Eligible patients fulfilled the following inclusion criteria: (1) Severe COPD according to GOLD guidelines , (2) home NIV initiated according to prevailing recommendations , (3) stable clinical condition defined as absence of exacerbation and of modifications of ventilator settings in the month preceding the study, (4) compliance to NIV of more than 4 h/day assessed by ventilator software, (5) presence of deventilation dyspnea defined as severe dyspnea occurring immediately after interruption of non-invasive ventilation and preventing the patient from getting out of bed and carrying out usual daily activities for more than 30 min after one night of NIV. Severe morning dyspnea was assessed by a chest therapist using a modified Borg scale. All patients with morning dyspnea score >4 were proposed to enter the study. Exclusion criteria were: (1) use of NIV in an acute exacerbation of COPD for hypercapnic ventilatory failure, (2) association of COPD with restrictive thoracic disorders (i.e. morbid obesity, cyphoscoliosis, neuromuscular disease, sequelae of tuberculosis), and (3) patients under 18 years old or not able to give informed consent.
Conventional polysomnography with patient under NIV included: standard electroencephalography (seven electrodes, F4–F3, O1–O2, C3–C4, Cz), left and right electrooculography, submental electromyography (EMG), measurement of airflow using a pneumotachograph, body position, thoracic and abdominal movements (Remlogic 1.1, Embla, Netherlands) and video recording. SpO2 was measured continuously with a pulse oximeter and a finger probe. PtcCO2 measurements were performed using a transcutaneous capnograph with an ear probe heated at 43°C (Tosca 500, Radiometer, Switzerland). Sleep was scored according to standard criteria using 30-s epochs . The investigator scoring the sleep study was blinded to the ventilator settings and subjective evaluation. The following sleep parameters were quantified: total sleep time (TST), sleep efficiency (SE—TST/total recording time × 100), percentage of each sleep stage, wake after sleep onset (WASO) and sleep latency. As indices of sleep fragmentation, we considered the index of micro-arousals (MAI). MAI were defined as a return to alpha or fast frequency, well-differentiated from the background EEG activity lasting >1.5 s.
Surface diaphragm EMG was recorded using silver electrodes in all subjects only to monitor inspiratory effort  and facilitate detection of ineffective inspiratory efforts. The electrodes were placed in the right and left anterior axillary line in the seventh intercostal space with a horizontal distance of 2 cm. Diaphragmatic EMG was sampled at 0.2 kHz, bandpass-filtered between 10 and 70 Hz  and recorded with all other parameters on the Embla recording system.
Patterns of patient–ventilator asynchrony included ineffective triggering (unrewarded inspiratory efforts), double-triggering and auto-triggering, as defined by Thille et al. . Unrewarded inspiratory efforts were identified as the occurrence of a definite diaphragmatic EMG signal, and/or presence of inspiratory thoracoabdominal movements on strain gauges without subsequent pressurisation by the ventilator. Auto-triggering was defined as a cycle delivered by the ventilator and not triggered by the patient. An unrewarded effort index was calculated as follows: [number of desynchronized cycles/total respiratory rate (i.e. cycles supported by the ventilator plus unrewarded efforts)] × 100 (expressed as a percentage) . Only epochs (3 min.) with an unrewarded effort index above 10% were scored as periods with asynchrony based on a study by Vittaca and al. . PVA index was calculated as the ratio of total sleep time spent with unrewarded efforts on TST .
Ventilator built-in software
Compliance, estimated tidal volume (VT), minute ventilation (VE), respiratory rate (RR), leaks and percent (%) of spontaneous inspirations were downloaded after both nights from built-in software of bi-level ventilators (ResScan software, version 1.3, ResMed, Sydney, Australia). Estimation of leaks and VT by ventilator software of VPAP III and VPAP IV ventilators (ResMed, Sydney, Australia) has been shown to be reliable in two recent independent studies [23, 24].
After both nights, to evaluate patient comfort and patients’ perception of patient–ventilator synchronisation, airflow, leaks and noise of ventilator, an eight-item visual analogue scale (VAS) was administered the morning following each sleep study . Higher values indicate better comfort. Patients were also asked to quantify their dyspnea on a visual analogue scale while withdrawing from pressure support in the morning (10—worst dyspnea ever experienced in the morning, 0—no dyspnea at all).
Patients were admitted to the sleep laboratory for two consecutive nights. On the first night, ventilator settings were those normally used by the patient, thus eliminating the need of an accommodation night. This first night aimed to detect and quantify PVA in highly symptomatic COPD patients on their usual NIV settings. In our institution, ventilation settings are adjusted by chest therapists, under the supervision of the attending pulmonary physician, aiming at patient comfort and optimal control of daytime and nocturnal SpO2 and PtcCO2. On the second night, the therapeutic target was to reduce sleep-disordered breathing and patient–ventilator asynchrony. The investigators adjusted ventilator settings using polysomnography and monitoring of SpO2/PtcCO2.
Our approach was as follows: in case of unrewarded inspiratory efforts, we first reduced pressure support, then, if necessary, we increased expiratory positive airway pressure (EPAP)  and set expiratory trigger at a higher percentage of peak inspiratory flow to avoid delayed cycling [26, 27] under close monitoring of PtcCO2. When time to peak pressure (pressurisation time) was set at 100 msec, we increased it by 50–100 msec. Back-up respiratory rate was increased if unrewarded inspiratory efforts persisted. Maximal inspiratory time (TIMAX) was calculated in order to maintain a physiological I/E ratio for COPD patients between 1:2.5 and 1:3.
If auto-triggering was identified, mask fit was adapted to prevent air leaks and inspiratory trigger sensitivity was reduced . Inspiratory time (TIMIN) was increased if double triggering was frequent . Special attention was given to mask leaks in order to limit their impact on patient–ventilator asynchrony.
Statistical testing was performed using Prism4 software (GraphPad, San Diego, CA, USA). When data were normally distributed (Schapiro–Wilk test), results were expressed as mean±SD. All variables between the two nights of the study protocol were compared using a paired t test. Differences were considered significant when probability of a type I error was <0.05.
The study protocol was approved by the Ethics Committee for Medical Research of Geneva University Hospital and written informed consent was obtained from all patients. The trial was registered at www.clinicaltrials.gov (Trial No. NCT01180439)