In this study, we performed OI trajectory analysis followed by regression analysis to investigate the role of OI dynamics as a predictor for BPD grades in preterm infants born before 30 weeks of gestation.
GAMM is a powerful technique for trajectory modeling [24, 25]. We have used the technique to model postnatal weight trajectories and probabilities of various BPD severity categories over time [26, 27]. Although the backbone technique, GAM, does not require a linear assumption between the independent and the outcome variables, the technique does still allow linear regression if a linear relationship pattern between the independent and the dependent variables does exist. In the presence of a non-linear pattern, overfitting is regulated by the smooth parameter. Mixed-effects modeling by GAMM allows the model to consider clusters of data points from each infant, rather than treating all data points independently as in GAM. Therefore, GAMM is more likely to accurately describe the trajectory. When mixed modeling was not used, we observed an upward skew in the modeled trajectory due of the nature of the dataset, as sicker infants who required higher respiratory support (higher OI) and more frequent blood gas assessments (more data points) predominated the later time points.
OI is used to characterize the severity of pulmonary illness. OI is calculated by taking the ratio between the degree of respiratory support and blood oxygen content. The higher the OI, the most severe the pulmonary condition is. OI has not been routinely used to describe the severity of respiratory distress syndrome of newborn, a disease of pulmonary immaturity, although it is commonly used in neonatal intensive care to assess the need for support escalation to extracorporeal membrane oxygenation. OI measurement requires an invasive procedure to obtain arterial blood gas. An arterial access may not be routinely performed in mildly to moderately ill newborns due to the risks associated with it. Studies have assessed the efficacy of using modified versions of OI, such as oxygenation saturation index (OSI, replacing PaO2 with blood oxygen saturation using pulse oximetry, or SpO2, in index calculation) or respiratory severity scores (RSS, the product of MAP and FiO2) as surrogate markers for pulmonary illness [28–34]. One recent study showed a strong correlation between OI and OSI in the neonatal population [32]. Another study showed a strong correlation between OI and RSS [28].
The OI trajectories for the No BPD, Low-Grade BPD, and High-Grade BPD subgroups provided a way to visually inspect the bulk movement of the OI values over the first 21 days of life. The modeled curves were not adjusted for perinatal or demographic factors. Therefore, the OI trajectory modeling was more hypothesis-generating than hypothesis-testing. An infant supported on CPAP of 5 cmH2O at 30% FiO2 with a PaO2 of 50 mmHg has an OI of 3; an infant supported on CPAP of 6 cmH2O at 30% FiO2 with a PaO2 of 45 mmHg has an OI of 4; an infant supported on CPAP of 6 cmH2O at 40% FiO2 with a PaO2 of 50 mmHg has an OI of 4.8. These examples are representative of the respective expected starting points of respiratory support in the three BPD grade subgroups based on the modeled curves. The infants in the No BPD group had stable OI in the first 14 DOL, followed by a downward trend between 14–21 DOL. On the other hand, infants who were diagnosed with BPD had an increase in OI in the first 14 DOL. However, those with High-Grade BPD continued to have an uptrending trajectory between 14 and 21 DOL, while those with Low-Grade BPD showed downtrending OI curves. These findings suggest that the degree of respiratory support during the third week of life may be key to predicting BPD grade. Interestingly, these findings coincided with the percentage of infants who received dexamethasone before DOL 21. It is possible that dexamethasone may have effectively improved respiratory status, leading to a decrease in OI, and subsequently affected BPD grade, although this project was not specifically designed to answer the question regarding the association between dexamethasone and BPD grade changes. Nonetheless, it may be desirable to introduce non-pharmacological or pharmacological measures during the third week of life to decrease OI and the risks of High-Grade BPD.
Based on the findings from the analysis of the OI trajectories, we developed regression models to quantify the contribution of OI dynamics in BPD grades. Additional potential covariates, including birth GA, BW-Z, and sex, were assessed for inclusion in the model by using AIC. AIC assesses the quality of the model by balancing maximum likelihood and the complexity of the model, with an attempt to control model overfitting. The selected model contained all five variables. Given the retrospective nature of the study, we were only able to assess average daily changes in OI from the available data. The disadvantage of this approach is the potential underestimation of the strength of the association because the OI trajectory patterns were not linear and cannot be easily summarized by the average daily changes. Prospective studies may allow a more accurate measurement of OI change velocity for BPD outcome correlation.
The most significant advantage of the BPD grading system is its data-driven approach by correlating BPD grades with respiratory and neurodevelopmental outcomes during the grading system development. Grade 3 BPD had significant worse respiratory outcomes and morbidities of prematurity [35]. The NICHD BPD outcome estimator was recently revised to accommodate for BPD grade prediction [15]. However, the model only provides fair performance, suggesting that there is a great need to identify additional predictors. Identifying factors which may be used to predict High-Grade BPD may allow early targeted interventions and an opportunity to at least “down-grade” BPD if not preventing it. Short-term improvement in BPD grade may translate into long-term neurodevelopmental and cardiopulmonary benefits. As arterial catheter placement may not be feasible or indicated in all at-risk infants, future investigation may focus on assessing whether OSI or RSS may serve as useful alternatives to OI as a predictor while validating our findings.
In conclusion, we characterize OI trajectories in a retrospective cohort and quantified the association between OI and BPD grades. These data shed light on the feasibility of adding OI dynamics to improve BPD grade prediction, providing a path towards early identification and early interventions.