Development and validation of a perioperative pediatric cardiac risk calculator

• In the pediatric population, there is a notable dearth of studies identifying pre-operative variables to predict the risk of post-operative cardiac arrest. • In this study the incidence of cardiac arrest within 30 days of surgery was 0.22% (910 patients). • Younger age, Ventilator dependence, Oxygen support, lung and airway abnormalities, nutritional support, cancer, and higher ASA classification were identified as factors that increase risk of post-operative cardiac arrest. • Logistic regression model with reduced variables demonstrated an excellent discrimination and calibration in predicting this rare event in a prospective analysis. Numerous tools for cardiac risk stratification exist for the adult population. While each has its strengths, there remains a need for a calculator that reliably predicts the cardiac risk across all types of pediatric surgery. We aim to develop and validate a novel risk stratification tool predicting the probability of 30-day post-operative cardiac arrest in pediatric surgical patients. Data from the 2020-2022 NSQIP Pediatric Database were analyzed. Forty nine risk factors were identified from the NSQIP database and used as potential predictors. Multicollinearity was evaluated by calculating the variance inflation factor for each predictor, and a reduced set of 21 variables was selected. Multiple models were created to assess the impact of different preoperative factors on the 30-day cardiac arrest risk using reduced and full variables. Calibration plots comparing the predicted probabilities with the actual outcomes across different probability bins were generated, and the Brier Score, C-Statistic, and H-L Values were calculated and used to compare the models’ performance. Validation was then performed using NSQIP Pediatric Data from 2018-2019, and sensitivity and specificity were calculated. From this data, an interactive application was created. Of all models developed and tested, the logistic regression model with reduced variables produced the best overall performance with a Brier score, C-statistic, and H-L value of 0.0021, 0.94, and 0.25, respectively. Upon validation, the sensitivity and specificity were 0.90 and 0.82, respectively. The developed and validated novel risk calculator accurately predicts the risk of post-operative cardiac arrest in pediatric patients.