Bone Mineral Density at Extremely Low Weight in Patients With Anorexia Nervosa

Pauline Bemer; Laura Di Lodovico; Ohanyan Haykanush; Hélène Théodon; Karine Briot; Robert Carlier; Marika Dicembre; Maéva Duquesnoy; Jean-Claude Melchior; Mouna Hanachi

Disclosures

Clin Endocrinol. 2021;95(3):423-429. 

In This Article

Methods

Design and Ethics

A retrospective, single-centre study was conducted on a population of consecutively admitted inpatients suffering from AN, in the nutrition unit of Raymond Poincaré Hospital in Garches (France) from January 2014 to April 2016.

This study was declared to the French national committee of informatics and liberties (CNIL) under the registration number 2029030vO. This research was conducted in accordance with the relevant French guidelines and regulations. Regular mailing sent individual information to all patients. Patient non-opposition was a prerequisite for the use of their data.

Population

A population of extremely undernourished inpatients with AN was included in this study. The following criteria should be fulfilled for inclusion: 1) a diagnosis of AN according to the DSM-5 criteria;[1] 2) female gender; 3) age≥16 years; and 4) having at least one BMD measure by dual-energy X-ray absorptiometry (DXA).[17] Patients were excluded if they presented any of the following conditions: BMI≥16 kg/m2; age older than 50 years; and any comorbidity associated with BMD loss or low bone mass. In total, 97 patients were included in the study.

Variables

All data were collected from medical records upon admission, before starting nutritional management.

Anamnestic data encompassed the type of AN, duration of illness, the existence of amenorrhea and its duration, age of menarche, weight history and active smoking. Clinical data collected were age, sex, height, weight, oral contraception and treatments susceptible to have an effect on bone density such as proton-pump inhibitors (PPI),[18] antidepressants[19] and bisphosphonates.[20] Weight was recorded on the first morning after admission. Height was recorded in metres using a wall-mounted stadiometer. The Hamwi equation was used to determine per cent of excess or deficit of the ideal body weight;[9,11] duration of illness was ascertained by patient self-reported age of onset of eating disorder behaviours.

The following biological data were collected at baseline: nutritional markers (transthyretin, albumin, CRP), 25-OH vitamin D3, sodium, calcium, phosphates, alkaline phosphatases and plasma concentrations of cortisol at 8 AM

DXAs were performed for each patient in the same conditions by Hologic® osteodensitometer model QDR 4500 with phantom daily control, calibrated on a Caucasian population. The following data were extracted directly from the device's information processing software (Hologic QDR for Windows version 12.6): BMD of lumbar spine, femoral neck and whole body, fat mass (FM) and fat-free mass (FFM). The severity of low BMD was estimated by Z-score measures. For each patient, Z-score was automatically calculated according to the formula "z = (x-μ)/σ", where x is the raw BMD value, μ is the mean of BMD of a healthy population of same age and gender preregistered in the software database, and σ is the standard deviation of this population.[21] This measure allows a comparison between BMD of patients and average values of healthy individuals of same age and gender. In this study, Z-scores were preferred to T-scores as the indicated measure for patients aged less than 50 years old by the International Society for Clinical Densitometry.[7,22,23] Low BMD was diagnosed for Z-scores ≤−2.0 SD.

History and topography of fractures were retrospectively assessed by systematic screening of patients' medical records.

Statistical Analysis

Variables were assessed for normal distribution by visual inspection of histograms and boxplots.

Pearson coefficients were calculated to identify correlations between BMD and clinical-biologic variables. Student's t test for independent samples was used to compare patients with fractures to those with no fracture. Linear and logistic regression models were built to find predictors of low BMD and fractures. Aberrant values were converted into missing data before statistical analyses. Missing data were treated by multiple imputation and chained equations.

All analyses were implemented in R software (version 3.5.1. 2018-07-02, x86_64-pc-linux-gnu, R Foundation for Statistical Computing, Vienna, Austria). A p-value lower than 0.05 was considered as statistically significant.

processing....