Association between dynamic digital radiography findings and post-extubation respiratory deterioration: A retrospective exploratory analysis of a prospectively collected ICU cohort

Introduction: Dynamic Digital Radiography (DDR) is a novel bedside imaging modality that enables real-time visualization of pulmonary motion with minimal radiation exposure. We evaluated whether DDR-derived parameters could assist in stratifying the risk of post-extubation respiratory deterioration in ICU patients. Methods: We analyzed a prospectively collected, single-center cohort of consecutive adults extubated after invasive mechanical ventilation between December 2024 and February 2025 (n = 56). Bedside DDR was performed immediately before and within 24 hours after extubation; respiratory deterioration was defined as any increase in oxygen supplementation before ICU discharge. We used logistic regression to address the initial objectives and added exploratory machine learning to explore predictive performance, with interpretability assessed using Shapley additive explanations (SHAP). Results: While no significant differences were found in directly measured DDR values between groups, age and pre-extubation respiratory rate differed significantly between patients with and without respiratory deterioration. In logistic regression analysis, post-extubation lung-area excursion was not significantly associated with respiratory deterioration after adjustment for age. In exploratory XGBoost/SHAP analysis, age, post-extubation respiratory rate, and post-extubation lung-area excursion showed relatively large model-based contributions, but these findings were interpreted as exploratory feature contributions rather than evidence of independent statistical association. Conclusions: In this exploratory ICU cohort, DDR-derived post-extubation lung-area excursion was not independently associated with respiratory deterioration after adjustment for age, but it showed a relatively large model-based contribution in exploratory XGBoost/SHAP analysis. DDR-derived dynamic information may provide complementary, hypothesis-generating information for post-extubation assessment. Larger prospective studies with external validation are required before DDR-derived indices can be considered clinically validated predictors.