Internal consistency and reliability

The first version of the LEAF-Q included 29 items distributed on five main variables (injuries, dizziness, cold sensitivity, gastrointestinal function and menstrual dysfunction (MD), including questions regarding the use of OC). Since these endurance athletes and dancers (n=37) were of English, Swedish and Danish origin, the LEAF-Q was, from the start, translated by two bilingual employees at the University of Copenhagen. A test–retest was performed to assess item performance and estimate reliability. The participants received the LEAF-Q twice within a 2-week period and a letter of information to which the participants were asked to reply within 2 days. After filling out the LEAF-Q the second time, the participants were interviewed to assess response bias. The participants were asked to comment whether questions or response categories were not clearly stated, whether they perceived any question as irrelevant, whether response options were inadequate and whether the volume of the questionnaire was appropriate. One variable, illness, was added.

Verification of self-reported symptoms

The version of the LEAF-Q used in the next part of the validation process included 14 items in the introduction, including occupation, type of sport, age, height, weight and training, followed by 30 items divided into six variables: dizziness, gastrointestinal function, cold sensitivity, illnesses (during the last year), injuries (during the previous year) and menstrual function (present and in the past). The variables for dizziness and cold sensitivity included one item assessed by self-reporting the frequency of the symptom on a Likert-type ordinal scale: (1) Yes, several times a day; (2) Yes, several times a week; (3) Yes, 1–2 times a week and (4) Rarely or never. The injury and illness variable contained three items for which the frequency and duration were assessed by ordinal scales and an open category to specify the types of injuries/illness. The open response for the variable of injuries was divided into two main types of injuries—accidental or overload. The variable for gastrointestinal function contained four items and was assessed by self-reporting gastrointestinal symptoms such as pain/cramps/bloating on a Likert-type ordinal scale. Nominal scales assessed stool frequency and consistency. Menstrual function included 12 items and was measured on dichotomous as well as ordinal scales. Age of menarche was measured on an ordinal scale and previous MD on a nominal scale. Bleeding pattern related to exercise was measured on dichotomous and nominal scales. The use of hormonal contraceptives contained six items and was assessed on dichotomous and nominal scales. After filling out the LEAF-Q, the endurance athletes (n=47) were invited to take part in a standardised verification programme to verify self-reported symptoms by assessment of EA, clinical verification of DE/ED, orthostatic BP, hypothyroidism, reproductive function and bone health.

Dietary intake and training intensity were recorded by the participants for seven consecutive days to assess EA.19 The participants were instructed to maintain and follow their normal eating pattern and training regime. Energy intake was calculated from weighed food records using a nutrient analysis programme, Dankost 2000 (Dankost, Copenhagen, Denmark), for Danish foods and Dietist XP (Kost och Näringsdata AB, Bromma, Sweden) for Swedish food items.

The participants were instructed to maintain and follow their normal training regime. Heart rate (HR) monitors (Polar RS400, Stockholm, Sweden) were used to assess exercise energy expenditure (EEE) based on individual prediction equations from measured HR and corresponding energy expenditure (EE) during an incremental maximal exercise test in the laboratory. The individual equation provided the basis for the calculation of EEE using HR measurement for each training session.20 Regression lines were calculated for the corresponding values of HR and EE during the incremental exercise test in the laboratory. HR were highly correlated with O₂ consumption during increasing workloads (r=0.94, 95% CI 0.93 to 0.96). EA was calculated by subtracting mean EEE from the mean energy intake, after calculating EEE by subtracting from total EEE the total EE during an equivalent time period without exercise. To calculate daily total EE, HR monitors (Polar RS400) were used to assess EE during bicycle transportation; actigraphy (ActiGraph GT3X, Pensacola, Florida, USA) and the data analysis software ActiLife 5 (ActiGraph) were used for the assessment of non-exercise physical activity thermogenesis. The participants were instructed to wear an accelerometer on the wrist during sleep, and on the hip from the time they wake up in the morning until bedtime, taking it off only during showering, swimming, bicycle transportation and training.

Eating behaviour was assessed using the Eating Disorder Inventory (EDI-3), a questionnaire assessing behaviour and attitudes related to DE behaviour.21 Participants were categorised as having DE behaviour when the subscale drive for thinness was ≥14 and/or body dissatisfaction ≥19.21 The Eating Disorder Examination (EDE-16.0)22 was used to determine whether participants met the criteria for ED according to the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) criteria for anorexia nervosa, bulimia nervosa and ED not otherwise specified. All interviews were performed by the same EDE-certified member of the research team.

Participants using OC were requested to stop for a minimum of six weeks prior to examination in order to secure a sufficient washout period for exogenous oestrogen and progesterone. Participants not recovering their menstrual bleeding within the six weeks were monthly contacted by the research team during a follow-up period of minimum of three months before gynaecological assessment. Menstruating athletes were examined in the early follicular phase, defined as the third to fifth day of menstruation. A pregnancy test was performed and menstrual function was examined by an experienced gynaecologist who performed a transvaginal ultrasound examination. The maximum number of ovarian follicles present in a single plane was counted and the total volume of these follicles was assessed. Sex hormone status (oestrogen, LH, follicle stimulating hormone (FSH) and androgens) and anamnestic assessment, including age of menarche, OC and previous MD were recorded. Participants were then classified as eumenorrhoea (menstrual cycles of 28 days±7 days); oligomenorrhoea (menstrual cycles >35 days); FHA (either primary; no menarche after 15 years of age, or secondary; absence of ≥3 consecutive menstrual cycles); other MD (anatomic defects, hyperprolactinemia or other dysfunctional ovarian conditions); polycystic ovary syndrome (PCOS; ≥2 of the following criteria fulfilled: (1) enlarged ovaries with a volume greater than 10 mL and/or ≥one ovary demonstrating ≥12 follicles in one plane, (2) oligomenorrhoea/amenorrhoea and (3) elevated androgen level or otherwise androgen stigmatised).

Body weight was measured, with the participants in underwear, on an electronic scale and height was measured using a fixed stadiometer. Dual-energy X-ray absorptiometry (DEXA; Hologic, Model Discovery 2009, Hologic Inc Waltham, Massachusetts, USA) was used to determine fat-free, fat and bone mass. All measurements and scans were performed in the fasted state between 7:30 and 9:00, and were assessed by the same technician and performed on the same scanner. Calibration of the DEXA was performed weekly using a phantom provided by the manufacturer. The participants were classified as having normal bone mineral density (BMD) when Z-scores were >−1 in all measured sites, low BMD when Z-score was −1 to −2 in at least one site and osteoporosis when Z-score was ≤−2.1

Changes in blood pressure (BP) when moving from a supine to a standing position were measured using a standardised tilt table and an electronic sphygmomanometer (Microlife BP A100, Widnau, Switzerland). After resting in a supine position for 7 min, BP was measured three times before (the mean was used) and once directly after tilting the participants to 70° standing position. Hypotension was defined as a supine systolic BP <90 mm Hg and/or a diastolic BP <60 mm Hg. Orthostatic hypotension was defined as a fall in systolic BP >20 mm Hg and/or a fall in diastolic BP >10 mm Hg when moving from a supine to a standing position.5 ,23

Blood samples were drawn from an antecubital vein in a resting state after an overnight fast. Oestrogen, FSH, LH and androgen were analysed using ADVIA Centaur Immunoassay Systems (Siemens Healthcare Diagnostics Products GmbH, Germany). The analytical sensitivity was 18.4–15 781 pmol/L for oestrogen, 0.3–200 IU/L for FSH, 0.07–200 IU/L for LH and 0.1–28.0 nmol/L for androgen. The intra-assay and interassay precision coefficient of variability (CV) for oestrogen was 2.6% and 4.1%, for FSH 1.2% and 2.0% and for LH 2.9% and 2.3%. The CV for androgen was 4.7% at 7.5 nmol/L. T₃ was analysed using the ARCHITECT system assay (Abbott Laboratories, Longford, Ireland) with an analytical sensitivity of ≤0.25–8 ng/mL and a total assay precision CV of <10%. Capillary blood glucose was analysed using Biosen C Line (EKF Diagnostic, Germany) with a measurement range between 0.5 and 50 mmol/L and a CV of 1.5% at 12 mmol/L. Cortisol was analysed using Roche Electro Chemiluminiscence Immunoassay (ECLI, Roche Diagnostic, Bromma, Sweden). Analytical sensitivity for the cortisol assay was 0.5–1750 nmol/L with an assay precision CV of 1.3–2.1%. Leptin was analysed using Quantikine ELISA (R&D Systems Europe Ltd, Abingdon, UK). The lower analytical limit for leptin was 1.56 ng/mL; lower concentrations are associated with some uncertainty, and therefore not automatically calculated. Seven participants had results <1.56 ng/mL, and in order to calculate their concentrations, the concentration was extrapolated using the formula for the standard curve: log₁₀ (ABS)=A × Log₁₀ (concentration) – B→concentration = 10^{log10 (ABS)+B))/A}. The intra-assay and interassay precision CV for leptin was 3.0–3.3% and 3.5–5.4%, respectively.

Statistical methods

Distributional features of variables were measured on a continuous scale (height, etc) and variable scores were examined for central tendency. Skewed data were log transformed before further analysis. The intraclass correlation coefficient was used to calculate the difference between the test and test–retest scores. In this group of female endurance athletes, oligomenorrhoea/FHA existed with and without the presence of low current EA (<125 kJ/kg FFM/day). Owing to this mismatch, the participants were divided into 50% with the highest current EA and 50% with the lowest current EA, to examine discriminant validity. The statistical procedure for questionnaire validation described by Black et al5 was used, where discriminant validity was assessed by testing the mean item score for each of the six variables for significant difference (two-sample t tests) between the groups with lower EA versus higher EA, MD versus eumenorrhoea and low BMD versus normal BMD. To measure concurrent validity and the degree of association among the total LEAF-Q score, LEAF-Q variables and Triad conditions, Pearson's correlation coefficient (r) was calculated. Furthermore, the contribution of LEAF-Q variables to the different Triad conditions was calculated using an ordinal logistic regression. Pearson's correlation coefficient (r) was also calculated to assess the degree of association between self-reported training hours per week and training hours registered during the week of data collection, as well as self-reported and measured body weight. For comparison of means between the participants categorised at risk for the Triad versus those categorised with low risk, Student's paired t test was used. To test whether there was a difference between the two kinds of classifications, for example, the number of participants categorised at risk for the Triad versus those categorised with low risk having MD, Fisher's exact test was applied. Statistical significance was declared for p<0.05. Internal consistency between questionnaire variables was examined by Cronbach's α.24 The items found to significantly predict lower EA, MD and/or low BMD were retained and summed up to provide an overall LEAF-Q score. To test the validity of the LEAF scale, sensitivity and specificity were calculated.