Automated Prediction of Cardiorespiratory Deterioration in Patients With Single Ventricle
Introduction.
In this study(1), Rusin et al. note that of all infants with congenital heart disease, those with a single ventricle circulation represent a population with the most brittle physiology and most likely to experience cardiorespiratory deterioration in the perioperative phase of care as a neonate. Despite advances in inpatient care since the initial stage 1 palliation in 1980, morbidity associated with clinical deterioration remains unacceptably in the inpatient care setting(2).
Objective.
The goal of this study was to validate a computer algorithm designed to recognize a physiologic signature of impending clinical deterioration in a unique population of patient prone to deterioration, with a recognition in a clinically actionable timeframe (i.e. 1-2 hours prior to an event).
Methods.
This was a single center, observational study undertaken over the course of 6 years through 2018 and included all neonates with single ventricle physiology who underwent a stage 1 palliation during the course of their hospitalization. Physiologic data (waveforms and vital signs) were continuously acquired within both the ICU and step-down environments, with endpoints defined as either a cardiac arrest or an unplanned endotracheal intubation. The authors designed a classification algorithm which incorporated physiologic signals including heart rate, heart rate variability (HRvar), peripheral arterial oxygen saturation, ST segment elevation, ST segment variability (STvar), PVCs and pleth variability index (PVI), of which HRvar, STvar, and PVI were processed from raw physiologic data. Study data was classified as either pre-deterioration or non-deterioration classes, and further categorized into a training (50%) and validation (50%) sets at random.
Results.
A total of 238 subjects met study inclusion criteria with over 300,000 hours of complete physiologic data available for analysis, of which a total of 138 clinical deterioration events were identified. The resulting model generating a Risk Index (RI) was both highly sensitive and specific for recognizing and differentiating pre-deterioration data, with an AUC of the training set of 0.960 (0.954-0.966) and 0.958 (0.950-0.965) for the validation subset. Authors describe a detailed sensitivity analysis demonstrating between 50%-70% of clinical deterioration predicted up to 120 minutes prior to the actual event. Mindful of alert fatigue, the authors also describe a thorough analysis of “Alarms per Day Per Patient” at varying risk index alert thresholds.
Conclusions.
Taken together the authors describe a validated, real-time algorithm with capability to identify imminent clinical deterioration in a population of patient with a brittle single ventricle physiology, within a time frame which allows for preventive clinical measures. The authors report that work is underway to assess generalizability and reproducibility through a multicenter trial.
References
1. Rusin CG, Acosta SI, Vu EL, Ahmed M, Brady KM, Penny DJ. Automated Prediction of Cardiorespiratory Deterioration in Patients With Single Ventricle. J Am Coll Cardiol. 2021 Jun 29;77(25):3184–92.
2. Norwood W, Lang P, Hansen D. Physiologic repair of aortic atresia-hypoplastic left heart syndrome. N Engl J Med. 1983;308(1):23–6.
