External Validation of the BREF Models

Introduction High-flow oxygen therapy is increasingly used to treat acute respiratory failure (ARF). It can reduce intubation rates without increasing mortality. Moreover, it is better tolerated than other non-invasive respiratory support therapies. However, patients with ARF often make strong breathing efforts, which may cause several harm. They can fatigue the respiratory muscles. They can increase whole-body oxygen consumption and carbon dioxide production. They can cause pulmonary edema. They might also injure the diaphragm and the lungs. For all these reasons, strong breathing efforts should be recognized and treated.

During spontaneous breathing, inspiratory muscle contractions produce parallel deflections of the pleural and esophageal pressure (ΔPes), which reflect the magnitude of the effort. Unlike changes in pleural pressure, ΔPes can be easily measured at the bedside using a catheter similar to a typical nasogastric tube. In healthy subjects, ΔPes is only a few cmH2O during quiet breathing but >10-15 cmH2O during vigorous exercise or carbon dioxide inhalation. In patients with ARF, the upper limit for a "safe" ΔPes is unknown. Nonetheless, according to experts, breathing efforts with a ΔPes >10-15 cmH2O are probably too strong to be tolerated for a long time.

However, esophageal manometry is not widely available. Estimating breathing efforts without it is complex, especially in non-intubated patients. Doctors mostly rely on their gestalt or overall impression. Therefore, it is unsurprising that they may disagree when rating their patients' breathing efforts or debating whether to proceed to intubation.

The investigators have recently developed two clinical prediction models for estimating the breathing effort of patients with ARF from a few variables readily available at the bedside. The first, "linear", model estimates the continuous value of ΔPes (in cmH2O) from the presence or absence of COVID-19, arterial base excess concentration (BEa) (in mmol/L), respiratory rate (in bpm), the ratio of the arterial tension to the inspiratory fraction of oxygen (PaO2:FiO2) (in mmHg), and the product term between COVID-19 and PaO2:FiO2. The calibration slope was 1, and the adjusted R2 was 0.39. The second, "logistic", model estimates the probability of ΔPes being >10 cmH2O (dichotomous outcome) from BEa (in mmol/L), respiratory rate (in bpm), and PaO2:FiO2 (in mmHg). When this model was tested on the same data set used to develop it (apparent performance), the area under the ROC curve (AUROC) was 0.79 (95% CI, 0.73-0.85). At internal validation (optimism-corrected performance), the AUROC was 0.76 (0.71-0.81). The investigators called these models BREF, which stands for BReathing EFfort, but also to the three main predictors: BEa (B), respiratory rate (RE), and PaO2:FiO2 (F).

Study aims The main goal of this study is to evaluate the BREF models' predictive performance in a new population. This process is known as "external validation". Additionally, there are two other secondary objectives. The investigators aim to update the BREF models by adding more variables unavailable in the development dataset (method "extension") and compare the accuracy of the BREF models with that of doctors who do not use them in assessing their patients' breathing efforts.

Study population

Inclusion criteria:

• adult (≥18 years of age) patients in the ICU
• treated with high-flow oxygen delivered via nasal cannula
• equipped with an esophageal balloon as per local clinical practice.

Exclusion criteria:

• history of chronic lung disease
• cardiogenic pulmonary edema
• >96 hours from admission to the participating unit.

Methods First, participants will record all the variables needed to estimate ΔPes using the original version of the BREF models. These are the presence or absence of COVID-19, BEa, respiratory rate, PaO2:FiO2, and the product term between COVID-19 and PaO2:FiO2. Moreover, participants will record other variables that may help predict ΔPes based on scientific reasoning. Next, the attending doctor (blinded to the actual ΔPes) will assess the patient's breathing effort based on clinical judgment. Third, the actual ΔPes will be measured with esophageal manometry. Finally, the participants will record the type of respiratory support provided to the patient in the 72 hours following the study, the length of stay in the unit, and the vital status of the patient (dead or alive) at discharge from the unit.

Sample size calculation This study aims to enroll 250 patients in approximately two years.