CLEAR Model for Predicting Ventilatory Liberation in Severe COPD

This study is a prospective, dual-cohort observational investigation designed to develop and internally validate the Clinical Load, Exchange, Ability of Respiration, and Reserve (CLEAR) model for predicting sustained liberation from ventilatory support in patients with severe chronic obstructive pulmonary disease (COPD). The study includes two parallel cohorts based on the mode of ventilatory support: an invasive mechanical ventilation (MV) cohort (CLEAR-MV) and a non-invasive ventilation (NIV) cohort (CLEAR-NIV). The model structure and domain composition are prespecified; parameter estimation involves applying a fixed framework rather than exploratory model development. In this study, "development" refers exclusively to parameter estimation within a prespecified and structurally locked model framework. The model structure, domain composition, and variable definitions were defined prior to outcome analysis and are not subject to modification. No variable selection, structural changes, or data-driven model building procedures will be performed. Validation refers to performance assessment in an independent cohort using the same fixed model structure. Eligible adult patients with COPD admitted to the intensive care unit (ICU) and receiving MV or NIV for acute respiratory failure will be enrolled. Patients will be assessed at the time of readiness for ventilatory withdrawal: during a spontaneous breathing trial (SBT) in the MV cohort and during a structured withdrawal or low-support trial in the NIV cohort. The CLEAR framework comprises four domains: Clinical Load (L), Exchange (E), Ability of Respiration (A), and Reserve (R). The Ability domain integrates diaphragm contractility assessed by ultrasound-derived diaphragm thickening fraction (DTF) and functional endurance during SBT or NIV withdrawal trials. The Load domain is assessed using the rapid shallow breathing index (RSBI) in the MV cohort and a composite Clinical Load Index (CLI) in the NIV cohort, incorporating respiratory rate, work of breathing, dyspnea, and ventilatory support requirements. The Exchange domain is based on arterial blood gas parameters, including pH, partial pressure of carbon dioxide (PaCO₂), and the change in PaCO₂ (ΔPaCO₂), combined into a Gas Exchange Index (GEI) in the NIV cohort and used as a safety assessment in the MV cohort. The Reserve domain reflects systemic muscle capacity and is assessed using ultrasound-derived thicknesses of the Rectus Femoris (RF) and vastus intermedius (VI). The primary outcome is successful liberation from ventilatory support within 72 hours, defined as the absence of reintubation in the MV cohort and the absence of NIV restart or escalation to invasive ventilation in the NIV cohort. Secondary outcomes include ventilatory failure within 7 days, ventilator- or NIV-free days at 28 days, and time-fixed 90-day outcomes including all-cause mortality, sustained ventilatory independence, and rehospitalization for respiratory failure. Comparative analyses will be performed against prespecified established indices. In the MV cohort, CLEAR-MV will be compared with the Rapid Shallow Breathing Index (RSBI) and, where available, the Integrative Weaning Index (IWI), which combines static compliance, arterial oxygen saturation, and RSBI. In the NIV cohort, CLEAR-NIV will be compared with the Heart rate, Acidosis, Consciousness, Oxygenation, and Respiratory rate (HACOR) score and the ratio of peripheral oxygen saturation to fraction of inspired oxygen divided by respiratory rate (ROX) index. Model performance will be evaluated using discrimination (area under the receiver operating characteristic curve), calibration (calibration intercept and slope), and clinical utility (decision curve analysis and net benefit). Separate multivariable logistic regression models will be estimated within a derivation cohort and evaluated in a validation cohort according to the prespecified CLEAR framework. Model performance will be evaluated directly within the study cohorts, without resampling-based internal validation. The independent and incremental contribution of the Clinical Load Index (CLI) and Gas Exchange Index (GEI) will be specifically assessed. Additional prespecified analyses will include domain correlation, model ablation, and prespecified ventilatory support state-transition analysis if longitudinal support-state data are available. No distinction exists between model development and structure specification; all structural elements are fixed prior to analysis. Model outputs will be translated into predefined bedside scoring representations consistent with the prespecified framework. The CLEAR framework and its component indices (Clinical Load Index and Gas Exchange Index) were prospectively defined and structurally locked before outcome analysis in an open-access research data repository (Zenodo), Digital Object Identifiers (DOIs): 10.5281/zenodo.19929675, 10.5281/zenodo.19931693, 10.5281/zenodo.19932153).