To be eligible for inclusion in this study, patients had to have been placed on NIV indication defined according to current criteria, 13 for at least 24 hours and up to 3 months (± 1 week). NIV had to be considered to be adequate, either immediately or after the first NIV adjustments 14 . Patients had to present signs of diaphragmatic dysfunction including respiratory pulse in the supine position.

Patients with ALS in whom NIV had been started in an emergency context were not eligible. Patients with a disease other than ALS likely to alter nutritional status and metabolic status (renal failure, diabetes or thyroid disease, recent episode of acute respiratory failure, active infection, chronic pancreatitis, chronic alcoholism, corticosteroid therapy, known malignancy) could not be included in the study.

All in all, over the study period, 24 patients were eligible among 106 ALS patients seen at the center. Two of those refused to participate in the study, technical problems occurred in one case, and recordings were missed in 5 cases.

The severity of the neurological deficit was evaluated by the revised ALS Functional Rating Scale (ALS-FRS-R), which rates the ability to perform activities of daily living from 0 (total inability) to 48 points (no limitation) and incorporates respiratory items (dyspnea, orthopnea, respiratory insufficiency). The bulbar section of the Norris scale was used to quantify bulbar impairment from 0 (no bulbar function) to 39 points (normal bulbar function). The date of onset of the symptoms and their initial level (bulbar or spinal), the date of confirmation of the diagnosis, and ongoing treatments (all patients received treatment with riluzole 50 mg, twice daily) were also recorded.

Respiratory function was evaluated by arterial blood gases with spontaneous breathing in room air and by overnight pulse oximetry (SpO₂) recording. Forced Vital Capacity (FVC) was measured in the supine and sitting positions with the EasyOne® ultrasound spirometer (NDD Medical Technologies, Andover, MA, USA). Inspiratory muscle strength was evaluated by measuring the maximal inspiratory pressure at the mouth (Pi_MAX) and at the nostril ("sniff nasal inspiratory pressure", SNIP) using a Micro-RPM® digital manometer (Micro Medical, Chatham, Kent, UK).

Optimization of nocturnal NIV was verified by arterial blood gases performed in the morning, one hour after disconnecting the ventilator, nocturnal SpO₂ recording on NIV and detection of leakages by the software integrated into the ventilator 14 15 . Of note, all the patients were equipped with home ventilators of the same make and model (Stellar®, Resmed, Bella Vista, Australia), with the spontaneous-timed pressure support mode (inspiratory and expiratory trigger).

Nutritional assessment comprised measurement of height and weight for calculation of body mass index (BMI), by adopting a cut-off of 20 kg/m² for malnutrition. Resting energy expenditure (REE) was measured by indirect calorimetry using a Quark RMR™ apparatus (Cosmed, Rome, Italy). Measurements were performed before 10 o’clock in the morning after fasting overnight and after resting for 20 minutes in a semi-sitting position in a quiet room heated to between 20°C and 24°C. Only values obtained at metabolic steady-state (less than 5% changes in V'O₂, V'CO₂ and RQ for at least 15 minutes) were taken into account.

During spontaneous breathing on room air (REE_SB), oxygen consumption (V'O₂) and CO₂ production (V'CO₂) were measured by a sensor fitted to the tip of a close-fitting oronasal mask. During NIV (REE_NIV), inspired air is carried from the respirator to the mask via a single tube. Expired air is delivered into the chamber via a calibrated leak distal to the sensor measuring V’O₂ and V’CO₂ (Figure  1). This methodology was developed in several healthy subjects prior to the study to ensure that measurement of REE on spontaneous breathing by the canopy method was equivalent to the that obtained by the cycle-to-cycle gas analyzer and that the calibrated leak in the NIV circuit did not induce any reduction of REE.

Predicted REE (REE_pred) was also calculated by the equation of Harris and Benedict 16 :

for males, REE = 66 + 1.38*weight(kg) + 5*height(cm)-6.8*age(years)

for females, REE = 655 + 9.7*weight(kg) + 1.8*height(cm)-4.7*age(years).

The contribution of respiratory muscles to REE or oxygen consumption at rest has been estimated in the literature to be between 1% 18 19 and 5-7% in normal individuals 20 21 22 23 and in quadriplegics 24 . These estimates were mostly based on back-extrapolations of the relationship between V'O2 and load-induced increases in the work of breathing 25 . This relationship is not linear and can underestimate the oxygen cost of unloaded breathing. To avoid this bias, the energy cost of breathing can be derived from comparisons between energy expenditure measured during spontaneous breathing and during mechanical ventilation in the same individuals 26 27 . This approach implies that respiratory muscles are actually passive during the measurements performed under mechanical ventilation, which is not always easy to ascertain. In patients with chronic obstructive pulmonary disease receiving NIV, Hugli et al. 28 estimated that respiratory V'O₂ was an average of 1.6% of V'O₂. This figure was underestimated because NIV mostly failed to abolish respiratory muscle electromyographic activity. In addition, the ventilation-induced reduction in V'O₂ was not normally distributed (1.6 ± 6%) and some patients had respiratory V'O₂ values of about 10% of V'O₂ 28 . In quadriplegic patients treated with phrenic stimulation, switching ventilatory assistance from controlled mechanical ventilation to diaphragm pacing increased REE by an average of 21% 29 . Because the patients in this study were overventilated, their respiratory-related energy expenditure was probably artefactually high. Hyperventilation-corrected values fell in the 10-15% range. To the best of our knowledge, the study with the most similar design to the present study compared energy expenditure during spontaneous breathing and mechanical ventilation in 9 tracheotomized patients with post-polio chronic respiratory failure 30 . In these patients, REE decreased from 1378 (958—1607) kcal/24 h during spontaneous breathing to 1086 (598—1579) kcal/24 h during pressure support ventilation (22%), and V'O₂ decreased from 207 (141—235) ml/min) to 159 (92—230)(23%). The REE reduction observed during NIV in our ALS patients (median 7%) and it V'O₂ counterpart are therefore in the upper range of available normal values for respiratory V'O₂ and in the lower range of reported values in patients. In addition to the possible underestimation discussed above, it must be emphasized that our patients mostly depended on their inspiratory neck muscles to breathe, as their diaphragm did not participate in ventilation.

In our patients, REE_SB was on average 10% lower than the value predicted by the Harris and Benedict equation. Of note, in the literature the prediction error for the Harris-Benedict equation has been reported to be as high as 18.6 + 14.9%, with limits of agreement showing that this equation could overestimate caloric expenditure by 591 kcal/d and underestimate requirements by 677 kcal/d 31 32 . The energy requirements in ALS is a complex issue see review in 33 and the details are beyond the scope of this discussion. Some studies have shown that one-half to two-thirds of ALS patients can be described as "hypermetabolic" 5 6 7 and remain so throughout the course of the disease 5 . In comparison with these studies, our patients had a slightly lower BMI, which, although fat-free mass was not measured, tends to suggest that they were somewhat malnourished. More importantly, the respiratory status of our patients was poorer than that reported in other studies, as our study appears to be the first to provide REE measurements during spontaneous breathing in ALS patients meeting the criteria for mechanical ventilation (Ellis and Rosenfeld measured REE in ALS patients requiring NIV, but it is unclear from their article whether the measurements were obtained during spontaneous breathing or under NIV) 32 . Our patients had a mean FVC of 42.7 ± 5.9% when REE_SB was determined, compared with a mean FVC close to 80% in the studies by Desport et al. 6 7 . In the longitudinal study by Bouteloup et al. 5 , the 28 patients with sequential REE measurements had a mean VC of 67% at the time of the last measurement. In the longitudinal study by Kasarkis et al 34 , performed before the NIV era, some patients were and remained hypermetabolic during the course of the disease, but the lowest observed FVC in this population was greater than 50%. It is therefore possible that the development of respiratory insufficiency in ALS has a masking effect on hypermetabolism, as suggested by Vaisman et al. 35 . This could occur via various mechanisms, including increasing malnutrition 33 35 due to the combination of ALS-related impairment of swallowing and/or eating-related dyspnea 36 , as described in other types of severe chronic respiratory insufficiency 37 38 . Of note, energy expenditure in ventilated ALS patients has been found to be low 8 39 , with results comparable to those observed in ventilator-dependent Duchenne patients 40 , but it has also been found to be increased. Sherman et al. observed hypermetabolism in ventilated patients 31 , and Ellis and Rosenfeld observed moderately increased REE values in patients requiring NIV 32 . In both cases, mean BMI values were about 25 kg.m^-2, vs. 22 in our patients, which might explain the discordant results. In any case, the present study showed that REE decreased under NIV in patients who were not hypermetabolic but who tended to be hypometabolic (possibly because of some degree or malnutrition): the amount of energy expenditure spared by NIV would probably have been even greater in better nourished patients. Also worthy of notice, the effect of NIV on REE was quite heterogeneous in our patients (Figure  2). We did not evidence any particular explanation for this finding. The reduction in REE was similar in the 10 patients studied upon NIV initiation and in the 6 patients studied somewhat later, and the correction of hypoventilation was excellent and homogeneous in the study population (Table  2). In addition, the REE response to NIV did not correlate with any of the anthropometric, respiratory or neurological variables collected.