4 Discussion
Our study confirms that a BPD prediction instrument has superior performance when developed in the place where it will be applied, demonstrating that individualized treatment has been increasingly gaining strength in neonatology.
We found a total incidence of BPD of 42% when considering the need for respiratory support at 36 weeks of GA, corresponding to what has been observed in recent years in the literature 16-21 and demonstrating the immense need for therapies that can reduce the world’s DBP rate.
The presence of ductus arteriosus and the need for surfactant were associated with moderate BPD, whereas early sepsis was associated with severe BPD. Several publications have associated these predictors with BPD 1-3,22-25. High oxygen requirement at 14 days of life and non-invasive or invasive ventilatory support were events strongly associated with the moderate and severe forms of the disease, confirming well-established data in the literature on the role of respiratory support in the development of BPD3,10,23,26-32.
The current foreign instrument showed low sensitivity, good specificity, and a good positive predictive value. However, it presented specificity values and a positive predictive value lower than those of our equations (Table 3), with a low agreement of this instrument with the instruments elaborated in this study, with Kappa indices below 0.50 (Table 4).
We also observed the impossibility of analyzing newborns with birth weights less than 501 g and SGA newborns in the external instrument currently used at the service. It shows that it is indicated for evaluating only patients born at more than 23 weeks and appropriate for gestational age, making our equations superior. The sample had 21.1% SGA, reflecting the number of developing countries where many SGA newborns occur 21.
Another crucial point was that the classification of our sample as “Hispanic” may limit the veracity of the results according to the instrument currently in use because the Brazilian has great miscegenation and genetic diversity 33.
The NICHD model published in 2011 was based on a population whose characteristics and neonatal care differ from ours, and the author himself recommends its careful use in other regions10.
NICHD published 2022 an update of the risk prediction instrument for BPD34. The model predicts the risk for the disease based on a new BPD classification proposal 35, no longer using ethnicity as a predictive factor, which may be helpful in other populations. However, this new instrument can also not predict risk in SGA patients. It is restricted to the population with GA up to 28 weeks34, limiting its use in places where BPD is still considerable in newborns with GA up to 30 weeks 21.
Some other risk prediction models for BPD have been published in recent years. Bhering published a model in Brazil that calculates the risks for BPD on the seventh day of life but without considering the classifications of severity 23, a fact that may limit its usefulness for the decision on the use of postnatal corticosteroids. In Turkey, Gursoy developed an instrument that assesses the risk for BPD through a scoring system, using variables in the first 72 hours of life, also without defining the severity of the disease but suggesting the use of postnatal corticosteroids when the score is higher than 636.
Sharma in the United States 29 and El Faleh in Switzerland 31 built practical and easy “web-based” instruments with prediction capable of identifying later lung deterioration phenotypes and predicting severity. However, these models may not reflect the reality of most neonatal units, ethnically representing only the local population.
Zhang published a nomogram in China that is easy to use in clinical practice. However, the risk calculation is only performed on the seventh day of life and using the serum level of NT-pro-BNP (N-terminal-pro brain natriuretic peptide) 30, a marker not readily available in many services, especially those with limited resources. Also, identification before 14 days of life may fail to identify later phenotypes of BPD development 37,38.
We sought to develop a prediction model using newborns in our population as a basis, enabling the appropriate indication of the use of postnatal corticosteroids, as they may be associated with side effects. Therefore, the indication must be precise, requiring good specificity and good positive predictive value, which are fundamental to guiding the choice of patients.
We observed that all the elaborated equations presented a superior performance than that estimated by the calculator currently in use at the service. However, Equation 2, which considered only the variables that presented p<0.05, had the best specificity and positive predictive value.
Our study was conducted retrospectively, but we emphasize that the data were collected prospectively and according to well-established criteria (Brazilian Neonatal Network/Vermont-Oxford Network)39. A sample from a single Center of Neonatology was also used in the research. However, this fact can be seen as a workforce, as we have a homogeneous population and uniform conduct.
Predicting the risk for a disease such as BPD is still a challenge. The literature suggests that predictive instruments external to the place should be validated, as it is the most appropriate way to use these calculators in clinical practice safely. It reinforces that, when possible, elaborating an equation specific to each service is still ideal 11,40.
Therefore, developing our prediction instrument, with specificity and positive predictive value higher than the currently used, will allow us to define more precise candidates for postnatal corticosteroids to prevent BPD. Prospective studies to evaluate its clinical impact have been designed.