376 - Investigating Vital Sign Trends Prior to Emergent Intervention in Patients with Bronchopulmonary Dysplasia

Publication Number: 1360.376

Addilyn B. Smith, Cincinnati Children's Hospital Medical Center, Cincinnati, OH, United States; Nathan M. Pajor, Cincinnati Children's Hospital Medical Center, Cincinnati, OH, United States

Background: Bronchopulmonary dysplasia (BPD) is a common pathology in preterm infants, associated with respiratory instability, need for prolonged respiratory support, and increased incidence of emergent interventions. Early identification of vital sign trends suggesting an impending need for emergent intervention may mitigate the risk of morbidity and mortality in this population. However, there is a paucity of models predicting short-term outcomes for BPD.

Objective: The goal of this research is to identify features in vital sign data prior to code events that differentiate those time periods from normal physiological data and to describe those patterns for use in predictive models for early identification of the need for emergent intervention.

Design/Methods: We conducted a retrospective analysis of events in which a code alarm was activated and an intervention was performed in patients diagnosed with BPD aged 3 months to 6 years (Table 1). A chart review was conducted to determine the intervention performed and the nature of the event and intervention. Events were classified into categories, such as endotracheal (ET) tube malpositioning and obstruction. Per-minute vital signs and ventilator data were queried for each event using BedmasterEX and assessed for consistent patterns. Additional complexity metrics-variability, self-similarity, and divergence - were also calculated and assessed for the 24 hours prior to the emergent intervention.

Results: Common trends included pulse and respiratory complexity metrics peaking in the 3-4 hours prior to emergent intervention, followed by a rapid return to baseline (Figure 1). These trends were most consistently present for events classified as progressive respiratory distress or airway obstruction. Ventilator data for clampdown events showed a decrease in maximum expiratory minute volume in the 2-4 hours preceding emergent intervention (Figure 2).

Conclusion(s): Our preliminary findings suggest the presence of patterns in vital sign data prior to events requiring emergent intervention. These characteristics of vital signs and ventilator data show promise as potential predictive features for machine learning models to predict the impending need for emergent intervention. However, additional research is necessary to prove the utility of these and other vital sign trends as meaningful features for predictive modeling.

Table 1 | Demographic Descriptors of Patient Population
Among our sample of 26 code events in 18 patients, the average age at the time of emergent intervention was 2 years 5 months ± 1 year 11 months. The average gestational age at time of birth for this population was 28 weeks + 1 day ± 2 weeks + 3 days. Average birth weight was 1178 g ± 413 g.

Figure 1 | Patient-level Complexity Metrics Surrounding Emergent Intervention
Mathematical representations of the time series data's complexity (Normalized Spectral Entropy), divergence (Rosenstein's Lyapunov Exponents), and self-similarity (Higuchi's Fractal Dimensions) were calculated and plotted for per-minute vital sign data. This example demonstrates these complexity metrics for a patient's respiratory rate in the 24 hours prior to and following an emergent intervention. Notably, this example demonstrates a sharp increase in one of the complexity metrics in the 3-4 hours prior to the emergent intervention, followed by a rapid return to baseline.

Figure 2 | Patient-level Ventilator Data Preceding Emergent Intervention
Mean, median, standard deviation, maximum, and minimum for per-minute vital sign and ventilator data were calculated for the 24, 12, 6, 4, 3, 2, and 1-hour periods prior to emergent intervention for each patient. This example demonstrates a dramatic decrease in the maximum expiratory minute volume in the 2-4 hour period prior to emergent intervention.