Predicting Duration of Invasive Mechanical Ventilation in Pediatric ICUs

Whereas invasive mechanical ventilation is crucial for supporting critically ill children, it is associated with several iatrogenic complications ranging from infection to mechanical lung injury and prolonged exposure to sedatives and neuromuscular blockade.¹ Consequently, the importance of early liberation from mechanical ventilation has been recognized as a key quality metric by the Society of Critical Care Medicine ICU liberation program and the recently published expert consensus on pediatric ventilator liberation.^2-3

Accurate prediction of mechanical ventilation duration can enhance clinical decision-making and achieve early liberation from the ventilator. Additionally, with the significant reductions in mortality rates within pediatric ICUs (PICUs) over the past few decades, researchers now consider the duration of mechanical ventilation as an alternative outcome measure.⁴ However, there is currently no standardized benchmark for the duration of invasive mechanical ventilation in the PICU. In this issue of Respiratory Care, Rogerson et al⁵ present a retrospective cohort study in which the researchers developed and validated a multi-center prediction model for the duration of invasive mechanical ventilation, aiming to establish a standardized ratio for this important measure.

In this study, the authors utilized registry data from 157 institutions in the Virtual Pediatric Systems database, encompassing encounters in the PICU between 2012–2021.⁶ Subjects included those who underwent endotracheal intubation and received invasive mechanical ventilation for > 24 h on the first day of PICU admission. Four different prediction models were developed and compared, including LASSO, random forest, gradient boosted machine, and Pediatric Risk of Mortality III (PRISM III). PRISM III is one of the most widely used and validated scoring systems in pediatric critical care for quantifying the severity of illness and risk of mortality.⁷ It employs 17 predetermined physiologic variables to quantify mortality risk and is commonly used in PICU research and quality improvement initiatives.⁸ Although PRISM III is not designed or tested for predicting the duration of mechanical ventilation, the authors included it as a reference to compare the more advanced models.

All the models performed well in predicting duration of mechanical ventilation at a large-cohort level, with observed-to-expected ratios close to one, indicating accurate predictions. Among the models, the random forest model demonstrated the best performance. However, when it came to predicting the duration of mechanical ventilation at the subject level, the models showed poor performance.⁵

The findings of this study have significant implications for quality improvement initiatives and benchmarking in PICUs. Standardizing the duration of invasive mechanical ventilation can provide a valuable metric for assessing and comparing institutional performance. By using the prediction model developed in this study, PICUs can identify institutions with low observed-to-expected ratios and investigate the best practices they implement. Additionally, the model can be used to track an institution’s performance over time after implementing changes in ventilator liberation practices.⁹

On the other hand, the predictive models cannot be utilized for making decisions at the patient level. Perhaps their predictions could be enhanced if provided with more detailed patient data pertaining to illness and pulmonary status. Further research is necessary to improve the performance of these models.

Despite these limitations, the study provides valuable insights into the variability in duration of invasive mechanical ventilation across institutions and highlights the need for standardization in PICU practices. The model developed in this study serves as a starting point for establishing benchmarks and promoting quality improvement in ventilator liberation efforts. As we strive to improve outcomes for critically ill children, it is crucial to continue refining and expanding our understanding of optimal ventilation practices. By adopting standardized benchmarks for duration of invasive mechanical ventilation in the PICU, health care providers can work together to minimize the risks associated with prolonged intubation and enhance patient outcomes.

Future research should focus on determining factors that affect the model accuracy and incorporating extubation failure and re-intubation predictions. Additionally, efforts should be made to develop patient-level prediction models for duration of invasive mechanical ventilation in pediatrics, as this remains an important but elusive goal.

In conclusion, the methodology employed by Rogerson et al,⁵ including the use of a large, multi-institutional database and machine learning models, provides a robust framework for developing an accurate predictive model for duration of invasive mechanical ventilation in the PICU. By addressing the challenge of variability in timing and duration of mechanical ventilation, this research contributes to the goal of improving ventilator liberation practices, optimizing patient outcomes, and reducing the morbidities associated with invasive mechanical ventilation in critically ill children.