Efficacy of Morphine in Reducing the Rate of Early Non-Invasive Ventilation Failure in Acute Exacerbation of Chronic Obstructive Pulmonary Disease, Phase I/IIa

Patients included in this dose-finding study will be enrolled consecutively after consent and verification of the selection criteria, with a minimum of 24 hours after completion of previous patient follow-up. Prior to the administration of morphine, the following exams will be performed: clinical, paraclinical and biological parameters. NIV will be started and performed in BiPAP pressure mode with facial mask. A Tidal Volume (Vt) to the maximum of 6 - 8ml/kg, an expiratory positive airway pressure (EPAP) of 5cm H2O and the FiO2 set at 35%. The settings will be adapted according to the SpO2 (88%≤SpO2≤92%) and the clinical tolerance. Thereafter, the patients included will receive a single open-label injection of morphine (T0) at the dose closest to that defined by the model (estimation by Bayesian method) from among the following 4 doses: 0.02; 0.04; 0.07 and 0.1mg/kg. A double check of the preparation of the dose administered will be carried out by a physician and a nurse in an emergency department. Each result is aimed to refine the gain function which will be used to estimate the optimal dose for the next patient. The dose to be administered to the next patient will be determined by the Clinical Investigation Centre based on the data of the previous patient. It will be determined using a gain function whose objective is to represent the best compromise between efficacy and safety. It assigns a value to the patient representing the benefit: in case of toxicity this value will be 0; otherwise it will be equal to the decrease in PaCO2. Constant monitoring for up to 4 hours will be done as part of this study to detect the occurrence of adverse events. In particular, the following will be monitored: disturbances of consciousness, vomiting or secretions requiring airway protection; hemodynamic instability; restlessness and/or inability to keep the mask on. Therefore, ongoing monitoring for routine patient care may be extended at the discretion of the clinical physician in charge of the patient. Follow-up of patients will be done in 15 minutes, 1 hour, 4 hours (or NIV stop) and 24 hours. Clinical, paraclinical, biological parameters and the occurrence of adverse events will be performed in 15 minutes, 1 and 4 hours follow-up. During 24 hours only the occurrence of adverse event will be reported. The trial shall be concluded when the maximum number of patients is reached (N=24), or when a sufficiently accurate estimate of the dose maximizing the benefit-risk balance is obtained while remaining within the toxicity limit set at a 30% probability of DLT. The precision of this estimate will be measured by the ratio of the upper and lower bounds of the confidence interval. If this ratio is less than 5, the precision is considered to be sufficient to stop the study. Four groups of data will be collected at different times: biological parameter, clinical and paraclinical parameters as well adverse event. Clinical parameter: demographic variable (gender, age), weight, height, BMI, Richmond Agitation- Sedation Scale, Visual analog scale of dyspnea (Borg scale), numerical rating scale for pain assessment. Paraclinical parameter: Vital parameters: heart rate, respiratory rate, blood pressure, ventilatory rate, SpO2, NIV settings, EPAP, IPAP and Tidal Volume. Biological parameter: Blood count, arterial blood gas, chemistry panel. outcome measure: Optimal dose is defined as the dose associated with the maximum of gain function, used to determine the best compromise between efficacy and toxicity. It combines: 1. The probability of dose - limiting toxicity (DTL), defined as the occurrence of one or more of the following criteria occurring within 4 hours of morphine administration: * Respiratory Rate ≤ 10/min; * Richmond Agitation-Sedation Scale (RASS) \< -2 * Vomiting * Naloxone administration. 2. Efficacy and toxicity defined by: * PaCO2 1 hour after morphine injection, according to the hormetic dose-response model ("J-curve": efficacy first increases with the dose, then reach a maximum and decreases in case of excessive bradypnea). Study design: Estimation of enrollment: Up to 24 participants. The calculation of the sample size was carried out by simulating several scenarios based on the prior to be used in the study. The simulated scenarios are as follows: one with low toxicity (5% at 0.02 mg/kg and 30% at 0.1 mg/kg), one with high toxicity (10% at 0.02 mg/kg and 40% at 0.04 mg/kg) and one where the toxicity is identical to the prior. Simulation results showed a number of subjects required up to 24, with a median of 6 to 11, according to the scenario. The investigators therefore chose to define a maximum number of patients to 24, which is higher than the average of the worst-case scenario and corresponds to the maximum number of patients required among the 100 simulations of the worst-case scenario. However, it is very likely that the inclusions will be stopped earlier, when the accuracy is satisfactory. Analysis: The method used will be an adaptive Bayesian dose-finding procedure for a binary response (DLT), and a continuous response (PaCO2 used as a biomarker of efficacy and toxicity, following a hormetic response), using a gain function which objectively represent the compromise between efficiency and safety. The investigators will use a function already described in the literature . No interim analysis planned for this study.