Diagnostic labels and clusters based on oxygen requirements in preterm infants with chronic lung disease: a data-driven exploratory cluster analysis in two independent cohorts.

Abstract

Diagnosis of bronchopulmonary dysplasia in very low birthweight infants is often only ascertained many weeks after birth. We aimed to investigate whether trajectories of the fractions of inspired oxygen (FiO₂) required to maintain O₂ saturation from birth reflect distinct clusters of preterm lung disease. In this data-driven exploratory cluster analysis, we used latent class trajectory modelling to derive clusters of FiO₂ fluctuations in the first 30 days of life among neonates with a birthweight of less than 1250 g, across two multicentre, prospective cohorts: the Discovery Bronchopulmonary Dysplasia Program (D-BPD; enrolment July 30, 2013, to Jan 1, 2020; n=376) in Argentina and the Prematurity and Respiratory Outcomes Program (PROP; enrolment Aug 1, 2011, to Nov 31, 2013; n=835) in the USA, with the PROP cohort being used for external validation. Eligible participants, in both the present and original studies, included infants with birthweight of less than 1250 g in the D-BPD cohort, and infants enrolled in the PROP study born at 23-28 weeks of gestation by best obstetrical estimate. After unsupervised clustering of patients, we evaluated cluster performance against conventional bronchopulmonary dysplasia classification, using mortality as the primary outcome. Of the 376 D-BPD infants, 190 (51%) were female (mean birthweight 968·1 g [SD 181·2]; mean gestational age 28·9 weeks [SD 2·3]) and 186 (49%) were male (mean birthweight 976·2 g [188·3]; mean gestational age 28·6 weeks [2·3]). Four clusters based on FiO₂ requirements were identified: persistently low requirement (PLR; 218 [58%] of 376), early high with subsequent improvement (EHRI; 60 [16%] of 376), late-onset high requirement (LOHR; 31 [8%] of 376), and persistently high requirement (PHR; 67 [18%] of 376). Mortality was more frequent in LOHR (six [19%] of 31) and PHR (ten [15%] of 67) clusters, with no deaths in PLR and one death in EHRI (one [2%] of 60; p<0·0001). Nine (53%) of 17 fatal cases occurred before 36 weeks' gestational age, precluding conventional bronchopulmonary dysplasia diagnosis; among infants who died later, most had severe bronchopulmonary dysplasia (six [67%] of nine). Results from the PROP cohort closely mirrored those of D-BPD, yielding the same number of clusters with similar trajectories and replicating patterns of mortality. Current bronchopulmonary dysplasia definitions might not fully capture the trajectories of lung disease of preterm infants. Data-driven approaches offer opportunities to identify infants at high risk earlier and to implement more precise interventions. US National Institute of Environmental Health Sciences, National Institutes of Health, Department of Health and Human Services, and the UK MRC.