2 Methods
A retrospective cohort study was conducted with patients registered in the Neonatal Database of a Brazilian tertiary hospital, born in the years 2016 to 2020 (5 years), with gestational ages between 23 and 30 weeks and birth weight between 401 and 1250 g.
Newborns with complex congenital heart disease diagnoses, major lung malformations, and deaths before 28 days of life were excluded from the study.
The analyzed antenatal variables consisted of maternal systemic arterial hypertension (chronic or pregnancy-induced arterial hypertension, with or without edema and proteinuria), maternal smoking, chorioamnionitis (considered through clinical or laboratory criteria: maternal leukocytosis, with no other identifiable infectious focus that would justify it), and use of antenatal corticosteroids (betamethasone, complete and incomplete cycles). The analyzed postnatal variables were gestational age (GA), birth weight in grams (BW), gender (male or female), adequacy for gestational age using Intergrowth-21st curves12, with birth weight below the 10th percentile for gestational age classified as small for gestational age (SGA), and birth weight above the 90th percentile classified as large for gestational age (LGA), need for surfactant, patent ductus arteriosus (clinical or echocardiographic diagnosis), early sepsis (clinical-laboratory criteria with the use of antibiotics for at least five days, starting up to the 3rd day of life), late sepsis up to the 14th day of life (clinical-laboratory criteria with the use of antibiotics for at least five days, starting after the 3rd day of life), necrotizing enterocolitis, defined using the Bell Criteria 13, and respiratory support and fraction of inspired oxygen (FiO2) on the 14th day of life.
Respiratory support was characterized as nasal oxygen catheter (simple or high-flow nasal catheter), nasal continuous positive airway pressure (CPAP) or non-invasive intermittent positive pressure ventilation (NIPPV), conventional invasive mechanical ventilation (MV), and high-frequency ventilation (HFV).
The evaluated FiO2 and respiratory support were the most frequently recorded on the day (higher needs that occurred transiently were discarded). The type of ventilation device and its manufacturer were not considered.
The analyzed outcomes were defined by the presence of bronchopulmonary dysplasia and their classifications, made according to the definitions of the National Heart, Lung and Blood Institute (NHLBI) in 2000: no bronchopulmonary dysplasia, mild bronchopulmonary dysplasia (patients requiring supplemental oxygen with 28 days of life or more but not at 36 weeks corrected gestational age), moderate bronchopulmonary dysplasia (Patients requiring supplemental oxygen at 28 days of age and maintaining a FiO2 requirement of less than 30% at 36 weeks of corrected gestational age), and severe bronchopulmonary dysplasia (patients in need of supplemental oxygen at 28 days of life and maintaining the need for FiO2 higher than or equal to 30% or positive pressure at 36 weeks of corrected gestational age)14,15.
The associations between the factors of interest and BPD and their classifications were made by adjusting single and multiple log-multinomial regression models, obtaining crude and adjusted relative risks (RRaj), and using gestational age and antenatal corticosteroids as covariates. The software SAS 9.4 was used.
For the elaboration of the predictive instruments, the database was randomly divided into a training sample (70%), used for the elaboration of the moderate and severe BPD prediction equations, and the rest of the sample (30%) was used for validation.
For the instrument’s construction, respiratory supports were classified into three categories: none, use of nasal oxygen catheter/CPAP/NIPPV, and conventional mechanical/high-frequency ventilation.
Four strategies were proposed for building the equations, as described in table 1. Thus, the probabilities of BPD were obtained, considering moderate and severe degrees versus mild and null, thus being a binary variable. A logistic regression model allowed each of the four variable selection strategies to result in BPD prediction probabilities (p).
The equations were obtained from the training sample and validated in the validation sample; that is, the probabilities were obtained from the data of the last-mentioned sample, from which the probability of BPD was also calculated using the NICHD instrument currently used in that institution. During validation, patients on an O2 catheter had FiO2 values classified as 29%; the entire sample was considered ”Hispanic” regarding ethnicity when using the NICHD instrument.
The results were compared with the actual status of the sample for BPD (confusion matrix), and then sensitivity, specificity, and predictive values were calculated. Kappa indices were calculated to compare the results obtained by the equations with those obtained by the NIHCD method, obtaining agreement between the methods, given that none of the methods can be considered a gold standard.
The project received ethical approval from the Institution’s Human Research Ethics Committee of the Ribeirão Preto Medical School (CEP FMRP), Opinion number: 1.903.783/CAAE 63764517.4.0000.5505).