The Future of Outcome Prediction for Preterm Infants in the Neonatal ICU.

Outcome prediction for prognosis or risk stratification is particularly important in the neonatal ICU as critically ill neonates are at high risk of mortality or long-term morbidity. There are several commonly used outcome prediction tools in neonatology. The majority of these tools use static measures such as data available at birth or within 24 hours of birth to predict the risk of subsequent mortality or neurodevelopmental impairment (1–3). Gestational age and birth weight are among the best predictor variables (4), such that clinical trials in preterm infants frequently use one or both of these variables for stratification to ensure balanced allocation. As early predictors do not take into account risk factors and complications that become apparent later during the hospitalization, tools incorporating respiratory support and selected postnatal morbidities were developed to better predict mortality, bronchopulmonary dysplasia, and neurodevelopmental impairment (5, 6). The use of prediction tools in neonatology to guide clinical decisions has not been tracked formally, and the impact on clinical care is unknown. A limitation of most of the currently available prediction models in neonatology is their reliance on data available at birth with limited or no sequential clinical and/or laboratory data.

In this issue of the Journal, Lavilla and colleagues (pp. 75–87) examined the discriminatory power of hourly changes in neonatal sequential organ failure assessment (nSOFA) scores from birth to predict death among 436 extremely preterm and extremely low-birth-weight infants. Hourly kinetics of the nSOFA score were strong predictors of mortality before discharge and within 24 hours after birth (7). The average score over the first 28 days was also associated with hospital mortality and major morbidities. Although mortality is rightly considered the most important outcome in neonatology, it is imperative to acknowledge that neurodevelopmental assessment at a later age was not addressed in the current prediction study. Neurodevelopmental impairment may be as important or more than some of the outcomes reported, including severe intraventricular hemorrhage, bronchopulmonary dysplasia, sepsis, necrotizing enterocolitis, and retinopathy of prematurity.

The nSOFA scoring system is based on the presence of mechanical ventilation and oxygen saturation as measured by pulse oximetry:FIO ratio for respiratory dysfunction, presence of vasoactive medications and/or corticosteroids for cardiovascular dysfunction, and platelet count for hematologic dysfunction. These measures may be dependent on clinical practice. For example, the nSOFA scores may be artificially lower in a clinical setting where hypotension is managed more conservatively (8). Similarly, in centers with more use of mechanical ventilation rather than continuous positive airway pressure, use of lower rather than higher oxygen saturation targets (9), or avoidance of higher FIO in favor of higher mean airway pressures, the model may not be as applicable. Thus, the generalizability of the results of this study may be limited and center-specific and may need to be validated externally.

The study by Lavilla and colleagues does not define if these are the best variables or weighting to predict the risk of adverse outcomes. It is possible that prediction might be improved using optimization of cutoff values or more granular continuous respiratory support data, additional vital signs measures, or more commonly collected laboratory values such as those available on standard blood gas analyzers. In the current study, the nSOFA score was calculated retrospectively. It needs to be determined if prospective nSOFA scores would improve clinical management decisions in neonatology. It is also important to note that although the nSOFA score is associated with poor outcomes, it does not follow that reducing the score by interventions would improve outcomes. The effect of artificially improving nSOFA scores by increasing platelet counts (10), limiting FIO exposure in favor of increases in mean airway pressures, or avoiding vasoactive medications could have unintended consequences in sick neonates. It is also likely that measures of critical illness severity leading to changes in the nSOFA score such as the need for mechanical ventilation or vasoactive drugs are readily identified and perceived by clinicians caring for extremely preterm infants and hence may not alter a physician’s intuitive assessment of prognosis (11).

Scoring systems need to be validated prospectively and in multiple settings before wider clinical use. The heart rate characteristics (HeRO) score that assesses measures of heart rate variability is an excellent example of a prediction system successfully tested from bench to bedside, including in a large multicenter randomized clinical trial (12) before its introduction into routine clinical care. A HeRO score that increases substantially above the baseline value or is high may indicate that the infant is developing sepsis. Clinicians provided with HeRO score data decreased the time to recognition and treatment of sepsis (12), which decreased the risk of mortality (13) in the randomized controlled trial. The HeRO score was associated with a lower risk of neurodevelopmental impairment compared with routine monitoring, likely because of early detection and treatment of sepsis (14). The authors reported recently that the HeRO score and nSOFA score provide complementary information on sepsis risk and sepsis-related mortality, with HeRO scores increasing before the blood culture, whereas the nSOFA scores increased at the time of culture (15).

However, the HeRO score relies on a single physiological value (heart rate variability), which may lead to low diagnostic accuracy in specific circumstances. For example, recuperating preterm infants taken off continuous positive airway pressure may develop increased episodes of bradycardia leading to an elevated HeRO score due to deescalation of respiratory support, whereas preterm infants with severe intraventricular hemorrhage may have a persistently elevated HeRO score. False negative results may occur among infants if the cause of deterioration is not due to sepsis, such as among infants with critical congenital heart disease. It is likely that future scoring systems that incorporate sequential physiological, laboratory, and clinical data may further improve the prediction or detection of infants at risk for death and major morbidities, including neurodevelopmental outcome. The advent of artificial intelligence into decision support tools, learning health systems, and evidence-based algorithms embedded into electronic medical records promises to improve prediction and real-time detection of infants at risk of adverse outcomes and inform clinicians so they can further improve the care of our most vulnerable infants.