This prevalence study on a large sample of Canadian adults presents a newly proposed measure of lifetime mechanical knee joint force based on hours in PA, bodyweight, and typical knee joint force for specific activities and relates it to self-reported knee OA. We provide evidence that while lifelong PA may generally be safe for the knee joint, that very high force from lifelong total force, and from high levels of total occupational force (men and women) and household activity (women), is a potential risk for knee OA. The results held after adjustment for known risk factors. These novel findings require confirmation in other populations and in longitudinal studies. Our results are consistent with previous studies that show overweight/obesity, age, female sex, and previous injury are significant risk factors for knee OA [47
The results of this study must be compared cautiously with previous studies due to its cross-sectional design, how subjects were assembled, and how exposures and outcomes were measured. For example, no other studies evaluating PA and knee OA have used the Internet for data collection or have completely classified PA in terms of a joint loading variable over the long term, perhaps partly explaining inconsistent results from these past studies [44
]. Of note, the cross-sectional design may have resulted in reverse causality potentially attenuating risk estimates. Conversely, subjects with OA may have overreported prior PA exposure because they perceive that PA caused their OA (recall bias), potentially increasing the risk estimates [56
]. Since this type of bias threatens most prevalence studies, emphasis should be placed on recent high-quality cohort studies evaluating the association between PA and OA.
A recent prospective cohort study with a 22-year-followup using physician-diagnosed OA, reported an adjusted OR for the heaviest category of physical demands at work compared with the lightest category of 18.3 for knee OA [62
]. Verweij et al. [63
], during 12 years of followup, recently reported that 463 of 1678 respondents (28%) developed clinical knee OA, and that a high mechanical strain score was associated with an increased risk of knee OA (HR 1.43, 95% CI 1.15–1.77) after adjustment for a number of covariates. Wang et al. [64
], in a prospective cohort study of approximately 40,000 Australians with an average 5 years of followup, reported a composite sport and occupational exposure (past 6 months, measured at baseline) and found a risk for total knee joint arthroplasty for the vigorous level of activity (HR 1.42, 95% CI 1.08–1.86). Several studies from the Framingham cohort suggest that job activities may cause as much as 15% to 30% of knee OA in men [65
]. Felson et al. [65
] reported that elderly persons (average age 70) in the highest quartile of PA at a baseline examination had over three times the risk of developing radiographic knee OA nine years later, when compared with those in the lowest quartile. McAlindon et al. [67
] using longitudinal Framingham data reported that the number of hours per day of heavy physical activity was associated with the risk of incident radiographic knee OA (OR = 7.0 for 4+ hours heavy physical activity/day). No effects were observed from moderate and light PA. In contrast, a study by Hannan et al. [68
] in the same cohort found no increase in the risk of knee OA with increasing physical activity. In the highest quartile of PA compared to the least active, the OR was 1.3 for men and 1.1 for women (both nonsignificant). Hart et al.[69
], using data from the Chingford study, followed 715 women (mean age: 54 years) for 4 years with no radiographic knee OA at baseline and included the PA categories of walking, occupation, and sport/recreation. They found no relationship between incident knee OA and PA, while walking protected against joint space narrowing (OR = 0.4, 95% CI 0.2–0.9).
It is evident from these often-cited reports that, despite the longitudinal cohort designs, large samples and lengthy followups, and estimates for the risk of PA on knee OA vary extensively. While differences in eligibility criteria, covariates included in multivariable models and small samples may account for some of the disparity, the most likely reason is the wide variation in PA exposure measurement. Of note, most studies have not measured the joint-force aspects of PA nor attempted to completely classify PA (including historic PA) from all three major activity domains. Apart from the Verweij et al's study [63
], none of the above studies considered PA from all three major activity domains or attempted to estimate the effect of activities in terms of joint force. The main finding of Verweij et al. was an OR of 1.43 (95% CI 1.15–1.77) for a high knee mechanical strain score, close to our reported OR's from the highest quintiles of total knee force, occupational force, and household force in women. Results are not directly comparable since apart from differences in design, the mechanical strain score was a ranking (1 to 4) of certain physical activities over the past 2 weeks (taken at baseline) and did not look at sex-specific differences in occupational and household activity. Consistent with the recent longitudinal cohort study of Toivanen et al. [62
] and a number of longitudinal and case control studies [55
], we did not find a relationship between sport/recreational activity and knee OA. Studies that have shown a relationship between sport and knee OA have generally been in populations of athletes in specific sports with high knee forces [54
] and not from population-based studies, or the association has been explained by joint injury [76
]. In population-based studies of lifetime activity, the highest sport/recreation rates typically occurs at a relatively young age, as it did in the current cohort, prior to the age of 25 [5
]. Thus, high forces from sport later in life, when the joint may be more vulnerable, were not well-represented in this sample and may contribute to the lack of association here.
The prevalence of knee OA in our study was 22.4%, 17.8% in men and 25.1% in women. These gender differences in prevalence are consistent with previous large population-based North American studies for this age group [82
]. Even though there is probably some misclassification, our definition, which required a medical diagnosis and the presence of pain on most days, is important since pain is usually the most important aspect of disease to patients and may precede X-ray change, potentially capturing earlier disease. We report the results of a validity study in a subsample of the current study comparing self-reported OA to clinical OA [41
It was important to measure and simultaneously adjust for PA-related force from all three major activity domains. Most previous PA-OA studies have investigated one or two domains (usually sport and/or occupation). Given the high levels of household and occupational PA reported in our previous paper [5
], omitting one or both of these domains leaves these studies vulnerable to confounding from the unmeasured domain(s).
In studying all three domains separately by sex, we also observed relationships of PA with very high occupational force in men and household force in women. Questionnaires used in many previous studies did not assess the frequency, duration, and intensity of PA actually performed by women [85
]. The majority of women's exposure to PA, particularly in older cohorts such as the current one, is due to accumulation of regular household activities [85
]. While household activity may generally not be considered vigorous from an energy expenditure perspective and is often ignored in epidemiologic study of OA, there are many repetitive motions (e.g., stair climbing, squatting, and kneeling) and activities (e.g., gardening, lifting, and carrying) that are associated with high knee joint forces [20
] but have low energy expenditure.
This is only the second study to measure lifelong household load at the knee joint and relate it to knee OA, and the first to quantify household knee joint force from historic activity for the assessment of dose response. In the previous study by Sandmark et al. [89
], exposure to physically demanding tasks at home was significantly associated with knee OA among women (but not men) and was the strongest risk factor for women among the physical load variables that were investigated in that study. Given that women have been shown to have higher PA than men when including household together with occupational and sporting activities [5
], and that reasons for the higher prevalence of knee OA in women are not clearly elucidated [51
], our findings provide preliminary evidence that the role of historic household PA requires further investigation. The role of occupational activity has received much more study, and is better understood in men than in women, in part because previous studies were based historically on male-dominated workforce cohorts [51
]. Even though women often spend forty or more hours a week at a full-time job and from twenty to forty-five hours per week working in the home, questionnaires used in many previous studies do not assess the frequency, duration, and intensity of PA actually performed by women [85
]. It has been shown that, when the definition of regular physical activity measured in surveys is expanded to include household activity, PA levels rise and associations with health outcomes are more evident—including protective relationships with cardiovascular disease and myocardial infarction [90
], cancer [91
], and an inverse relationship with all-cause mortality [85
]. Our finding of an increased risk of knee OA for the top quintile of occupational joint force is generally consistent with previous studies [62
] and several systematic reviews [92
The CPFI, a quantitative joint force measure, together with a large sample allowed for evaluation of a dose-response relationship between lifelong force and knee OA. In the models where a significant relationship with knee OA was found (total force, occupation, and household), there was an increasing, significant trend in the ORs from lower to higher levels of CPFI, though only the ORs for the highest (5th) quintile reached statistical significance. While this requires confirmation and further delineation in future studies, the presence of dose response strengthens evidence for a causal relationship [95
This study had several strengths, including a large sample drawn from the population and a sufficient number of cases to adjust for a number of covariates, the separate analyses of men and women (equivalent to including interactions with gender for all variables), and assessment of dose-response. Another strength was the use of detailed information on the duration, frequency, and joint loading aspects of historic activities, from all three main physical activity domains, allowing for relatively complete classification of the total volume of PA. Historic PA is a potentially important exposure in OA etiology given the lengthy induction and asymptotic latency period. Measuring current or recent levels of PA does not capture long-term joint forces, may miss etiologically important periods of exposure and is a poor proxy for cumulative lifetime exposure [96
]. Lastly, many studies have used advanced disease markers, such as total joint arthroplasty or moderate-to-marked radiographic change as the outcome in assessing the role of PA. It is not clear whether the relationship with PA for early, symptomatic cases is the same as that observed for advanced or radiographically defined OA. Our cases definition allowed us to capture information on earlier stages of disease. This may be important in understanding modifiable risk factors that could play a role in a prevention strategy for OA, something not currently available.
There are a number of limitations that are important in interpreting the results of this study. Self-report of knee OA may lead to misclassification. In our examination of the measurement properties of our case definition [41
], we noted that specificity was very high. This is critical for studies of risk factors, since low specificity (inclusion of many false positives among cases) causes a greater attenuation of effect than low sensitivity. PPV was also high, another important measure indicating that the vast majority of the cases identified in the survey were true cases. Nevertheless, we did not use radiography as part of the classification criteria for knee OA. Radiographic OA in the presence of symptoms is thought to represent the best definition of OA. However, X-ray change is associated largely with moderate-to-advanced disease [97
], and there is only moderate agreement between pain and symptoms and X-ray changes [98
]. Wu et al., in a study using a validated outcome instrument for knee OA based on arthroscopic visualization, suggest that the ACR clinical classification criteria can be used to identify patients with early articular cartilage loss, before any radiographic changes are evident [97
]. However, it is probable that the false positives include not only subjects with early OA not captured by the ACR criteria [97
], but also other causes of knee symptoms.
The results from this study may not be broadly generalizable. The subjects were fairly well-educated, predominantly Caucasian Canadians with access to public health care and Internet users. This method of data collection may not be as effective in low-income populations, and those with decreased access to medical care—important since we asked about medically-diagnosed osteoarthritis as part of the case definition. Further, since recruitment and enrolment of subjects was via the Internet, subjects were largely self-selected. Self-selection implies that the nature of the bias cannot be known with certainty [99
]. Studies of subjects who participate in online research reveal that they are more likely to be older, females and have higher socioeconomic status [100
]. Also, response rates for online recruitment and enrolment vary from traditional rates. Many more individuals potentially view invitations to participate in research, with most declining to participate, making validity of results more challenging to interpret. In online surveys, there is no single response rate-multiple metrics for calculating a response rate have been defined such as the participation rate and completion rate [101
]. However, the goal of this study was not to describe characteristics of the population at large, but to assemble subjects to test hypotheses about PA-related knee joint force and knee OA in a large sample of individuals who met criteria for a disease and those who did not, sampled in the same way (internal validity).
The limitations of our measure of self-report of PA measures and construction of the CPFI variable have been discussed elsewhere [4
]. In short, self-reported PA measures require cautious interpretation because of large within-person variability and problems with recall [102
], that may lead to nondifferential misclassification and attenuation of the effect size in analytic studies using the exposure. In particular, this attenuation may have contributed to a lack of a significant finding from the sport domain, since the highest sport levels occurred in the distant past (prior to age 25) for most subjects, are not part of the generic memory pattern (shown to have better recall) [105
] and thus may been imprecisely recalled. Despite these limitations, it has been repeatedly shown that PA questionnaires are both practical and valid when used appropriately for large-scale epidemiologic studies [86
The CPFI, a time-force-bodyweight product, was a stronger predictor of knee OA than any of its component parts alone and is a new measure of PA-related force measured in joint loading units. However, the CPFI does not separately and specifically capture elements of activity-related force that may be most injurious such as shear, rapid deceleration, or high-impulse loads. Activities where those elements of force occur (e.g., cutting and pivoting sports, jumping sports, and carrying heavy loads) were captured, indirectly measuring these harmful types of load, but the strength of a potential signal from these forces may have been blunted. Another potential limitation related to recall is the possibility of recall bias, where the ability to recall past exposure is dependent on outcome status. Of note, subjects with OA at baseline may have overreported prior PA exposure, attributing their OA to their past activity. This could lead to increased risk estimates, and while justified for the reasons outlined previously, the results remain vulnerable to this type of bias. However, risk estimates for sport and occupational exposure as well as other covariates were generally in the expected direction and consistent with the literature including prospective data [47
], lending validity to the findings. Regardless, the possibility of this bias must be acknowledged, and study results interpreted in light of this.
Although this study provides evidence of an association between high levels of lifelong joint force, overweight/obesity, previous injury, BMI, and knee OA, the cross-sectional design makes the determination of a cause and effect more challenging. However, the time window used for the main PA exposure (prior to age 50) captures the ages [30
] with the highest level of lifetime force and is separated in time from knee OA diagnosis for the vast majority of cases. Supporting this, most of the risk estimates for covariates reported in this study were in the expected direction and effect sizes consistent with the literature [47
], and the period prevalence design included incident cases. Lastly as this study is the first attempt to examine the effect of a new exposure measure (quantitative lifelong joint load from all three primary activity domains) on knee OA, a cross-sectional approach is reasonable.
The finding that most PA-related force is not related to knee OA, but that the highest levels of joint force are, is biologically plausible and fits within the conceptual framework of causation. Under normal physiological conditions, the transmission and distribution of joint loads can occur for decades with little or no wear [108
]. However, when normal joint physiologic mechanisms are overwhelmed via excessive local mechanical force, biologic events are triggered which destabilize the normal coupling of degradation and synthesis of articular cartilage and subchondral bone [109
]. Animal studies clearly illustrate that high joint force from PA affects cartilage metabolism and plays a role in the development of OA [110
In summary, a newly proposed measure of lifetime mechanical knee force was used to estimate the risk of self-reported knee OA. While it must be interpreted cautiously because of the cross-sectional design and the possibility of recall bias, this study suggests that lifelong physical activity is generally safe. High levels of lifetime knee force from occupational activity in men and women, and household activity in women were associated with knee OA. Obesity and previous injury were also a significant risk, consistent with previous studies. Prevention efforts may best be directed at occupations requiring high physical demands, at weight-control programs and injury prevention. Future research should further investigate the potential role of household activity, improve the estimation and validity of knee force measurement in new populations, and apply these measures in longitudinal studies.