Osteoporosis International

, Volume 15, Issue 8, pp 611–618

Nutritional and exercise-related determinants of bone density in elite female runners


  • Jane H. Gibson
    • Olympic Medical InstituteNorthwick Park Hospital and Clinical Research Centre
    • Rheumatic Diseases Centre, Sir George Sharp UnitCameron Hospital
  • Angela Mitchell
    • Olympic Medical InstituteNorthwick Park Hospital and Clinical Research Centre
    • Department of MedicineAddenbrooke’s Hospital
  • Mark G. Harries
    • Olympic Medical InstituteNorthwick Park Hospital and Clinical Research Centre
    • Northwick Park and St Mark’s Hospital
    • Olympic Medical InstituteNorthwick Park Hospital and Clinical Research Centre
    • Department of MedicineAddenbrooke’s Hospital
Original Article

DOI: 10.1007/s00198-004-1589-2

Cite this article as:
Gibson, J.H., Mitchell, A., Harries, M.G. et al. Osteoporos Int (2004) 15: 611. doi:10.1007/s00198-004-1589-2


Although the female athletic triad is widely recognized clinically, there have been few studies quantitating the effect of disordered eating on bone mineral density. The purpose of this study was to explore the mechanisms through which disordered eating might influence the skeleton in nationally or internationally competitive runners. Fifty British national or higher standard middle and long-distance female runners aged under 36 years were recruited; 24 had amenorrhea (AM), nine had oligomenorrhea (OL) and the others were eumenorrheic (EU). Bone mineral density (BMD g.cm−2) of the proximal femur (femoral neck and trochanter) and lumbar spine (L2−L4) was measured by dual energy X-ray absorptiometry (DXA) and compared with population-based European reference data. Dietary eating patterns were assessed with the Eating Attitudes Test (EAT26) and Bulimia Investigatory Test Edinburgh (BITE) questionnaires. High eating disorder scores were common; the EAT26 score predicted menstrual disorders (P=0.014) and correlated with body mass index (BMI). BMD was generally low in the AM group, but was raised in the proximal femur in the EU group. In the AM group, younger age at start of training was associated with higher trochanteric BMD. In addition, years of eumenorrhea were positively associated with spine BMD. Although a high EAT26 score was associated with lower BMD in the proximal femur, this could be explained by the intermediary effect of menstrual disorders. Osteocalcin, a marker of bone formation, was reduced in the AM group and was also reduced by high VO2max and high BITE score, consistent with a central (hypothalamic) pathway for suppressing osteoblastic bone formation. Eumenorrheic runners had increased femoral BMD compared with European controls, consistent with a positive effect of increased mechanical loading. The effect of disordered eating to reduce BMD could be explained by its association with menstrual dysfunction. Lumbar spine BMD was reduced most in those athletes who menstruated for the shortest time in adolescence.


AmenorrheaBone mineral densityEating disordersElite runnersOsteocalcin


Physical activity has site-specific effects on bone mineral density (BMD). For instance, in tennis players, bone density is higher in the dominant arm than the non-dominant arm [1] and rowers have higher lumbar spine bone density associated with their greater back strength [2]. However, intensive exercise is associated with secondary amenorrhea, a condition associated in turn with low bone density, particularly in the lumbar spine [3]. Concern has been raised that these athletes may be at risk of premature osteoporosis or fracture and a large literature has been generated [4,5,6,7]. Much work in runners has assessed BMD of the lumbar spine and non-weight-bearing areas such as the radius. The proximal femur has been less studied. There is also an extensive literature on menstrual disorders and their association with disordered eating patterns in female athletes [8,9,10]. However, it is not clear to what extent the risk of low bone density or stress fractures is determined by hypoestrogenization or alternatively directly to nutritional factors associated with disordered eating. In part, the uncertainty concerning the direct or indirect role of nutritional factors in determining menstrual status and bone density is due to the paucity of quantitative studies in which disordered eating has been rigorously assessed with well-validated instruments.

In animals, externally applied mechanical loading that is not damaging (i.e. does not exceed the bone’s so-called elastic limit and does not lead to identifiable microdamage) can stimulate bone formation on some bone surfaces while reducing bone resorption on others [11]. The result is to generate a more positive balance between the two processes and a net increase in bone mass. Recent studies have focused on the role of osteocytes as bone’s mechanosensors and the signaling molecules they generate, such as prostaglandins and nitric oxide [12]. In previous work, we found that suppression of endogenous estrogens in premenopausal women could lead to osteocyte death by apoptosis [13]. On the other hand, non-damaging mechanical loading was both associated with increased generation of NO by bone cells [14] and also enhanced the resistance of osteocytes to apoptosis [11]. That the preservation of bone cell signaling in response to mechanical loading could be jeopardized by amenorrhea was supported by the results of Stacey et al., who found reduced NO production in amenorrheic athletes [15]. Recently, Beals and Manore [9] have reported quantitative questionnaire-based assessment of disordered eating in college-level athletes and their association with bone injuries. However, there has been very little published work on the relationship between disordered eating as quantitated with validated instruments and bone mineral density in runners.

In this cross-sectional study, we investigated site-specific statistical effects of eating patterns on BMD and determined whether the effect of disordered eating was demonstratively independent of other determinants of reduced femoral or spinal BMD in a sample of elite runners. We developed statistical models describing in elite runners the determinants of disordered menstruation including the EAT26 and BITE questionnaire-based measures of disordered eating, and at the same time assessed markers of bone remodeling as potential determinants of reduced bone density seen in amenorrheic runners. Generally, our results supported the concept of the “female athletic triad”, with menstrual dysfunction and its associated endocrine basis as the key intermediary variable linking disordered eating with reduced bone density in athletes. However, certain interesting interactions between mechanical loading patterns and propensity to bone loss were suggested in comparing BMD outcomes across measurement sites.

Materials and methods


Fifty national and international standard Caucasian runners aged 17–35 years were recruited from advertisements in running magazines. They were required to have trained and competed in middle or long-distance running for at least the last 3 years, to have trained for a minimum of 3 h per week and to have run at least 25 miles per week. Exclusions included a current or past history of respiratory disease, diabetes, metabolic bone disorders, rheumatoid arthritis, thyroid or parathyroid disease, malignancy, cardiac, renal or inflammatory bowel disease or taking oral glucocorticoids (corticosteroids). The study was approved by the Harrow Health District Ethical Committee and each participant gave informed written consent to her participation. The subjects were the same as the “young athletes” group reported previously in our descriptive study of BMD in athletes [20].

Clinical assessments

Questionnaires were administered which gave details of menstrual history as well as training status and history. Athletes were categorized according to their menstrual history; eumenorrheic (EU) 10–13 cycles per year since menarche, oligomenorrheic (OL) 4–9 cycles per year for at least the last 18 months and amenorrheic (AM) 0−3 cycles per year for at least the last 18 months. Informed consent to participate was obtained as required by the Harrow Health district Ethical Committee. Weight and height were measured and body mass index (BMI, weight/height2) was calculated. Percentage body fat was assessed from skin fold thicknesses measured at mid-biceps, mid-triceps, suprailiac and subscapular sites on the right side of the body using Harpenden skin fold callipers. Percentage body fat (%BF) was calculated using the equation derived by Durnin and Womersley [16]. Maximal oxygen uptake (VO2max) was measured during a multistage continuous running test to exhaustion on a Powerjog treadmill using the Jaeger EOS Sprint on-line gas analysis system.

Current and past dietary intake was assessed using the Calquest 2.1 dietary calcium questionnaire. This questionnaire was previously validated against 5-day and 7-day weighed dietary inventories with reported correlation coefficients of r=0.76 and r=0.69, respectively [17].

Assessment of eating disorders

Eating habits were assessed by the Eating Attitudes Test (EAT26) and the Bulimia Investigatory Test Edinburgh (BITE) questionnaires, which have been validated for anorexia [18] and bulimia [19], respectively. A score of >30 on the EAT26 score represents a high likelihood of anorexia and a score of 15–30 represents a sub-clinical group with disordered eating habits and anorectic attitudes. The EAT26 was also divided into three further factors: factor 1 is related to a preoccupation with being thinner; factor 2 relates to bulimia and food preoccupation; and factor 3 relates to self-control of eating. These subdivisions were also analysed in our sample. In the BITE questionnaire, a “symptom score” of >20 represents highly disordered eating and defines the presence of binge eating and a score of 15–20 reflects disordered eating habits. A score of 10 or less is considered normal. The “severity” scale on the BITE questionnaire assesses the degree of any binge eating.

Bone mineral density

Bone mineral density (BMD) in g/cm2 of the 2nd 3rd and 4th lumbar vertebral bodies (LS) and the left hip (neck of femur, FN and trochanteric region, FT) was measured in each subject by dual energy X-ray absorptiometry (DXA) using a Hologic QDR lOOOW densitometer (Hologic Inc., Waltham, Mass. USA). Scan analysis was performed by technicians with daily experience in DXA analysis and who were blind to the menstrual status of the athlete. In order to determine the degree to which BMD rates were higher or lower than in a young reference secondary population, BMD values for each region of interest were compared with a young European reference population as described by Gibson et al. [20]. T-scores were calculated which express the degree to which a particular BMD at a measurement site is higher or lower than the young normal mean value in standard deviation units (e.g. mean for the reference population=0.0; mean+1 SD=+1.0).


Serum levels of follicle stimulating hormone, prolactin, testosterone, dihydroepiandrosterone and thyroid stimulating hormone and urinary progesterone and human chorionic gonadotrophin were measured to exclude other causes of secondary amenorrhea. Serum parathyroid hormone, calcium and alkaline phosphatase levels were within normal limits. On one occasion per subject, a second fasting early morning void of urine was sent for analysis of urinary creatinine and hydroxyproline and the results expressed as the ratio of hydroxyproline to creatinine [21]. Our method had a precision error of approximately 1 mmol Ca resorbed from bone per day when assessed against our “gold standard” radioisotopic method [22]. Plasma osteocalcin was measured as described previously using an Incstar kit [21]. This had a similar precision error for bone formation [23]. The ranges of bone formation and resorption observed in normal subjects were 1.1–8.3 mmol calcium incorporated into or removed from bone per day [24]. Estradiol was measured on one occasion unrelated to the menstrual cycle in the routine hospital assay.

Statistical analysis

Two-tailed tests of significance were used. The EAT and BITE data were not normally distributed, so for ANOVA comparisons between groups was by Wilcoxon/Kruskal-Wallis tests. To compare groups, data were analysed by single factor analyses of variance (ANOVA) across menstrual groups; this was followed, if the ANOVA proved significant (P<0.05), by Student’s t-tests to explore between-group differences in factors of the EAT26 and BITE. BMD T-scores are presented with their 95% confidence intervals.

To investigate the dependence of menstrual status on behavioural factors, menstrual status was defined as an ordinal variable with three levels: eumenorrheic (0), oligomenorrhea (1) and amenorrhea (2). Menstrual status was made the dependent variable in simple and multiple logistic regression analyses. A backwards, stepwise approach was used to eliminate variables (at P>0.05 in sequence, beginning with the least significant) that did not contribute significantly to explaining menstrual status from among: age began competitive running; VO2max; miles run per week; BITE score; and EAT score. Then the dependence of menstrual status on anthropometric measures was similarly investigated.

Next, we investigated the dependence of bone density on other variables that have been postulated to influence bone density alongside menstrual status. This was an extension of the analysis we published previously, which included older athletes up to their 7th decade of age [20] and tested hypotheses generated by this study and other investigators. Candidate determinants included biochemical markers of bone remodeling, diet-related variables and duration of high-level sporting activity, as well as other variables previously reported on. Simple regression analysis was employed to select variables for inclusion among candidate determinants of BMD, following which a backwards stepwise approach was used in multiple linear regression analyses with the BMD T-scores as the outcomes of interest. The T-scores were used because we included runners up to 35 years of age and we wished to avoid any confounding effect of age. Our two markers of bone remodeling, osteocalcin for bone formation and the hydroxyproline:creatinine ratio for bone resorption, were also made outcome variables to explore the dependence of bone remodeling on measures of endocrine status and physical activity. Significance was indicated throughout at P<0.05 (two-tailed test). Finally, repeated measures multiple analysis of variance (MANOVA) was used to contrast and compare the effects of determinants of bone density across the three measurement sites. All statistical evaluations were undertaken using JMP v 4.0 (SAS Institute, Carey, S.C., USA).


Training and eating status

Physical, menstrual and training characteristics are presented in Table 1. Eating habits differed across groups for the EAT26 (P<0.05) and EAT factor 3 (P<0.025) and borderline differences were found for EAT factor 1 and BITE (P<0.1) (Table 2). In the AM group, one athlete had an EAT26 score of >30 (defined as clinical anorexia) and eight had scores of 15–30 (disordered eating). One subject under clinical treatment for anorexia did not complete her questionnaire and one subject with independently verified anorexia and bulimia did not fully complete her questionnaire. One EU subject did not return her questionnaire, without giving a reason. The distribution of EAT26 scores is shown in Fig. 1. The BITE symptom score revealed unusual patterns of eating (score of 10–19) in 45% (AM athletes), 33% (OL athletes) and 25% (EU athletes). One OL athlete had severe binge eating. The EAT26, factor 1 and factor 3 scores were all negatively correlated with weight (r=−0.37, −0.32 and −0.44, respectively, P all <0.02). The BITE score was not correlated with weight.
Table 1

Physical, menstrual and training characteristics of athletes by menstrual group. Mean (SD). Menstrual groups: AM 0–3 menses per year; OL 4–9 menses per year; EU 10–13 menses per year


AM (n=24)

OL (n=9)

EU (n=17)

Age (years)


25.44 (6.7)

26.29 (6.4)

Height (cm)

164.6 (4.5)

163.7 (4.3)

165.9 (4.5)

Weight (kg

51.3 (5.1)**

53.9 (5.6)

57.5 (5.4)

Body fat (%)

17.6 (4.5)*

21.3 (7.3)


BMI (kg/m2)

18.8 (1.8)**

20.1 (2.1)

20.9 (1.3)

Miles per week (MPW)

54.3 (20.5)

47.3 (14.25)

44.4 (16.3)

V02max ml.min−1.kg−1

58.9 (7.3)

57.9 (7.6)

58.2 (5.5)

Age at menarche

13.8 (1.3)

14. 1 (1.8)


Years of eumenorrhea

6.4 (4.8)**

7. 1 (5.2)


Years of OL/AM

7.0 (3.7)

4.9 (3.8)

*P<0.01, **P<0.005 for AM vs EU

Table 2

Quantitative aspects of diet mean (SD). Menstrual groups: AM 0–3 menses per year; OL 4–9 menses per year; EU 10–13 menses per year. Two athletes in the AM group and one athlete in the EU group failed to return questionnaires


EAT 26 “Symptom score”

BITE “Symptom score”

Dietary calcium intake (mg/day)


Factor 1

AM (n=22)

12.3 (9.4) **

7.7 (6.3)*

7.6 (3.7)

830 (388)

OL (n=9)

10.4 (8.6)

7.4 (5.8)

9.2 (6.1)

808 (485)

EU (n=16)

5.6 (6.0)

3.8 (3.8)

5.4 (3.7)

1056 (441)

*P<0.05, **P<0.02 for AM vs EU

Fig. 1

Logistic relationships between the order of menstrual dysfunction (right ordinate scale: 0, plus signs eumenorrhoea, 1, crosses oligomenorrhoea, 2, dots amenorrhoea) and (top) total EAT score (P=0.014 for model) and (bottom) body mass index (P<0.001 for model). Left ordinate scale is a scale of probability. The right ordinate scale shows the proportions of subjects in each category

Statistical determinants of menstrual status

In the backwards stepwise regression of menstrual status on behavioral factors, all explanatory variables were eliminated with the exception of the EAT score (χ2=6.0, P=0.014: Fig. 1 top). Body mass index (BMI; χ2=13.6, P<0.0002: Fig. 1 bottom) or alternatively deviation from the ideal BMI (dBMI, χ2=15.2, P<0.0001) displaced all other anthropometric measures (height, weight, %fat). When the EAT score was added to BMI in the logistic regression, it had only a borderline significant additional effect (P=0.08).

Bone mineral density: descriptive data

BMD values in g.cm−2 are shown in Table 3 alongside biochemical marker data. Comparison of BMD with age-matched European data (T-scores) is shown in Fig. 2. Between-group analysis showed that the AM and OL athletes had significantly lower T-scores in all regions than the EU athletes (P<0.001, P<0.01, respectively). Although the T-scores were higher in the OL group compared to the AM athletes this did not reach significance. The AM athletes were significantly lighter than the other athletes.
Table 3

BMD and biomarkers by menstrual group. Mean (SD) and T-score (95% confidence intervals). FN neck of femur, FT femoral trochanteric region, LS lumbar spine, L2−L4 Hyp:Creat hydroxyproline:creatinine ratio, OC osteocalcin, E2 estradiol. E2 presented as median and interquartile range (IQR)


FN BMD (g.cm−2)

FT BMD (g.cm−2)

LS BMD (g.cm−2)

Hyp:Creat (μmol/mmol)

OC (ng/ml)

E2 (pmol/l)

AM (n=24)


0.66 (0.11)

0.92 (0.10)



115 (IQR 74,157)

T=−0.56 (−0.87 to –0.26)

T=−0.57(−0.96 to –0.17)

T=−1.27(−1.65 to –0.88)

OL (n=9)

0.85 (0.12)

0.73 (0.07)

0.98 (0.07)



153 (IQR 77,707)

T=−0.23 (−0.99–0.53)

T=+0.19 (−0.45–0.83)

T=−0.63 (−1.26 to –0.01)

EU (n=17)

0.97 (0. 11)

0.82 (0.07)

1.09 (0.10)



253 (IQR 160,442)

T=+0.98 (0.50–1.46)

T=+1.13 (0.66–1.60)

T=+0.41 (−0.05–0.86)

Fig. 2

BMD T-scores in the lumbar spine (L2–L4), femoral neck and femoral trochanter. Note that the mean diamonds for each group show the group mean values and 95% confidence intervals for each mean (point of diamonds)

Statistical determinants of BMD

Simple linear regressions of BMD T-score on anthropometric parameters revealed significant associations in all regions between BMD and weight and height (at each site, P<0.02 and P<0.05, respectively). Additional correlations are shown in Table 4; LS BMD was correlated with BMI, number of years of eumenorrhoea and percent body fat (P<0.05, P<0.002, P<0.05, respectively). FN BMD was correlated with BMI and EAT26 score (P<0.02, P<0.05, respectively). FT BMD was correlated with age at which training began and EAT26 score (P<0.05, P<0.01, respectively).
Table 4

Student’s t-values for simple correlations of selected continuous variables with BMD T-scores. When P>0.05 they are shown as NS. Note that inverse correlations are shown as negative t-values. With 48 degrees of freedom, |t|>2.01, P<0.05; |t|>2.68, P<0.01

Independent variable




EAT 26


% fat

Age start




Diet Ca2+


Years eumen

T femoral neck














T femoral trochanter














T spine L2–L4














When the BMD data were modeled with all the above variables included in backwards stepwise regressions, only menstrual status remained significant across all three measurement sites as an independent determinant (P<0.0001). However, FT BMD (but not FN BMD) was also associated with later age at which training began (with a 0.05 reduction for each year start of competitive training was delayed, P<0.025); or, alternatively, inversely with years of eumenorrhea (P=0.014). In contrast, LS BMD was positively associated with height (P=0.005) and positively (P=0.038) with years of eumenorrhea. After adjustment for menstrual status, there were no significant relationships between BMD (any region) and dietary calcium, miles run per week, VO2max, age at menarche or BITE score. In the MANOVA repeated measures analysis, statural height (P=0.013) as well as menstrual status (P<0.0001) had similar (P>0.5) effects across all three measurement sites, the effect of age at which training began was no longer significant and the effect of years of eumenorrhea was strongly contrasting, so that it had a negative effect at the femoral trochanter and a positive effect at the lumbar spine.

The two larger groups (AM and EU) were then considered separately to determine within group correlation of BMD with height, weight, (or alternatively BMI), percent body fat and age at which training began, using MANOVA repeated measures analysis to compare measurement sites. In the AM group, BMD was significantly dependent on height at all regions (P=0.04). However, age at which training began was a determinant of BMD only in the trochanter, contrasting significantly with the other two measurement sites (P<0.01). These associations were not evident in the EU group. Within the AM group considered alone, years of amenorrhea did not correlate significantly with BMD at any site, nor did body weight or BMI.

Statistical determinants of markers of bone formation and bone resorption

There was no significant effect of menstrual status on the hydroxyproline:creatinine ratio, as shown in Table 3. However, the effect of menstrual status on osteocalcin was highly significant (P=0.009; R2adj=0.15) with the two groups having disordered menstrual function having lower values [EU: mean 3.53 (SE 0.26); OL: mean 2.62 (0.37); AM: mean 2.44 (0.22); difference between EU and AM P=0.0055, Dunnett’s test using EU as the control group]. In univariate testing, there was no significant determinant of the hydroxyproline:creatinine ratio, but the total BITE score was inversely correlated with osteocalcin (P=0.03) as was VO2max (P=0.05), whereas body weight was positively correlated (P=0.04). These variables were entered into backwards stepwise regressions alongside menstrual status. Only body weight dropped out, and the final model included VO2max (P=0.015, Fig. 3) and BITE score (P=0.047) as inverse determinants of osteocalcin level alongside menstrual status (P=0.04). In the whole model 26% of the variance was accounted for.
Fig. 3

Leverage plot describing the effect of VO2max on osteocalcin level after adjusting for order of menstrual dysfunction and BITE score


In this cross-sectional study, we have shown that disordered eating was strongly correlated with disordered menstruation in elite young female runners. As expected, those with high Eating Attitudes Test (EAT) scores (measuring relative anorexia) tended towards low body weight, whereas those with a high Bulimia Investigatory Test Edinburgh (BITE) score did not necessarily have low body weight compared with the other athletes. Among the behavioral determinants of menstrual status, the EAT score displaced all other variables examined, but was itself largely displaced by the body mass index in multiple logistic regression. This is consistent with BMI (as a marker for low body fat) acting as an intermediary variable on the causative pathway between disordered eating behavior and menstrual disturbance. However, to the extent that this study was carried out before their ready availability, we could not assess markers related to leptin regulation.

In examining the determinants of bone density, we found as expected that menstrual status was the most important determinant of BMD T-score at all three measurement sites. However, in the amenorrheic group two interesting determinants emerged which were additional to menstrual status. Height was a positive determinant of BMD at all sites. The reason for this effect is not obvious, since height in most previous work has proved a much less strong determinant of BMD than weight. Maybe it is a direct result of the fact that because BMD is measured as BMC/area rather than BMC/volume, larger subjects will, other things being equal, have higher BMDs. The age at which serious training began had a more complex effect, with a younger start having a positive influence on femoral trochanter BMD, no significant influence on femoral neck BMD and a negative effect on spine BMD. This also presents some difficulty of interpretation, because, not surprisingly, age at which training began correlated inversely with years of eumenorrhea (r2=0.36). Therefore, at the trochanter, years of eumenorrhea also had an adverse effect on BMD, which is counter-intuitive, has no reasonable biological explanation and takes effect only when a large adjustment is made for current menstrual status. It seems more likely that the athletes who were amenorrheic and who experienced the longest training durations experienced some cumulative benefit to BMD at the femoral trochanter from long term exposure to continued high training levels.

We were able to confirm the findings of previous investigators that in amenorrheic subjects, our chosen marker of bone formation was suppressed relative to levels in eumenorrheic athletes [25]. EU athletes tend toward having higher marker levels for bone formation than sedentary controls [26]. However, besides menstrual status, VO2max and BITE score also emerged as significant determinants of osteocalcin; and in particular VO2max had a sizeable effect, reducing osteocalcin by 13% for each 1 SD increase in VO2max, after adjusting for menstrual status. Since osteocalcin concentration is only a marker for osteoblast function, that is also influenced by its own rates of entry into and disposal from the circulation, this suggests, but cannot prove, that VO2max might influence osteoblastic function through a local or a central pathway. VO2max is itself dependent on constitutional factors and relative leanness as well as the capacity of the individual’s own tissues to undertake oxidative respiration rapidly.

Physical exercise has been demonstrated to have beneficial effects on bone mineral density and the maximum effect appears to occur at regions of greatest mechanical stress [1,2]. But studies on BMD in the lower limb have given variable results [27,28]; although Wolman et al. have demonstrated higher femoral shaft BMD in young runners compared to sedentary controls, dancers and rowers [29]. Our results show that BMD of the neck of femur and trochanteric region is substantially higher in eumenorrheic runners compared to an age-matched European reference population, who had similar BMD values to the NHANES III American population [20]. Lumbar spine bone density was also higher, but not significantly, than the European reference mean in this group.

If increasing peak bone mass reliably reduces the lifetime risk of fracture, the eumenorrheic athletes might be at reduced risk of hip fracture in old age. The effect of high exercise levels on bone in the athletes with disordered menses is less certain; for one thing, it is not clear that they would regain normal eating patterns or menstrual function if they were to reduce their athletic activity. Over the last decade, it has been shown that prolonged amenorrhea in athletes is associated with a reduction in BMD, particularly in the lumbar spine [3,30,31]. However, because exercise is an osteogenic stimulus in itself, it is possible that bone loss due to amenorrhea may be offset in areas of high mechanical stress. In one study on amenorrheic rowers, lumbar spine BMD was not as low as in amenorrheic runners [2] suggesting that rowing may partially counteract the reduction in BMD, presumably by the muscular strains placed on the spine by this form of exercise. If this is so, the proximal femur in runners may be relatively protected from bone loss during episodes of amenorrhea, despite a relatively high trabecular bone content (up to 50% in the neck of femur). There are somewhat few reports on the influence of athletic amenorrhea on the proximal femur. Drinkwater et al. [32] showed a reduction in BMD in the femoral shaft but not the femoral neck in amenorrheic runners but Wolman et al. [29] were unable to detect a difference in BMD of the femoral shaft between amenorrheic runners, rowers and dancers and their eumenorrheic counterparts. Others have also found preservation of BMD in the leg in these hypoestrogenic athletes [28,33]. Our results suggest that running may afford some protection at the proximal femur but that this was not sufficient to prevent significant decreases in BMD by comparison with the eumenorrheic runners or control European values (Fig. 2).

Like Drinkwater [32], we were unable to demonstrate any relationship between the duration of amenorrhea and BMD. It may be that during athletic amenorrhea, as after menopause, the greatest rate of loss of bone mineral occurs during the first few years. Our cohort of amenorrheic women had a mean of 7.75 years of amenorrhea. The duration of amenorrhea appeared to be more important for the lumbar spine BMD in our study. This suggests that in the spine the longer the skeleton is exposed to normal sex hormone levels after menarche, the more BMD is increased towards its peak adult value and the less the absolute value of BMD is reduced below the anticipated peak BMD by prolonged amenorrhea subsequently. If this is true, then it will be even more important to protect young athletes from amenorrhea in the first few years after menarche.

Height, weight, percentage body fat and body mass index were significantly associated with BMD at all sites. These variables were all lower in the amenorrheic group and when menstrual group was accounted for, only height continued to contribute significantly to the variance in BMD. Within-group analysis of these parameters showed that in the amenorrheic group, height was an independent predictor of BMD, whereas this was not demonstrated in the other groups. These results are somewhat different to those of Drinkwater et al., who showed that BMD was correlated with body weight, with a stronger relationship in those athletes with the greatest menstrual dysfunction [32].

The prevalence of eating disorders was much greater in the amenorrheic group and a high EAT26 score was associated with low body weight, body mass index and percentage body fat. Eating disorders are common in athletes [4,5] and pubertal girls with higher EAT26 scores had lower BMD and gained bone slower that girls with normal EAT26 scores [34]. In addition to low caloric intake in relation to energy expenditure, athletes tend to avoid foods high in fat, often dairy products containing calcium. This was a possible explanation for the finding of Wolman et al. [35] relating calcium intake positively to BMD. Although 30% of our cohort consumed less than 600 mg/day calcium, we found no differences in BMD at any region between those consuming either less than 600 mg/day or more than 1000 mg/day compared to the remainder. The apparent absence of eating disorders in about half of the AM group should be viewed in the light of their high caloric needs and the EAT26 questionnaire having been designed for sedentary subjects. Within the amenorrheic subgroup with normal EAT26 scores (<15), the overall EAT26 score was principally determined by factors 1 (staying thin) and 3 (self control), rather than by food preoccupation (factor 2), as it was in the amenorrheic group as a whole.

Cross-sectional studies of volunteers can be criticized for not eliminating selection bias. It is possible that those women who developed menstrual dysfunction had a constitutional tendency towards lower BMD. To answer this criticism will require longitudinal cohort studies that follow adolescent athletes for a number of years. Furthermore, it is also unknown whether amenorrheic athletes continue to have low BMD up to and after the menopause. There are suggestions that there may be a gain in BMD when menses return or when training is reduced [36,37]. In our study on older athletes, BMD in previously amenorrheic athletes did not differ from the age matched sedentary control means [20]. However, given the commitment to their sport of those among our runners with amenorrhea or oligomenorrhea, it seems questionable that they will ever reach their maximum potential peak bone mass. Until this question is resolved, athletic amenorrhea should not be considered benign. The use of the older biomarkers hydroxyproline:creatinine and the early Incstar osteocalcin kit reflects both the time when the study was done and the investment we had made in validating these methodologies for our needs at the time.

In conclusion, many athletes in our study suffered from disordered eating patterns and this was quantitatively associated with risk of menstrual disturbance. In the lumbar spine, which is not so heavily loaded by running, the reduction in BMD with menstrual disturbance was more severe in those runners who had experienced the shortest history of post pubertal eumenorrhea. Interestingly, the vertebral epiphyses fuse late in puberty. On the other hand, the femoral trochanter had higher BMDs in those amenorrheic runners who started training at a younger age. The mechanism for the failure to achieve high BMD levels in those with menstrual disorders may have been through lowered bone formation rates, as suggested by our osteocalcin results, confirming previous data. There was only indirect evidence in support of lowered bone formation being influenced through a central pathway [38], modulated by training intensity (marked statistically by VO2max) and eating patterns (BITE score). Plasma leptin is reportedly low [39] in athletes with inappropriately low energy intakes. Another mechanism with the potential to reduce bone formation in amenorrheic athletes is lowered nitric oxide production, by osteocytes or other cells affected by low oestrogen levels [15].

Copyright information

© International Osteoporosis Foundation and National Osteoporosis Foundation 2004