Evaluation of risk prediction model for perioperative respiratory adverse events in pediatric anesthesia

Abstract

Background: Perioperative respiratory adverse events are among the most common critical incidents in pediatric anesthesia. Three risk prediction models have been developed to predict occurrence of such adverse events in children. However, these tools were only internally validated, limiting generalization. The Perioperative Respiratory Adverse Events in Pediatric Ambulatory Anesthesia risk prediction tool developed by Subramanyam et al. consists of five predictors: age ≤ 3 years, ASA physical status II and III, morbid obesity, preexisting pulmonary disorder, and surgery. Aims and Methods: We aimed to evaluate the suitability of Subramanyam’s model in predicting the occurrence of perioperative respiratory adverse events in a more general tertiary care pediatric population, including anesthesia for both outpatient and inpatient procedures. Therefore we validated this scoring system in a tertiary care cohort of 204 children included in the APRICOT study at our hospital through retrospective analysis of this data. Secondarily, we aimed to study the incidence of perioperative respiratory adverse events in our hospital. Results: Overall incidence of perioperative respiratory adverse events in our sample was 19,6%. Applying Subramanyam’s prediction model to our cohort, we found no patients categorized as low risk, 76 patients as intermediate risk and 128 patients as high risk. Discriminatory ability of the risk scoring system was modest, with AUC of the simplified model 0,65 (95% CI 0,57-0,74) and AUC of the original logistic regression model 0,66 (95% CI 0,57-0,75). Calibration of the simplified model was rather poor, with Brier score 0,49. The original logistic regression model calibrated better, with Brier score 0,18. A subgroup analysis considering solely ambulant patients in Ghent-APRICOT yielded comparable results. Conclusions: We conclude that the overall performance of Subramanyam’s risk prediction tool in our cohort was moderate. Modest discrimination and calibration suggest that the risk score may not reliably predict perioperative respiratory adverse events in individual patients in our tertiary care pediatric population. Therefore the clinical relevance of the implementation of this scoring system in our tertiary hospital would be negligible, which leaves us with the lack of good scoring systems to predict perioperative respiratory adverse events in our population. In addition, we found the incidence of these adverse events in our hospital to be markedly higher as compared to the sample of Subramanyam.