Denosumab Increases Spine Bone Density in Women with Anorexia Nervosa: A Randomized Clinical Trial

Abstract

Objective

Anorexia nervosa is complicated by high bone resorption, low bone density (BMD), and increased fracture risk. We investigated whether off-label antiresorptive therapy with denosumab increases BMD in women with anorexia nervosa.

Design

Twelve-month randomized, double-blind, placebo-controlled study

Methods

Thirty ambulatory women with anorexia nervosa and areal BMD (aBMD) T-score <−1.0 at ≥1 sites were randomized to 12 months of denosumab (60 mg subcutaneously q6 months)(n=20) or placebo (n=10). Primary endpoint was postero-anterior (PA) lumbar spine aBMD by DXA. Secondary endpoints included femoral neck aBMD; tibia and radius volumetric BMD and bone microarchitecture by high-resolution peripheral quantitative CT; tibia and radius failure load by finite element analysis; markers of bone turnover.

Results

Baseline mean (±SD) age [29±8 (denosumab) vs. 29±7y (placebo)], BMI (19.0±1.7 vs. 18.0±2.0kg/m²), and aBMD (PA spine Z-score −1.6±1.1 vs. −1.7±1.4) were similar between groups. PA lumbar spine aBMD increased in the denosumab vs. placebo group over 12 months (p=0.009). Descriptively, the mean (95% CI) increase in PA lumbar spine aBMD was 5.5 (3.8–7.2)% in the denosumab group and 2.2 (−0.3–4.7)% in the placebo group. Femoral neck aBMD was similar between groups. Radial trabecular number increased, radial trabecular separation decreased, and tibial cortical porosity decreased in the denosumab vs. placebo group (p≤0.006). Serum C-terminal telopeptide of type I collagen and procollagen type I N-terminal propeptide decreased in the denosumab vs. placebo group (p<0.0001). Denosumab was well tolerated.

Conclusions

Twelve months of antiresorptive therapy with denosumab reduced bone turnover and increased spine aBMD, the skeletal site most severely affected in women with anorexia nervosa.

Introduction

Anorexia nervosa, a psychiatric disease characterized by caloric restriction, fear of gaining weight, and body image distortion, is increasing in prevalence among women in the United States, with a current estimated lifetime prevalence of 1.4%.^{1, 2} One of the most common and serious medical comorbidities of anorexia nervosa is low bone mineral density (BMD), particularly at the spine, and associated fractures. Approximately 80% of women with anorexia nervosa have a BMD Z-score <−1 at one or more skeletal sites, and 39% have a BMD Z-score <−2 at one or more skeletal sites, as measured by dual energy x-ray absorptiometry (DXA).³ Recent data have also shown that women who are not low weight but meet psychiatric criteria for anorexia nervosa are also at risk for low BMD.³ Anorexia nervosa is also associated with impairments in cortical and trabecular bone microarchitecture, as measured by high resolution peripheral quantitative computed tomography (HR-pQCT).⁴ Moreover, fracture risk is higher in women with anorexia nervosa than in healthy age-matched controls.⁵ Although anorexia nervosa is a chronic disease in the majority of women,⁶ low BMD and increased fracture risk often persist, even for those women who experience recovery.^7-9 The ideal window to intervene on low BMD in anorexia nervosa may therefore be early when the disease in active.

Although several therapies have been investigated to treat low BMD in women with anorexia nervosa, none are FDA-approved. Since bone metabolism in women with anorexia nervosa is characterized by high bone resorption and low bone formation,¹⁰ therapies targeting the high resorptive state have been studied. Available data strongly suggest that oral contraceptives are not effective at increasing BMD in this disorder.^11-13 Transdermal estrogen increased lumbar spine and hip BMD in female adolescents with anorexia nervosa, but the increase was not different from that observed in healthy female adolescents not on transdermal estrogen, suggesting that low BMD in adolescent females with anorexia nervosa may not catch-up.¹⁴ Transdermal estrogen also increased lumbar spine, but not total hip or femoral neck, BMD in a small, 6-month, open-label study in women with anorexia nervosa.¹⁵ The bisphosphonate risedronate increased BMD by 3.2% and 1.9% at the spine and hip, respectively, in women with anorexia nervosa compared with placebo after 1 year,¹⁶ but mean PA spine Z-score remained <−1.0. Denosumab, a RANK ligand inhibitor, is a more potent antiresorptive therapy than estrogens or bisphosphonates. In postmenopausal women, it has been demonstrated to increase lumbar spine BMD by 5.3% compared with alendronate after 1 year,¹⁷ and by 9.2% compared with placebo after 3 years, in addition to a significant reduction in vertebral and non-vertebral fractures.¹⁸ Denosumab has also been demonstrated to improve bone microarchitecture¹⁹ and reduce fracture risk²⁰ in postmenopausal women. The effects of denosumab on BMD and bone microarchitecture in women with anorexia nervosa are not known, but denosumab may be an ideal therapy in this population because increased RANK ligand signaling is thought to contribute to the high bone resorption characteristic of this disease.

Our primary objective was to determine whether off-label antiresorptive therapy with denosumab increases BMD (primary endpoint: PA lumbar spine aBMD at 12 months) and improves bone microarchitecture and estimated bone strength (secondary endpoints at 12 months) compared to placebo in ambulatory women with anorexia nervosa. We hypothesized that denosumab would increase lumbar spine aBMD and improve bone microarchitecture and estimated bone strength compared to placebo. Our secondary objective was to determine the change in aBMD within groups over 12 months.

Materials and Methods

Thirty ambulatory women with anorexia nervosa were recruited from local eating disorder centers or referred by clinical providers between October 2017 and August 2019. Women were randomly assigned (2:1) to either denosumab or placebo at a single site (Massachusetts General Hospital, Boston, MA) in this double-blind study (ClinicalTrials.gov NCT03292146) (Figure 1). Mass General Brigham Human Research Committee institutional review board approved the study, and all participants gave written informed consent before study participation. None of the following data have been previously reported.

Figure 1.

Consort diagram.

All participants met criteria for low-weight anorexia nervosa or atypical anorexia nervosa, either restrictive or binge-purge subtype, according to the Structured Clinical Interview for Diagnostic and Statistical Manual of Mental Disorders, 5^th edition.²¹ The study included both women with low-weight anorexia nervosa (BMI <18.5 kg/m²) and atypical anorexia nervosa, who are not low weight (BMI ≥18.5 kg/m²) but meet psychiatric criteria for the disease, since low BMD is a common comorbidity in both disorders³. Other inclusion criteria included: age 20-60 years; BMD T-score <−1.0 at the spine, hip, or radius; 25OH vitamin D >30 ng/mL; normal serum calcium level; dental exam within the past 12 months; and agreement to use a medically effective form of contraception throughout the study. Exclusion criteria included: diseases elicited on medical history that are known to affect bone including untreated thyroid dysfunction, Cushing’s syndrome, or renal failure; medications known to affect bone metabolism within 3 months of the study, including corticosteroids but excluding exogenous estrogen (bisphosphonates must have been discontinued for at least 1 year prior to participation); immunodeficiency; immunosuppressive therapy; serum potassium <3.0 mEq/L; serum alanine aminotransferase (ALT) >3 times the upper limit of normal; estimated glomerular filtration rate (eGFR) <30 mL/min; diabetes mellitus; active substance use disorder; history of malignancy; osteonecrosis of the jaw (ONJ) or risk factors for ONJ, such as an invasive dental procedure within the past 6 months or planned within the next 12 months or other at-risk pre-existing dental disease; pregnancy, breastfeeding, or planning to become pregnant within 7 months after the end of the study.

Eligibility was determined at a screening visit, which included a medical history and physical examination. Weight and height were measured and body mass index (BMI) was calculated. Blood was drawn for comprehensive metabolic panel (CMP), complete blood count (CBC), and 25OH vitamin D, and a urine pregnancy test was completed. Dual energy x-ray absorptiometry scans of the lumbar spine, hip, and total body were performed unless the participant had had a clinical DXA scan performed within 6 months of the screening visit (n=14, DXA scan performed a mean of 9 weeks before screening visit).

Eligible participants were randomized by the research pharmacy to denosumab or placebo using a random allocation sequence. Two randomization lists were created, one for BMI <18.5 kg/m² and one for BMI ≥18.5 kg/m². Each randomization list contained 63 randomization numbers organized into 21 blocks of 2:1 denosumab: placebo, but not all blocks were used. The randomization list for BMI <18.5 kg/m² used 14 randomization numbers (4 and 2/3 blocks), while the randomization list for BMI ≥18.5 kg/m² used 16 randomization numbers (5 and 1/3 blocks). Only the pharmacy staff was unblinded to each participant’s treatment assignment. Amgen, Inc. provided an unblinded kit assignment indicating which package IDs were denosumab and which were placebo. Unblinded pharmacy staff referred to the randomization list, confirmed treatment assignment, and retrieved the appropriate package ID corresponding to each participant’s treatment assignment. Unblinded pharmacy staff labeled the syringe in a blinded manner, i.e., “denosumab or placebo 60 mg/mL prefilled syringe”. Placebo was identical in appearance to active denosumab.

The baseline visit was completed within 10 weeks of the screening visit, except for 2 participants who baselined 13 weeks after their screening visit due to low 25OH vitamin D levels that required significant supplementation. At the baseline visit, study staff updated each participant’s medical history and performed a physical exam, including height and weight. Blood was drawn for 25OH vitamin D, calcium, ALT, and potassium, and a urine pregnancy test was completed. If a DXA scan had not been performed at the screening visit or within 4 weeks of the baseline visit, lumbar spine, hip, and total body DXA scans were performed. High-resolution peripheral quantitative computed tomography (HR-pQCT) of the ultra-distal radius and tibia was performed. The Paffenbarger exercise questionnaire was administered, and hours of moderate or vigorous physical activity per week was reported.²² Participants received the following intervention based on their randomization assignment:

“Denosumab” group: Active denosumab 60mg subcutaneous injection (Amgen Inc., Thousand Oaks, CA) administered at 0 and 6 months.

“Placebo” group: Placebo subcutaneous injection administered at 0 and 6 months.

In addition, all participants received calcium 1200 mg and vitamin D3 800 IU to take daily for the duration of the study.

Ten days after each subcutaneous injection, safety labs including a serum calcium from a CMP, were measured. A clinical evaluation; nutrition evaluation including weight in gown, height, and calculation of percent ideal body weight and BMI; urine pregnancy test; and blood work for CBC and CMP were performed at 1, 3, 6, 7, 9, and 12 months after the baseline visit. At 6 and 12 months, DXA and HR-pQCT scans were repeated, and blood was drawn for markers of bone metabolism. Due to the nature of participants’ psychiatric disease, fasting for blood draws was not required. To ensure participant safety in the study, all participants were required to have an active treatment team, and participants’ weight and safety labs were monitored by a research nurse practitioner during the study. Drop criteria during the study were pregnancy, medical instability as determined by study staff, a serious study-related adverse event, or an invasive dental or periodontal procedure.

Standardized clinical methods were used to measure CBC and CMP. A liquid chromatography/tandem mass spectrometry (LC/MS-MS) assay certified by the National Institutes of Health Vitamin D Standardization Certification program was used to measure serum 25OH vitamin D. Chemiluminescence assays were used to measure serum procollagen type I N-terminal propeptide (P1NP) (sensitivity <1.0 ng/mL, inter-assay variability 5%, intra-assay variability of 3%) and C-terminal telopeptide of type I collagen (CTX) (sensitivity 0.023 ng/mL, inter-assay variability 6%, intra-assay variability 3%) (Immunodiagnostic Systems, Tyne & Wear, UK), which were sent in batch at the end of the study from samples stored in a −80°C freezer.

Body composition (i.e. total body lean and fat mass) and aBMD at the PA lumbar spine and hip were assessed by DXA with a precision of 0.02 g/cm² at the lumbar spine for aBMD²³ and <2% for lean mass²⁴ (Hologic Horizon A, Hologic, Inc., Waltham, MA). A co-investigator (MSH) reviewed all DXA scans and excluded vertebrae from all timepoints for that participant if deformities were present. Reference data were as follows: Hologic (PA lumbar spine BMD),²⁵ National Health and Nutrition Examination Survey (NHANES) phase II dataset (total hip BMD),²⁶ NHANES phase I dataset (femoral neck BMD),²⁷ and 1999-2004 NHANES dataset (body composition).²⁸ Least significant change (LSC) at 95% confidence level, which is the smallest difference that can confidently be discerned from precision error from two separate scans in an individual patient, were obtained from the Massachusetts General Hospital Translational and Clinical Research Center.

High resolution peripheral quantitative CT (HR-pQCT, XtremeCT I; Scanco Medical AG, Brüttisellen, Switzerland) was used to measure ultra-distal radial and tibial volumetric BMD and microstructure with an isotropic voxel size of 82 μm³, as previously reported.^{29, 30} The non-dominant arm or leg was scanned; the contralateral side was scanned if there had been a prior fracture in that region. Microfinite element analysis (FEA) was used to estimate the bone’s biomechanical properties in the setting of simulated axial compression.³¹ Failure load (in Newtons) was estimated by scaling the resultant load from a 1% apparent compressive strain until 2% of all elements reached an effective strain >7000 μstrain per previously published methods.³² All scans were scored 1 (no movement) to 5 (severe motion) for motion artifacts, and those with scores of 4 to 5 were excluded from further analyses. Eight radius and no tibia scans were excluded due to movement artifact. Follow-up scans were matched to baseline using 2D registration, as provided by the manufacturer’s software. The median overlap was 90% at the radius and 96% at the tibia. Zero scans were excluded for matching < 75%. Scan acquisition, analysis and reporting followed published guidelines.³³ Least significant change at 95% confidence level were obtained from published literature.³⁴

Statistical analysis

Statistical analyses were performed using JMP Statistical Discoveries (JMP 14Pro, SAS Institute, Cary, NC). Baseline characteristics are reported as mean ± SD. The primary analysis was an intent-to-treat mixed effect model with randomization assignment and timepoint as fixed effects, participants as random effects, and variance component as the correlation structure for repeated measures; data from interim visits were used in participants who dropped out before the 12-month visit. The assumption of normality of the error term in the mixed model was checked using conditional residual plots. The pre-specified primary endpoint of the primary analysis was the difference between groups in PA spine aBMD at 12 months. Pre-specified secondary endpoints of the primary analysis included: total hip, femoral neck aBMD; radial and tibial total, cortical, and trabecular BMD; radial and tibial cortical thickness, cortical porosity, trabecular thickness, trabecular number, trabecular separation; and radial and tibial failure load. Exploratory subgroup analyses of the primary analysis comparing effect sizes ± 95% confidence intervals between the following groups were performed: baseline PA lumbar spine aBMD Z-score <−1.5 vs. Z-score >−1.5, baseline amenorrhea (no menstrual period for ≥ 3 months not due to contraceptive method) vs. eumenorrhea (or on exogenous estrogen), and weight gain vs. weight loss over 12 months.³⁵ A pre-specified secondary analysis of change in aBMD within groups was performed using paired t-tests (results were then scaled from absolute change in aBMD to percent change in aBMD in Figure 3 in order to be able to visually compare changes in spine and hip aBMD); multivariate linear regression was performed within groups to assess for determinants of change in aBMD. Two outliers, 1 in the denosumab group and 1 in the placebo group, identified using quantile analysis were excluded from radial bone microarchitecture analyses. Multiple comparisons were controlled for using Benjamini-Hochberg’s method. Differences in adverse event incidence were analyzed using the Fisher exact test. Statistical significance was defined as a two-tailed P value of ≤0.05. Baseline data are presented as mean ± SD. Effect sizes are presented as mean (95% CI).

Figure 3.

(A) Postero-anterior (PA) lumbar spine aBMD increased by 5.5 (3.8 – 7.2)% in the denosumab group and 2.2 (−0.3 – 4.7)% in the placebo group over 12 months. (B) Femoral neck aBMD increased by 3.3 (1.4 – 5.1)% in the denosumab group and 2.3 (−0.2 – 4.9)% in the placebo group over 12 months. In the box and whisker plots, boxes represent the median and interquartile range, and whiskers represent the minimum and maximum values.

An a priori power calculation for the primary endpoint of the primary analysis was as follows: group sample sizes of 20 (denosumab) to 10 (placebo) for a total n=30, accounting for a 10% drop-out rate, provided 87% power to reject the null hypothesis of equal means when the population mean difference is 4.5 with a standard deviation of both groups of 3.4 and with a significance level of (alpha) of 0.05 using a 2-sided 2-sample equal-variance t-test. This was based on the difference in percent change in spine aBMD between the denosumab and placebo groups in McClung et al.³⁶

Results

Of the 43 women who provided informed consent, 30 were eligible and randomized and 13 were excluded because they were ineligible, declined to participate, or were lost to follow-up (Figure 1). Participants who declined to participate or were lost to follow-up between signing informed consent and randomization had shorter disease duration than participants who were randomized, but all other characteristics were similar between the groups (Appendix Table 1).

In the denosumab group (n=20), 2 participants discontinued the intervention early, and 18 participants completed the 12-month study. In the placebo group (n=10), 1 participant discontinued the intervention early, and 9 participants completed the 12-month study.

Table 1 shows the clinical characteristics of the study participants at baseline. The age range was 20-51 years. Overall, 50% of participants had atypical anorexia nervosa (BMI ≥ 18.5 kg/m²) and 30% were amenorrheic. None of the participants had taken IV or oral bisphosphonates in the past. Over 12 months, PA lumbar spine aBMD increased in the denosumab vs. placebo group (Figure 2A). Descriptively, the mean (95% CI) increase in PA lumbar spine aBMD was 5.5 (3.8 – 7.2)% in the denosumab group and 2.2 (−0.3 – 4.7)% in the placebo group (Figure 3A); the absolute mean change in PA lumbar spine aBMD exceeded the LSC of 0.022 g/cm² in the denosumab group (0.046 ± 0.028 g/m²) but not the placebo group (0.016 ± 0.033 g/m²). Least squares mean (95% CI) PA lumbar spine aBMD Z-scores increased from −1.6 (−2.1 – −1.2) to −1.2 (−1.6 – −0.7) in the denosumab group and from −1.7 (−2.4 – −1.1) to −1.5 (−2.2 – −0.9) in the placebo group. Femoral neck aBMD was not different between the groups over 12 months (Figure 2B). Descriptively, the mean (95% CI) increase in femoral neck aBMD was 3.3 (1.4 – 5.1)% in the denosumab group and 2.3 (−0.2 – 4.9)% in the placebo group (Figure 3B); the absolute mean change in femoral neck aBMD did not exceed the LSC in either group. Least squares mean aBMD Z-scores increased at the femoral neck from −1.2 (−1.6 – −0.8) to −1.0 (−1.4 – −0.6) in the denosumab group and from −1.7 (−2.3 – −1.1) to −1.5 (−2.1 – −1.0) in the placebo group. Within groups, the following variables were not associated with change in PA lumbar spine or femoral neck aBMD over 12 months: baseline age, lean mass, amenorrhea, and moderate + vigorous physical activity per week, and change in weight over 12 months.

Table 1.

Baseline characteristics

	Denosumab (n=20)	Placebo (n=10)	Overall P-value
Clinical characteristics
Age, y	29.2 ± 8.0	29.4 ± 6.8	0.91
Caucasian, n (%)	19 (95)	10 (100)	1.00
BMI, kg/m2	19.0 ± 1.7	18.0 ± 2.1	0.23
Lowest past BMI, kg/m2	15.9 ± 1.2	14.6 ± 1.5	0.05
Duration of anorexia nervosa, y	14.3 ± 9.1	14.8 ± 8.5	0.39
Restrictive subtype, n (%)	16 (80%)	8 (80%)
Physical activity, moderate or vigorous, hrs/wk	25.8 ± 18.3	26.8 ± 19.0	0.90
Amenorrheic and not on exogenous estrogen, n (%)	6/20 (30%)	4/9 (44%)*	1.00
Serum calcium, mg/dL	9.4 ± 0.3	9.6 ± 0.3	0.06
Serum 25OH vitamin D, ng/mL	45.8 ± 12.4	45.0 ± 14.5	0.88
Body composition
Lean mass, kg	37.2 ± 4.3	33.8 ± 4.7	0.08
Fat mass, %	26.4 ± 5.5	29.6 ± 3.5	0.05
Areal BMD
Postero-anterior spine, g/cm2	0.86 ± 0.09	0.85 ± 0.16	0.92
Postero-anterior spine, Z-score	−1.6 ± 0.9	−1.7 ± 1.4	0.83
Total hip, g/cm2	0.81 ± 0.12	0.75 ± 0.12	0.20
Total hip, Z-score	−1.0 ± 0.9	−1.5 ± 0.9	0.18
Femoral neck, g/cm2	0.71 ± 0.11	0.65 ± 0.12	0.22
Femoral neck, Z-score	−1.2 ± 1.0	−1.7 ± 1.0	0.18
Bone Metabolism Markers
CTX (0.03-6.0), ng/mL	0.36 ± 0.24	0.29 ± 0.18	0.66
P1NP (2-230), ng/mL	80.5 ± 41.3	60.5 ± 26.0	0.12
Osteocalcin, ng/mL	14.3 ± 6.1	12.8 ± 4.4	0.44
HRpQCT
Radius
Total bone mineral density, mg/cm3	306.1 ± 70.7	266.2 ± 55.6	0.11
Cortical bone mineral density, mg/cm3	883.5 ± 54.2	846.2 ± 57.9	0.11
Cortical thickness, mm	0.78 ± 0.18	0.65 ± 0.16	0.07
Cortical porosity, %	0.82 ± 0.39	0.79 ± 0.32	0.79
Trabecular bone mineral density, mg/cm3	137.7 ± 32.8	122.0 ± 21.3	0.13
Trabecular bone volume to tissue volume ratio	0.11 ± 0.03	0.10 ± 0.02	0.14
Trabecular number, 1/mm	1.72 ± 0.30	1.69 ± 0.19	0.69
Trabecular separation, mm	0.53 ± 0.11	0.54 ± 0.08	0.65
Trabecular thickness, mm	0.07 ± 0.01	0.06 ± 0.01	0.07
Failure load, N	3369 ± 648	2844 ± 391	0.01
Tibia
Total bone mineral density, mg/cm3	279.1 ± 54.2	255.2 ± 60.0	0.30
Cortical bone mineral density, mg/cm3	890.1 ± 36.4	875.8 ± 44.6	0.39
Cortical thickness, mm	1.08 ± 0.18	0.99 ± 0.22	0.27
Cortical porosity, %	3.33 ± 1.59	2.77 ± 2.04	0.32
Trabecular bone mineral density, mg/cm3	152.9 ± 45.0	136.6 ± 45.6	0.36
Trabecular bone volume to tissue volume ratio	0.13 ± 0.04	0.11 ± 0.04	0.36
Trabecular number, 1/mm	1.64 ± 0.38	1.55 ± 0.31	0.53
Trabecular separation, mm	0.57 ± 0.17	0.60 ± 0.17	0.56
Trabecular thickness, mm	0.08 ± 0.02	0.07 ± 0.01	0.33
Failure load, N	9226 ± 1875	8016 ± 1619	0.08

Values reported as mean ± SD.

Abbreviations- BMI: body mass index. PINP: amino-terminal propeptide of type I procollagen. CTX: C-terminal telopeptides of type I collagen. HR-pQCT: high resolution peripheral quantitative computed tomography.

^*1 participant with progestin intrauterine device (IUD) was excluded from amenorrhea analysis as her reproductive status could not be determined

Figure 2.

(A) Postero-anterior (PA) lumbar spine areal BMD (aBMD) by DXA increased in the denosumab vs placebo group over 12 months. (B) Femoral neck aBMD did not differ between the groups over 12 months. Data are displayed as LS mean and SEM. P-values represent differences between groups.

Radial cortical BMD by HR-pQCT decreased less in the denosumab vs. placebo group over 12 months (Figure 4A), which was not significant after controlling for multiple comparisons. There were no significant increases in radial trabecular BMD, or tibial cortical or trabecular BMD, in the denosumab vs. placebo groups over 12 months (Figure 4B and Figure 5A-B). Radial trabecular number increased, and trabecular separation decreased, in the denosumab vs. placebo group over 12 months (Figure 4C-D). Tibial cortical porosity decreased in the denosumab vs. placebo group (Figure 5C). There were no differences between groups in any of the other HR-pQCT variables or failure load. The mean percent change in HR-pQCT variables in either group did not exceed the LSC reported in the literature.

Figure 4.

Radial cortical (A) and trabecular (B) volumetric BMD (BMD) by high resolution peripheral quantitative computed tomography (HR-pQCT) did not differ between the denosumab vs. placebo group over 12 months, but radial trabecular number (C) increased, and radial trabecular separation (D) decreased, in the denosumab vs. placebo group over 12 months. Radial total BMD, cortical thickness, cortical porosity, trabecular thickness, and failure load did not differ between the groups (data not shown). Data are displayed as LS mean and SEM. P-values represent differences between groups.

Figure 5.

Tibial cortical (A) and trabecular (B) volumetric BMD (BMD) by high resolution peripheral quantitative computed tomography (HR-pQCT) did not differ between the denosumab vs. placebo group over 12 months, but tibial cortical porosity (C) decreased in the denosumab vs. placebo group over 12 months. Tibial total BMD, cortical thickness, trabecular thickness, trabecular number, trabecular separation, and failure load did not differ between the groups (data not shown). Data are displayed as LS mean and SEM. P-values represent differences between groups.

As expected, serum CTX and P1NP decreased in the denosumab vs. placebo groups over 12 months (Figure 6A-B). Least squares mean (95% CI) serum CTX and P1NP at 12 months were 0.07 (0 – 0.18) ng/mL and 21.9 (8.3 – 35.5) in the denosumab group, and 0.34 (0.22 – 0.45) and 67.5 (48.2- 86.8) ng/mL in the placebo group, respectively.

Figure 6.

Serum C-terminal telopeptide of type I collagen (CTX) (A) and procollagen type I N-terminal propeptide (P1NP) (B) decreased in the denosumab vs placebo group over 12 months. Data are displayed as LS mean and SEM. P-values represent differences between groups.

A series of stratified analyses were performed to investigate predictors of change in aBMD. 95% CI of the effect sizes over 12 months in all bone variables overlapped between the following groups: (1) participants who had a baseline PA lumbar spine BMD Z-score <−1.5 (n=9 denosumab, n=7 placebo) vs. those who did not, (2) participants who were amenorrheic at baseline (n=6 denosumab, n=4 placebo) vs. those who were not, and (3) participants who gained weight (n=13 denosumab, n=6 placebo) vs. those who did not. Mean weight increased by 2.7 (−0.8 – 6.3)% and 5.5 (−2.0 – 12.9)% in the denosumab and placebo groups, respectively (p=0.71).

For the primary endpoint of PA lumbar spine aBMD, two sensitivity analyses were performed One included only women ≤40 years old (n=28) in order to remove potential confounding effects of perimenopause and menopause, as menopausal status can be difficult to assess in women with anorexia nervosa and a history of irregular menses/amenorrhea or on hormonal replacement. The primary endpoint of PA lumbar spine BMD remained higher in the denosumab vs placebo group at 12 months (p=0.004). The second sensitivity analysis included only women who did not have a change in hormonal replacement during the study (n=26). The primary endpoint of PA lumbar spine BMD remained higher in the denosumab vs placebo group at 12 months (p=0.047).

Denosumab was well tolerated. The frequency of adverse events was similar between the groups. Hypocalcemia was not detected in any participant assigned to the denosumab group. In the denosumab group, one participant experienced a stress fracture of the right tibia in the setting of running, one participant reported chronic low back pain and new sciatica (which lumbar spine MRI revealed was due to a herniated disc), and three participants reported constipation.

Discussion

In this study, we demonstrated that 12 months of denosumab increased PA lumbar spine aBMD, our primary study endpoint, more than placebo in women with anorexia nervosa. This is clinically salient since the lumbar spine is the most common and most severely affected site of bone loss in women with this disease.³ As a RANK ligand inhibitor, denosumab addresses an underlying pathophysiologic mechanism of low BMD in anorexia nervosa, namely the elevated soluble RANK ligand levels from paradoxically elevated marrow adipose tissue, both of which have been associated with abnormal bone metabolism.³⁷ Consistent with its mechanism of action, denosumab significantly reduced bone resorption compared to placebo. These data suggest that in women with anorexia nervosa, antiresorptive therapy with denosumab may be an effective therapy to increase BMD.

Twelve months of denosumab in this study increased PA lumbar spine aBMD and improved some bone microarchitecture variables, specifically radial trabecular number and separation and tibial cortical porosity, in women with anorexia nervosa. The mean percent increase of 5.5% in PA lumbar spine aBMD in the denosumab group was similar to studies in postmenopausal osteoporosis, in which denosumab increased PA lumbar spine aBMD by 3.0-6.7% over 12 months^{17, 36} and reduced vertebral and nonvertebral fractures.²⁰ Unlike studies in women with postmenopausal osteoporosis, we did not detect a significant increase in femoral neck aBMD, which may have been a type II error. In contrast to our results, previous studies investigating the effects of 12 or 24 months of denosumab on bone microarchitecture in women with postmenopausal osteoporosis demonstrated increases in radial and tibial total, cortical, and trabecular BMD, and cortical and trabecular thickness, but no change in radial or tibial cortical porosity, trabecular number, or trabecular separation.^{19, 38} This may be because mechanisms of bone loss differ in anorexia nervosa vs. postmenopausal osteoporosis, e.g., bone formation is suppressed in women with anorexia nervosa but not in postmenopausal women. The radial HR-pQCT variables that demonstrated improvement with denosumab in our study are known to be impaired in women with anorexia nervosa compared to healthy controls,⁴ which suggests these findings may be of clinical relevance.

We demonstrated improvements in more radial vs. tibial HR-pQCT variables with denosumab in women with anorexia nervosa. This is in contrast to previous studies of antiresorptive therapies in postmenopausal women which saw improvements in more tibial vs. radial HR-pQCT variables.^38-42 It is thought that weight-bearing and mechanical loading on the tibia reduces sclerostin production by osteocytes,⁴³ thereby better maintaining osteoblast function at the tibia during antiresorptive therapy. In women with anorexia nervosa, despite physical activity, the protective effect of weight-bearing on the tibia may be reduced given their current, or history of, low weight and other endocrine effects of chronic undernutrition. Previous literature demonstrating that women with anorexia nervosa have more impairments in tibia vs radial HR-pQCT variables compared to healthy controls is consistent with this.⁴

As expected given its potent antiresorptive properties,¹⁷ denosumab suppressed bone resorption (and thus also bone formation) compared with placebo. Of note, although anorexia nervosa in adult women is typically characterized by low bone formation (and high bone resorption), there was significant variability in baseline mean P1NP levels in this study, which is consistent with another recent study in women with anorexia nervosa.⁴⁴ Such variability in P1NP values could be due to differences in nutritional, weight, or menstrual status. One of the major concerns with denosumab therapy is the rebound in bone turnover with discontinuation of therapy, which results in bone loss⁴⁵ and an increased risk of vertebral fractures⁴⁶ if denosumab discontinuation is not followed with additional therapy. Of the 27 women who completed 12 months of denosumab or placebo, 26 consented to receive 12 months of open-label alendronate (1 participant did not consent as she was planning pregnancy; her randomization assignment was unblinded at the 12-month visit and she had been on placebo). Results of that extension study are pending.

The changes in the placebo group were as expected. Weight and aBMD did not significantly change in the placebo group over 12 months, despite the study requirement for all participants to have an active treatment team. This is consistent with prior literature that reported that both weight gain and restoration of menses are needed to increase PA lumbar spine aBMD in women with anorexia nervosa.⁴⁷ The requirement to have a treatment team and the extra healthcare interactions received in the research study may have prevented the placebo group from losing BMD, as observational studies that shown that women with active disease lose lumbar spine BMD at a rate of 2.6%/year.⁴⁷ These data highlight the need for FDA-approved therapies for low BMD in this population above and beyond treatment of the underlying disease.

Our findings should be considered in light of some limitations. Due to the nature of participants’ psychiatric disease, fasting blood draws were not required. However, feeding has been reported to decrease P1NP by only about 4%.⁴⁸ The study was powered to detect a difference in PA lumbar spine aBMD between groups, but we were underpowered to detect a difference in many HR-pQCT measurements given the inherent variability in these measurements, and the mean percent change in HR-pQCT variables in either group did not exceed the LSC reported in the literature. Changes that were seen in DXA and HR-pQCT variables could also be due at least in part to contraction of the reversible remodeling space, increased mineralization density, and/or edge detection changes with denosumab vs placebo. Although descriptive stratified analyses suggest that the effect of denosumab vs. placebo on bone variables may be similar regardless of baseline PA lumbar spine BMD, baseline menstrual status, or weight gain during the study, the study was not powered to test for such interactions, and thus conclusions from these exploratory analyses cannot be made. A larger study with a longer duration of follow-up is needed to determine whether denosumab increases BMD, microarchitecture, and estimated strength by HR-pQCT.

The generalizability of our findings is enhanced by the inclusion of both women with anorexia nervosa and women with atypical anorexia nervosa. However, bone metabolism is different in adolescents with anorexia nervosa, namely that both bone resorption and bone formation are suppressed. Therefore, the results of our study in women cannot be generalized to adolescent girls. Women with anorexia nervosa are often overwhelmed by the requirements of therapy and the illness itself. Therefore, a treatment such as denosumab, which can be administered every 6 months, may have the benefit of high compliance, and therefore effectiveness.

In conclusion, in this randomized, placebo-controlled study, 12 months of antiresorptive therapy with denosumab reduced bone turnover and increased spine aBMD, the skeletal site most severely affected in women with anorexia nervosa. Larger, longer-term studies are needed to determine whether denosumab increases BMD, microarchitecture, and estimated strength by HR-pQCT. In addition, whether bisphosphonates can consolidate gains in BMD, and prevent rebound bone loss, after denosumab therapy is an important clinical question.