Volume 14, Issue 9 pp. 1628-1633
Research Article
Free Access

A Cross-Sectional Study of Smoking and Bone Mineral Density in Premenopausal Parous Women: Effect of Body Mass Index, Breastfeeding, and Sports Participation

Graeme Jones

Corresponding Author

Graeme Jones

Menzies Centre for Population Health Research, Hobart, Tasmania, Australia

Address reprint requests to: Graeme Jones, M.D., Menzies Centre for Population Health Research, GPO Box 252-23, Hobart, Tasmania, 7001 AustraliaSearch for more papers by this author
Fiona Sara Scott

Fiona Sara Scott

Menzies Centre for Population Health Research, Hobart, Tasmania, Australia

Search for more papers by this author
First published: 02 December 2009
Citations: 43

Abstract

The objective of this cross-sectional study was to describe the relationship between cigarette smoking, effect modifiers, and bone density in premenopausal parous women. We studied a sample of 276 women (mean age 33 years) from Southern Tasmania. The study factors were cigarette smoking, body mass index (BMI), sports participation, and breastfeeding history. Bone mineral density was measured utilizing an Hologic QDR 2000 densitometer and converted to Z scores using the group mean and variance. There were 118 current smokers and 158 nonsmokers. Smokers had lower bone mass at all sites (femoral neck, −0.32 SD, 95% confidence interval [CI] −0.60 to −0.04; lumbar spine, −0.49 SD, 95% CI −0.76 to −0.22; total body, −0.40 SD, 95% CI −0.66 to −0.14). Stratifying by BMI revealed that this association was only present, but greater in magnitude, for those with a BMI <25 kg/m2. Smokers who had breastfed at least one child had an additional deficit in bone mass (femoral neck, −0.48 SD, 95% CI −0.89 to −0.07; lumbar spine, −0.39 SD, 95% CI −0.80 to 0.02; total body, −0.37 SD, 95% CI −0.77 to 0.06) while smokers who took part in competitive sport had significant increments in bone mass (femoral neck, 0.74 SD, 95% CI 0.31 to 1.17; lumbar spine, 0.48 SD, 95% CI 0.03 to 0.93; total body, 0.42 SD, 95% CI 0.00 to 0.84). Neither of these two associations were present in nonsmokers. In conclusion, current smoking was associated with substantial deficits in bone mass in our sample of women, particularly those with a BMI <25 kg/m2. In addition, smoking may prevent the usual postweaning recovery phase of bone after breastfeeding while sports participation may offset the negative effect of smoking on bone mass. These observations need to be confirmed in longitudinal studies but they imply that past studies of smoking in this age group may have missed important associations as they did not consider possible effect modifiers.

INTRODUCTION

FRACTURES DUE TO OSTEOPOROSIS are a major public health problem. Bone density is one of the major predictors of these osteoporotic fractures1 and is the result of the amount of bone gained in early life (i.e., peak bone mass) and subsequent bone loss.2 Cigarette smoking has been implicated as a risk factor for osteoporosis in both males3 and postmenopausal females where a recent meta-analysis concluded that smoking was responsible for one in eight hip fractures4 and that the mechanism is most likely through increased postmenopausal bone loss because there was no adverse effect premenopausally. However, while the effect in postmenopausal women has been extensively studied, there have been relatively few studies in premenopausal women,5-12 which have generally been of small sample size. While the overall point estimate from the meta-analysis indicated no deficit in bone mass, individual studies have been somewhat contradictory with some reporting a deficit5-9 and others not.10-12 This suggests that there may be heterogeneity between different populations with regard to the effect of smoking. Possible reasons for this may include an interaction with body size which has previously been reported in the elderly13-16 or modification of the smoking effect by other factors such as lactation or physical activity, but studies to date have not considered these potential modifiers of the smoking effect.

In this study therefore, we asked the following research questions: Is cigarette smoking associated with deficits in bone mass in a sample of premenopausal parous women from Southern Tasmania? Is the smoking association modified by body composition, lactation, and/or physical activity?

MATERIALS AND METHODS

In 1988, approximately 20% of children born in Tasmania were enrolled in a study of risk factors for Sudden Infant Death Syndrome. In southern Tasmania, there were 656 women (with a total of 696 children) enrolled in this study (95% of those eligible). In 1996, these women and their children were followed up for a study into determinants of bone density in children. In brief, children were selected on the basis of a risk score comprised of their birth weight, season of birth, gender, maternal age, and maternal intention to breastfeed.17 The result of these selection criteria was a sample consisting of younger mothers who were less likely to breastfeed and more likely to smoke cigarettes and have male children as compared with Tasmanian women as a whole. The current study relates to factors associated with bone mass in the mothers only. This study was granted ethics approval by the University of Tasmania Ethics Committee (Human experimentation).

Pre-existing data of relevance to the current study was available on maternal weight, smoking during pregnancy in 1988 (yes/no and dose by trimester), breastfeeding habit for that child (intention and actual up to 3 months), and diet during the third trimester of pregnancy (by food frequency questionnaire [FFQ]).

Bone mass was assessed utilizing the technique of dual-energy X-ray absorptiometry at the femoral neck, lumbar spine (L1–L4), and total body. The machine utilized was a QDR 2000 densitometer on fan beam setting (Hologic, Waltham, MA, U.S.A.). Bone mineral density (BMD) only is reported. Precision estimates in vivo are not available in our subjects. The longitudinal coefficient of variation for our machine during 1996 using daily measurements of a spine phantom was 0.54%. Weight was measured by electronic scales (calibrated at the beginning of the study by the manufacturer). Height was measured by a stadiometer. Current smoking was assessed by questionnaire completed by the mother (yes/no and number of cigarettes per day). Current smoking was defined as smoking one or more cigarettes per day. The above measures were assessed at the time of bone density measurement from February to December 1996.

Subjects were then resurveyed in October 1997 to assess number of children, lactation history (number of children breastfed and duration for each child, time since last breastfeeding episode, and return of menses after childbirth), contraceptive usage, calcium intake (using FFQ)18 and physical activity using a validated questionnaire modified for Tasmanian sports which includes questions on duration, frequency, and intensity of exercise as well as television watching and recreational and competitive sport.19 Because bone density information was provided at the time of measurement of BMD, we also asked the subjects to report, in the questionnaire, whether they had changed their physical activity.

Statistics

To facilitate comparisons of effect sizes, BMD was converted to a standardized score by subtracting the group mean from each individual measure and dividing by the standard deviation for that measure, creating a standard normal distribution or Z score for each BMD site. A difference in this score of one is equivalent to a 1 SD difference in BMD. Unpaired t-tests were utilized for comparison of means (where appropriate) in both the whole sample and strata. Linear modeling was utilized for multivariate analysis using an iterative approach. Standardized BMD was regarded as the dependent variable. The association with smoking was then adjusted for age, dietary calcium intake, ever breastfeeding status, competitive sports participation, and self-reported change in physical activity by their addition to the model. To examine if the smoking association was mediated by body weight or body fat, these variables were then added separately to a BMD model containing the smoking term only. Effect measure modification was assessed by stratified analysis and by constructing a model containing the two main effect factors and the product of these two factors. A p value of <0.05 (two-tailed) or a 95% confidence interval (CI) not including the null point was regarded as statistically significant. The original study was planned on this level of significance in the whole sample so a correction for multiple comparisons was not performed. All statistical calculations were carried out on SPSS version 6.1 for Windows (Chicago, IL, U.S.A.).

RESULTS

There were 278 subjects who had bone density measured (representing 90% of the mothers of the children after allowing for 22 multiple births) or 42% of those in the original study. A total of 263 of these women (95%) returned completed questionnaires in 1997. There were 118 current smokers and 158 nonsmokers (with missing data on the smoking status of two subjects). In the 8 years between interviews, 13 women had commenced and 27 had ceased smoking. Demographic and study factors of interest are presented in Table 1. Smokers and nonsmokers were of similar height. Smokers were younger and weighed substantially less than nonsmokers. Most of this was due to a reduction in fat mass rather than lean mass. Smokers were less likely to have breastfed any children but had similar levels of calcium intake, competitive sports participation, as well as light and strenuous physical activity.

Table Table 1.. Demographic and Study Factors by Smoking Status
image

Current smoking was associated with a 4–5% decrease in bone mass at all sites (Table 1). The deficit was most marked in those with a body mass index (BMI) <25 kg/m2, while in those with a BMI >25 kg/m2 there was no evidence of a detrimental effect of smoking (Fig. 1). Similarly, there was evidence of a dose response in those with a BMI of <25 kg/m2, with a statistically significant negative linear relationship at all three sites (Fig. 2). There was no evidence of a dose response in overweight or obese women using the same categories (femoral neck, p = 0.62 for trend, lumbar spine, p = 0.78 for trend, and total body, p = 0.92 for trend). In multivariate analysis, in a model including current smoking and BMI, the interaction term between smoking and BMI was significant at the lumbar spine (p = 0.04) and total body (p = 0.01) with a consistent trend at the femoral neck (p = 0.06) indicating that the negative association with smoking became greater in magnitude as BMI became lower.

Details are in the caption following the image

Smoking and bone mass in premenopausal women: interaction with BMI. In women with a BMI below 25 kg/m2, current smoking is associated with deficits in bone mass at all sites as compared with nonsmokers. In contrast, current smoking is not associated with deficits in bone mass at any site in overweight or obese women (BMI ≥25 kg/m2). Results are presented as Z scores ± 95% CI.

Details are in the caption following the image

Dose of smoking and bone mass in normal or underweight women. There is evidence of a dose response at all sites with statistically significant decrements in bone mass with increasing daily cigarette consumption. Results are presented as mean ± SE.

After adjustment for ever breastfeeding, sports participation, calcium intake, age, and self-reported change in physical activity in multivariate analysis, current smoking was more strongly associated with bone mass in the whole sample (Table 2). For women with a BMI <25 kg/m2, the relevant figures were femoral neck (−0.46 SD, 95% CI −0.80 to −0.12), lumbar spine (−0.59 SD, 95% CI −0.94 to −0.24) and total body (−0.51 SD, 95% CI −0.82 to −0.20). These reduced by a similar magnitude after adjustment for body weight (femoral neck −0.31 SD, 95% CI −0.63 to 0.00; lumbar spine −0.40 SD, 95% CI −0.72 to −0.08; total body −0.37 SD, 95% CI −0.66 to −0.08) or body fat (femoral neck −0.28 SD, 95% CI −0.60 to 0.04; lumbar spine −0.40 SD, 95% CI −0.74 to −0.06; total body −0.35 SD, 95% CI −0.65 to −0.05) but remained statistically significant at the lumbar spine and total body. In women with a BMI >25 kg/m2, smoking was not significantly associated with bone mass at any site in multivariate analysis (data not shown).

Table Table 2.. Multivariate Regression Coefficients of Smoking, Breastfeeding, and Sports Participation on Bone Density in Premenopausal Women
image

As compared with current smokers, women who ceased smoking between 1988 and 1996 (n = 27) had higher BMI (27.6 vs. 25.1 kg/m2, p = 0.047), body weight (72.5 vs. 65.3 kg, p = 0.056), and spinal BMD (0.52 SD, p = 0.021) with consistent trends at the femoral neck (0.17 SD, p = 0.33) and total body (0.31 SD, p = 0.09). If smoking had a cumulative effect premenopausally one might also expect to see an age-related gradient among smokers reflecting the duration of use. However, this was not the case at any site (femoral neck β = −0.001 g/cm2/year, p = 0.68; lumbar spine β = −0.001 g/cm2/year, p = 0.77, total body β = 0.002 g/cm2/year, p = 0.46).

In univariate analysis in the whole sample, ever breastfeeding was associated with small nonstatistically significant deficits in bone mass at all sites (femoral neck −0.23 SD, 95% CI −0.51 to 0.06; lumbar spine −0.22 SD, 95% CI −0.49 to 0.06; total body −0.11 SD, 95% CI −0.38 to 0.16), which remained statistically nonsignificant in multivariate analysis (Table 2). There was no relationship with time since breastfeeding or total duration of breastfeeding (data not shown). Stratifying by smoking status revealed that nonsmokers who had breastfed had small nonsignificant reductions in bone mass, while smokers had larger significant reductions in bone mass at one site with borderline results at the other two (Table 3). In multivariate modeling, after including ever breastfeeding and current smoking, the interaction term for smoking and ever breastfeeding was not statistically significant at any site (femoral neck, −0.32 SD, p = 0.31; lumbar spine −0.16 SD, p = 0.63; total body, −0.27 SD, p = 0.29).

Table Table 3.. Effect of Breastfeeding and Sports Participation Stratified by Smoking Status
image

Competitive sports participation during the previous 12 months was associated with higher bone mass at all sites in multivariate analysis (Table 2). Again, there were differences between smokers and nonsmokers, with the magnitude of benefit being apparently greater in smokers (Table 3). In multivariate modeling, after including sports participation and current smoking, the interaction term for smoking and sports participation was significant at the femoral neck but not at the other sites (femoral neck, 0.59 SD, p = 0.03; lumbar spine 0.42 SD, p = 0.13; total body, 0.25 SD, p = 0.35).

Age was not associated with bone mass at any site (femoral neck β = −0.0025 g/cm2/year, p = 0.10; lumbar spine β = 0.0015 g/cm2/year, p = 0.34; total body β = 0.0003 g/cm2/year, p = 0.86). Current calcium intake assessed 12 months after the measurement of bone density was not associated with bone mass at any site (femoral neck r = −0.03, p = 0.69; lumbar spine r = −0.09, p = 0.18; total body r = −0.12, p = 0.07).

DISCUSSION

This study has shown that current smoking in our sample of young, predominantly Caucasian, premenopausal women is associated with substantial deficits in bone mass, particularly in those with a BMI <25 kg/m2. Furthermore, the combination of smoking and breastfeeding was associated with an additional deficit in bone mass, while, conversely, competitive sports participation may offset the deficit from smoking. The magnitude of the deficits with smoking and breastfeeding and/or lack of competitive sports participation in smokers are substantial and would be expected to increase fracture risk in later life by bringing forward the age at which fractures will occur with minimal trauma.

Smoking has been estimated to account for one in eight hip fractures in women, making it a major preventable cause of osteoporosis.4 Recent evidence has suggested that this is due to an increased rate of bone loss after the menopause, resulting in a cumulative decrease over time of 0.45 SD in comparison with nonsmokers.4 The effect on premenopausal women remains controversial with individual studies suggesting anywhere between a 0.4 SD deficit to a 0.4 SD increase in smokers. Our study found a 0.5–0.6 SD reduction in bone mass at all sites that was only evident in women with a BMI <25 kg/m2 who made up approximately half of our sample. No effect was evident in overweight or obese women, suggesting that the original observation by Daniell of osteoporosis in the slender smoker may also apply premenopausally, although he did not actually observe this in his small sample of premenopausal women using a less accurate measure of bone density.13 Since then there have been other reports confirming this interaction with BMI in postmenopausal women but none in premenopausal women that we are aware. Why might smoking be detrimental only in thin women? It may be mediated by alterations in body composition. It is possible that body fat may protect against the negative effect of smoking by two mechanisms. First, the extra mechanical load on bones may negate the smoking effect. This appears to be a continuous effect rather than a threshold effect since modeling for a smoking BMI interaction was also significant. This also indicates that the slope of the relationship between BMI and BMD is greater in smokers compared with nonsmokers, suggesting that body fat may be a more important determinant of BMD in smokers and that reductions in body fat may be associated with greater deficits in bone mass. Second, body fat may protect against bone loss by extra-ovarian synthesis of estrogen. The former hypothesis appears more attractive, given the current controversy relating to smoking and estrogen status.20-23 However, even after adjustment for body fat or weight, smoking still had a smaller but significant effect on bone mass which may be mediated by other factors such as a direct toxic effect on bone24 or other as yet undescribed mechanisms.

Smoking may also be associated with deficits in lifestyle such as diet and physical activity. There was, however, little difference between smokers and nonsmokers in these exposures in our sample and, furthermore, adjustment for these factors did not substantially alter the associations between smoking and bone mass in multivariate analysis, suggesting that they were not acting as confounders in this case. Furthermore, the observation that smoking interacts with BMI does not support this hypothesis because one would expect to see greater deficits in overweight people who would be expected to have poorer diets and lower levels of physical activity. Similarly, the evidence for a dose response does not support this hypothesis.

It is appropriate to discuss why smoking appears to have varying effects in different populations of premenopausal women. Our data would suggest that it may be due to the modification of the smoking effect by BMI which has rarely been considered. Part of the variation may be due to the small sample sizes studied to date or the fact that we studied parous women only where the demands of pregnancy and breastfeeding may place greater demands on the skeleton. Yet another reason may be variations in the timing of uptake of smoking in different populations. A limitation of this study is the lack of direct data on time of uptake, which, for females, is on average 15 years in Tasmania at present (L. Blizzard, personal communication). However, with some limitations, age would be a reasonable surrogate for duration of use. If smoking had a cumulative effect, as in postmenopausal women, one would expect to observe an age-related decrement in bone mass with age, which is not what we observed. Furthermore, the small number of women who had given up smoking between 1988 and 1996 had generally higher weight and BMD, which might indicate a reversible effect of smoking premenopausally.

Breastfeeding appears to lead to transient decreases in bone mass which recovers, in most cases, within 12–18 months of cessation.25-27 Longer term effects are uncertain, with some reporting a decrease, others no change, and others an increase.28-31 While smoking is often adjusted for as a confounder, we are aware of no studies that examine if smoking acts as an effect modifier of breastfeeding as we have observed in the current study, suggesting that the effect of breastfeeding did not reverse with time as over 80% of these women had not breastfed within the 2 years prior to the measurement of their bone mass. While we do not have direct data on smoking habits while breastfeeding, it would appear most likely, in a biological sense, that this may be when smoking is most harmful either through diet at this time, body composition, or, possibly, as an exaggeration of the normal low estrogen state at this time.27 If confirmed in longitudinal studies in breastfeeding smokers, this effect has important implications for breastfeeding in this group who will not be able to be reassured about the lack of a long-term detrimental effect. Similarly, but somewhat surprisingly, we observed that competitive sports participation also appeared to be modified by smoking status with greater increments in bone mass in smokers, while smokers who took part in sports had similar bone mass to nonsmoking subjects who did not take part in sports. Furthermore, the breastfeeding and sports associations were largely independent of each other. There was some inconsistency between the stratified and multivariate analysis as to the presence of effect measure modification. However, the coefficients for the interaction terms for both breastfeeding and sports participation would be regarded as clinically significant.

This study has a number of limitations. It is cross-sectional in design and while it can demonstrate association it cannot establish cause and effect. Confirmation of these observations will require longitudinal studies in a cohort of early postpartum women who both smoke and plan to breastfeed their child with objective assessment of physical activity. In addition, the women in this study are not representative of Southern Tasmanian women as outlined in the Materials and Methods. Interestingly, as we have previously reported, subjects who were lost to follow up or did not participate were under represented in terms of breastfeeding and over represented in terms of smoking and younger mothers,32 suggesting possible lower bone mass in nonparticipants and a possible shift toward a more representative final sample given the biases introduced by the initial risk score. Indeed, the bone mass distribution in our subjects is very similar to the NHANES population currently utilized in the Hologic reference database. Furthermore, the net effect of these biases has been to increase the efficiency of the study since the prevalence of smoking was 42%. While this selection bias may influence inferences about prevalence in a population, this is less likely but still possible in an etiologic study. Miettinen states that for exposure outcome associations to be generalizable to other populations, three key criteria need to be met regarding definition of eligibility, sample size, and adequate distribution of study factors,33 all of which are met by this study.

In conclusion, current smoking is associated with substantial deficits in bone mass in our sample of premenopausal parous women, particularly in those women with a BMI <25 kg/m2. Smoking may prevent the usual postweaning recovery phase of bone after breastfeeding, while sports participation may offset the negative effect of smoking on bone mass. Last, these results imply that past studies of smoking in this age group may have missed important associations as they did not consider possible effect modifiers of the smoking effect.

Acknowledgements

Special thanks to the Denise Kaye and the staff of the Medical Imaging Department, Royal Hobart Hospital and Kate Evans. This work was supported by the Royal Hobart Hospital Acute Care Program, the Tasmanian Dairy Industry Authority, and Blundstone Pty. Ltd.

      The full text of this article hosted at iucr.org is unavailable due to technical difficulties.