In this large population of postmenopausal women, self-reported arthritis is associated with significant fracture risk increase in women reporting OA and RA. After controlling for several covariates, the RA group had a highly significant increased risk of all fractures studied (HRtotal = 1.49, HRspine = 1.93, HRhip = 3.03) in comparison to the non-arthritis group. Modest significant increases in total (HR = 1.09) and spine (HR = 1.17) fracture risk were seen in the OA group in comparison to the non-arthritis group, but no significant increase in hip fracture risk was seen. The associations found between arthritis and fracture were not modified by race, ethnicity, or glucocorticoids use in this study.
The RA findings from this study are consistent with the literature showing an increased risk of fractures in RA patients (9
). Recently, incidence of any, spine and hip fracture in the Consortium of Rheumatology Researchers of North American (CORRONA) Registry were reported to be 3.71, 0.78, and 0.66 per 100 person-years, respectively (27
). The age-adjusted fracture rates for the RA group were 3.64, 0.49, and 0.50 per 100 person-years for total, spine, and hip fracture in the WHI, and though the CORRONA registry includes premenopausal women and men, the incidence rates of the nationwide CORRONA registry are comparable to the rates found in the WHI.
General lifestyle and demographic osteoporosis risk factors, such as age, smoking, and physical activity, play a significant role in fracture risk (1
), but the primary risk factor for fracture is low BMD. It has been well documented that RA patients have lower BMD at many skeletal sites compared to various control populations (6
), and though BMD was not examined in this study, it is highly probable that the associations seen are in part, driven by BMD. A sensitivity analysis in the participants from three WHI clinical centers with available BMD measurements was proposed, however, could not be adequately completed due to the low frequency of fractures (total fractures n=22; spine fractures n=2; hip fractures n=3) in the smaller RA group (n=78).
The risk of sustaining any clinical fracture and a spinal fracture was modestly but significantly increased in the OA group compared to the non-arthritic controls. It is likely that the effects of OA on fracture rate are being underestimated in this study due to the misclassification inherent in self-reporting OA. As previously mentioned, the association between OA and fracture has been mixed in the literature. The most recent study to suggest OA increases the risk of fractures reported by Arden and colleagues found that after adjusting for falls and the use of walking aids, clinician diagnosed knee OA patients had a significant risk for non-vertebral fractures [1.48 (1.00, 2.19)], and no significant association was seen between clinician diagnosed knee OA patients and risk for hip fracture [1.84 (0.78, 4.34)] (18
). Though the results of our study are in agreement with the Arden study, the use of clinically diagnosed, site-specific OA patients yielded higher fracture estimates than those found in our study using self-reported OA cases.
In contrast, the most recent study showing a protective effect of OA on fracture risk was a case-control population-based study conducted in Denmark. After adjustment for several variables, Vestergaard and colleagues found an significant risk reduction for any fracture, hip, and spine fractures in participants with OA duration greater than two years (29
). Population demographics could be the primary explanation for the difference associations seen between the Vestergaard study and the current, as the Danish population used was almost 20 years younger than the WHI population.
Though a consensus has not been reached, several biological mechanisms have been proposed relating OA to fracture. Like RA, the increase in fracture in OA patients could be driven through a BMD pathway. Studies have shown that BMD in OA populations is typically higher than non-arthritic populations (30
); therefore, this argument does not provide a good explanation for increases in fracture. Though OA patients have a higher BMD or bone quantity, quality or strength of the bones may be compromised compared to other arthritic and non-arthritic populations. Javaid and colleagues assessed hip structural geometry, as a marker of bone strength, in a group of OA patients and found alterations in geometry precede OA diagnosis (33
), suggesting a biological process involved in OA potentially alters bone strength.
Falling is another proposed OA fracture mechanism. OA, especially at sites like the knee and hip, is associated with increased pain, decreased postural stability, and decreased muscle strength, all which have been shown to be significant contributors to fall risk (34
). Falling is a well-documented risk factor for fractures (1
), and early studies have shown that the self-report of OA is associated with increased risk of falls (37
). More recently, Foley and colleagues did not see increased risk for falls in knee and hip radiographic OA cases, but did see that report of pain is highly associated with falls and OA patients reported more pain (39
One last possibility is that our results represent the consequences related to behavioral and physiologic changes that occur in individuals that perceive articular discomfort they classify as arthritis. Self-reported health status has been shown to be an independent risk factor for fractures in many studies (40
). It is possible that self-reported arthritis in the WHI is a measure of autoperception that encompasses a variety of health domains, such as pain, balance confidence, self-efficacy, and functional status.
Strengths and limitations
This study has several limitations related to the arthritis exposure. Regarding OA, the limitations associated with self-report and the use of a proxy measure of OA within the WHI previously described by Wright and colleagues apply to this analysis (23
). Walitt and colleagues also found that self-reported OA in the WHI was very sensitive (95.0%), but not particularly specific (23.4%), and only had fair agreement between self-reported OA and chart review (kappa = 0.23) (unpublished data). The potential for the moderate amount of misclassification in the OA group would bias the results of this study to the null. People experiencing joint pain due to a previous injury, have other soft-tissue condition such as tendonitis, or other non-inflammatory arthritic conditions, may report having OA though not clinically diagnosed. This could also lead to a moderate amount of misclassification, again, biasing the estimates towards the null. Not having site specific or radiographically confirmed OA cases is another limitation of this study. Fracture risk is probably different for persons with OA of the hip compared to persons with knee, hand, or spine OA. The OA affected area may have a higher BMD, whereas regions without OA have normal or low BMD, potentially altering overall fracture risk.
Regarding the RA classification, the use of medication in the RA definition probably captured true RA cases, but these may represent the more severe cases, potentially overestimating the effect of RA on fracture risk. This study did not take into account incident cases of arthritis and the effect it has on fracture risk, and it also did not also account for the additive or multiplicative effect of having both conditions on fracture risk.
The use of self-reported fracture outcomes can also be seen as a limitation. Sensitivity analyses were performed using adjudicated fractures only. Slight changes in the point estimates were observed with the smaller sample size, however, the overall conclusions did not change. Chen and colleagues found high agreement between self-report and adjudicated fractures in WHI sub-study (45
), assuring high quality of the fracture data used in this study.
This study adjusted for several covariates, but was unable to adjust for GC use, as it was used in the definition of the RA group. To test the possible interaction of GCs in the relationship between arthritis and fracture, a categorical variable was created capturing users and non-users in each arthritis group (data not shown). Though no interaction was present, the point estimate of the fracture risk was higher in GC users compared to non-users, and by not adjusting for GCs, the true fracture risk for women not taking GCs was overestimated and the risk was underestimated for women using GCs.
Though limited by the above mentioned factors, there are many strengths of the study. The most notable is the size of the WHI, and the size of each of the exposure groups. Having over 63,000 women in the OA group gave more than adequate power to estimate the effects OA have on fracture outcomes. Though not clinically ascertained, the prevalence of OA in the WHI population was approximately 43%, which is comparable to the 42% prevalence of radiographic OA in the hands, knees, and hips found in the women 60 years and older participating in National Health and Nutrition Examination Survey (NHANES)-III (46
). The OA limitations presented would have resulted in estimates being biased toward the null; however, significant association remained in our study. Though not reaching general population prevalence estimates, the RA group sample size was large enough to confirm the association between RA and fracture. The WHI also had a larger percentage of women from minority groups, which allowed for examination of effect modification by race and ethnicity. The women of the WHI were followed on average almost 8 years, ensuring adequate numbers of fracture outcomes, especially for the more rare hip fracture outcome.
Arthritis and osteoporosis are import public health conditions for older adults. OA and RA affect over 25 million adults in the United States and fractures costs billions of health care dollars annually. The increase in fracture risk found in this study confirms the importance of fracture prevention in both patients with RA and OA.