The genetic epidemiology of joint hypermobility: A population study of female twins




To estimate the genetic influence on joint hypermobility in an unselected population using a classic twin study design.


A self-report questionnaire on joint hypermobility as well as data on age, height, weight, estrogen replacement therapy, and menopause status were obtained from 483 monozygotic (MZ) and 472 dizygotic (DZ) unselected female twin pairs ages 21–81 years who were registered with the St Thomas' Adult Twin Registry in the UK.


The overall prevalence of hypermobility was 19.5% in MZ twins and 22.1% in DZ twins. The prevalence of hypermobile joints declined with age, falling from 34% in subjects ages 20–30 years to 18.4% in those ages 60 years or older. Significantly greater concordance for joint hypermobility was observed in the MZ twins when compared with the DZ twins (60% versus 36%), consistent with a genetic influence. In variance components analysis, the age- and body mass index–adjusted heritability of joint hypermobility was estimated to be 70% (95% confidence interval 57–89%).


Genetic factors have a substantial contribution to joint hypermobility in the adult female population.

Joint hypermobility is common in the adult population and is an under-recognized cause of morbidity. Approximately 30% of adults can be defined as having hypermobile joints. Moreover, associations between joint hypermobility and the risk of soft tissue injury, chronic widespread pain, and early degenerative disease have been reported.

The factors that determine joint hypermobility have not been extensively studied. Racial and ethnic variation is widely recognized; for example, joint hypermobility is more common in Asian and African population groups compared with whites (1). However, the extent to which this might be accounted for by genetic factors, as opposed to environmental and lifestyle factors, has not been established. Possible environmental factors that might contribute to the onset of joint hypermobility include hormone status or physical training.

The relative contribution of genetic and environmental factors to traits and diseases can be separated using the classic twin study design. Herein we report the results of the first twin study to assess joint hypermobility among healthy volunteers who were participants in a national twin registry in the UK.


Study population.

The study subjects comprised unselected female–female twin pairs recruited from the St Thomas' Adult Twin Registry (London, UK). The twins enrolled in the registry were all healthy volunteers drawn from the UK population and were recruited through successive media campaigns. These twins have been shown to be representative of the UK population with respect to their physical characteristics and prevalence of common musculoskeletal diseases (2).

Subjects who were still enrolled in the registry in July 2003 and who had replied to surveys in a previous study of soft tissue disorders (3) were mailed a questionnaire that included questions designed to classify the presence of joint hypermobility. This subgroup of the registry was invited to participate because there was an intention to also explore associations between soft tissue disorders and hypermobility (results not reported herein). Questions about hypermobility were included in conjunction with questions pertaining to the twins' general health; the respondents would have been unaware of any specific hypothesis related to the study of hypermobility.

Classification of hypermobility.

Joint hypermobility is traditionally classified using the Beighton score, which is derived by examination. In this scheme, a cutoff point of 4 of 9 points is widely used to define the presence or absence of hypermobility (4). Because of its reliance on examination, this scoring method may be unsuitable for use in large-scale epidemiologic studies. We have recently demonstrated an alternative scheme that is entirely question-based and adequately encompasses all of the information included in the Beighton scoring method. Classification relies on the responses to a set of 5 questions, each designed to elicit evidence of specific characteristics of hypermobility in the subjects' history (Table 1). Affirmative responses to 2 or more questions had a sensitivity of 80–85% and specificity of 80–90% when compared with the cutoff point of 4 of 9 on the Beighton score (5).

Table 1. Classification of joint hypermobility*
  • *

    Affirmative responses to 2 or more questions were used to classify subjects as having joint hypermobility.

1. Can you now (or could you ever) place your hands flat on the floor without bending your knees?
2. Can you now (or could you ever) bend your thumb to touch your forearm?
3. As a child did you amuse your friends by contorting your body into strange shapes OR could you do the splits?
4. As a child or teenager did your shoulder or kneecap dislocate on more than one occasion?
5. Do you consider yourself double-jointed?

The validity of the questionnaire in this study was further assessed in a separate survey of 80 unselected individual twins attending the Twin Research Unit. Each patient had undergone a physical examination for joint laxity, which was assessed using the Beighton score, and had completed the questionnaire. The sensitivity and specificity of the questionnaire for hypermobility in the twins were similar to those reported in the previous study (5).

Zygosity and other questionnaire-derived variables.

Zygosity was previously determined by a separate questionnaire, and had been confirmed by DNA fingerprinting in a proportion of subjects who had attended the unit for earlier studies. Data on age, height, weight, estrogen replacement therapy, and menopause status were collected.

Statistical analysis.

The similarity of characteristics among the twins was assessed, first, by casewise concordance. This is a measure of the probability of the cotwin of an affected twin also expressing the trait. For the ascertainment method in this sample, casewise concordance was expressed as a ratio, calculated as 2C divided by (2C + D), where C is the number of concordant pairs and D is the number of discordant pairs in the sample. Higher concordance in monozygotic (MZ) twins when compared with dizygotic (DZ) twins is indicative of a genetic effect. The significance of the difference between MZ and DZ heritability was estimated from a likelihood-based formula described by Witte et al (6).

Genetic influences on a trait can also be estimated by measuring heritability. This provides a measure of the extent to which variation in a trait in the population can be accounted for by genetic factors (7). In this analysis, the presence or absence of hypermobility was considered to be a discontinuous trait and the analysis focused on the heritability of the underlying susceptibility to hypermobility (8). The heritability of hypermobility was estimated using an implementation of the DeFries-Fulker regression method that has been described elsewhere (9). This approach allows a comparison of models that include a combination of additive genetic, common environmental, and unique environmental variance components. Heritability can be estimated from the relative contribution to the model of genetic variance that exhibits the most appropriate explanation of the data; this is determined through a combination of model fit and parsimony. The approach was extended to account for the influence of covariates, which included age, body mass index (BMI), menopause status, and concomitant use of estrogen replacement therapy as potential confounders.


We sought to obtain data on 2,600 twins. Data were available on 483 MZ twin pairs and 472 DZ twin pairs. Six hundred ninety cotwins had not replied to the questionnaire, yielding a response rate of 70%.

The prevalence of joint hypermobility was similar between the MZ and DZ groups (19.5% in MZ twins versus 22.1% in DZ twins). These values were comparable between singletons and twin pairs. The characteristics of age, BMI, menopause status, and use of estrogen replacement therapy were also similar between the zygosity groups. The mean ± SD age was 53.5 ± 12.5 years (range 21–81 years), and the mean ± SD BMI was 22.9 ± 4.1 kg/m2 (range 13.3–48.2 kg/m2) (Table 2). These characteristics were similar to those of the total cohort of 3,600 twins reported previously (3). The prevalence of hypermobility declined with age, with those twins ages 20–30 years reporting a prevalence of hypermobility of 34% and those ages 60 years or older reporting a prevalence of 18.4%.

Table 2. Characteristics of the cohort, by zygosity*
VariableMZ twinsDZ twins
  • *

    MZ = monozygotic; DZ = dizygotic; BMI = body mass index.

Age, years  
 Mean ± SD53.3 ± 12.853.7 ± 11.7
Age group, %  
 20–30 years64
 31–40 years1110
 41–50 years1820
 51–60 years3131
 61–70 years2226
 71+ years129
BMI, kg/m2  
 Mean ± SD22.5 ± 4.123.1 ± 4.1
Menopause status, %  
Current oral contraceptive use, %9.76.2
Estrogen replacement therapy, %  
Prevalence of hypermobility, %19.522.1

The distribution of affirmative answers, by age, to the 5-part questionnaire was similar between the zygosity groups. The MZ and DZ twin populations answered each component of the questionnaire in the affirmative with similar frequency, ranging from 50% for question 1 to 3–4% for questions 4 or 5. The number of affirmative answers per individual was also similar across zygosity, with 35% of individuals answering yes to only 1 question, 14% answering yes to any 2, and 5% answering yes to any 3. The casewise concordance for each question was higher in MZ twin pairs than in DZ twin pairs, but this was most significant for questions 2, 3, and 4, for which MZ:DZ concordance ratios were 1.75, 1.62, and 3.6, respectively. Ninety percent of the population prevalence for hypermobility and the casewise concordance was accounted for by an affirmative answer to any combination of 2 or all 3 of questions 1–3.

Casewise concordance for the presence of hypermobility was significantly greater in the MZ twin pairs (at 60%) than in the DZ twin pairs (at 36%). A chi-square test for the difference in concordance was significant (χ2 = 16.9, P = 0.03).

In the variance components modeling analysis, additive genetic and unique environmental components provided the best explanation of the data. In this model, the heritability for hypermobility was estimated at 75% (95% confidence interval 59–91%). When age and BMI were included in the analysis, the heritability was adjusted to 70% (95% confidence interval 57–89%). The addition of menopause status or other hormone parameters as potential confounders had no further effect on this estimate (Table 3).

Table 3. Distribution of twin pairs concordant and discordant for hypermobility, and heritability estimates*
 MonozygoticDizygoticAnalysis, heritability estimates
  • *

    95% CI = 95% confidence interval; BMI = body mass index; ERT = estrogen replacement therapy.

Number of pairs concordant for presence of hypermobility6039 
Number of pairs discordant for presence of hypermobility83148 
Casewise concordance, %6036 
Significance of difference, chi-square (P)  16.90 (0.03)
Heritability, % (95% CI)  75 (59–91)
Age- and BMI-adjusted heritability, % (95% CI)  70 (57–89)
Age-, BMI-, menopause-, and ERT-adjusted heritability, % (95% CI)  70 (56–89)


Our study confirms that joint hypermobility has a substantial heritable component. This is the first formal demonstration of a heritable component to this condition through the study of twins. The high level of heritability supports observations of joint hypermobility within families, which have led to the consensus that the condition is autosomal-dominant in character.

In interpreting this finding, several considerations merit further discussion. We used a self-report questionnaire for case definition. The questionnaire was originally developed for use in the clinic setting as a screening tool for hypermobility. Epidemiologic studies of joint hypermobility have, to date, used the Beighton 9-point score (4). Although this remains the recognized standard, it should be noted that the cutoff point of 4 of 9 for positivity on the Beighton score is arbitrary, requires a physical examination, takes no account of previous agility, and excludes assessment of common sites of hypermobility including the neck, shoulders, hips, and ankles. As such, the use of the Beighton score may, for practical reasons, limit the size of a population study and could generate a significant number of false-negatives. A self-report questionnaire may be justified as an alternative in situations where a physical examination is impractical, particularly if it also does not rely entirely on specifying particular joints and takes account of agility in a historical context.

We consider the questionnaire to have performed well in this study for several reasons. First, it is unambiguous and simple, and was completed in full by all respondents. Second, it determined a prevalence value that was consistent with that in other large population studies (1). Third, judging by the sensitivity and specificity of the findings, this instrument behaved in the twin population in the same way as it had done in repeatability studies of clinic populations (5).

We found an age effect in the prevalence data that was not expected given that the questionnaire was designed to take account of loss of agility secondary to aging. Younger women tended to report the presence of hypermobility more than older women. The finding was, however, similar in both the MZ and the DZ population and had only a small confounding effect on the heritability estimate. This may reflect a true change in prevalence with aging, or it may be the consequence of recall bias. However, it is important to stress that the variables included in the scoring system represent discrete features that are unlikely to be recalled equivocally in different age groups. Furthermore, recall and repeatability of self-reported data have been found to be similar between MZ and DZ twins. We have shown both MZ individuals and DZ individuals to have high levels of repeatability in a study that compared subjects' replies to a number of domains at baseline and at 3 years (Hakim AJ, et al: unpublished observations). The domains evaluated included distant, fixed events such as age having moved away from the family home, as well as levels of activity and reported illnesses, allowing individuals to consider the possibility of a change over the last 3 years. Kappa values for equivalent responses at baseline and year 3 were high, at between 0.8 and 0.95, in both MZ and DZ twins. There was no evidence of a significant difference in casewise concordance by zygosity, and age was not associated with failure to repeatedly recall an event or date, regardless of the distance in time.

The age effect seen in our present study could have led to an underestimate of the prevalence of hypermobility. However, given a similar distribution, by age, of responses between zygosity, and the observation that recall was similar between the MZ and DZ twin pairs, we would not expect this effect to bias the heritability estimate.

Differences in the shared environment of MZ and DZ twins have the potential to bias the interpretation of a heritable effect (10). Few environmental factors would be considered to have any major confounding on the heritability estimates. We accounted for the influence of hormone factors. Exercise and learned agility may be another factor that was not considered. From a previous study, we know that there were no significant differences in the reported levels of activity either at home, leisure, or at work between the MZ and DZ twins (3). The number of individuals training to such a degree that they would increase their agility is likely to be small.

The variance could also be partly explained by a greater awareness of cotwin health in the MZ twins than in the DZ twins. The assumption is that MZ twins are closer to each other, socially and psychologically, and that an MZ individual is more likely to question his or her own health given the presence of a disorder in the cotwin. Further work from our group (Hakim AJ, et al: unpublished observations) suggests that MZ twins are no more likely to be aware of the illnesses of their cotwins than are DZ twins.

We have previously explored twin–cotwin awareness of the presence of back pain, osteoarthritis, migraine, and asthma, in 150 MZ twin pairs and 150 DZ twin pairs. The participants were asked to declare whether they believed they had any of these conditions, whether they thought their cotwin had any of these conditions, and how they rated their own knowledge of their cotwin's health. With regard to knowledge of cotwin health, 23% of the MZ twins responded that their knowledge was “fair” and 76% of the MZ twins responded with “good,” compared with 30% and 68%, respectively, of the DZ twins. For each condition, in both zygosities, ∼60–65% of individuals were correct in identifying the presence or absence of the disorder in their cotwin, 20–25% of individuals reported that they did not know, and 5–10% got it wrong. No significant difference was observed between the MZ and DZ pairs with regard to casewise concordance for the correct identification (whether present or absent) of disease for any of the 4 conditions studied (Hakim AJ, et al: unpublished observations). We would argue, therefore, that awareness of cotwin health is unlikely to be a significant confounder.

Joint hypermobility may remain asymptomatic, and indeed, for athletes, dancers, and musicians, this condition is likely to be an asset. It is also, however, associated with several musculoskeletal disorders, including chronic pain disorder, joint hypermobility syndrome, premature osteoarthritis, and soft tissue injury (1). It has yet to be determined whether there are specific environmental or genetic factors that increase the risk of developing such disorders in the hypermobility population.

The demonstration, in our study, of a large genetic component to hypermobility should encourage a wider exploration of abnormalities of collagen and other connective tissue proteins. To date, there are no reports of genetic studies in the literature pertaining to hypermobility in the general population. However, the range of potential candidate genes of interest is large and includes, for example, those encoding the protein or processing enzymes in the production of collagens, elastin, fibrin, or tenacins.

In summary, joint hypermobility in the population is a common condition and is strongly genetically determined. Knowledge of the individual genetic abnormalities that lead to hypermobility may offer greater insight into the pathophysiology of this condition, and enable better understanding of the risk factors associated with other musculoskeletal disorders.