Introduction

Anorexia nervosa (AN) is a pernicious psychiatric disorder that is associated with serious medical consequences including bradycardia [1, 2], low-bone mineral density (BMD) [3,4,5], and increased fracture risk [6,7,8]. More severe malnutrition is associated with greater deficits in BMD, fat mass (FM), and lean body mass (LBM) [9, 10]. Premenopausal low BMD often is a complication of long-term energy deficiency with or without the presence of an eating disorder (ED), in which the endocrine alterations due to the low energy availability contribute to bone loss [11, 12].

In healthy adolescents and young adults, exercise is associated with higher BMD [13], possibly through the direct effect on bone accrual by impact and the indirect effect of increased muscle mass and thereby muscle strain [14, 15]. Sport-specific impact on BMD have been observed; for instance middle- and long-distance running and non-impact sports (e.g., swimming and bicycling) do not have the same positive effect on BMD as high-impact training (e.g., gymnastics) or odd-impact training (e.g., soccer) [16]. Low BMD is two- to three times more prevalent in non-athletic premenopausal females than in elite athletes [17].

The effect of exercise on BMD and body composition in adolescents with AN has not been well characterized, in part because excessive exercise may have a role in AN psychopathology and be limited during treatment. A study of 70 adolescents with AN found that exercise was positively associated with dual-energy X-ray absorptiometry (DXA) hip BMD Z-score, the hip being an area consisting primarily of cortical bone; the association was not observed in the spine which contains more trabecular bone [9]. This study was limited to females and did not include measures of body composition. Although adolescents with AN or bulimia nervosa who report team sports participation present with a lower heart rate and are more likely to meet hospital admission criteria than those who do not report team sports participation [18], the association of participation on team sports with bone density and body composition has not been addressed in adolescents with AN.

The objective of this study was to examine associations among exercise, BMD, and body composition in adolescents with AN. We hypothesized that duration of exercise and team sports participation would be associated with greater cortical bone such as hip BMD and whole body BMC through accrual of LM, but not greater trabecular bone such as spine BMD, among adolescents with AN.

Methods

In this retrospective, cross-sectional study, we reviewed the medical records of all patients, ages 9–20 years, presenting for an initial outpatient or inpatient evaluation to the Eating Disorders Program at Stanford between March 1997 and February 2011. We included patients with a Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) diagnosis of AN [19], available DXA results obtained using a Hologic 4500 bone densitometer within three months of presentation, and exercise data (Online Fig. 1). The study was approved by the Committee on Human Research (IRB) at Stanford University.

Psychological and clinical characteristics of patients were re-reviewed and classified into a DSM-5 ED diagnosis since patients were initially diagnosed using Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) criteria [20]. Median body mass index (mBMI, kg/m2) was defined as the 50th percentile BMI for age using the Centers for Disease Control and Prevention growth curves [21]. Percentage mBMI (%mBMI) was defined as the patient’s BMI on presentation divided by the mBMI multiplied by 100 [22]. Follicle-stimulating hormone (FSH), luteinizing hormone (LH), and estradiol were obtained in females.

The intake for all patients in the Stanford Eating Disorders Program includes a standardized semi-structured clinical interview with questions about exercise [18]. This information was extracted from the medical record using a codebook by two trained independent research abstractors, with dual data entry. Exercise duration (approximate hours per week in the 30 days prior to presentation) and team participation in the 30 days prior to presentation were recorded. Subjects who reported any team involvement for any sport or exercise were classified as participating in team sports. Subjects who reported any involvement in exercise that involved running, walking, hiking, team sports involving running (i.e., soccer, basketball, softball, baseball, volleyball), ballet/dance, gymnastics, or weight-lifting were classified as participating in weight-bearing exercise.

The semi-structured interview also included questions about menstrual status, self-reported tobacco use in the past month [23] and current use of antidepressants such as serotonin selective receptor uptake inhibitors (SSRI) [24] or tricyclic antidepressants (TCA) [25].

All bone density and body composition assessments were performed by DXA (Hologic 4500, Hologic, Waltham, MA). The DXA whole body bone mineral content (BMC) and lumbar spine, total hip, and femoral neck BMD measurements were converted to sex-, race-, and age-specific Z-scores using reference curves by the Bone Mineral Density in Childhood Study (BMDCS) [26]. The Z-scores were further adjusted for height Z-score at the time of presentation using the method developed by BMDCS investigators [27]. Height-normalized FM index (FMI, kg/m2) and LBM index (LBMI, kg/m2) were converted to sex, race and age-specific Z-scores using reference values from the National Health and Nutrition Examination Survey (NHANES) [28].

Data were analyzed using STATA 15.0. Differences in participant characteristics were calculated using independent samples t tests or Pearson’s chi-square tests. Linear regression analyses were used to identify associations among exercise variables (team sports and approximate hours per week) and DXA Z-scores, adjusting for age, %mBMI (for BMC and BMD only since related to FMI and LBMI), duration of illness (log-transformed due to skewness), tobacco use, and antidepressant use. For all bone density measures that were significantly associated with exercise variables, we added LBMI and FMI Z-scores to the models to see if they might account for the relationship. p < 0.05 was considered statistically significant.

Results

A total of 188 subjects with AN (178 females and 10 males) met eligibility criteria (Online Fig. 1). Although a higher proportion of males were excluded, there were no statistically significant differences in other demographic or anthropometric characteristics (Appendix A). Demographic and health characteristics of the included sample are listed in Table 1. A majority of subjects participated in weight bearing (93.1%) and team exercise (64.3%).

Table 1 Demographic, health, bone, and body composition characteristics of adolescents with anorexia nervosa by sex
Full size table

In linear regression models, greater exercise duration (approximate hours per week) was associated with higher BMD Z-score at the total hip (B = 0.15, p = 0.005) and higher LBMI Z-score (B = 0.10, p = 0.018), controlling for listed covariates (Table 2). Participation in team sports was associated with higher whole body BMC Z-score (B = 0.39, p = 0.001), higher BMD Z-score at the total hip (B = 0.53, p = 0.001), and greater deficits in FMI Z-score (B = − 1.06, p = 0.007), controlling for listed covariates. These associations remained significant when additionally adjusting for months since last menstrual period, FSH, LH, and estradiol in the subsample of females (Appendix B). The coefficients (B) for the associations between exercise variables and whole body BMC and total hip BMD Z-scores were attenuated and no longer statistically significant after adjusting for LBMI Z-score, suggesting that the relationships may also be partially explained by muscle (Appendix C).

Table 2 Multivariate linear regression analyses of exercise duration and team sports participation as the independent variables and dual-energy X-ray absorptiometry (DXA) Z-scores as the dependent variables, adjusting for covariates
Full size table

Lower %mBMI was significantly associated with lower BMC and BMD Z-scores at all sites (Table 2); however, duration of illness, tobacco use, and antidepressant use were not associated with BMC or BMD Z-scores in adjusted models (data not shown).

Discussion

Among adolescents with AN, we found that duration of exercise and participation in team sports were associated with higher BMD at the hip and team sports were associated with higher whole body BMC. This is consistent with recent findings of females with AN reported by Divasta et al., 2017, although our sample was over twice as large and additionally included males with AN. Since the hip was one of the sites with the greatest bone deficits in our sample as well as in other studies of adolescents with AN [29, 30], the potentially protective effect of duration of weight-bearing exercise and team sports participation on BMD at the hip was clinically relevant. These findings are consistent with the literature from healthy athletes, where participation in high-impact sports such as gymnastics and team sports such as soccer, field hockey, and softball was associated with higher BMD at the pelvis [31]. Exercise may be particularly beneficial for areas of cortical bone such as the hip and whole body in adolescents [32, 33]. Athletes are expected to have higher BMD than non-athletes and BMD Z-scores < − 1.0 are considered abnormal in athletes [34]. Patients with restrictive eating disorders who exercise should still be screened for low bone density if clinically indicated, such as amenorrhea for > 6 months in females [35].

However, given the potential BMD-related benefits conferred by exercise in those with AN, clinicians must also weigh the potential medical risks (bradycardia, caloric expenditure, amenorrhea [36, 37]) with the potential benefits (improved BMD and management of mood or anxiety) of exercise in the management of eating disorders. Some literature has shown that supervised exercise in the treatment of eating disorders may reduce drive for thinness and body dissatisfaction [38]. Data from a meta-analysis indicated that while single-group studies of supervised exercise in AN showed improvement in weight and body fat, pooled randomized controlled trials and quasi-randomized studies showed no significant effect of supervised exercise training on selected anthropometric measurements [39]. A systematic review of physical therapy interventions for patients with eating disorders found that aerobic and resistance training resulted in significantly increased muscle strength, body mass index, and body fat percentage in patients with AN and that aerobic exercise, yoga, massage, and basic body awareness therapy significantly lowered scores of depression and eating pathology without negative effects on weight restoration [39, 40]. However, a crucial consideration relates to the role of exercise in psychological symptomatology of AN. The propensity for compulsive exercise in AN has been well noted [41], and is thought to serve as a compensatory mechanism in offsetting caloric intake. As such, even in the context of broader BMD-related benefits, engagement in compulsive exercise likely represents a detriment to crucial weight restoration, and may challenge overall psychological outcomes.

Adolescents with AN have greater deficits in FM than LBM [7, 42, 43]. We find that participation in team sports was associated with further FMI deficits in adolescents with AN. Female undergraduates who participated in club or varsity team sports reported higher rates of body dissatisfaction than those who were non-exercisers or independent exercisers [44]. EDs are particularly high among female athletes competing in esthetic sports (42%), endurance sports (24%), technical sports (17%), and ball sports (16%), which all may have team components [45].

Hours per week of exercise was positively associated with LBM, indicating that despite a state of malnutrition, adolescents with AN may be able to accrue some muscle. Furthermore, LBM (i.e., muscle) may explain the positive association between exercise and whole body BMC and total hip bone density. Prior studies have shown that LBM was independently associated with bone density, even when adjusting for FM [14, 15, 46].

Our study had several limitations that merit acknowledgement. The retrospective, cross-sectional study design precluded causal inferences. The exercise questions were asked regarding the previous 30 days; a longer time frame may be expected for exercise to impact BMD. However, given their high levels of activity, it is likely that these adolescents had been participating in exercise for over 30 days. Although we had some data on exercise type and were able to comment if subjects engaged in any weight-bearing exercise, most subjects engaged in multiple exercise types and we were unable to isolate how many hours per week were dedicated exclusively to weight-bearing, high-impact, or odd-impact exercise. Future research may more precisely isolate the effect of particular exercise types (such as high-impact or odd-impact sports) on site-specific changes in BMD in the setting of AN. Although we included FSH, LH, and estradiol laboratory measures in females, future research could determine the association of androgens (in males), growth hormone (GH), insulin-like growth factor 1 (IGF1) [47], and Vitamin D in relation to exercise and BMD. We did not find any significant sex differences in BMD or body composition similar to prior studies [5, 42], although our sample of males was smaller than our sample of females. The higher proportion of males than females who were excluded may be attributed to the fact that there are no clear guidelines for when to obtain DXA in males with AN [35]. Clinicians may, therefore, be less likely to obtain DXA in males with AN. However, there were no significant differences in other demographic characteristics of those who were excluded versus included. Although there is some concern for measurement of changing landmarks at the hip in younger subjects, the hip may still be clinically relevant as it was shown to be associated with incident fracture in multivariable models using BMDCS data [48].

Strengths of this study included the evaluation by a specialized eating disorder team with systematic data collection. We included males in our sample; previous studies on bone density and exercise in AN have not included this understudied population. We used the most up-to-date DSM-5 eating disorder criteria as well as the BMDCS and NHANES reference curves which are the most robust adolescent reference values for bone density and body composition, respectively.

Conclusion

Exercise may be protective of BMD at the hip and whole body BMC, while participation in team sports was associated with greater FMI deficits among adolescents with AN. LBMI Z-score was positively associated with duration of exercise and may explain the relationship between exercise and bone outcomes. In determining the transition from abstinence to return-to-exercise in adolescents with AN who are medically and psychologically stable, clinicians should consider potential bone-related benefits in addition to other risks and benefits of exercise. Future research may explore the effect of exercise on medical and psychological recovery in eating disorders, including effects on bone density and body composition, to provide guidance for the management of exercise in the setting of adolescent eating disorders.