Reproduction Life History and Hip Fractures HANS CHR. PETERSEN, PHD, BERNARD JEUNE, MD, JAMES W. VAUPEL, PHD, AND KAARE CHRISTENSEN, MD, PHD PURPOSE: The present study assesses the association between reproduction and hip fractures. METHODS: We used two surveys on elderly Danish twins of both sexes and the Danish National Register of Patients. In a cross-sectional study of 2045 twins aged 75–98 years in 1995, we studied the association between different aspects of reproduction and hip fractures leading to hospitalization between 1977 and 1994. In a prospective study, 3057 twins aged 66–99 years in 1977 were followed for a total of 29,112 years, and the association between number of children and incidence of hip fractures was investigated. RESULTS: In the cross-sectional study, as well as the prospective study, we found, for both sexes, that having no children was associated with a higher risk of hip fracture compared to having at least one child. When excluding persons without children, we found no effect of number of birth events/children. In the cross-sectional study, we found no effect of birth interval length or age at first birth. For women, we found an effect of age at last birth. CONCLUSIONS: No “cost-of-reproduction” in terms of hip fractures was observed. On the contrary, we found that having one or more children was associated with a lower risk of hip fracture. Ann Epidemiol 2002;12:257–263. © 2002 Elsevier Science Inc. All rights reserved. KEY WORDS: Hip Fractures, Reproduction, Number of Children, Men, Women. INTRODUCTION A recently published twin study (1) showed an association between number of biological children and tooth loss in women, confirming a proverb known in several countries: “a tooth per child”. No such association was found in men, indicating a biological rather than an environmental explanation. The child-tooth loss association in women is one example of what in evolutionary terms is called a trade-off between reproduction and somatic maintenance. Another possible trade-off related to the skeletal system is a connection between reproduction and osteoporosis, mediated by calcium depletion in the mother. At present, there is no clear evidence that reproductive effort increases the risk of osteoporosis (2, 3). Kolthoff and coworkers (3) have shown that women in general tend to reach a low level of bone density during pregnancy and especially during the first months after delivery. However, prepregnancy bone density levels were reached about 24 months after conception. Some studies (e.g., 4) have shown that nulliparous women have a higher risk of experiencing hip fractures, compared to women with at least one child, and a recent study (5) has shown that women with six or more children have the same bone mineral content as nulliparous women. A possible association between reproduction and osteoporosis could thus still be mediated by accumulating calcium depletion in the long run, if a woman have many children with repeatedly short birth spacing. As osteoporosis leads to a higher risk of experiencing hip fractures (6–8), these fractures can be used as proxies for osteoporosis in studies of associations between reproductive life history and osteoporosis. An advantage of such an approach is that fractures can be considered the ultimate clinical outcome of osteoporosis, with a direct relation to well-being and survival (6–8). The present study gives results pertaining to this association based on two comprehensive Danish twin surveys, including information on reproduction. Information on fractures was obtained by linkage to a nationwide register of hospital admissions. MATERIALS From the Danish Center for Demographic Research (H.Chr.P., J.W.V., K.C.) and Epidemiology (H.Chr.P., B.J., K.C.), Institute of Public Health, University of Southern Denmark, Odense C, Denmark; Terry Sanford Institute, Duke University, Durham, NC, USA (J.W.V., K.C.); and Max Planck Institute for Demographic Research, Rostock, Germany (J.W.V.). Address reprint requests to: Hans Chr. Petersen, Ph.D., University of Southern Denmark, Sdr. Boulevard 23A, DK-5000 Odense C, Denmark. Received April 18, 2001; revised May 14, 2001; accepted July 10, 2001. © 2002 Elsevier Science Inc. All rights reserved. 655 Avenue of the Americas, New York, NY 10010 The data employed in the present study originate from two surveys on Danish male and female twins and the Danish National Register of Patients. It should be emphasized that we do not intend to do a twin study, but rather assume that the results obtained can be generalized to the population in general (9). Males were included in order to evaluate the 1047-2797/02/$–see front matter PII S1047-2797(01)00275-7 258 Petersen et al. REPRODUCTION AND HIP FRACTURES Selected Abbreviations and Acronyms BMI  body mass index ICD8  The International Statistical Classification of Diseases and Related Health Problems, eighth revision ICD10  The International Statistical Classification of Diseases and Related Health Problems, tenth revision CI  Confidence interval impact of factors not related to childbearing, e.g. environmental factors. Furthermore, osteoporosis and hip fractures in men are receiving increasing attention (10). The surveys and their merging with the Danish National Register of Patients have been approved by the local ethical committee. The Longitudinal Study of Aging Danish Twins 1995—the Cross-Sectional Survey The Longitudinal Study of Aging Danish Twins 1995 survey included all Danish twins aged 75 years in 1995. The survey, comprizing 2401 individuals, has been described in detail elsewhere (11, 12). The survey interview included questions pertaining to hip fractures (two categories: ever/ never), number of birth events of biological children, and date of birth for these. The number of birth events was divided into four categories (0, 1–2, 3–4, 5), with twin and triplet births counting as one birth event. In the following, number of birth events will also be called parity, even for men. For persons with two or more birth events, mean time span between birth events was used as an indicator of birth spacing. Furthermore, the number of times a person had children with less than 24 months between births was defined as the number of short birth intervals (three categories: 0, 1, 2). Age at first birth event and age at last birth event were also included (each with four categories). Age at interview (four categories) was included in all analyses. Several risk factors for hip fracture have been identified (6–8, 10, 13), and these could potentially confound a parity-hip fracture association. A number of such possible confounders were included; all treated as categorical variables. The confounders, with number of categories in square brackets, were: marital status [2: ever/never married], social group [2], smoking status [3], BMI [4], alcohol consumption [4], and diabetes [2], and—for women—age at menopause [4], and hormone replacement therapy [2]. Unfortunately, information on lactation in women was not available. Number of teeth [4] was also included as a possible confounder. Due to missing values, data on 2045 individuals (85% of the participants in the survey), with median age in 1995 being 80 years (range: 75–98 years), were available for the present study. For a summary of sociodemographic and health AEP Vol. 12, No. 4 May 2002: 257–263 characteristics, and definitions of categories, see Table 1. Information on self-reported hip fractures was available for 2044 of the 2045 individuals. Age at first birth event and age at last birth event are only relevant for persons with children. Furthermore, there were here some missing values. The numbers of persons included in the analyses of these two variables were thus 1637 and 1162, respectively. As birth span measures are only relevant for persons with at least two birth events, and as there were here some additional missing values, the number of individuals in the analysis of birth span effect was 1209. In the following the TABLE 1A. Sociodemographic characteristics of the crosssectional study Males a Females %b N Total Age at interview, years 75–76 77–80 81–84 85–98 No. of birth events none 1–2 3–4 5 Age at first birth event, yearsc 24 24–26 27–30 31 Age at last birth event, yearsc 30 30–33 34–37 38 No. of short birth intervalsc none 1 2 Age at menopause, yearsc 45 45–49 50–51 52 Social classc Lower Upper Ever marriedc Yes No 830 40.6 1215 59.4 160 254 217 199 19.3 30.6 26.1 24.0 236 385 316 278 19.4 31.7 26.0 22.9 151 350 265 64 (650) 83 139 212 216 (637) 90 158 163 226 (474) 303 119 52 18.2 42.2 31.9 7.7 18.6 43.7 29.2 8.5 63.9 25.1 11.0 226 531 355 103 (987) 377 240 242 128 (975) 316 269 221 169 (735) 475 181 79 — — — — — — — — 221 349 332 313 18.2 28.7 27.3 25.8 689 141 83.0 17.0 971 244 79.9 20.1 753 77 90.7 9.3 1117 98 91.9 8.1 12.8 21.4 32.6 33.2 14.1 24.8 25.6 35.5 N a Characteristics a % b 38.2 24.3 24.5 13.0 32.4 27.6 22.7 17.3 64.6 24.6 10.7 For some variables, the number of persons is less than the total, due to irrelevance and/or missing values. In these cases, N is given in brackets. b For total, % refers to sex; for all other variables, % refers to within-sex distribution. c Variables regarded as possible confounders for the effect of number of birth events on hip fracture occurrence. Longitudinal Study of Aging Danish Twins 1995 survey will be referred to as the cross-sectional study. The Danish Twin Survey 1953–1963—the Prospective Survey These data form part of the Danish Twin Register, containing information on twins born between 1870 and 1910 (14). In the period 1953–1963 a questionnaire was sent out to twins born before 1911, including a question about the number of biological children (here four categories: 0, 1–2, 3–4, 5). The criterion for inclusion in the present study was that a twin should be alive on January 1, 1977, after which date the Danish National Register of Patients was in operation (see below). The twins were followed-up until the hospitalization date of the first registered hip fracture, emigration, death, or end of study period (December 31, 1994), whichever event came first. A total of 3057 persons with median age on January 1 1977 being 73 years (range: 66–99 years), was available for this analysis. For a summary of demographic characteristics, see Table 2. In the following sections, the Danish Twin Survey, 1953–1963 will be referred to as the prospective study. The Danish National Register of Patients—the Patient Register The Danish National Register of Patients, in the following referred to as the patient register, includes patient specific information on practically all non-psychiatric in-patients in Danish hospitals, with discharge from 1977 onwards. Our study was based on a linkage between the two twin surveys and the patient register, with information from January 1, 1977 to December 31, 1994, just before the cross-sectional survey. All Danes alive on April 1, 1968 or later have a unique personal identification number. These numbers were used in the linkage. From the patient register information on hip fractures was extracted (820.00–820.09 cf. ICD8 (15), January 1977–December 1993; S72.0–S72.2 cf. ICD10 (16), January 1994–December 1994). For orthopaedic diagnoses, in general the agreement between the content of the patient register and the medical records of the source hospitals is 83% when evaluated at the five digit level of the ICD-codes (17), and for hip fractures it could be expected to be even higher, as it is a usually well-defined condition. As more than one entry into the patient register with the same diagnosis can be related to the same hip fracture, fracture occurrence was regarded as a dichotomous variable (no fractures versus at least one fracture). The date of the first hospitalization with hip fracture was regarded as date of event in the prospective study. METHODS The analyses were performed separately for the cross-sectional and the prospective study. 259 Petersen et al. REPRODUCTION AND HIP FRACTURES AEP Vol. 12, No. 4 May 2002: 257–263 TABLE 1B. Health characteristics of the cross-sectional study Males Characteristics Hip fracture, patient register None One or more Body mass indexc 20 20–24 25–28 29 Alcohol consumption, drinks per weekc None 1–5 6–10 11 Smoking habitsc Never smoked Former smoker Present smoker Diabetes mellitusc No Yes Hormone replacement therapyc No Yes No. of teeth leftc None 1–9 10–19 20 N a Females % b Na %b 789 41 95.1 4.9 1121 94 92.3 7.7 46 404 275 105 5.5 48.7 33.1 12.7 206 599 287 123 17.0 49.3 23.6 10.1 126 427 162 115 15.2 51.4 19.5 13.9 305 723 135 52 25.1 59.5 11.1 4.3 136 411 283 16.4 49.5 34.1 677 246 292 55.7 20.2 24.0 768 62 92.5 7.5 1126 89 92.7 7.3 — — — — 1044 171 85.9 14.1 499 152 84 95 60.1 18.3 10.1 11.4 745 147 146 177 61.3 12.1 12.0 14.6 a Males: 830, females: 1215, for all variables. For all variables, % refers to within-sex distribution. c Variables regarded as possible confounders for the effect of number of birth events on hip fracture occurrence. b The Cross-Sectional Study The linkage between the cross-sectional survey and the patient register was used to investigate the agreement between self-reported fracture occurrence and fractures recorded in the patient register. The comparison was made for both sexes combined, using Kappa statistics, . A moderate  was expected, primarily because the patient register only lists fractures leading to hospitalization between 1977 and 1994. The association between parity and fractures was analysed by sex-wise logistic regression models, with the group having 1–2 children as reference group (2), with parity treated as a categorical variable, as an association between fractures and parity was not necessarily linear, and with age as a covariate. The odds ratios for parity groups in this model were compared to odds ratios for parity in models where the potential confounders (see Table 1) were added one at a time to the model with parity and age. If parity odds ratios changed 10% or more, the variable in question was regarded as a confounder (18). 260 Petersen et al. REPRODUCTION AND HIP FRACTURES AEP Vol. 12, No. 4 May 2002: 257–263 TABLE 2. Sociodemographic characteristics and hip fracture experience of the prospective study Males Characteristics Total Age at January 1 1977, years 66–68 69–72 73–77 78–99 No. of biological children None 1–2 3–4 5 Hip fractures None One or more N RESULTS The Cross-Sectional Study Females % a N % a 1349 44.1 1708 55.9 379 387 303 280 28.1 28.7 22.5 20.8 418 470 390 430 24.5 27.5 22.8 25.2 293 523 352 181 21.7 38.8 26.1 13.4 482 601 408 217 28.2 35.2 23.9 12.7 1267 82 93.9 6.1 1388 320 81.3 18.7 a For total % refers to sex, for all other variables, % refers to within-sex distribution. In order to investigate the possible effect of birth span, sex-wise logistic regressions with age, and mean birth span and number of short birth spans respectively, were performed. These logistic regressions were performed partly stratified by parity group in order to look for an effect of birth spacing per se, partly with parity as a covariate in order to see whether birth spacing was a confounder for parity. The effects of age at first birth and age at last birth were investigated separately by sex-wise logistic regression with age as a covariate, and these two variables were also analysed for a possible confounding of the effect of parity. The Prospective Study The prospective survey provided the basis for an incidence study. The statistical analyses were performed as sex-wise Cox regressions, with first hospitalization with hip fracture as event, and time from start of follow-up to first hospital admission date with hip fracture diagnosis as time to event. Individuals who had emigrated or had died before December 31, 1994 were regarded as censored at the date of emigration or death. Individuals surviving to December 31, 1994 without having had a fracture were regarded as censored at the end of follow-up. The group having 1–2 children was used as reference group (2). Age on January 1, 1977 was included as a covariate. No additional covariates were available for this study. All statistical analyses were made using Stata 6.0 (19). The fact that we used twin data could lead to an underestimation of standard errors of effect estimates if the outcome and exposure are highly correlated in the twins. However, the point estimates are not biased. In order to obtain reliable standard errors and thus confidence intervals, the option “cluster” in Stata was used (19), with twin pair identification number being the cluster variable. In the patient register data 135 hip fractures were recorded for the 2045 persons in this study. In the self-reported data 161 hip fractures in 2044 persons were recorded. The agreement between self-reported hip fractures and hip fractures registered in the patient register was substantial,   0.80. Because of the degree of agreement between the two types of hip fracture registration, and because the prospective survey did not contain self-reported information on fractures, only results based on fractures in the patient register are presented in the following sections. None of the logistic regressions including only persons with children showed any effect of number of birth events. The analysis of persons with at least two birth events showed no effect of birth span variables, neither birth span per se nor birth span as a confounder for the effect of parity. The analyses including age at first birth showed no effect of this variable, neither per se nor as a confounder of parity. However, age at last birth showed an effect for women, in that persons having an age at last birth between 30 and 33 years had a smaller risk of hip fracture than persons in the other age-intervals [age adjusted odds ratios with 95% confidence interval (CI), 29 years or younger: reference group, 30–33 years: 0.34 (0.16–0.74), 34–37 years: 0.66 (0.34–1.28), and 38 years: 0.86 (0.44–1.66)]. Furthermore, age at last birth event was a confounder of the effect of parity for both sexes, where the already non-significant differences between parity groups were further diminished. Parity was not a confounder of the effect of age at last birth event in women. The logistic regressions including all individuals showed no confounding for any of the variables evaluated for men, and only for BMI for women. Thus, specifically number of teeth was not a confounder of the parity–hip fracture association. The results of the logistic regression with hip fractures as outcome and with age and number of birth events, and for women also BMI, as covariates are shown in Table 3. For both sexes the association with number of births had a tendency towards a U-shape, and hypotheses of a linear effect were rejected at the 10% significance level in likelihood ratio tests. However, tests specifically aimed at evaluating the upturn to the right (20), showed no significance. In analyses comparing persons without children and persons with at least one child, odds ratios for the latter group were, for women, with BMI (the only confounder) as a covariate: 0.77 (95% CI: 0.47–1.27), and for men (no confounders): 0.41 (95% CI: 0.21–0.79). The Prospective Study A total of 402 fractures was registered for the 3057 persons, with a total risk-time of 29,112 years. In total 326 persons, of whom 39 had experienced hip fractures, figured in the cross-sectional study as well as in the prospective study. Petersen et al. REPRODUCTION AND HIP FRACTURES AEP Vol. 12, No. 4 May 2002: 257–263 TABLE 3. Number of persons from the cross-sectional study, number of hip fractures, odds ratios and their 95% confidence intervals No. of persons Women (N  1215) No. of birth events 0 1–2 3–4 5 Age at interview, years 75–76 77–80 81–84 85–98 Body mass index 20 20–24 25–28 29 Men (N  830) No. of birth events 0 1–2 3–4 5 Age at interview, years 75–76 77–80 81–84 85–98 No. of hip fractures Odds ratio 95% confidence interval 226 531 355 103 24 42 19 9 1.18 1.0a 0.69 1.22 0.39–1.23 0.56–2.65 236 385 316 278 14 14 24 42 1.0a 0.58 1.20 2.68 0.26–1.28 0.59–2.42 1.35–5.29 206 599 287 123 26 52 11 5 1.0a 0.74 0.28 0.35 0.45–1.22 0.14–0.59 0.13–0.94 151 350 265 64 14 14 9 4 2.44 1.0a 0.86 1.48 160 254 217 199 6 8 13 14 1.0a 0.77 1.41 1.81 0.69–2.02 1.13–5.29 0.36–2.03 0.48–4.55 0.25–2.31 0.53–3.77 0.69–4.76 261 TABLE 4. Person-years of follow up from the prospective study, number of hip fractures, Cox hazard ratios and their 95% confidence intervals Characteristics Women (N  1708) No. of biological children 0 1–2 3–4 5 Age at January 1 1977, years 66–68 69–72 73–77 78–99 Men (N  1349) No. of biological children 0 1–2 3–4 5 Age at January 1 1977, years 66–68 69–72 73–77 78–99 Personyears of folow up No. of hip fractures Cox hazard ratio 95% confidence interval 4553 6443 4124 2082 108 113 64 35 1.28 1.0a 0.88 0.96 0.98–1.68 0.65–1.19 0.66–1.42 5727 5366 3599 2510 66 78 89 87 1.0a 1.39 2.63 4.83 1.00–1.92 1.89–3.67 3.43–6.79 2342 4830 3120 1619 17 41 14 10 0.91 1.0a 0.54 0.71 0.30–0.98 0.36–1.41 4336 3620 2449 1505 15 25 17 25 1.0a 2.20 2.55 7.42 1.15–4.18 1.30–5.00 3.76–14.68 0.51–1.61 a Reference group. a Reference group. For women there was a decrease in incidence rate with increasing number of children, see Table 4. A hypothesis of linearity of the effect of number of children was accepted, and in a model with number of children as a continuous covariate the hazard ratio for number of children was 0.88 (95% CI: 0.78–0.99). However, the effect of number of children disappeared when excluding individuals without children. In a comparison of women without children and women with at least one child, the latter group had a hazard ratio of 0.74 (95% CI: 0.59–0.94). For men no clear effect of number of children could be shown in an analysis with number of children as a categorical covariate, see Table 4. In a model comparing men without children and men with at least one child, the hazard ratio of the latter group was 0.89 (95% CI: 0.51–1.52). DISCUSSION The purpose of the present study was to investigate the association between reproduction and hip fractures on the basis of a cross-sectional study and a prospective study. An important aspect of the present study is whether findings from studies on twins can be generalized to the general population. Previous studies on mortality (9) have shown that except for the first few years of life, twins are not different from the general population, and as childbearing and osteoporosis related hip fractures occur in adult life there is no reason to suspect that the present results can not be generalized. Furthermore, the possible impact of twin correlations has been taken into account in the statistical procedures employed. In the part of the cross-sectional study, delimited to persons with one or more birth events, no effect of number of birth events, birth spacing, or age at first birth was found. For men this was not unexpected, and for women this result implies that a possible calcium depletion during pregnancy and lactation is compensated well before reaching high age. The analysis of age at last birth showed an effect for women where having the last child either early or late was associated with a higher risk of hip fracture. Women having their last child in the early third decade of life might benefit from pregnancy induced high levels of estrogen at the time when peak bone mass is achieved (6), whereas women having the last child earlier in life do not have this benefit, and women having their last child later in life might experience ad- 262 Petersen et al. REPRODUCTION AND HIP FRACTURES ditional bone mass reduction due to pregnancy-induced calcium demands. Furthermore, for both sexes, age at last birth was a confounder for number of children. When taking age at last birth into account, the effect of parity diminished, corroborating the finding that among persons with children parity is not important for hip fracture risk. The U-shape of the association between hip fracture and number of children should probably not be given too much consideration as it was not even supported by a post-hoc test. However, as mentioned in the introduction, a recent study (5) has shown that women with six or more children have the same bone mineral density as nulliparous women. It would be interesting to have data on a larger group of persons with five or more children, in order to decide whether the non-significance of the upturn to the right is a real phenomenon or due to the small sample size of this parity group. The most important difference between groups with regard to hip fracture risk was between women with children and women without children. This finding is in agreement with the results from other studies (e.g., 4). The cause for this difference could be that having children, no matter how many, protects from hip fractures by way of maintaining a high level of estrogen, whereas having no children implies medical or physiological problems in general, problems that could be of biological or environmental origin, or both. Endocrinologically determined low estrogen levels, implying low fertility and inducing low bone mass density is a possible explanation. An interesting aspect of the results is the fact that the difference between persons with and without children showed up for men in the cross-sectional study and also to some extent in the prospective study. A possible contributor to this association is that hypogonadism is a risk factor for osteoporosis (10) as well as an indicator of reduced fertility. Thus, the finding of an association between hip fractures and having no children in both sexes may be influenced by variation in two sex-specific hormone-related phenomena. In the cross-sectional study, the odds ratio is lower for the age group 77–80 years compared to the age group 74–76 (Table 3). This is at variance with the general pattern of increasing hip fracture risk with age. A possible explanation for this pattern is that the patient register commenced in 1977. Some of the persons 77–80 years old at interview might therefore have been hospitalized with a hip fracture before 1977, and therefore be more precautious, whereas the persons 74–76 years old experienced their (first) hip fracture hospitalization in the time from 1977 onwards. The pattern for persons older than 80 years is dominated by the age effect and could also be affected by recurrence of hip fractures. The strength of the association between hip fractures AEP Vol. 12, No. 4 May 2002: 257–263 and having no children was different in the cross-sectional study and the prospective study. One possible explanation for this difference is that the persons in the two studies to some extent represent two different birth cohorts, the median birth year being 1904 for the prospective study and 1914 for the cross-sectional study. This difference leaves room for impact of changes in social and environmental factors affecting health in general, and thus also hip fracture risk, and reproductive patterns. Furthermore, the persons in the prospective study is on average 10 years older than the persons in the cross-sectional study at the time when the patient register commences. The most prominent finding of the present study is that having no children was associated with a higher risk of experiencing a hip fracture compared to having children. Therefore, no “cost-of-reproduction” in terms of hip fractures was observed. This result means that especially women should be reassured that having children does not lead to a higher risk of hip fractures later in life. Support for this study was provided by the Danish National Research Foundation, the National Institute on Aging (grant: NIA-PO1AG08761), the Danish Interdisciplinary Research Council. REFERENCES 1. Christensen K, Gaist D, Jeune B, Vaupel JW. A tooth per child? Lancet. 1998;352:204. 2. Sowers M. Pregnancy and lactation as risk factors for subsequent bone loss and osteoporosis. J Bone Miner Res. 1996;11:1052–1060. 3. Kolthoff N, Eiken P, Kristensen B, Nielsen SP. Bone mineral changes during pregnancy and lactation: A longitudinal cohort study. Clin Sci. 1998;94:405–412. 4. Hoffman S, Grisso JA, Kelsey JL, Gammon MD, O’Brien LA. Parity, lactation and hip fracture. Osteoporos Int. 1993;3:171–176. 5. Henderson PH, Sowers M, Kutzko KE, Jannausch, ML. Bone mineral density in grand multiparous women with extended lactation. Am J Obstet Gynecol. 2000;182:1371–1377. 6. Cooper C, Melton LJ. Epidemiology of osteoporosis. Trends Endocrinol Metabol. 1992;3:224–229. 7. Cumming RG, Nevitt MC, Cummings SR. Epidemiology of hip fractures. Epidemiol Rev. 1997;19:244–257. 8. Wark JD. Osteoporotic fractures: Background and prevention strategies. Maturitas. 1996;23:193–207. 9. Christensen K, Vaupel JW, Holm NV, Yashin AI. Mortality among twins after age 6: Fetal origins hypothesis versus twin method. BMJ. 1995;310:432–436. 10. Seeman E. The dilemma of osteoporosis in men. Am J Med. 1995; 98(Suppl 2A):76S–88S. 11. Christensen K, Holm NV, McGue M, Corder L, Vaupel JW. A Danish population-based twin study on general health in the elderly. J Aging Health. 1999;11:49–64. 12. Christensen K, McGue M, Yashin A, Iachine I, Holm NV, Vaupel JW. Genetic and environmental influences on functional abilities among Danish twins aged 75 years and older. J Gerontol A Biol Sci Med Sci. 2000;55(8):M446–M452. AEP Vol. 12, No. 4 May 2002: 257–263 13. Forsén L, Meyer HE, Midthjell K, Edna TH. Diabetes mellitus and the incidence of hip fracture: Results from the Nord-Trøndelag Health Survey. Diabetologia. 1999;42:920–925. 14. Kyvik KO, Christensen K, Skytthe A, Harvald B, Holm NV. The Danish twin register. Dan Med Bull. 1996;43:467–470. 15. Danish National Board of Health. Classification of Diseases. 8th revision. Copenhagen: Danish National Board of Health; 1986. 16. Danish National Board of Health. Classification of Diseases. 10th revision. Copenhagen: Danish National Board of Health; 1996. Petersen et al. REPRODUCTION AND HIP FRACTURES 263 17. Mosbech J, Jørgensen J, Madsen M, Rostgaard K, Thornberg K, Poulsen TD. Landspatientregistret. Evaluering af datakvaliteten. (Summary in English). Ugeskrift for Lager. 1995;157:3741–3745. 18. Rothman KJ, Greenland S. Modern Epidemiology. 2nd ed. Philadelphia: Lippincott-Raven Publishers; 1998. 19. Statacorp. Stata Statistical Software: Release 6.0. College Station, Texas: Stata Corporation; 1999. 20. Goetghebeur EJT, Pocock SJ. Detection and estimation of J-shaped risk-response relationships. J Roy Stat Soc. 1995;A-158:107–121.