ABSTRACT 413

Abstract

Background and aims: 59.7% of admissions to Melbourne’s Royal Children’s Hospital (RCH) ICU require intubation. 10.3% of these fail extubation and require unplanned reintubation. Accurate prediction of children at risk of extubation failure by clinicians may help avoid the increased morbidity and mortality associated with this outcome. No studies to date have clinically investigated the accuracy of clinician predictions of extubation failure. Aims: To identify the accuracy of clinician predictions of extubation outcome and the clinical predictors they rely upon. Methods: A 6-month prospective cohort study and 12-month retrospective chart review was conducted at RCH ICU. Extubation failure was defined as unplanned reintubations <72rs or <2ks if severe respiratory distress/steroids/nebulised adrenaline commenced <72rs and continued until failure. Confidence of success was recorded before extubation and clinician seniority was noted. Clinical parameters were measured <1hr pre-extubation. Logistic regression and area under receiver operating characteristic curves (AUROC) were used to generate and compare accuracy of predictive models. RCH ethics approved the study and waived the need for informed consent. Results: Air-leak, sedation level and ventilator mode were most commonly used to predict outcome in children at perceived risk of failure, though were poorly accurate (AUROC 0.60). Clinician predictions were moderately accurate (AUROC 0.73). Accuracy did not significantly vary with increasing seniority. Conclusions: Clinicians predicted extubation outcome with accuracy greater than the clinical predictors they reportedly used. This reflects an excellent ability of ICU clinicians to interpret clinical information. Predictive accuracy and patient outcomes may be improved by clinician education regarding strong clinical predictors of extubation outcome.