Early Fluid Balance Dynamics and Short-Term Mortality in Critically Ill Patients with Acute Exacerbation of Chronic Obstructive Pulmonary Disease: An Interpretable Prediction and Trajectory Analysis

Abstract Background Acute exacerbation of chronic obstructive pulmonary disease (AECOPD) is a leading cause of intensive care unit (ICU) admission and short-term mortality. Fluid management is critical, yet the prognostic significance of dynamic fluid changes especially the shift from resuscitation to stabilization remains unclear in this population. Methods We conducted a retrospective cohort study using the MIMIC-IV database, including adult ICU patients with AECOPD. Group-based trajectory modeling (GBTM) was used to identify latent patterns of fluid balance over the first three ICU days. An interpretable prediction model for 28-day mortality was built using multivariable logistic regression. Model performance was evaluated with discrimination (AUC), calibration, and decision curve analysis. Feature importance was assessed using Shapley Additive Explanations (SHAP). Results A total of 639 patients were included. The prediction model showed good discrimination in both the training (AUC 0.747) and validation cohorts (AUC 0.753). SHAP analysis revealed that fluid balance on ICU day 3 was a key, non-linear, and potentially modifiable predictor of mortality. GBTM identified five distinct longitudinal fluid balance patterns. Trajectories with persistent positive fluid balance were independently linked to the highest 28-day mortality. In contrast, a "rapid correction" trajectory marked by early fluid accumulation followed by prompt removal was associated with survival rates similar to those of patients with stable fluid balance, despite higher baseline illness severity. Conclusions In critically ill patients with AECOPD, failure to clear positive fluid balance by ICU day 3 is a strong warning sign. The reversibility of early fluid accumulation, rather than the absolute initial volume, appears to determine survival. These findings support a dynamic risk-stratification approach and underscore the importance of timely de-resuscitation in this vulnerable population.