The age-adjusted estimated heritability (black bars), shared environmental influence (gray bars), and nonshared environmental influence (white bars) of hip fracture (A), osteoporotic fracture (B), and all fractures (C) by age at first fracture within each twin pair.
Michaëlsson K, Melhus H, Ferm H, Ahlbom A, Pedersen NL. Genetic Liability to Fractures in the Elderly. Arch Intern Med. 2005;165(16):1825–1830. doi:10.1001/archinte.165.16.1825
The genetic impact on the causation of osteoporotic fractures is unclear. A large twin study is ideally suited to determine the genetic liability to categories of fracture at various ages.
A cohort of all 33 432 Swedish twins born from 1896 to 1944 was used to evaluate the genetic liability to fracture occurrence in the elderly. The Swedish Inpatient Registry and computer-assisted telephone interviews enabled us to identify 6021 twins with any fracture, 3599 with an osteoporotic fracture, and 1055 with a hip fracture after the age of 50 years.
Genetic variation in liability to fracture differed considerably by type of fracture and age. Less than 20% of the overall age-adjusted fracture variance was explained by genetic variation. The age-adjusted heritability of any osteoporotic fracture was slightly greater (0.27; 95% confidence interval [CI], 0.09-0.28), and for hip fracture alone, it was 0.48 (95% CI, 0.28-0.57). Heritability was not attenuated after further adjustment for several known osteoporotic covariates but was considerably greater for first hip fractures before the age of 69 years (0.68; 95% CI, 0.41-0.78) and between 69 and 79 years (0.47; 95% CI, 0.04-0.62) than for hip fractures after 79 years of age (0.03; 95% CI, 0.00-0.26).
The importance of genetic factors in propensity to fractures depends on fracture site and age. The search for susceptibility genes and environmental factors that may modulate expression of these genes in younger elderly patients with hip fracture, the most devastating osteoporotic fracture, should be encouraged. Prevention of fractures in the oldest elderly should focus on lifestyle interventions.
Osteoporotic fractures constitute a tremendous and growing problem worldwide. These fractures have a profound impact on quality of life: only one third of patients with osteoporotic fracture regain their prefracture level of function.1,2 There is also a substantial increase in risk of death after these fractures.3- 5 Established lifestyle risk factors can explain only a modest proportion of the liability to osteoporotic fractures.6 The importance of genetic factors in fracture risk is unclear. Twins offer a natural study population for evaluating genetic risk because monozygotic twins are genetically identical, whereas dizygotic twins share half of their segregating genes.7 Thus, if heritable factors contribute to fracture risk, concordance should be greater in monozygotic than in dizygotic twin pairs. We found only one previously published study evaluating the heritability of fracture.8 The authors concluded that the genetic influence on liability to osteoporotic fracture risk was low. On the contrary, heritability for bone mineral density (BMD), the best established risk factor for osteoporotic fractures,9,10 is substantial (50%-85%).11,12 The genetic liability to osteoporotic fractures might reflect that for BMD, given the importance of BMD in osteoporotic fractures. On the other hand, concomitant diseases, balance disorders, and lifestyle habits not directly related to BMD but to fracture risk13- 15 that reflect environmental influences may counterbalance genetic influences due to BMD. A better understanding of the genetic etiology of fractures would enhance development of proper strategies for intervention and prevention. If genetic susceptibility is low, lifestyle intervention approaches should be encouraged, whereas research and primary prevention efforts should focus on families with high risk for fractures when genetic influences are greater.
We therefore examined information concerning outpatient- and inpatient-treated fractures in the population-based Swedish Twin Registry to determine the genetic liability to different categories of fractures.
Subjects were ascertained from the Swedish Twin Registry, currently the largest twin registry in the world.16,17 Twin pairs born in 1896 to 1944 of whom both co-twins were alive in 1972 were considered eligible for this study, for a total of 33 432 subjects. A computer-assisted telephone interview was conducted between 1998 and 2000 that included a number of items regarding diseases and symptoms, prescription and nonprescription medication use, occupation, education, physical activity, anthropometric measures, and consumption of alcohol and tobacco. Efforts were made to interview members of a pair within a month of each other to minimize the risk of biasing the results by differential age effects. Of the base sample, 4156 died between 1972 and the telephone interview, leaving 29 276 eligible twins for the interview. Of these, 24 598 (84%) agreed to participate in the interview. The records of all twins in the base sample were matched to the inpatient, cancer, and death registries for identification of diagnoses through December 31, 2000, regardless of participation in the interview. In addition, like-sex twin pairs participated in mailed questionnaire surveys in 1967 (for those born in 1896-1925) and 1972 (for those born in 1926-1944). Zygosity information was obtained at the time of registry compilation on the basis of questions about childhood resemblance and updated during the most recent screening for those with previously uncertain diagnoses. Four separate validation studies using serologic testing and/or genotyping have shown that, with these items, 95% to 98% of twin pairs are classified correctly.17
Fractures were ascertained through 2 sources, the Inpatient Register and self-report. The Inpatient Register, started in 1964, covered 83% of the Swedish population in 1972 and all inpatient care in Sweden since 1987. It is updated annually and is valid for identifying cases of fracture.18 The incidence of pathological fractures and fractures related to high-energy trauma among the elderly in Sweden is approximately 1%.19,20
Because there is no national register for outpatient diagnoses in Sweden, we assessed outpatient-treated fractures by asking the twins during the telephone interview to recall type and number of fractures after the age of 50 years. In an effort to validate these answers, the orthopedic hospital and radiography records of all twins (n = 647) living in one county were scrutinized for records of treated orthopedic fractures. The reviewer was blinded to the telephone self-report of fractures and zygosity. One hundred eight twins with fractures were identified by the hospital records. Of these subjects, 99 reported a fracture at the correct site during the telephone interview, leaving 9 subjects not reporting a fracture. Of these 9 twins, 4 had had finger or toe fractures, 2 proximal upper-arm fractures, 1 a fracture of the capitulum radii, 1 a clavicular fracture, and 1 a fracture of the scapula. There were no false-positive reports; all self-reported fractures were confirmed. The positive and negative predictive values were thus 1.00 and 0.98, respectively.
The similarity in occurrence of fractures among monozygotic and dizygotic twins was summarized as the probandwise concordances with 95% confidence intervals (CIs). This analysis was restricted to like-sex twins. In addition to any fracture, separate analyses were performed for fractures specifically designated as osteoporotic, ie, hip, pelvis, spine, forearm, and humerus,21 as well as for hip fractures alone. If we assume C to be the number of concordant pairs and D the number of discordant pairs, the pairwise concordance is the relative number of twin pairs with fractures among both twins calculated by C/(C+D) and the probandwise concordance is the ratio of fractures in concordant pairs over the total number of twins with a fracture (2C/[2C+D]). The overall cumulative risk of fracture was calculated by dividing the number of fracture cases by the total number of individuals. The relative fracture risk was calculated by dividing the probandwise concordance by the overall cumulative fracture risk.
Heritability is an estimate of the relative importance of additive genetic differences for variance in the population. Shared environment reflects experiences that contribute to familial (twin) similarity. Nonshared environment refers to the contribution of environment experiences not shared by family members (eg, twins). Estimates of heritability were based on liability-threshold models of all twin types, which assume a latent, normally distributed liability to disease that is manifest as a categorical phenotype, ie, affected or not affected.17,22 Expectations for fitting the observed data to a genetic model are based on the fact that genetic similarity is half as great for dizygotic as for monozygotic twin pairs, while shared environmental influences contribute equally to making monozygotic and dizygotic pairs similar. Models were fit to raw data files that included information about fracture status and age at follow-up for each of the twins in a pair for male and female monozygotic twins, male and female like-sex dizygotic twins, and unlike-sex twins by using the Mx program.23
We speculated that the age at time of fracture might be important for the parameter estimates. Therefore, we additionally stratified the analysis by the tertile distribution of age of the first inpatient-treated hip fracture within fracture-affected twin pairs. The same age cutoffs were used for all fractures and osteoporotic fractures. Because we had access to the date of hospitalization (normally the date of fracture) through the Inpatient Register, we could calculate an accurate age at fracture. Self-reported fracture cases were not included in this analysis.
We further considered a multivariate logistic regression model, conditional on pair membership, to estimate odds ratios (ORs), with 95% CIs, as measures of relative risk of a fracture in the second twin if the first twin had had a fracture. The etiologic fraction was calculated by the following formula: ([OR − 1]/OR) × p, where p is the number of twins with a co-twin who had a fracture divided by all twins with fractures, which equals the probandwise concordance. This is an estimate of the proportion of all fracture cases that could be attributed to a twin partner with a fracture. Larger differences in etiologic fractions between monozygotic and dizygotic twins indicate a stronger genetic component. In the multivariate model we included, in addition to age at the end of the follow-up period, body mass index (calculated as weight in kilograms divided by the square of height in meters) and menopausal age (all continuous), hormone therapy (ever vs never), smoking status (never, former, or current), and leisure-time physical activity (low, medium, or high) from the telephone interview. From the central registries, we further included dichotomous variables reflecting malignancies, endocrine disorders, cardiovascular diseases, psychiatric disorders, neurologic disorders, gastrointestinal disorders, musculoskeletal diseases including rheumatoid arthritis, and renal failure or other urinary tract diseases. The primary purpose of the multivariate adjustment was to evaluate whether differential changes of the ORs for the zygosity groups could be detected. Differential changes in the ORs after multivariate adjustment and, hence, attenuated differences in etiologic fractions between monozygotic and dizygotic twins may suggest that the age-adjusted heritability estimates are inflated by the risk factors included in the models.
Because our aim was to study fractures in old age, a fracture before the age of 50 years (n = 499) was categorized in all analyses as no fracture. When analyses restricted to osteoporotic fractures were performed, twins with only nonosteoporotic fractures after the age of 50 years were treated as missing values. Similarly, only nonfracture cases and hip fracture cases, with or without other types of fractures reported, were included in the analysis of hip fracture risk.
We excluded from the analysis the 942 twin pairs (5.6%) with an uncertain zygosity designation, leaving 3724 monozygotic, 6314 like-sex dizygotic, and 5736 unlike-sex dizygotic twin pairs. Descriptive characteristics are reported in Table 1. Only modest differences could be detected between zygosity groups in anthropometric measures, lifestyle habits, and prevalent diseases.
A total of 6021 fracture cases were identified (Table 2); 1 of 5 of the twins had had a fracture after the age of 50 years, with a higher proportion among women (23%) than among men (14%). Forty percent of the twins with a fracture had the fracture identified only by the telephone interview. More than half of the fracture cases (n = 3599 [60%]) were classified as osteoporotic. The most important osteoporotic fracture, hip fracture, was recorded for 1055 twins. There were no differences in fracture frequencies by zygosity.
The probandwise concordance for any fracture, based on analysis of like-sex twins, was 31% in monozygotic twins compared with 27% in dizygotic twins (Table 3). The crude relative overall fracture risk was 1.50 (95% CI, 1.43-1.56) for monozygotic twins compared with 1.33 (95% CI, 1.28-1.38) for dizygotic twins. The difference in relative risk of hip fracture was more pronounced: 7.91 (95% CI, 7.00-8.62) for monozygotic twins compared with 3.75 (95% CI, 3.12-4.29) for dizygotic twins.
We consequently found a strong fracture site–dependent influence on the heritability estimates (Table 4). The age-adjusted heritability for overall fracture risk was modest (0.16; 95% CI, 0.11-0.21), whereas the age-adjusted heritability for osteoporotic fractures was 0.27 (95% CI, 0.09-0.28). When hip fracture cases were excluded from the osteoporotic category, the result remained similar, with a heritability of 0.25 (95% CI, 0.02-0.37). The strongest genetic influence was evident for hip fractures. Nearly half of the variance in liability to these fractures was attributable to genetic factors. Even though there was a tendency toward lower heritability estimates of fractures among the men, no statistically significant sex-specific differences in genetic effects could be detected. The shared (familial) environmental component was negligible.
The heritability of inpatient-treated fractures was 0.23 (95% CI, 0.07-0.30) and strongly dependent on the age at which the first twin had had the first fracture (Figure). If the first hip fracture within a twin pair occurred before the age of 69 years, heritability was 0.68 (95% CI, 0.41-0.78), and between the ages of 69 and 79 years, it was 0.47 (95% CI, 0.04-0.62); in contrast, after age 79 years it was 0.03 (95% CI, 0.00-0.26). The same pattern of attenuated heredity estimates with increasing age was also evident for the other fracture categories.
The crude and multivariate ORs of fracture given that the co-twin also had a fracture compared with subjects without fracture are presented in Table 5. In general, ORs and etiologic fractions were higher in monozygotic than in dizygotic twins. The differences in etiologic fractions between monozygotic and dizygotic twins were even more pronounced after multivariate adjustment.
The results of our study demonstrate that heritability for fracture is dependent on site and age. These findings are important for efforts to target effective interventions against osteoporotic fractures. Our results indicate that hip fracture prevention strategies should be focused on lifestyle intervention in the oldest elderly. On the other hand, especially hip but even other types of osteoporotic fractures at younger ages seem to be strongly genetically influenced. This observation should encourage the search for fracture susceptibility genes and gene-environment interaction analysis. Furthermore, assessment of osteoporotic fracture risk by a clinical examination may be recommended for relatives of patients with hip fractures before the age of 80 years.
Several studies have shown that a family history of osteoporotic fractures increases the risk of these fractures.24- 26 In one family study of women who had previously participated in bone density studies,27 the heritability of distal forearm fractures was 25%. Unlike twin studies, family studies cannot separate genetic from common familial environmental influences on fracture risk. We found one previous twin study of fracture. On the basis of analyses of data within the Finnish twin cohorts, Kannus et al8 concluded that osteoporotic inpatient-treated fractures are not strongly influenced by genetic factors. The limited power of the study, although composed of more than 15 000 twins with 786 fracture cases, precluded determination of the best model fit, age effects, and analysis of specific fractures. In contrast, our study, with twice as large a study base, identified 8 times as many fractures. Thus, our power to evaluate the genetic liability to fracture was greatly enhanced, although analyses of this sort are population dependent. Our study was performed in a high-incidence area of osteoporotic fractures, whereas Finland, which has a different ancestry,28 has lower fracture rates.29 These genetic differences may result in different fracture propensities; however, the 30% lower hip fracture rate in Finland compared with Sweden is moderate compared with the 15-fold variation between regions of the world.29 Thus, we believe that our results are robust and representative of, at the very least, Nordic populations.
Bone mineral density and content, bone size, bone turnover, and hip axis length, all of which independently predict future fracture risk,30- 32 are strongly influenced by genetic factors.11 Although variation in these bone measures seems to be predominantly inherited, polymorphisms of known candidate genes have hitherto explained only a small proportion of this variation.11,33 On the other hand, a maternal history of hip fracture doubles hip fracture risk independent of bone density,14 indicating that other potentially inherited factors, such as those affecting propensity to falling, are important contributors to osteoporotic fracture risk.
The main advantages of our investigation are the large and complete set of twins followed up for more than 25 years within a country that has a relatively homogeneous population; the considerable number of fractures that enabled us to examine fracture-, sex-, and age-specific heritability; and extensive covariate information. The differences in etiologic fractions between monozygotic and dizygotic twins were not attenuated after multivariate adjustment. This indicates that the heritability estimates are not overestimated because of the influences of our conceivable covariates. Despite these strengths, there are several potential limitations of our study. Incomplete fracture ascertainment might lead to overestimation of the environmental component and underestimation of the genetic component in our study. The Swedish Inpatient Register has a high validity of fractures, although it had incomplete (83%) coverage at the beginning of the follow-up in 1972. Since the mean ageat entry to this cohort was 40.6 years and only 18% were older than 50 years, the proportion of inpatient-treated fractures overlooked is probably low. Sixty percent of the fractures were identified by the Inpatient Register. The remainder of the fractures were identified by the telephone interview. However, restricting our analyses to twins participating in the telephone interview showed estimates almost identical to those from analyses with all twins included. Finally, only clinically overt fractures were included in this study. A substantial number of vertebral fractures, particularly those with mild clinical symptoms, might have been overlooked if no radiographic examination had been performed. Nonetheless, we find it unlikely that this potential detection bias differed by zygosity.
We conclude that the genetic influence on susceptibility to fractures is dependent on type of fracture and age at fracture event. The heritability of osteoporotic fractures is stronger than has been previously estimated, especially for early-occurring osteoporotic fractures. A search for genes and gene-environmental interactions that affect early osteoporotic fracture risk is likely to be fruitful, but fracture-prevention efforts at older ages should be focused on lifestyle habits.
Correspondence: Karl Michaëlsson, MD, PhD, Department of Surgical Sciences, Section of Orthopaedics, Uppsala University Hospital, S-751 85 Uppsala, Sweden (firstname.lastname@example.org).
Accepted for Publication: January 3, 2005.
Financial Disclosure: None.
Funding/Support: This study was supported by grants from the Swedish Research Council, Stockholm; National Institute on Aging, Bethesda, Md (grant AG 08724 to Margaret Gatz, PhD); and Uppsala University Hospital, Uppsala, Sweden.