Detection of pulmonary hypertension in preterm infants with bronchopulmonary dysplasia using oxygen saturation data.

Abstract

Background: Pulmonary hypertension (PH) complicates the clinical course of bronchopulmonary dysplasia (BPD) in preterm infants, increasing risk of mortality and other morbidities. Early detection of PH may provide the clinical opportunity for earlier initiation of PH targeted therapies.

Methods: We analyzed variability patterns of oxygen saturation data in 41 preterm infants with BPD and associated PH and 57 infants with BPD alone using irregularity indices to identify PH. Irregularity of oxygen saturation (SpO₂) variability time series was characterized by sample entropy over multiple time scales, and characteristic features were derived. Multivariable logistic regression models with entropy-based indices and significant clinical risk factors as features were developed and their diagnostic accuracy metrics were evaluated.

Results: SpO₂ measures significantly differed between groups by PH status. The model with signal-based measures and clinical features of birthweight, sepsis, and presence of patent ductus arteriosus had discriminative ability for PH with 82% area under ROC (95% CI: 75%, 88%) and accuracy, sensitivity, and specificity of 81%, 78% and 82% respectively.

Conclusion: Irregularity profiles of oxygen saturation variability are distinct in infants with BPD and PH. Signal based measures derived from routinely collected SpO₂ data can be leveraged as continuous bedside markers to detect PH in preterm infants.

Impact: Early detection of pulmonary hypertension (PH) may improve outcomes in infants with bronchopulmonary dysplasia (BPD). Bedside markers which can track evolving PH, using routinely collected data in the NICU are hence important. Analysis of the irregularity exhibited by oxygen saturation variability data from preterm infants with BPD identified distinctive behavior in the presence of PH. We demonstrated detection of PH using signal measures accounting for known clinical risk factors using data acquired prior to clinical diagnosis. This study expands the possibility of implementing signal based models for early identification of PH in this vulnerable population.