]+[\mathit{1}\ast (\#\mathit{Days}\mathit{Full}\mathit{Home}\mathit{Exercise}\mathit{Program}\\ \mathit{Completed})+\mathit{0.5}\times (\#\mathit{Days}\mathit{Partial}\mathit{Home}\mathit{Exercise}\mathit{Program}\mathit{Completed})]/\mathit{84}\mathit{days}\end{array}$$

Since a variety of exercises under varying conditions have been shown to improve clinical outcomes among patients with degenerative knee disease, our outcome of interest was total adherence to exercise, rather than adherence to home exercise or adherence to physical therapy alone. Thus the numerator, bi-weekly adherence data from each individual participant, was summed. We assumed that study patients would not perform home exercises on the days they had physical therapy. However, to account for potential redundancy arising from reporting home exercise and PT attendance on the same day, the outcome variable was capped at 1 for each patient. We developed conservative algorithms for augmenting data in instances of partially missing data on physical therapy sessions and home exercises (Appendix 1 provides the algorithms). Study patients not providing any data on physical therapy or home exercise data were excluded because the outcome variable could not be calculated (n=26, 7%).

Statistical Analysis

We assessed each of the potential predictors of exercise non-adherence (listed in Table 1) individually, in bivariate analyses using the chi-squared test for categorical predictors and the t-test for continuous variables. We advanced to multivariate analyses those variables that had relative risks of over 1.5 or under 0.67, or p-values less than or equal to 0.25. We identified independent predictors of non-adherence with multivariable models. In order to provide risk ratios rather than odds ratios, we utilized a modified Poisson regression approach with robust error variance [42]. We conducted sensitivity analyses at both 4- and 8-weeks post-randomization by examining the association between the same predictors identified at 12-weeks and exercise non-adherence using the same modeling approach used for the 12-week analyses. SAS v9.4 was used to perform all statistical analyses.

Results

Cohort Demographics

The study was conducted in 325 of the 351 patients in the MeTeOR study. Twenty-six patients from the MeTeOR cohort were excluded from the analysis due to completely missing data from which to derive an outcome variable.

Of the patients included in this analysis, 57% were female and 86% were white (Table 2). Sixteen of patients reported living alone. In general, the cohort was well educated with 84% having attended college. Most patients were employed, with 10% of the group reporting earning less than $30,000 annually.

Table 2.

Baseline Demographic and Clinical Characteristics of Cohort (N=325)

Variable	Frequency	Percent
Study Arm
Physical Therapy	172	52.9
APM surgery	153	47.1
Demographics
Age (years)
45–54	120	36.9
55–64	146	44.9
65–74	54	16.6
≥75	5	1.5
Female
No	139	42.8
Yes	186	57.2
White Race
No	45	13.8
Yes	280	86.1
Education Level
High School Graduate or Less	51	16.0
Some College	51	16.0
Graduated from Technical/Professional School	36	11.3
College Graduate	181	56.7
Annual Income
≤$29,000	32	9.8
$30,000–$99,999	152	46.8
≥$100,000	114	35.1
Missing	27	8.3
Living Situation
Alone	51	15.9
With Spouse or Partner	238	74.1
With other Family Members or Friends	32	10.0
BMI (kg/m2)
Missing	17	5.2
≤25.0	62	19.1
25.1–30.0	109	33.5
≥30.1	137	42.1
KL Score
0	68	24.4
1	53	19.0
2	84	30.1
3	74	26.5
Knee Symptoms
Catching
No	156	49.2
Yes	161	50.8
Clicking
No	107	33.6
Yes	211	66.3
Giving Away
No	156	49.5
Yes	159	50.5
Pain with pivoting/twisting
No	23	7.2
Yes	298	92.8
Popping
No	166	52.4
Yes	151	47.6
Episodic Swelling
No	80	25.2
Yes	238	74.8
Quality of Life
SF-36 Physical Activity Subscale
0–39	14	4.3
40–59	136	42.0
60–79	117	36.1
≥80	57	17.6
KOOS Activities of Daily Living Subscale
0–39	39	12.1
40–59	96	29.7
60–79	131	40.6
≥80	57	17.6

APM = arthroscopic partial meniscectomy

Clinical Characteristics and Self-Reported Symptoms

Knee OA severity varied: 24% had Kellgren-Lawrence grade 0 (normal plain radiographs), 19% had grade 1 (questionable osteophyte), 30% had grade 2 (definite osteophyte), and 26% had grade 3 (define joint space narrowing, < 50%) [43]. Thirty-three percent of the cohort was overweight (BMI 25–30) and 42% was obese (BMI > 30). At baseline, 21% of patients reported being anxious or depressed currently. Twenty-nine percent expected a moderate or great improvement in symptoms as a result of the intervention to which they were assigned while 71% anticipated only a little improvement in symptoms. At baseline, 83% reported their current level of pain or discomfort as moderate and 13% as extreme. Twenty-eight percent reported having an intra-articular injection in the past 3 months and half of patients reported daily or near daily use of Tylenol, anti-inflammatory medications or stronger pain medicine for their knee pain.

Functional Characteristics

The baseline functional characteristics of the cohort were as follows: on the physical activity subscale of the SF-36 (possible range 0–100 with 100 best), 14 (4%) reported a score of 0–39, 253 (78%) reported a score of 40–79 and 57 (18%) reported a score of ≥ 80. Thirty-nine patients (12%) endorsed a score of 0–39 on the KOOS activities of daily living subscale (possible score 0–100, 100 best), 227 (70%) endorsed a score of 40–79 and 57 (18%) endorsed a score of ≥ 80.

Total Cohort Adherence

Finally, with respect to the primary outcome, 124 patients (38%) were less than 50% adherent to exercise (non-adherers) over 12 weeks.

Bivariate Analysis

In the bivariate analysis, 3 intrinsic patient factors – male sex, non-white race and less pain with pivoting and twisting, reached or approached statistical significance as predictors of non-adherence. Men had 1.2 times the risk of non-adherence compared to women (95% CI 0.92, 1.60). Non-white study participants had 1.3 times the risk of non-adherence compared to white study participants (95% CI 0.96, 1.88). Patients without pain with pivoting and twisting had 1.7 times the risk of non-adherence compared to those with painful meniscal symptoms (95% CI 1.17, 2.41). One extrinsic factor (lower annual income; Table 3) was also associated with non-adherence. Compared to those who earned greater than $100,000 annually, those who earned ≤$29,000 annually had 1.5 times the risk of non-adherence (95% CI 1.01, 2.19), while those who earned between $30,000 and $99,000 annually had 15% lower risk (95% CI 0.61, 1.19) and those who did not report income data had 1.4 times greater risk of non-adherence to exercise over 12 weeks (95% CI 0.89, 2.12). In addition, compared to patients randomized to the physical therapy arm of the study, patients randomized to the surgical arm of the study had 1.6 times the risk of non-adherence (95% CI 1.17, 2.06).

Table 3.

Bivariate Associations between Candidate Predictors of Non-Adherence at 12 Weeks

Variable	Adherence Level				Parameter Estimates
	<50% (n=124)		≥50% (n=201)		Relative Risk	95% CI	P-value
	N	% or Mean (SD)	N	% or Mean (SD)
Sex							0.17
Female	65	34.9	121	65.0	1.00	Reference
Male	59	42.4	80	57.5	1.21	0.92–1.60
White Race							0.09
Yes	102	36.4	178	63.6	1.00	Reference
No	22	48.9	23	51.1	1.34	0.96–1.88
Study Arm							0.002
PT	52	41.9	120	59.7	1.00	Reference
APM	72	58.1	81	40.3	1.56	1.17–2.06
Annual Income							0.046
≥$100,000	43	37.7	71	62.3	1.00	Reference	--
$30,000–$99,999	49	32.2	103	67.8	0.85	0.61–1.19
≤$29,000	18	56.2	14	43.7	1.49	1.01–2.19
Missing	14	51.8	13	48.1	1.37	0.89–2.12
Pain with pivot							0.005
Yes	108	39.1	190	63.8	1.00	Reference
No	14	60.9	9	36.2	1.68	1.17–2.41
Age (years)							0.84
45–54	48	40.0	72	60.0	1.12	0.75–1.69
55–64	55	37.7	91	62.3	1.06	0.71–1.58
≥65	21	35.6	38	64.4	1.00	Reference

Additional factors included in Table 1 were not associated with exercise non-adherence at OR < 0. 67 or > 1.5 and are therefore not shown

Multivariate Analysis

The final model (Figure 1) included both intrinsic (pain, age) and extrinsic (socioeconomic status) factors as predictors of non-adherence after adjusting for study arm and age. Compared to those who earned $100,000 per annum or more, those who earned less than or equal to $29,000 per annum had 1.6 times the risk of non-adherence (95% CI 1.10, 2.43). Those not experiencing pain with pivoting or twisting had 1.6 times the risk of non-adherence (95% CI 1.14, 2.25) as compared to those who endorsed such pain.

Figure 1.

Multivariate Correlates of Non-Adherence at 12 Weeks

Model also includes randomization assignment

Sensitivity Analysis

To test the strength of select variables to predict non-adherence at various time points following enrollment, we examined predictors of non-adherence at shorter time intervals (4 and 8 weeks post-randomization; Table 4). The models comprised the same variables from the 12-week analysis including annual income, pain with pivoting and twisting and age and were further adjusted for study arm. Lower annual income and no pain with pivoting or twisting became more strongly associated with non-adherence over longer duration of follow-up. Compared to those who earned ≥$100,000 annually, those who earned ≤$29,000 annually had 1.3 times the risk of non-adherence at 4 weeks but 1.6 times the risk at 12 weeks (95% CI 0.88, 1.96 and 1.10, 2.43, respectively). Compared to those who experienced pain with pivoting and twisting, those who did not experience painful meniscal symptoms had the same risk of non-adherence at 4 weeks (95% CI 0.70, 1.52) but 1.6 times the risk of non-adherence at 12 weeks (95 % CI 1.14, 2.25; Table 4). The APM group had greater risk of non-adherence at all 3 time points (data not shown).

Table 4.

Risk Ratios for Select Variables at 4, 8, and 12 Weeks to Test the Effects of these Variables on Exercise Non-Adherence over Time

Risk Factors	4-Weeks			8-Weeks			12-Weeks
	RR	95% CI	P-value	RR	95% CI	P-value	RR	95% CI	P-value
Age, per 5 years ↓	1.01	0.94–1.08	0.82	1.05	0.95–1.15	0.32	1.08	0.98–1.19	0.10
Annual income (versus≥$100,000)			0.52			0.34			0.016
≤$29,000	1.31	0.88–1.96	0.19	1.25	0.76–2.07	0.37	1.64	1.10–2.43	0.015
$30,000–$99,999	0.96	0.75–1.23	0.74	0.88	0.64–1.21	0.43	0.82	0.59–1.14	0.24
Missing	1.09	0.75–1.58	0.66	1.26	0.81–1.95	0.31	1.33	0.88–2.02	0.18
No pain with pivoting/twisting	1.03	0.70–1.52	0.86	1.25	0.81–1.93	0.32	1.60	1.14–2.25	0.006
C-statistic	0.72			0.66			0.66

Discussion

We examined the relationship between intrinsic and extrinsic individual-level factors and non-adherence to prescribed exercise over 12 weeks in a cohort of patients with knee OA and concomitant meniscal tear (the MeTeOR trial patients). Of the factors studied, lower socioeconomic status (extrinsic) and no pain with pivoting and twisting (intrinsic) were associated with higher risk of exercise non-adherence.

One key finding of this research is that patients with lower income were significantly less likely to be adherent. These patients may have had more difficulty attending outpatient physical therapy (e.g. because of difficulty finding transportation, affording co-payments, etc.). This finding conforms to published data among similar cohorts (e.g. well-educated adults or adults with lower extremity osteoarthritis), where decreasing socioeconomic status was negatively associated with physical activity level [44, 45]. The same association has been described among older adults [46].

Another key finding is that patients with less baseline pain with pivoting and twisting had worse adherence. Considering the types of activities that require pivoting and twisting of the knee such as golf and basketball, participants who reported pain with pivoting and twisting may have been more active at baseline than those who did not report this symptom. These patients may have had greater desire to get back to usual activities, leading them to be more adherent than those who were less active at baseline. The finding that patients with more severe pain with twisting and pivoting may have been more adherent is also consistent with prior OA literature showing that patients with more severe symptoms were more likely to adhere to their exercise programs [31]. We did not observe a relationship between overall lower extremity pain as measured by WOMAC and adherence, suggesting this relationship may depend on the type of pain.

In multivariate and sensitivity analyses, study arm was associated with adherence status. Patients randomized to surgery were most likely to be non-adherent. This finding may simply reflect the longer duration of time after randomization required for patients to undergo surgery, get scheduled for PT, and receive PT. Post-surgical pain may have played a role. In addition, patients randomized to surgery may have unwittingly assumed that a high-intensity intervention such as surgery could more swiftly and completely improve their symptoms, obviating their need to participate in a low-intensity intervention like exercise. Thus, if the surgical patients assumed that surgery would be curative and they did not need to exercise, they may have exercised less compared to patients randomized to PT.

The observation that many of the intrinsic and extrinsic factors that we examined were not associated with exercise adherence is consistent with prior literature. Studies of exercise adherence that have focused on healthy adults [25], and musculoskeletal patients [47] have documented few predictors of adherence.

Limitations

Our study was limited by only using data available in the secondary analysis of an RCT. For example, due to the a priori study design of the MeTeOR trial, during bi-weekly telephone calls, research staff did not ask patients to provide reasons why they did not adhere to their exercise programs. Similarly, while the original MeTeOR study allowed patients to use additional medications as needed, data on medication use during the trial was not available for this secondary analysis. If use of medications influenced patient motivation to exercise, this may have confounded our results. Finally, Focht and colleagues demonstrated that osteoarthritic patients randomized to a cognitive behavioral therapy-based exercise program featuring intrinsic factors like self-regulation and goal-setting and extrinsic factors like group identity and social support had increased adherence compared to those randomized to a traditional, center-based exercise program [48]. In the same way that we were not able to examine self-reported reasons for non-adherence, we were not able to measure factors such as self-regulation and goal-setting.

Conclusion

The number of high-quality studies examining exercise adherence in patients with lower extremity musculoskeletal disease is increasing [34], reflecting the key role exercise adherence plays in clinical outcomes within this group. Our findings suggest that socioeconomic status may be an important predictor of non-adherence. Strategies to understand the specific barriers to adherence associated with low income are merited, as are studies of interventions to overcome these barriers. Finally, as our study and other studies emphasize, the intrinsic and extrinsic factors that lead patients with degenerative knee disease to initiate, maintain and progress in an exercise program may be difficult to measure. To optimize clinical outcomes, clinicians need a better understanding of measurable and modifiable predictors of exercise adherence. Focht and colleagues’ have suggested that future interventions should include elements of behavioral psychology, such as exercise stage of change as candidate predictors of adherence [48].

Appendix 1. Algorithms for imputing missing data on physical therapy sessions and home exercise

If a patient had missing data on the number of physical therapy (PT) sessions attended over a specific 2-week period, we imputed the number of PT sessions attended with the average number of PT sessions attended for that patient over the entire study period, calculated based on non-missing data for that person. If the average number of PT sessions attended over 12 weeks was missing (i.e. there was no data available across the entire 12 week span), we assumed the average number of PT sessions attended for that patient over the entire study period was 0. If a patient had missing data on the number of PT sessions assigned over a specific 2-week period, this was imputed by either the average PT sessions attended for that patient or the average PT sessions assigned for that time interval (across all patients) whichever value was smaller. If a patient had missing data on number of PT sessions assigned over 14 days, this was imputed by either the average number of PT sessions attended for that patient over the entire 12-week period or by the average assigned PT session data for that time interval across all study participants, whichever was larger.