Classifying multicenter approaches to invasive mechanical ventilation for infants with bronchopulmonary dysplasia using hierarchical clustering analysis.

Introduction: Evidence-based ventilation strategies for infants with severe bronchopulmonary dysplasia (BPD) remain unknown. Determining whether contemporary ventilation approaches cluster as specific BPD strategies may better characterize care and enhance the design of clinical trials. The objective of this study was to test the hypothesis that unsupervised, multifactorial clustering analysis of point prevalence ventilator setting data would classify a discrete number of physiology-based approaches to mechanical ventilation in a multicenter cohort of infants with severe BPD.

Methods: We performed a secondary analysis of a multicenter point prevalence study of infants with severe BPD treated with invasive mechanical ventilation. We clustered the cohort by mean airway pressure (MAP), positive end expiratory pressure (PEEP), set respiratory rate, and inspiratory time (Ti) using Ward's hierarchical clustering analysis (HCA).

Results: Seventy-eight patients with severe BPD were included from 14 centers. HCA classified three discrete clusters as determined by an agglomerative coefficient of 0.97. Cluster stability was relatively strong as determined by Jaccard coefficient means of 0.79, 0.85, and 0.77 for clusters 1, 2, and 3, respectively. The median PEEP, MAP, rate, Ti, and PIP differed significantly between clusters for each comparison by Kruskall-Wallis testing (p < 0.0001).

Conclusions: In this study, unsupervised clustering analysis of ventilator setting data identified three discrete approaches to mechanical ventilation in a multicenter cohort of infants with severe BPD. Prospective trials are needed to determine whether these approaches to mechanical ventilation are associated with specific severe BPD clinical phenotypes and differentially modify respiratory outcomes.