OA causes substantial morbidity and disability, as well as economic costs [2]. Consequently, substantial efforts have been made to identify factors that can affect the incidence and progression of OA. Mendelian genetic disorders that have precocious joint failure as a component feature are an uncommon cause of OA [3], but their existence suggests that other genetic variants may contribute to common forms. Further evidence supporting a genetic contribution to common OA derives from studies looking at aggregation of OA within families [4-10] and within ethnic/geographic groups [11-14], and in twins [15-17].

In light of the evidence supporting a genetic contribution to common OA, investigators have sought to identify genetic variants, or chromosomal regions predicted to contain variants, that contribute to common forms of OA. Many variants have been suggested as risk factors for common OA based on association studies [18-25]. Similarly, genome-wide searches using relatives concordant for OA have suggested several chromosomal regions as containing susceptibility loci [14,26-29]. Importantly, statistical evidence in favor of chromosomal regions (or genes) increased when the original patient cohorts were stratified into subgroups based on sex, affected joint site, number of affected joints, and severity of joint disease, such as having undergone joint replacement [26,27,30-32]. Chromosome 4 and 16 linkages are specific to female affected families with hip OA who had undergone total hip replacement (THR) [30]; chromosome 2q linkage was found in distal interphalangial joint OA [33], knee OA [34] and hip OA [35]; and 11q linkage is specific to female OA [36] and hip OA [37]. When the linkage signals are analyzed after the OA affected individuals are stratified based on knee/hip joint arthritis and sex, it is observed that most of the linkages are consistent for hip OA [32,35,38,39], but they are not consistent for knee OA or sex specific.

However, to date, few putative disease-predisposing variants identified by one genetic method have been confirmed using another genetic method. These include secreted frizzled related protein 3 for hip OA [35], matrillin-3 for hand OA [28], and a locus on chromosome 6p for hip OA in women [40]. These recent data are intriguing, but their replication in other cohorts is required to confirm their causality and to distinguish them from positive associations due to type I error (for instance, a false-positive result).

In the present study, using joint replacement surgery for idiopathic OA as a qualitative measure of joint failure, we sought to determine how hip joint failure and knee joint failure cluster in families. Finding familial aggregation for one or both of these traits would support their usefulness for finding and testing candidate genes. Conversely, failure to find familial aggregation would suggest that these traits are not useful in genetic studies. As an appropriate control group for comparison, we chose to use the siblings of the probands' spouses. Other studies have used spouses as control individuals [6,8] presumably because they are more likely to be matched for ethnic/geographic ancestry and socioeconomic status. We used siblings of spouses, which comparably match for ethnic/geographic ancestry and socioeconomic background but they better enable us to delineate confounding variables such as sex and environment [41].

We found that hip arthroplasty is significantly increased in the siblings of THR probands when compared with the siblings of spouses, even after controlling for age and sex. Previous reports [39,48-50] have suggested that increased BMI is associated with hip and knee arthroplasty. Therefore, assuming that increased BMI could be a cause of the familial aggregation, we also collected BMI data from some of the study participants to determine the correlation of BMI with knee and hip arthroplasty. There were no significant increases in BMI with hip arthroplasty, whereas knee arthroplasty did exhibit an increase. These data lend further support to the contention that genetic factors contribute to the aggregation of end-stage hip OA but not knee OA. However, the data cannot preclude other shared factors (vocation, exercise habits) as being responsible for the familial aggregation rather than shared genes. In contrast, we found that knee arthroplasty was not increased among the siblings of TKR probands after controlling for age and sex. This finding argues against either genetic or shared environmental contributions to end-stage knee OA in families.

Our study agrees with prior studies that found increased familial aggregation for hip arthroplasty [6,7,14] and lends support to studies that found increased aggregation of hip OA defined by other measures [5,13]. Although our results do not support an earlier study that suggested increased familial aggregation for knee arthroplasty [6], they are consistent with those of another study [39] that also did not find familial aggregation for knee OA. It is important to emphasize that because this study recruited patients who underwent joint replacement surgery, it only addresses the role of genetics in severe forms of knee and hip OA, because we looked at arthroplasty rates and not at other measures of arthritis employed in prior studies [8,10].

The most important outcome of our study is that, when considering arthroplasty as a phenotypic trait, only hip arthroplasty is likely to be under genetic influence. Few gene association studies have been performed that specifically looked at genetic variants as risk factors for joint arthroplasty [35,38]. Most association studies compared allele rates between cases and controls (defined by the presence and absence of clinical or radiographic arthritis changes) or between severely and mildly affected individuals (defined by Kellgren-Lawrence or other OA scales) [18-20,23,51]. Therefore, our finding does not allow us to comment definitively on the conclusions of these studies. In contrast, several studies of excess allele sharing among concordant sibling pairs utilized participants with arthroplasty for idiopathic OA as the phenotypic trait; these studies found increased support for linkage between chromosomal regions and arthroplasty after stratification of the data by site of arthroplasty and sex [27,32,52,53]. Furthermore, the linkage signals increased when female or male hip affected sibling pairs were analyzed separately [30,31,54]. Although stratifying cohorts into smaller groups may increase the ability to detect linkage by decreasing heterogeneity, it can also lead to type I errors because of the multiple additional hypotheses being tested [55]. Our findings suggest that these investigators' stratification based upon hip replacement is appropriate. Although our data do not support stratifying hip replacement cohorts based on sex, our study design might not have been able to detect sex-dependent familial aggregation. Hawker and coworkers [56] found that, despite having a higher prevalence of severe arthritis (odds ratio = 1.76; P = 0.001), women significantly under-utilize arthroplasty as a treatment compared with men. Because our cohorts were ascertained only when probands underwent arthroplasty, our results may show less risk for female familial OA.

In the hip sibling cohort, in addition to family history, we identified age as an independent risk factor. This agrees with other studies [57] and the common observation that the prevalence of OA increases as humans age [58]. In the knee sibling cohort, we identified age and sex as independent risk factors, which also agrees with previous studies [57-59].

We speculate that shared genes account for familial aggregation of hip arthroplasty. However, other factors, such as access to medical care and communication between family members who may have experienced improved quality of life following arthroplasty, could also contribute to this clustering. These two explanations seem unlikely to account for the increased rate of hip replacement in our study because siblings did not have increased rates of knee replacement compared with control individuals. One would expect overall increases in the rate of arthroplasty if health insurance, access to health care, and 'word of mouth' were important factors in influencing a sibling's decision to undergo arthroplasty.

Increased BMI has been reported to correlate consistently with knee OA, although correlations with hip OA have yielded inconsistent results [50]. In this study correlations between BMI and arthroplasty were inconsistent. This may relate to the fact that the mean BMIs in our probands were in the overweight to mildly obese range (25.8–30.2 kg/m²) and that we collected BMI data from only a small subset of the siblings. Accordingly, skewed BMI distributions and small sample size in our cohort might have lessened our ability to demonstrate a clear correlation between BMI and arthroplasty risk. However, we did find a significant correlation between BMI and knee arthroplasty in the female siblings of hip OA probands and a trend toward significance in the male siblings, which supports prior studies describing BMI as a fundamental risk factor for knee joint failure [60,61].

In the TKR proband sibling cohort, a prior history of hip replacement was an independent risk factor for having a knee arthroplasty. Several explanations may account for this result. First, these individuals may have had severe forms of OA that affected several joints concurrently. To address the question of whether a more severe form of OA could account for this observation, we stratified probands based on their having unilateral versus bilateral arthroplasty. Arthroplasty rates were higher in siblings of probands who had bilateral hip replacement but not bilateral knee replacement. Because we excluded probands having both a hip and a knee arthroplasty from this study, we cannot comment on whether a more generalized form of OA that causes end-stage OA in multiple joints will also cluster in families. Second, arthritis in the second joint may have arisen as a consequence of disability caused by arthritis in the first joint [44]. Third, the threshold for having a second arthroplasty may be reduced in individuals who had a satisfactory result from their first replacement.

Among the strengths of our study is that it ascertained probands who underwent only hip arthroplasty or only knee arthroplasty, and recruited a control group that did not share a common household but was comparable in terms of age and ethnicity. We also applied a statistical analysis (SEGREG) that allowed for sibling correlations. Prior association studies of hip and knee OA compared affected and unaffected family members or siblings of affected individuals with the general population. These types of studies are more difficult to control for shared environment in families, population stratification, and/or site(s) of joint involvement.