Prediction of high-flow nasal cannula failure in patients with sepsis by a preliminary risk stratification model based on the ROX index.

Abstract

Background: High-Flow Nasal Cannula (HFNC) carries the risk of delayed intubation and invasive mechanical ventilation (IMV), often associated with adverse outcomes. Therefore, it is crucial to predict the risk of HFNC failure as early as possible.

Objective: To explore the value of the Respiratory Rate-Oxygenation (ROX) index and a Preliminary Risk Stratification Model based on the ROX Index for prediction of HFNC failure in patients with sepsis.

Methods: This retrospective study was based on the Medical Information Mart for Intensive Care (MIMIC) database. Independent risk factors for HFNC failure were explored using Least Absolute Shrinkage and Selection Operator (LASSO) and multivariable logistic regression analyses, with discrimination shown using Receiver Operating Characteristic Curves (ROC). To evaluate the discriminatory ability of the ROX index at different time points, we performed Landmark Analysis. The relationship between the ROX index and the risk of HFNC failure was confirmed through logistic proportional hazards model. Kaplan-Meier (K-M) curves were employed to illustrate the association between the ROX index and the likelihood of HFNC failure. Additional subgroup analyses were conducted to further substantiate the robustness of the findings. Finally, a Preliminary risk stratification model for predicting HFNC failure was established based on the ROX index, with ROC curves used to demonstrate its discriminatory power.

Results: A total of 316 sepsis patients who met the criteria were included, of whom 116 patients failed HFNC and required intubation and IMV. LASSO and multivariable logistic analyses demonstrated the ROX index as an independent risk factor for HFNC failure, with an area under the ROC curve of 0.721 (0.658-0.784) after 12 h of HFNC use, indicating good predictive value. Logistic proportional hazards model and K-M analysis demonstrated that patients in the low ROX group (based on the established cutoff value) exhibited a significantly elevated risk of failure of HFNC therapy. This finding was consistent with the results of subgroup analyses. Based on the ROX index, a Preliminary risk stratification model with a total score of approximately 16 points was established by combining Glasgow Coma Scale (GCS), Body Mass Index (BMI), and flow rate, and the area under the ROC curve was 0.801 (0.7516-0.8509), indicating better predictive value.

Conclusion: The ROX index has good predictive value for HFNC failure in sepsis patients, and the Preliminary risk stratification model established based on the ROX index shows better predictive potential, but requires prospective external validation.