Performance of published scoring tools for predicting the risk of perioperative respiratory adverse events in children – An evaluation in a large paediatric cohort

Résumé fourni par la source

• Perioperative respiratory adverse events (PRAE) are a main cause of morbidity and mortality in paediatric anaesthesia; therefore, clinicians need to be able to predict their patients’ risk of PRAE to plan their care. • Large-scale assessment of the predictive performance of six common risk prediction tools. • While relating to an increased incidence of PRAE, the tools had overall a poor discriminative ability. • Best performing tool was the OLDS score, selecting predictor variables based on clinician knowledge rather than data mining or statistical significance thresholds. Perioperative respiratory adverse events (PRAE) are a main cause of morbidity and mortality in paediatric anaesthesia. Clinicians need to be able to predict their patients’ risk of PRAE to plan their care. Clinical risk prediction tools have been developed to assist with pre-operative risk stratification; however, validation outside the contexts of their development is limited. In this study, we test the ability of common risk prediction tools to identify patients at high risk of PRAE in general anaesthesia. In this post-hoc secondary analysis, six risk prediction scores were evaluated in 12,364 cases of general anaesthesia in children. Area Under the Receiver Operator Characteristic curves (AUC) were calculated for each as the primary measure of predictive performance, along with Positive Predictive Value and Negative Predictive Value. The rate of PRAE in our sample of 12,364 cases was 11.9% for any PRAE or 2.0% for severe PRAE (bronchospasm and/or laryngospasm). Although each of the tools assessed was positively associated with the occurrence of PRAE, we found poor to moderate predictive ability for each of the risk prediction tools assessed. AUCs ranged from 0.562-0.649 for the prediction of any PRAE, and 0.509-0.614 for the prediction of severe PRAE. The best performing tool was the OLDS score (AUC 0.649) adapted from Lee 2018, which selected predictor variables based upon the beliefs of clinicians rather than on data mining or statistical significance thresholds. The discriminative ability of existing PRAE risk prediction tools was poor in our test group. These scales would need to be carefully adjusted for use in clinical environments other than those in which they were developed. There remains a gap for a well-validated and user-friendly PRAE prediction tool to fit into clinical practice. Australian and New Zealand Clinical Trials Registration (ACTRN12624000018516)

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.