Knee osteoarthritis (OA) is a leading cause of disability and joint pain. Although other risk factors of knee OA have been identified, how physical activity affects incident knee OA remains unclear.
Knee osteoarthritis (OA) is a leading cause of disability and joint pain. Although other risk factors of knee OA have been identified, how physical activity affects incident knee OA remains unclear.
Using data from the first (1999–2004) and second (2005–2010) followup periods of the Johnston County Osteoarthritis Project study, we tested the association between meeting physical activity guidelines and incident knee outcomes among 1,522 adults ages ≥45 years. The median followup time was 6.5 years (range 4.0–10.2 years). Physical activity at baseline (moderate-equivalent physical activity minutes/week) was calculated using the Minnesota Leisure Time Physical Activity questionnaire. Incident knee radiographic OA (ROA) was defined as the development of Kellgren/Lawrence grade ≥2 in a knee at followup. Incident knee symptomatic ROA (sROA) was defined as the development of ROA and symptoms in at least 1 knee at followup. Weibull regression modeling was used to estimate hazard ratios (HRs) and 95% confidence intervals (95% CIs) for interval-censored data.
In multivariable models, meeting the 2008 Department of Health and Human Services (HHS) physical activity guidelines (≥150 minutes/week) was not significantly associated with ROA (HR 1.20 [95% CI 0.92–1.56]) or sROA (HR 1.24 [95% CI 0.87–1.76]). Adults in the highest level (≥300 minutes/week) of physical activity had a higher risk of knee ROA and sROA compared with inactive (0 to <10 minutes/week) participants; however, these associations were not statistically significant (HR 1.62 [95% CI 0.97–2.68] and HR 1.42 [95% CI 0.76–2.65], respectively).
Meeting the HHS physical activity guidelines was not associated with incident knee ROA or sROA in a cohort of middle-aged and older adults.
Knee osteoarthritis (OA) affects an estimated 19–27% of adults ages 45 years and older () and remains a leading cause of disability and pain (). Some of the major consequences of knee OA are immobility, requirement for knee arthroplasty, high health care costs, and poor quality of life (). Annually, knee pain is reported in almost half of adults ages 50 years and older; an estimated 23% report severe and disabling knee pain ().
Obesity, older age, history of knee injury, and female sex are risk factors that have been associated with the development of knee OA (). The role that physical activity might play has been considered, but the association between physical activity and incident knee OA appears complex and is not well understood. Experimentally, animal studies have shown that weight-bearing exercise activities may prevent OA development ([5, 6]). Observationally, several longitudinal studies in humans have evaluated these associations with conflicting results ([7-13]). While some studies have indicated that physical activity is protective () or has no impact ([9, 12]) on knee OA development, other studies have suggested that torsional loading and high-impact activities are risk factors for knee OA ([7, 8, 10, 13]). Moderate volumes of low-impact activities (i.e., walking) may be protective for OA because joint loading and movement are necessary for optimal bone and joint health (). Furthermore, dynamic compression can stimulate chondrocyte biosynthesis and augment the cartilage matrix, which suggests that regular physical activity could be a protective factor for the development of knee OA (). Adding to the enigma, physical activity is recommended as a first-line nonpharmacologic intervention for OA patients because it has been proven to decrease pain, improve function and mood, and delay disability ().
Prior longitudinal studies ([7-13]) have been limited by study populations from convenience samples with restricted generalizability, crudely measured physical activity variables that failed to capture a broad range of activities, inadequate adjustment for potential confounders, and methods that do not account for uncertainty in the onset of disease (i.e., interval censoring) during the long observation intervals typically found in such studies. The majority of these studies have also used predominately white samples of participants. Comparison of these findings presents several challenges including different definitions of physical activity and OA, disparate populations, and varying followup times.
The purpose of our study was to examine the association between meeting the Department of Health and Human Services (HHS) physical activity guidelines and incident knee OA among middle-aged and older community-dwelling adults using methods that address some of the limitations identified in previous studies.
We analyzed data from the Johnston County Osteoarthritis (JoCo OA) Project, an ongoing population-based prospective cohort study of knee and hip OA in Johnston County, North Carolina. The survey methods are described in detail elsewhere (). In brief, the JoCo OA Project at T0 (1991–1997) enrolled 3,068 community-dwelling, noninstitutionalized white and African American residents ages ≥45 years who were physically and mentally capable of completing the study protocol and had resided in one of 6 townships in Johnston County for at least 1 year. At the 1999–2004 baseline for our analyses, there were 2,573 participants who completed both the clinic visit and home interview, including 1,590 from the first followup of this original cohort (T1; 1999–2003) and 983 from a newly enrolled cohort recruited to replace losses and enrich the cohort for African American participants (T1*; 2003–2004). Of these 2,573 participants, 1,528 completed both a clinic visit and home interview during the subsequent followup visit (T2; 2005–2010) (Figure 1). The median followup time (from T1/T1* to T2) of participants was 6.5 years (range 4.0–10.2 years). This study was approved by the Institutional Review Boards of the University of North Carolina School of Medicine and School of Public Health and the Centers for Disease Control and Prevention. All participants gave written informed consent at the time of recruitment.
We measured physical activity at our study baseline using the Minnesota Leisure Time Physical Activity (MLTPA) questionnaire, which collects self-reported information on physical activity during the previous year (). A previous comprehensive evaluation of the MLTPA concluded that the validity (direct validation measures 0.21 ≤ r ≤ 0.75) and reliability (1-month test–retest r = 0.92 and 1-year test–rest r = 0.69) of this instrument were reasonably good (). The MLTPA consists of 65 activities partitioned into 10 categories. We used the frequency and duration of 62 activities classified as moderate or vigorous activities (≥3 metabolic equivalents) to estimate the average minutes of moderate-equivalent physical activity per week (1 minute of vigorous aerobic activity = 2 minutes of moderate aerobic activity). We excluded 6 participants with extreme outlying values for physical activity (based on the statistical criterion of <3 SDs from the mean [i.e., 6,063.8 minutes/week or 14.4 hours/day]), leaving a final analytical sample of 1,522 participants.
We classified physical activity in 2 ways. First, we defined physical activity as meeting or not meeting the HHS physical activity guidelines of ≥150 minutes of moderate-equivalent physical activity per week (). Second, we defined physical activity using the 4 health benefit levels from the 2008 HHS Physical Activity Guidelines for Americans (inactive = no benefits [0 to <10 minutes/week], low = some benefits [10 to <150 minutes/week], medium = substantial benefits [150 to <300 minutes/week], and high = additional benefits [≥300 minutes/week]) ().
Posteroanterior paired reads of knee radiographs were performed at baseline and followup by a single bone and joint radiologist (JBR) using the conventional Kellgren/Lawrence (K/L) scale (). Previous interrater and intrarater reliability were high (weighted κ = 0.86 and 0.89, respectively) ().
Incident knee radiographic OA (ROA) was defined as K/L grade ≥2 or knee replacement at followup in a knee with K/L grade <2 at baseline. Participants with a baseline K/L grade ≥2, a knee replacement, missing radiographic data, or a non-OA diagnosis (i.e., radiographic evidence of inflammatory arthritis [e.g., rheumatoid arthritis]) in either knee were excluded.
Incident knee symptomatic ROA (sROA) was defined as a knee with ROA and symptoms at followup among participants without both ROA and symptoms in that specific knee at baseline. Knee symptoms were defined as a “yes” answer to the survey question “on most days, do you have pain, aching or stiffness, in your knee,” asked separately for the left and right knees. Because participants who had ROA without symptoms, had symptoms without ROA, or had neither ROA nor symptoms in a knee at baseline were included, we refer to this outcome as incident knee sROA (less restrictive definition). We performed a secondary analysis of incident knee sROA restricted to those with neither ROA nor symptoms in a knee at baseline (restrictive definition).
We also analyzed joint space narrowing (JSN; graded on a scale ranging from 0–3) as an outcome in our analysis. We defined incident knee JSN as at least a 1-grade increase on either the medial or lateral compartment of the tibiofemoral joint from baseline to followup. The JSN analysis was limited to participants with K/L grade <2 in both knees at baseline.
The potential confounders examined at study baseline were self-reported age, sex, race, education (no college versus attended college), and history of knee injury. Body mass index (BMI) was calculated from measured weight and height using the formula weight (kg)/height2 (m2) and was treated as a continuous variable. Because the MLTPA does not include occupational activity, we calculated a composite self-reported occupational activity variable (range 0–9) that estimated the total number of tasks from 9 occupational activities (light work standing, sitting, heavy work standing, kneeling, walking, hand motion, lifting 10 kg, lifting 20 kg, or lifting 50 kg) performed at least 50% of the time for a particular job.
The unit of analysis was the person, so an incident outcome needed to occur in only 1 knee to be considered an event for that person. To assess attrition bias, we compared those who did or did not complete followup using chi-square and 2-sample t-tests for baseline characteristics. Similarly, characteristics by physical activity level and incident ROA status were examined using the same statistical inference tests.
Interval-censored data are common in longitudinal time-to-event studies and reflect uncertainty as to the exact time that an event (failure) occurred in a known interval of time. The use of semiparametric Cox regression is problematic for interval-censored data, due to its dependence on ordering of event times. Weibull parametric regression can accommodate variable followup times and interval-censored data, so we used this procedure to estimate hazard ratios (HRs) with 95% confidence intervals (95% CIs) (). All multivariate analyses were adjusted for age, sex, race, BMI, education, occupational activity, and prior knee injury. We also evaluated the potential for interactions between physical activity and the variables BMI, sex, and race. However, the interaction terms were not significant (P < 0.05 for all models); therefore, stratified analyses were not performed. Because the association between physical activity and knee sROA differed by the definition used (restrictive/less restrictive), sensitivity analyses were performed to better understand what factors might be influencing these discordant findings (see Supplementary Appendix A, available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.22120/abstract).
The complex sample design of the JoCo OA Project was fully accommodated in our analyses. Specifically, population-calibrated sampling weights were applied in all analyses to account for oversampling of African Americans and differential rates of nonresponse, thereby enabling our estimates to be more representative of the target population. Furthermore, to correct for sampling without replacement, and to reduce sampling error, a finite population correction was computed in conjunction with the jackknife resampling method that was used to account for intracluster correlation (i.e., correlation between participants residing within the same street) in estimating SEs. The significance level was set at a P value less than 0.05 for 2-tailed tests. All analyses were performed using Stata 12.0 software. The Stata module INTCENS was used to perform the interval-censored survival analysis ().
Table 1 shows baseline characteristics of the 1,528 participants who completed followup and the 1,045 who were lost to followup. Compared with completers, the participants who were lost to followup were older, were more often men, were more likely to be African American, more likely to not attend college, had lower BMI, had higher occupational activity, were less likely to meet the physical activity guidelines, were more likely to be inactive, and had a higher prevalence of knee ROA.
|Baseline characteristics||Completed followup||Pa|
|Yes (n = 1,528 [59.4%])||No (n = 1,045 [40.6%])|
|Age, mean ± SE years||59.8 ± 0.3||63.6 ± 0.5||< 0.01|
|Men, %||40.0||45.1||< 0.01|
|African American, %||20.5||26.9||< 0.01|
|Attended college, %||33.9||22.8||< 0.01|
|BMI, mean ± SE kg/m2||30.8 ± 0.2||30.3 ± 0.2||0.045|
|History of knee injury, %||21.5||26.9||< 0.01|
|Occupational activity (range 0–9), mean ± SEb||2.21 ± 0.05||2.53 ± 0.07||< 0.01|
|Meets HHS physical activity guidelines, %c||55.9||50.7||< 0.01|
|HHS physical activity health benefits level, %d||< 0.01|
|Inactive (<10 minutes/week)||10.1||16.7|
|Low (10 to <150 minutes/week)||34.0||32.6|
|Medium (150 to <300 minutes/week)||21.8||19.4|
|High (≥300 minutes/week)||34.1||31.3|
|Prevalent knee ROA, %||27.2||35.3||< 0.01|
|Prevalent knee sROA, %||14.5||15.9||0.27|
Figure 2 shows the distribution of all leisure time physical activities (minutes/week) by the 10 MLTPA categories. Walking, conditioning exercises, and lawn and garden activities were the most reported types of physical activities, composing 62.8% of all activities.
Most participants (55.9%) met the physical activity guidelines, with whites, those attending college, and those with a lower BMI being significantly more likely to meet there guidelines (Table 2). The participants who met the physical activity guidelines had higher occupational activity; however, this association was not statistically significant (P = 0.06). Physical activity status did not differ significantly by age, sex, or history of knee injury.
|Baseline characteristics||Met physical activity guidelines (n = 1,522)a||Incident knee ROA (n = 993)|
|No (n = 684)||Yes (n = 838)||Pb||No (n = 742)||Yes (n = 251)||Pb|
|Age, mean ± SE years||59.9 ± 0.5||59.6 ± 0.4||0.62||58.2 ± 0.4||61.8 ± 0.6||< 0.01|
|Race, %||< 0.01||0.52|
|Attended college, %||< 0.01||0.17|
|BMI, mean ± SE kg/m2||31.8 ± 0.3||30.0 ± 0.2||< 0.01||29.3 ± 0.3||30.8 ± 0.3||< 0.01|
|Occupational activity, mean ± SEc||2.30 ± 0.07||2.13 ± 0.07||0.06||2.21 ± 0.07||2.20 ± 0.08||0.95|
|History of knee injury, %||0.70||0.79|
Of the 993 participants without ROA at baseline, 251 had incident knee ROA at followup. Older adults, men, and those with a higher BMI were significantly more likely to develop ROA. Incident ROA did not differ significantly by race, education, occupational activity, or history of knee OA (Table 2).
In terms of multivariate analysis, Table 3 shows the adjusted association between physical activity variables and incident knee ROA, incident knee sROA, and incident knee JSN. Meeting the recommended physical activity levels (≥150 minutes/week) was not associated with incident ROA (HR 1.20 [95% CI 0.92–1.56]) or incident sROA (less restrictive definition; HR 1.24 [95% CI 0.87–1.76]), but there was a borderline significant association (HR 2.20 [95% CI 0.99–4.90]; data not shown) with incident sROA (restrictive definition). The sensitivity analysis was limited by relatively low statistical power, but suggested that the association between physical activity and sROA (restrictive definition compared with the less restrictive definition) was potentially modified by sex, knee injury, and having a combination of both knee injury and being obese/overweight (see Supplementary Appendix A, available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.22120/abstract).
|Incident knee ROA (n = 981)a||Incident knee sROA (less restrictive definition; n = 1,114)b||JSN (n = 981)a|
|Meets HHS physical activity guidelinesc|
|Yes||1.20 (0.92–1.56)||1.24 (0.87–1.76)||1.42 (1.10–1.82)|
|HHS physical activity health benefits leveld|
|Low||1.25 (0.75–2.10)||1.07 (0.56–2.05)||1.22 (0.73–2.04)|
|Medium||1.14 (0.67–1.97)||1.16 (0.60–2.25)||1.22 (0.72–2.07)|
|High||1.62 (0.97–2.68)||1.42 (0.76–2.65)||1.97 (1.20–3.26)|
Participants in the highest level (≥300 minutes/week) of physical activity had a higher risk of incident ROA and incident sROA when compared with those who were inactive; however, these associations were not significant (HR 1.62 [95% CI 0.97–2.68] and HR 1.42 [95% CI 0.76–2.65], respectively) (Table 3). Both meeting the physical activity guidelines and having the highest level of physical activity were significantly associated with incident JSN (HR 1.42 [95% CI 1.10–1.82] and HR 1.97 [95% CI 1.20–3.26], respectively) (Table 3).
In our prospective cohort study, meeting the physical activity guidelines was not associated with incident knee ROA or sROA over 6.5 years among community-dwelling middle-aged and older adults. When analyzed by level of physical activity, high levels (≥300 minutes/week) were suggestive of an increased risk of knee ROA and sROA, although these differences were not statistically significant. Additionally, meeting physical activity guidelines and high levels of physical activity were associated with an increased risk of knee JSN.
For knee ROA, comparing studies with similar aims and objectives is difficult because of disparate and crude measurements of physical activity. No other studies used meeting the HHS guidelines as a measure. Instead, these studies compared levels of physical activity. Two fairly recent studies ([9, 12]) found that physical activity neither protected nor increased the risk of knee ROA. The Framingham Study () used crude measures of physical activity (i.e., walking and sweating), while the Nord-Tr⊘ndelag Health Study () estimated self-report of physical activity per week but did not ask about individual activities. Other studies ([7, 8, 10]) have found that high levels of physical activity are associated with an increased risk of knee ROA. Cooper et al found that participants engaging in weekly sports participation for ≥10 years had a <3-fold risk of knee ROA development compared to others (), although this finding was only significant when knee OA was defined using a K/L grade threshold of 1, rather than the conventional K/L grade of 2. Felson et al found that physical activity significantly increased the risk of knee ROA by 3.3 times for those in the highest quartile compared with the lowest quartile (). The study by Felson et al used the Framingham Physical Activity Index (), which assigns a level of activity (i.e., slight, moderate, and heavy) to different items and derives a weighted sum of activity over 24 hours. Despite including many activities, the Framingham Physical Activity Index does not estimate physical activity over a year or account for seasonality of certain activities, which the MLTPA does. McAlindon et al used the same Framingham Physical Activity Index and similarly found that the highest risk of knee ROA was among participants in the highest physical activity categories (). What appears to be consistent from the studies mentioned above is that comprehensive measures of physical activity (which account for many individual activities) are more likely to show an increased risk of ROA at high levels of activity compared with crude measures of physical activity.
Knee sROA has major public health and clinical implications, but few studies have attempted to address this outcome ([9, 10]). McAlindon et al found that higher physical activity was a risk factor for knee sROA (using the same restrictive definition) (). Conversely, a more recent Framingham Study found no link between recreational physical activity and knee sROA (using the same less restrictive definition) (). We classified sROA using 2 different definitions as indicated above. In our primary analysis, meeting the HHS physical activity guidelines did not significantly increase the risk of knee sROA (less restrictive definition), whereas in our secondary analysis, meeting the physical activity guidelines was borderline associated with an increased risk of sROA (restrictive definition). The sudden onset of both symptoms and ROA suggested greater exposure to a major risk factor for the outcome. Perhaps the concordant findings regarding higher physical activity and increased risk of sROA (restrictive definition) may be explained by an interaction between physical activity and risk factors such as history of an injury or overweight/obesity (shown in our sensitivity analysis).
JSN, often a surrogate for cartilage loss (), although there may be other causes, was significantly associated with both meeting the physical activity guidelines (≥150 minutes/week) and high levels (≥300 minutes/week) of physical activity in our analysis. Using cruder measures of physical activity, the Framingham Study found a null association of physical activity with JSN (), while Hart et al found that walking significantly decreased the odds of JSN (). The disparate measures of physical activity may explain these discrepant findings. Additional studies are needed to sort out these differences.
This study has several strengths. First, we used a population-based study that is generalizable to a community population rather than a convenience sample. Second, we used a comprehensive physical activity measure that captures a broad range of leisure time activities and accounts for the seasonality of some activities (i.e., shoveling snow or mowing a lawn). Third, we controlled for a variety of well-known confounders (i.e., age, sex, race, BMI, education, and knee injury) and for another measure of physical activity (i.e., occupational activity) to allow for an assessment of the independent effect of leisure time physical activity. Finally, we addressed interval censoring by using the Weibull distribution rather than using cumulative incidence, which does not account for time in the analysis or taking the midpoint time at followup, and this can produce widely varying HRs when time intervals vary greatly (). The Weibull distribution also has 2 advantages: it allows for increasing, decreasing, and constant hazard rates as opposed to the exponential distribution, and Weibull parameter estimates are based on the maximum likelihood estimates and not partial likelihood estimates produced by the semiparametric Cox proportional hazards model, which contain an unspecified baseline hazard function ().
There were also some limitations to this study. First, we measured leisure time physical activity at baseline only, and as a result, physical activity levels that are most important to knee OA etiology may have occurred at different time points (i.e., prior to baseline or followup). Physical activity is dynamic and may change individually over time, although a study from The Netherlands examining older adults showed that walking time (which composed a large portion of physical activity in our study) was stable over a period of 10 years (). Second, physical activity was self-reported and therefore subject to misclassification of the physical activity levels. Over-reporting of physical activity may have occurred because of social desirability or recall bias; nonetheless, a recent Centers for Disease Control and Prevention report () showed that, for the US population, ∼52% met the HHS aerobic guidelines, which compares well with our findings. Additionally, to our knowledge, there is no prior evidence to suggest that physical activity misclassification would be differential by OA status or even demographic variables (), and if anything, our results would likely be biased to the null. Third, attrition bias may have occurred because those who did not complete the followup differed in some characteristics from those who completed the study. Differential loss to followup would have occurred if the association between physical activity and ROA among participants lost to followup differed from the association among participants not lost to followup, but we did not have ROA status at followup for the missing adults, so the direction of bias (if any) could not be determined. Fourth, the followup time was for only a median of 6.5 years. Continuing followup on this cohort will extend the observation time. Fifth, it would have been worthwhile to explore the effect of more vigorous activities (i.e., running and swimming) individually on knee OA, but these activities were performed at low frequency and duration in our cohort. Sixth, because our study used only 2 time points, we were unable to estimate the mediating effect of weight change on the onset of knee OA outcomes. For instance, higher physical activity may lead to lower weight gain, which in turn reduces the risk of OA. Finally, there were only 42 participants that developed sROA when applying the restrictive definition. Consequently, the secondary analysis was limited to the association between meeting physical activity guidelines and sROA because of limited statistical power.
In summary, meeting the HHS leisure time physical activity guidelines was not associated with a significantly increased risk of incident knee ROA or sROA in a cohort of middle-aged and older community-dwelling adults over a median followup time of 6.5 years. These findings and studies in the literature suggest that engaging in moderate levels of physical activity does not increase the risk of knee OA, and that activities such as walking, conditioning exercises, and household activities (that amount to moderate levels of physical activity) can continue to be encouraged in the population. Conversely, we and others found an elevated risk of knee OA outcomes among persons in the highest level of physical activity, so high levels of physical activity may have negative implications.
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 published. Dr. Barbour 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. Barbour, Hootman, Helmick, Schwartz, Kalsbeek, Jordan.
Acquisition of data. Renner, Jordan.
Analysis and interpretation of data. Barbour, Hootman, Helmick, Murphy, Theis, Schwartz, Kalsbeek, Renner, Jordan.
We would like to acknowledge researchers Yiling Cheng, Barbara Do, and Yvonne Golightly for their feedback, which has helped improved the quality of this manuscript.