Purpose: Very low birth weight (VLBW < 1500 grams) infants in the Neonatal Intensive Care Unit (NICU) are at risk for respiratory deterioration requiring endotracheal intubation and mechanical ventilation, with associated morbidities. Methods for predicting impending respiratory failure are needed. Our objective was to develop automated algorithms for continuous analysis of cardiorespiratory waveforms and vital signs to predict respiratory failure requiring intubation in VLBW infants.
Methods: We collected continuous cardiorespiratory and demographic data, and types and times of respiratory support on all VLBW infants admitted to the University of Virginia NICU from January 2009-June 2011. We identified non-elective intubations that were followed by mechanical ventilation for at least 12h. Over 25 physiological measures were tested for their relationship to intubation. A multivariate logistic regression model was used to develop a respiratory failure index, the relative risk of urgent intubation in the next 24 hours.
Results: Of 287 VLBW infants admitted, 96 urgent intubations in which there were at least 12h of waveform data occurred in 51 patients. Oxygen saturation and its correlation with heart rate, correlation of heart and respiratory rate, and apnea burden were independent predictors and were included in the model, which has a ROC area of 0.84. The figure shows the median and quartiles of the model output, normalized to the relative risk of intubation in the next 24 hours. There was a significant rise in the respiratory failure index over the 24h prior to intubation. The large dots on the median at 36 and 12 hours prior to intubation are significantly different using the signed rank test (p=0.001).
Conclusion: Continuous computer analysis integrating multiple cardiorespiratory waveform patterns and vital signs can predict respiratory failure as much as 24h prior to urgent intubation, perhaps allowing more aggressive preventive interventions
Funded by NIH 1RC2HD064488