Presented in part at the 23rd Annual Meeting of the American Society for Bone and Mineral Research in Phoenix, AZ, October 12-16, 2001.
Whether nulliparity increases fracture risk is unclear from prior studies, which are limited by small samples or lack of measured bone mineral density. No study has evaluated whether the effect of parity differs by skeletal site. We prospectively analyzed the relationship of parity to the risk of incident nontraumatic hip, spine, and wrist fractures in 9704 women aged 65 years or older participating in the Study of Osteoporotic Fractures to determine if parity reduces postmenopausal fracture risk, and if so, if this risk reduction is (1) greater at weight-bearing skeletal sites and (2) independent of bone mineral density. Parity was ascertained by self-report. Incident hip and wrist fractures were determined by physician adjudication of radiology reports (mean follow-up, 9.8 years) and spine fractures by morphometric criteria on serial radiographs. The relationship of parity to hip and wrist fracture was assessed by proportional hazards models. Spine fracture risk was evaluated by logistic regression. Compared with parous women, nulliparous women (n = 1835, 19%) had an increased risk of hip and spine, but not wrist, fractures. In multivariate models, parity remained a significant predictor only for hip fracture. Nulliparous women had a 44% increased risk of hip fractures independent of hip bone mineral density (hazards ratio, 1.44; 95% CI, 1.17-1.78). Among parous women, each additional birth reduced hip fracture risk by 9% (p = 0.03). Additionally, there were no differences in mean total hip, spine, or radial bone mineral density values between nulliparous and parous women after multivariate adjustment. In conclusion, childbearing reduces hip fracture risk by means that may be independent of hip bone mineral density.
Studies conflict about the association of nulliparity with an increased risk of fractures.(1–12) Additionally some,(12–14) but not all,(15) studies evaluating reproductive correlates of bone mineral density (BMD) have found that nulliparity is associated with decreased BMD, which could account for differences in fracture risk as low BMD is a strong predictor of increased fracture risk.(16) To our knowledge, no study has evaluated whether the effect of parity on fracture risk differs by skeletal site.
In the Study of Osteoporotic Fractures (SOF), our prospective cohort study of 9704 postmenopausal women over age 65, we assessed the longitudinal relationship of parity and incident hip, spine, and distal forearm fractures to determine whether nulliparity is associated with an increased risk of nontraumatic fractures at these skeletal sites and whether this relationship is independent of site-specific BMD.
MATERIALS AND METHODS
The study population consisted of 9704 postmenopausal community-dwelling women 65 years of age or older enrolled in the SOF.(16) Participants were recruited between 1986 and 1988 from population-based listings in four areas of the United States: Baltimore County, Maryland; Minneapolis, Minnesota; Portland, Oregon; and the Mononghela Valley near Pittsburgh, Pennsylvania. Black women were excluded because of their low incidence of osteoporotic fractures. Also excluded were women who were unable to walk without assistance and those with bilateral hip replacements. All women provided written informed consent, and the study was approved by the Institutional Review Boards at each site. Parity status and number of live births were ascertained by self-report at the baseline exam in 9699 women, and these women represent the final sample for the current study.
Measurement of bone mass
Distal radial BMD (g/cm2) was measured by single-photon absorptiometry (OsteoAnalyzer; Siemens-Osteon, Wahiawa, HI, USA), and anteroposterior (AP) spine and total hip BMD (g/cm2) were measured by DXA (QDR 1000; Hologic, Inc., Waltham, MA, USA). Details of measurement and quality control procedures for this cohort are published.(17)
Ascertainment of incident fractures
Details of our method for identifying new fractures during follow-up are published.(18) Briefly, for hip and non-spine fractures, we contacted participants every 4 months by postcard or telephone to ask whether they had sustained a fall or fracture. More than 98% of these follow-up contacts were completed. If a fracture occurred, participants were interviewed to determine how the fracture occurred, and a fracture was confirmed by physician review of a radiology report that had to specifically mention the occurrence of an acute fracture. We excluded fractures that occurred from major trauma such as motor vehicle accidents.
Although spine compression fractures are the most common osteoporotic fracture, only one-third are clinically recognized.(19) Therefore, we defined incident vertebral fractures as a 20% reduction in height of any vertebral body and a >4-mm change between a baseline lateral spine radiograph and a subsequent radiograph (mean follow-up, 3.7 years).(20)
The women were examined in clinical research centers. In addition to self-reported number of live births, we ascertained participants' age at menarche, age at their last menstrual period, height and nonpregnant weight at the age of 25, change in weight since age 25, calcium intake as milk from ages 12 to 18 and 18 to 50, physical activity at the ages of 25 and 50, parental history of fractures, smoking status, exercise,(21) number of hours spent sitting and lying down daily, and self-rated health. We asked about physician-diagnosed fractures since age 50, hyperthyroidism, diabetes and age of onset, hysterectomy, and oophorectomy. Current medication use was verified by review of participants' pill bottles at the clinic visit. Current dietary calcium intake was assessed by a food-frequency questionnaire,(22) and caffeine intake was estimated.(23) Current calcium intake included a total of both dietary and supplement intake obtained from medication review.
We measured weight, height (by stadiometer), and resting pulse rate. Tests of neuromuscular function included whether the participant could rise up from a chair (without using her arms) five consecutive times. Tests of visual function included corrected visual acuity, contrast sensitivity,(24) and depth perception using the Howard-Dohlman device.(25)
We used Student's t-test and χ2 analyses for comparisons of mean and proportional differences, respectively, between the nulliparous and parous groups. We used Cox proportional hazards models for hip, wrist, and non-spine fractures, and logistic regression for spine analyses because date of fracture was unknown. Non-spine fractures are defined as any fracture (including hip and wrist) except for vertebral. To determine the final multivariate model for parity and fracture, we tested variables known to predict fracture,(16) as well as other variables and potential interactions we thought could confound the relationship of parity and fracture. Interaction terms tested included the interaction of each significant covariate with parity. For the multivariate models including BMD as a covariate, we also tested for potential interactions of each of the other significant factors and BMD. All covariates included in the final model were obtained at the baseline exam (1986-1987), with the exception of DXA BMD, which was obtained at visit 2 (1989-1990) when DXA was available.
We first evaluated the relationship of parity to fracture risk adjusting only for age and weight. Then, to identify independent predictors (p < 0.05) of fracture for each of the four fracture sites, we performed manual backward stepwise Cox regression with evaluation of model fit and covariate significance at each step before eliminating a variable. We required age and weight to remain in every model. To assess the influence of BMD on the parity and fracture relationship, we limited all of the multivariate models to participants who had a measured BMD. Because site-specific BMD is the best predictor of site-specific fracture risk,(26–27) the multivariate models for spine and wrist fracture used spine and distal radial BMD, respectively, instead of total hip BMD. To determine if BMD modified the parity-fracture relationship, we performed subsequent analyses that excluded the BMD measurement and its interactions from the final multivariate models.
All statistical analyses were conducted using the SAS Statistical Analysis System, version 6.12 (SAS Institute, Cary, NC, USA). Adjusted hazards and odds ratios were obtained from the Cox proportional hazards and logistic regression analyses, respectively. To assess the relationship of increasing parity on hip fracture risk, we calculated the probability of hip fracture with a proportional hazards model for five levels of parity. Confidence intervals for the odds ratios are based on the asymptotic normality of the parameter estimates.(28) All the statistical tests that we report are two-sided, and the term statistically significant implies a p value less than 0.05.
Nulliparous women (n = 1835, 19%) were slightly older, leaner, gained less weight since age 25, had lower daily calcium intake, and had less estrogen use than parous women (Table 1). Nulliparous women also did less weight-bearing activity (assessed by on feet ≤4 h/day), had more proximal muscle weakness (assessed by using arms to stand from a chair), and had slightly poorer vision (assessed by low-frequency contrast sensitivity and distant depth perception). There was no difference by parity in history of maternal hip fracture (Table 1).
Table Table 1. Baseline Characteristics
Almost one-third of the women had an incident nonvertebral fracture after the baseline exam (mean follow-up, 9.8 years; Table 2). Nulliparous women had a much higher age-adjusted incident rate of hip fracture (1447 vs. 1144 per 100,000 person-years). Similarly, nulliparous women had an increased rate of incident spine fractures (mean follow-up between radiographs, 3.7 years).
Table Table 2. Incident Nontraumatic Fractures Since Study Baseline in Nulliparous (n = 1835) vs. Parous (n = 7864) Women
After adjustment for age and weight, nulliparous women had a 27% increase risk of spine and 37% increased risk of hip fractures compared with parous women. After further adjustment for other potential confounders, nulliparity increased the risk of hip fracture by 44%, and the increased risk of spine fracture was no longer statistically significant (Table 3). The relationship of nulliparity and a 44% increased risk of hip fracture remained significant with BMD in the model (Table 3).
Table Table 3. Risk of Nontraumatic Fracture for Nulliparous (n = 1835) vs. Parous (n = 7864) Women
The final significant covariates and interactions for each of the multivariate models are noted in Table 3. The inclusion of site-specific BMD in each of these models did not alter the significance of the parity and fracture relationship, although BMD was a significant covariate related to fracture risk in each of the models (Table 3). Thus, the relationship of nulliparity and increased fracture risk remained significant independent of BMD (Table 3).
We built the final stepwise models from the age-weight adjusted models, and thus, both age and weight remained in the final multivariate models in addition to other significant covariates. However, age was significantly associated with parity and fracture risk (p < 0.05) only in the hip and nonvertebral multivariate models. Moreover, we were surprised that weight was not significantly associated (p < 0.05) with parity and fracture risk in the multivariate models for the hip, spine, or wrist. Furthermore, weight change since age 25, noted previously in SOF to be a predictor of fracture,(16) was not significantly associated with the parity and fracture relationship. We also evaluated a model with body mass index (BMI) instead of body weight, and this did not change our results.
Other variables tested that were not significantly associated with parity and fracture risk either on univariate or multivariate adjustment include previous hyperthyroidism, current use of long-acting benzodiazepine, current use of anticonvulsant drugs, walks for exercise, weight change since age 25, calcium intake from milk at ages 12-18 or 18-50, prior history of oral estrogen use, active in sports at age 30, age at menarche, years of menstruation, level of education, and a history of lactation.
To better understand the interrelationships of parity and BMD with fracture, we also compared BMD measurements for hip, spine, and radius by parity status and found no differences in BMD between the groups after multivariate adjustment (Table 4). Furthermore, there was no significant interaction with parity and BMD in any of the multivariate fracture models. Thus, BMD does not seem to modify the effect of nulliparity on increased hip fracture risk that we found.
Table Table 4. BMD Values in Nulliparous (n = 1835) vs. Parous (n = 7864) Women
To further explore the relationship of parity and reduced hip fracture risk, we stratified parity status into five groups: nulliparous, one child, two children, three children, and four or more children. The probability of hip fracture decreased as parity increased across these five groups, which were approximate quintile groups for parity status (Fig. 1; hazards ratio [HR], 0.87; 95% CI, 0.81-0.94; p = 0.0002). To determine if this relationship could be primarily affected by nulliparous women (because the reason for nulliparity was unknown and in some cases could be caused by infertility), we restricted this analysis to parous women only. Among parous women, the risk of hip fracture decreased by 9% with each additional birth (HR, 0.91; 95% CI, 0.84-0.99; p = 0.03).
We found that childbearing significantly decreases the risk of postmenopausal hip fractures. Compared with parous women, nulliparous women had a 44% increased risk of hip fracture independent of BMD. Furthermore, there was a “dose-response” relationship with increasing parity to further reduce hip fracture risk. Nulliparous women also had a 27% increased risk of spine fracture, but this relationship was of borderline significance in the multivariate model. There were no differences in wrist or non-spine fractures by nulliparity.
Most,(1–11, 13) but not all,(12) prior studies focusing on the relationship of parity and fracture risk have been limited by small sample size or lack of BMD measurements before incident fracture. The Dubbo Osteoporosis Study of 1091 postmenopausal women found a higher risk of all combined atraumatic fractures in nulliparous women independent of BMD.(12) However, this study did not assess the association of parity and fracture risk by fracture type. Our results suggest that there is something uniquely different about the effect of parity to protect from future fracture at weight-bearing sites because parity had the greatest effect in reducing hip fracture.
What are potential differences between nulliparous and parous women that might explain the marked difference in hip fracture risk we observed? We expected to find fracture risk differences because of differences in BMD for several reasons. Estrogen levels rise 20-30 times higher during pregnancy above their peak in the menstrual cycle.(29, 30) The additional estrogen exposure with each additional pregnancy might reduce fracture risk. Furthermore, some nulliparous women could be subfertile with reduced estrogen production during the menstrual cycle and thus might be at risk for fracture. The long-term effect on BMD from the high maternal bone turnover that occurs during pregnancy and lactation would be expected, if anything, to result in lower BMD among parous women.(31–34) Pregnancy-associated weight gain may also result in higher BMD later in life.
Some,(12–14) but not all,(15) previous studies evaluating correlates of BMD have found a cross-sectional relationship with parity and increased BMD. All of these positive studies were on a smaller sample of women, including the study by Fox et al.,(13) a subsample of 2230 women participating in the SOF Baltimore site. Similar to the earlier work by Fox et al.,(13) which assessed only radial BMD, we found that parity was significantly associated with age- and weight-adjusted distal radial BMD, but this was the only skeletal site that was significant before our multivariate adjustment (Table 4). In our current study of 9699 women, we found no association between parity and total hip, spine, or distal radial BMD after adjustment for potential confounders (Table 4).
Furthermore, our findings that childbearing reduces hip fracture risk were independent of BMD, so other factors must contribute. During pregnancy and childbirth, many changes occur to hip and pelvic alignment that could permanently alter hip structure and protect from future fracture risk. Other factors could also be associated with pregnancy and/or delivery of a live-birth to uniquely reduce hip fracture risk. The effect of parity to reduce hip fracture risk does not seem to be related to the anatomic location at which the hip fracture occurs, because we found no difference in the proportion of women with femoral neck versus intertochanteric hip fractures by parity status (data not shown). Whatever the etiology, our finding of parity to reduce hip fracture risk is strengthened by the relationship we saw of increasing parity to further reduce hip fracture risk (Fig. 1). Furthermore, when we excluded nulliparous women to determine if something differed in this group (such as subfertility for other reasons), this relationship of decreasing fracture risk with increasing number of children persisted.
Our large prospective cohort study has many strengths: the robust size of almost 10,000 postmenopausal women with excellent fracture outcome ascertainment that allows us to assess differences in parity status and fracture risk at different skeletal sites; assessment of multiple potential confounders, including measurements of site-specific BMD; long duration of follow-up; and physician-adjudicated fractures.
Our study also has limitations. The mean duration of follow-up between vertebral X-rays was 3.7 years, and a longer duration of follow-up might have possibly increased our power to see a significant relationship with spine fracture risk. However, the magnitude of the association is still clearly highest for hip fracture, and it would be unlikely for the point estimates to markedly change with longer follow-up. Because the age distribution for wrist fractures in women is younger than for hip fractures, it is also possible that we underestimated a potential effect of nulliparity to increase wrist fracture risk in our older cohort that was recruited after age 65 if a high proportion of wrist fractures occurred before enrollment. To assess for this potential bias, we did additional multivariate analyses for wrist fracture that included women with self-reported wrist fracture after age 50, but before the baseline SOF exam, and still found no difference in risk of wrist fracture by parity status (data not shown). Also, as our cohort is primarily white women, results may not be generalizable to other ethnic groups. Finally, we are unable to assess if nulliparity was because of choice or inability to conceive in some women, and the nulliparity effect may be greater in those women with infertility. However, the prevalence of nulliparity in our population (19%) is similar to that of all U.S. white women of this age group.(35, 36)
What are the clinical implications of our findings? Our findings show that nulliparous women are at increased risk of hip fracture independent of BMD and that, among parous women, each additional birth reduces the risk of future hip fracture. Clearly women will not alter childbearing choices based on these results. However, while considering prevention and treatment of osteoporosis for an individual patient, it may be important to identify nulliparous women. Our results suggest this identification by clinical history may be particularly important, because BMD may not identify this increased risk for hip fracture in nulliparous women.
Finally, our findings are in a cohort of women born in 1889-1923, and the mean number of total live births for the entire cohort is 2.24 children. The total fertility for U.S. white women was an average of 2.2-2.6 children per woman during World War II, increased to 3.2-3.6 children in the 1950s, and then steadily decreased to less than 2 children (1.8) by 1980.(37, 38) As the number of children women bear continues to decrease in the United States(39) with the increasing presence of women in the workforce, we will see an increasing impact of reduced parity on hip fracture risk over the next several decades.
In summary, we found that nulliparous women have an increased risk of postmenopausal hip fracture independent of BMD. Furthermore, the risk of postmenopausal hip fracture decreased with increasing number of children. More research is needed to determine why the effect of parity to reduce fracture risk independent of BMD is greatest at weight-bearing sites.
We wish to thank the following contributors: Jeanne Reinhardt for excellent technical assistance and Martie Sucec for editorial review. This research was supported by the National Institute of Arthritis and Musculoskeletal and Skin Diseases and the National Institute on Aging (Public Health Service Grants AR35583, AG05407, AR35582, AG05394, and AR35584).