O1004: From Watchful Waiting to Urgent Action: Modeling Risk of Cardiac Intensive Care Unit (CICU) Escalation in Pediatric Cardiac Care

Background: In pediatric cardiac care, the cardiac acute care unit (CACU) serves as a bridge between home and intensive care. However, some patients directly admitted to the CACU experience an unexpected clinical deterioration requiring emergent transfer (ET) to the CICU. These patients who receive interventions such as vasoactive support or intubation within one hour of transfer have significantly increased in-hospital mortality and length of stay, underscoring the need for effective, early risk stratification. Identifying clinical markers preceding deterioration may support timely intervention. This study used logistic regression (LR) to identify predictors of CACU-to-CICU ET with the long-term goal of informing real-time risk models.

Methods: This is a single-center retrospective study using local Pediatric Acute Care Cardiology Collaborative/Pediatric Cardiac Critical Care Consortium (PAC3/PC4) registry data. We analyzed CACU-to-CICU transfers between February 1, 2019 and May 9, 2024. We excluded patients who had been to the operating room or cardiac catheterization lab prior to CICU admission. ET was defined as requiring vasoactive support or intubation within 60 minutes of CICU arrival. CACU encounters were divided into 12-hour shifts; those ending in ET were labeled “ET shifts” and compared to non-ET shifts. Predictor variables included respiratory decline (e.g., initiation of high-flow nasal cannula, CPAP, intubation), diagnosis grouping, surgery status, chromosomal/genetic abnormalities, and weight. We performed univariate and bivariate analysis, followed by a mixed-effects LR model with patient ID included as a random intercept to account for multiple shifts per patient.

Results: We identified 8,472 CACU encounters, 84 (1.0%) of which resulted in an ET. In the multivariable model, respiratory decline in the prior 12 hours was associated with a 15.6-fold increase in odds of ET (OR 15.55, 95% CI 2.61–92.57, p=0.003). Univentricular heart disease was a significant predictor (OR 4.67, 95% CI 4.64–4.69, p< 0.001), and there was a statistically significant interaction between diagnosis grouping and weight. Specifically, lower weight was associated with higher odds of ET in patients with more complex CHD (OR 0.63, 95% CI 0.63–0.63, p< 0.001). Notably, chromosomal/genetic abnormalities were associated with decreased odds of ET (OR 0.23, 95% CI 0.06–0.98, p=0.047). The model demonstrated strong within-subject correlation (intraclass correlation coefficient = 0.94), indicating substantial variability explained by patient-level differences.

Conclusion: Our LR model identified several key predictors of emergent CACU-to-CICU transfer, including recent respiratory decline, univentricular physiology, and lower weight. The association between chromosomal/genetic abnormalities and lower odds of ET was unexpected and warrants further investigation. Although the model demonstrates high between-patient variability, the low prevalence of events underscores the complexity of forecasting deterioration at a population-level, highlighting the need for incorporation of additional granular data to enhance the model’s discriminative power. Future work will explore machine learning to improve predictive accuracy and support early warning systems in CACU settings.