Association of physical function and physical activity in women with rheumatoid arthritis
To explore the associations between measures of physical activity (PA) and measures of physical function (PF) in women with rheumatoid arthritis (RA). We hypothesized that the strength of the associations between PA and PF would be moderate, and that after controlling for social and biomedical characteristics, the associations would decrease.
Women with RA (n = 47, mean ± SD age 56.5 ± 7.0 years) participated in the cross-sectional analysis of this study. Social and biomedical characteristics explored included age, ethnicity, disease duration, marital and educational status, height, weight, comorbidity, and disease activity. PF was measured by the self-reported Health Assessment Questionnaire (HAQ) and by a battery of performance-based measures that included self-selected gait speed, the 5 chair rise test, and the single leg stance test. PA was measured by a portable activity monitor worn for 10 days, and was characterized in 2 ways: daily average number of steps and daily energy expenditure during moderate levels of PA.
Correlations between measures of PA and PF were small to moderate (zero-order correlations = 0.189–0.479). After controlling for social and biomedical characteristics, the correlations became smaller (semi-partial correlations = 0.095–0.277) and only HAQ score remained significantly associated with PA.
Associations between measures of PA and measures of PF were explained, in part, by social and biomedical characteristics in women with RA. The results indicate that measures of PF and PA may represent different constructs and support the need to measure PA in rehabilitation research in RA.
Patients with rheumatoid arthritis (RA) have fatigue, pain, limited joint mobility, impaired muscle strength, and decreased aerobic fitness, all of which limit their functioning (1–3). Rehabilitation treatments are prescribed to counteract the functional limitations of these patients. As a result, functional measures have been a useful benchmark by which to evaluate the effectiveness of rehabilitation in RA. Measures of physical function (PF) have been used as the primary end point in research and clinic rehabilitation. The use of PF measures in patients with RA has received official support. PF is a proposed criterion in the definition of improvement to be used in RA trials by the American College of Rheumatology and the international committee Outcome Measures in Rheumatology Clinical Trials (4). In clinical practice, the Centers for Medicare and Medicaid Services launched the Physician Quality Reporting Initiative to financially reward providers for their ability to assess several quality measures in RA, including patient functional status (5). Although PF measures are informative about the patient's impairments, limitations related to movement, and ability to perform everyday activities, they do not provide information about the amount of daily activity a patient performs, defined as physical activity (PA).
PA includes all body movements that result in energy expenditure. It includes recreational and occupational activities, sports, structured exercises, activity during leisure time, household activities, and any activity that requires the action of skeletal muscles. The benefits of PA are well known (6). Regular PA improves general health, prevents cardiovascular and other chronic diseases (such as type 2 diabetes mellitus [DM], hypertension, depression, and osteoporosis), and reduces mortality from all causes (7–9). Although the information about the benefits of PA in patients with RA is limited compared with that in the general population (10–13), the increased morbidity and mortality from cardiovascular disease in this group (14–17) justifies a focus on fitness in general and on PA specifically as an important goal of rehabilitation. Consequently, to make sure that these goals are met, these patients' levels of PA should be assessed. Regardless of the overwhelming benefits of PA and the apparent need for further investigation of the effects of PA in patients with RA, measures of PA are rarely used in this population.
The underutilization of measures of PA may be due to evidence from studies that reported significant associations between PF and PA (17–22) and generated the impression that measures of PF offer relevant information about PA. Clinicians may think that the increased pain, stiffness, and limited mobility captured by measures of PF may adversely affect the amount of PA. If this is true, measures of PF should predict the amount of PA and could perhaps serve as an indirect measure of PA. To date, the apparent distinction between PF (a patient's ability to perform activities) and PA (the amount of activities or movements) has not been investigated using concurrent measures of PF and PA in subjects with RA. Furthermore, being that PF and PA are both affected by social and biomedical characteristics such as age, level of education, obesity, and disease activity (15, 23, 24), their associations may be affected by these social and biomedical patient characteristics. The purpose of this study was to explore the associations between measures of PA and PF. On the basis of previous literature in other populations (15, 23, 24), we hypothesized that the strength of the associations between PA and PF would be moderate, and that after controlling for social and biomedical characteristics the associations would decrease.
PATIENTS AND METHODS
This was a cross-sectional study. Subjects were recruited from an original cohort of 104 women with RA (25). The study investigated cardiovascular disease and associated risk factors. This ancillary study was planned before the 5-year followup visit of the parent study, and all women who came back for the followup visit were invited to participate. Of the 56 returning for followup, 53 agreed to participate. The 3 women who declined stated that they had insufficient time to complete the study. The study took place from November 2007 to July 2008. Eligibility criteria included age older than 30 years, diagnosis of RA in accordance with the American College of Rheumatology (formerly the American Rheumatism Association) criteria (1) for at least 2 years, and no cardiovascular events (myocardial infarction, angina, or stroke) prior to recruitment. Study participants signed informed consent approved by the University of Pittsburgh Institutional Review Board.
Subjects participated in 1 testing session. Trained study personnel collected demographic and history data, administered self-reported questionnaires, and performed phlebotomy. A certified physical therapist (GJMA) administered the performance-based tests of PF and instructed the subjects in how to wear a portable PA monitor. Social variables included age, ethnicity, and marital and educational status. Biomedical factors included body mass index (BMI; calculated as mass/height2 in kg/m2), disease duration, disease activity, and comorbidities. Data on comorbidities were recorded using questions originally designed for the parent trial and were thus weighted more heavily on items related to cardiovascular disease. The number of comorbidities was calculated as the total number out of 8 conditions: heart attack, surgery on the arteries of the legs, stroke, transient ischemic attack, high blood pressure, DM, joint surgery for RA, and cancer. Disease activity was measured by the modified Disease Activity Score in 28 joints (DAS28) (26). The DAS28 includes 4 parameters in its calculation: number of joints tender to the touch (out of 28 joints), number of swollen joints (out of 28 joints), erythrocyte sedimentation rate using the Westergren technique, and the patient assessment of disease activity using a 100-mm visual analog scale. The DAS28 is generally accepted as a reliable, valid, and responsive measure of disease activity in patients with RA (26).
Measures of physical function.
PF was measured by performance-based and self-reported measures to better capture the broad dimension of the PF construct. Whereas self-reported measures assess one's perception of the ability to perform functional tasks, performance-based measures assess one's ability to complete a task. We used the Health Assessment Questionnaire (HAQ) as the self-reported measure of PF. The HAQ is a widely used and validated tool to quantify functional disability in RA (27). The HAQ disability index makes queries about 20 activities of daily living, including dressing and grooming, rising, eating, walking, hygiene, reach, grip, and community activities. It is expressed on a scale from 0 to 3 (where 0 = no functional disability and 3 = severe functional disability).
Performance-based PF was measured by 3 tests that were selected for ease of performance in a clinical setting: self-selected gait speed, the timed chair rise test, and the single leg stance test. Self-selected gait speed was measured by recording the time each subject needed to pass 2 tape markers on the floor placed 4 meters apart, located in the central part of a longer path of 7 meters. This configuration was used to avoid measurement during the acceleration or deceleration phases of the task. Participants were timed twice, and the faster speed was recorded. For the timed chair rise, participants were seated in a chair without armrests with their arms crossed over the chest. They were timed during 5 repetitions of rising to a full upright position and sitting back down in the chair without assistance. The single leg stance test consisted of recording the time that participants balanced on 1 leg while keeping their hands on their hips. The test lasted up to 30 seconds and was stopped if 1) the swing leg touched the floor, 2) the tested foot displaced on the floor, 3) the swing lower leg touched the tested limb, or 4) the arms swung away from the hips. If a subject was not able to balance on 1 leg, the score was 0 seconds. The single leg stance score of the 2 legs were averaged. The performance-based tests cover important domains of lower extremity physical function such as walking ability, muscle strength and power, and balance. They have been shown to be reliable and responsive to interventions, and they have the ability to discriminate from low to high functional ability in individuals at various ages and functional levels (28–32).
Measures of PA.
PA was objectively measured by the SenseWear Professional, version 6.1 (Body Media). The SenseWear armband (SWA) is a portable activity monitor that combines multisensor data, such as accelerometry, heat flux, skin temperature, and galvanic signal data. The SWA has shown to yield reliable and valid measures of PA (33–36). Participants were asked to wear the SWA on the back of their right arm (over the triceps muscle area) for 10 consecutive days. They took the SWA off only during showering or water exercises, period in which they recorded the activity performed in a log in order to input into the software accurate information about the off time. After the 10-day period, subjects returned the SWA and the log by mail and the data were downloaded. PA was characterized in 2 ways. First, we used physical activity energy expenditure (PAEE), which is the averaged daily energy expenditure during moderate PA and represents the amount of calories burned during moderate activities (activities performed at the 3 metabolic equivalents level or greater). Second, we used the daily average number of steps, representing the amount of lower extremity movement.
Multiple lineal regressions were used to test the hypotheses. Measures of PA (PAEE and number of steps) were the dependent variables and measures of PF (HAQ score, gait speed, 5 chair rise test, and single leg stance test) were the independent variables, whereas social and biomedical factors (age, ethnicity, marital and educational status, BMI, disease duration, disease activity, and comorbidities) were potential covariates. PA variables were checked for normality using the Shapiro-Wilk test. PAEE was not normally distributed and was square root transformed, which resulted in normal distribution. Pearson's or Spearman's correlations were used to describe the bivariate relationships between the potential covariates and PA and PF. Covariates were controlled in the multiple linear regressions only if the bivariate correlation between them and both PF and PA was significant.
Separate multiple linear regression models were then built for each of the 2 dependent variables; 4 models with PAEE as the dependent variable and 4 models with number of steps as the dependent variable. Each model had 2 steps. In step 1, we entered the PF measure. In step 2, we entered the covariates. Goodness-of-fit was evaluated by testing the residuals for normality and homoscedasticity. To test the hypothesis that the strength of the associations between PA and PF was moderate, we observed the zero-order correlation (the same as the Pearson's correlation) of step 1, whereas to test whether the associations between PA and PF would be partially accounted for by social and biomedical characteristics, we observed the semi-partial correlations of step 2. The 2-step approach was needed to derive the P values of t statistics for the contribution of PF during each step. The semi-partial correlation represents the explained variance in PA after PF was controlled for social and biomedical factors. We chose to assess the semi-partial correlation rather than the standardized regression coefficients so that its value could be squared and interpreted as the percentage of variance in PA uniquely accounted for by PF (37). Due to the small sample size, we based our findings on the strength of the associations and the percentage of explained variance rather than on statistical significance. The SPSS statistical software (SPSS) was used for all calculations.
Characteristics of the sample are reported in Table 1. From the 53 subjects who entered the trial, we were able to obtain complete data on 47 subjects. The 6 subjects with incomplete data did not wear the SWA for a full 7 days; their data were considered not representative of a full week of PA and thus they were excluded from analysis. The bivariate correlations between social and biomedical characteristics and PA and PF are presented in Table 2. Age, education, comorbidities, and disease duration were the variables most commonly associated with PA and PF and controlled in the regression models.
Table 1. Social, biomedical, physical function, and physical activity sample characteristics (n = 47)*
|Social and biomedical|
| Age, mean ± SD years||56.5 ± 7.0|
| BMI, mean ± SD kg/m2||27.9 ± 6.5|
| Education, mean ± SD years||15.7 ± 2.7|
| Ethnicity, no. (%) white||45 (96)|
| Marital status, no. (%) married||35 (74)|
| Comorbidities, no. (%)|
| None||16 (34)|
| 1||23 (49)|
| 2||8 (17)|
| Disease activity, mean ± SD DAS28 score||3.0 ± 0.81|
| Disease duration, mean ± SD years||14.3 ± 8.4|
| HAQ score, mean ± SD||0.74 ± 0.58|
| Gait speed, meters/second||1.2 (1.1–1.4)|
| 5 chair rise score, seconds||12.3 (11.0–16.0)|
| Single leg stance score, seconds||19.0 (6.4–26.6)|
| Daily average PAEE, kcal/day||199 (103–317)|
| Daily average number of steps, mean ± SD||7,151 ± 2,637|
Table 2. Bivariate correlations between social and biomedical characteristics and measures of physical activity and physical functioning (n = 47)*
The linear regression results for the PF variables predicting PA are shown in Table 3. The associations between measures of PA and PF in step 1 were small to moderate. The zero-order correlations ranged from 0.189 to 0.479 (absolute values), explaining no more than 23% (square of 0.479) of the variance in PA. The correlations indicated that subjects with better PF were more physically active. The only correlations not statistically significant were for the 5 chair rise test to predict PAEE, and for the single leg stance test to predict the number of steps. The results show that after controlling for the social and biomedical factors, the correlations became smaller (absolute values of semi-partial correlations ranged from 0.095 to 0.277). The only semi-partial correlations that remained significant were for HAQ score predicting the number of steps, suggesting that subjects with low HAQ scores (better PF) had higher daily numbers of steps. Visual observation of the residuals plots revealed that the data fit the linear model assumptions.
Table 3. Results of the multiple linear regression models for physical function variables to predict PAEE and number of steps (n = 47)*
| HAQ||−0.376||0.009||Age, education||−0.269||0.056|
| Gait speed||0.365||0.012||Age, education||0.248||0.080|
| 5 chair rise||−0.189||0.204||Education||−0.095||0.513|
| Single leg stance||0.308||0.035||Age||0.250||0.084|
|Number of steps|
| HAQ||−0.478||0.001||Age, education, comorbidity, disease duration||−0.277||0.041|
| Gait speed||0.386||0.007||Age, education||0.264||0.060|
| 5 chair rise||−0.479||0.001||Education, comorbidity, disease duration||−0.254||0.059|
| Single leg stance||0.222||0.134||Age||0.170||0.248|
To our knowledge, this is the first study in RA that purposely investigated the association between measures of PF and PA while accounting for social and biomedical factors. The findings indicate that measures of PF provide little information about PA, especially after accounting for social and biomedical factors. The findings suggest the need for inclusion of measures of PA with measures of PF in rehabilitation research of patients with RA. This may be particularly important, being that efforts to reduce cardiovascular risk in patients with RA focus on improving aerobic fitness.
Despite the vital importance of PA on health benefits (7, 8, 10–12), measures of PA are seldom used in arthritis. Conn et al conducted a meta-analysis on studies testing interventions to increase PA in patients with arthritis (38). They identified only 38 studies (from which 16 were in patients with RA) that tested PA interventions on PA behavior. The inclusion criteria were broad and accepted studies that measured PA in several ways: self-reported (e.g., diary, questionnaire), directly or indirectly measured (e.g., doubly-labeled water, indirect calorimetry, activity monitors), and even measurement of subsets of PA such as episodic exercise. Results indicated a moderate positive effect from PA interventions on PA behavior. The authors stated that their meta-analysis included only a small number of trials because most studies did not include measures of PA.
Most published studies in PA in the general population have been longitudinal and have demonstrated that increased PA is associated with less functional decline and reduction of adverse health outcomes and mortality (18, 24, 39–42). In longitudinal studies, the outcomes of PA and PF have not always been parallel. For example, a randomized trial in patients with RA investigated the effect of a 1-year coaching program for healthy PA. They reported improvements in perceived health status and muscle strength, but no change in self-reported PA (43). Cross-sectional studies have reported inconsistent associations between PF and PA in various patient populations. In patients with total hip replacement, 3 studies reported associations from small to moderate (r = 0.14–0.62) between self-reported PF and PA measured by a portable activity monitoring system (20) and by questionnaires (18, 44). In subjects with multiple sclerosis, the associations between self-reported PF and PA (measured by questionnaire and activity monitor) ranged from small to moderate (r = 0.21–0.54) (21). In older adults, the association between performance-based PF and PA measured by doubly-labeled water was moderate (r = 0.68) (19). In RA, the use of different statistics to calculate associations and cutoffs to determine PF and PA have limited direct comparison. A study in patients with RA from 21 countries reported that physical inactivity was more prevalent in patients with functional limitations (risk ratio 2.4). Functional limitation was defined as a HAQ score ≥1, whereas PA was self-reported by querying about the frequency of exercise. A response of weekly exercise once or more per week was considered regular exercise, and less than once per week was considered physical inactivity (17). In that study, the risk ratio was calculated from a univariate generalized linear model adjusting for age and sex. Perhaps the associations between PF and PA in that study would have been attenuated if the authors adjusted for additional factors such as education, disease activity, and disease duration.
The strength of the associations in these combined studies ranged from small to moderate (r = 0.14–0.68). The highest association indicates that measures of PF explain, at most, 46% (0.68 squared) of the variance of measures of PA, suggesting that perhaps they do not represent the same construct. In our study, the associations between PF and PA ranged from 0.19 to 0.48, explaining, at most, 23% of the variance in PA. Unique in our study was the investigation of the associations after controlling by social and biomedical factors that have previously been shown to relate to PF or PA (15, 23, 24, 45). When controlling for these factors, the associations were considerably tapered (semi-partial correlations ranged from 0.095 to 0.277), indicating that some associations were in part due to these factors. Although they were tapered, one should be careful not to judge the findings based on the statistical significance. With a larger sample size, the unique contribution of PF to PA in the final steps would likely be statistically significant. However, the larger semi-partial correlation after accounting for social and biomedical factors explains, at most, 8% of the variance in PA (0.277 squared), which seems to support the statement that measures of PF provide little information on PA.
There are several explanations for the variability in the ranges of associations between studies. The relatively low association between PF and PA in our study could be explained by either our having a somewhat young group of subjects (mean age 57 years) or only women in our sample. With regard to aging, one study assessed PF and PA in a group of nonagenarians compared with a group of subjects 20 years younger. They reported higher correlation coefficients for PF and PA for the nonagenarians group (r = 0.78) than for the younger group (r = 0.52) (19). With respect to sex, Manini et al used a group of older community dwellers to determine whether higher activity energy expenditure, assessed by using doubly-labeled water, was associated with a reduced decline in mobility limitation (24). Across sex-specific tertiles of activity energy expenditure, men in the lowest activity group experienced twice the rate of mobility limitation as men in the highest activity group. Conversely, women in the lowest and highest activity groups exhibited similarly high rates of mobility limitation. The findings did not change after adjustment for potential confounders. Therefore, it is likely that the results of our study would have been different if we had included men.
An alternative reason for the different magnitude of associations may be the difference in the methodology used to assess PA and PF. Some studies have used self-reported instruments to measure both PF and PA. When using the same type of instrument (i.e., self-reported measures), the associations may be inflated by the problem of common method variance. Moreover, for measures of PF, the use of performance-based versus self-reported methods to measure PF is also relevant. Both methods have pros and cons: self-reported methods are easy to use and not influenced by the tester, but are affected by pain and psychological factors such as expectations, cognitive status, and education level; performance-based methods identify early deficits in PF and are less affected by pain and psychological factors, but are criticized for measuring PF in an artificial situation and being influenced by the subject's motivation (46–48). Although there is continuous debate favoring one or the other method, there is consensus that performance-based and self-report measures only correlate modestly and probably measure different constructs of the PF domain (46, 49, 50). As a result, the method chosen to measure PF may influence its association with PA.
Regarding the measures of PA, the method most commonly employed in the studies has been use of self-reported questionnaires and diaries. Although questionnaires and diaries are inexpensive and easy to use, their limitations include inconsistent patient recall, overestimation of PA, underestimation of sedentary pursuits, and seasonal variation, all of which result in inadequate psychometrics (51). Questionnaires may also be complex, and subjects may not understand the phenomenon investigated. A recent qualitative study described variation in the understanding of the intensity of PA among patients with RA (52). The authors reported discrepancies between subjects and investigators regarding the understanding of PA intensity. They suggested that health professionals and patients with RA should reach a common understanding of ways to determine PA intensity to accurately prescribe and assess PA using questionnaires and diaries. Although the measurement method may influence the associations between PA and PF, we are not aware of any studies, including the ones reviewed in this discussion, that have tested whether different associations are a function of the tools selected to measure PF and PA. We have tried to compare the strength of associations across studies that used different methods and could not identify any pattern. This is an area that needs further investigation.
The fact that only HAQ score remained significantly associated with PA after controlling for social and biomedical factors in the current study was intriguing. A recent study disputed the notion that the HAQ mainly assesses functioning. Hakkinen et al performed a study in which they linked the HAQ items with components of the World Health Organization International Classification of Functioning, Disability and Health (ICF) instrument (53). According to the ICF classification, disability comprises 3 main components: body functions and structures, activity limitations, and participation restriction. The results of the study demonstrated that 16 of the 20 items included in the HAQ belong to the activity component, whereas the other items fall within the participation component. They concluded that the use of the HAQ instrument gives a rather narrow perspective on functioning in comparison with the ICF classification. Therefore, the associations between the HAQ and PA may have remained significant because they both measure the activity limitations component, whereas the other measures of PF used in this study are related to the body functions and structures component. This latter observation may also help to explain the significant association between self-reported PA and PF measured by the HAQ in the study performed in patients with RA from 21 countries discussed above (17).
The current study had limitations. Because our sample only included women, the findings should not be generalized to men with RA. The age of this sample is typical of many adults with RA, although future studies should investigate the associations of PA and PF in young and elderly subjects with RA. The cross-sectional design precludes ascertainment of temporal and causal relationships. Longitudinal studies should determine whether improvement in PF will increase PA. We may also not have accounted for all of the factors that may affect the associations. Larger studies with broader inclusion criteria should improve our understanding of the associations between measures of PF and PA in subjects with RA.
We report herein the new finding that the associations between PF and PA in RA are small and explained in part by a subject's social and biomedical characteristics. To date, measures of PA have been infrequently utilized in this population. Replication of our findings will further justify measuring PA in patients with RA in rehabilitation research and clinical practice.
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be submitted for publication. Dr. Piva had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study conception and design. Piva, Wasko.
Acquisition of data. Piva, Almeida.
Analysis and interpretation of data. Piva, Wasko.