A prediction model for the presence of moderate/severe obstructive sleep apnoea in childhood based on clinical findings and oximetry.

Abstract

Introduction: Formally defining the presence of obstructive sleep apnoea (OSA) in a snoring child requires polysomnography (PSG), a test with limited availability. In this study we aimed to examine which combination of clinical factors, oximetry and actigraphy measures provide the most effective model for prediction of moderate-severe OSA (MS OSA) for possible use as a tool outside specialist centres. Method(s): Children referred for PSG for possible OSA were recruited prospectively and underwent clinical investigations including: OSA-18 questionnaire, physical examination, and home overnight oximetry (Masimo Radical 2-sec averaging time) with concurrent actigraphy. Variables assessed for inclusion in the predictive model were: age, gender, OSA-18 (total, sub-scale and individual question scores), BMI z-score, tonsil size, Mallampati score, Friedman pharyngeal score, Angle's dental malocclusion score, oximetry parameters (SpO2 nadir, dips below 90%/h, dips of 4%/h, pulse rate (PR) standard deviation and PR increases of 15 bpm/h), and movement fragmentation index from actigraphy. The primary outcome variable dichotomised the obstructive respiratory disturbance index from PSG into <5/h (primary snoring/ mild OSA) and >=5/h (MS OSA). Stepwise forward elimination logistic regression was used to define significant variables (p > 0.1) and a model was then developed to predict MS OSA. Result(s): 89 children (38F; mean +/- SD age 5.5 +/- 2.5 y) with full PSG results were included. 37 (42%) had MS OSA. OSA-18 Q16 (parent concerned child not getting enough air), dental malocclusion score, SpO2 nadir and PR increases of 15 bpm/h were identified as significant predictor variables and the model was developed on these parameters, dividing SpO2 nadir and PRI15 into tertiles and OSA-18 Q16 scores into two categories. ROC analysis gave an area under the curve of 0.88 using this model, indicating very good discrimination. 81% of children were correctly classified, with sensitivity 82%, specificity 81%, positive predictive value 77% and negative predictive value 85%. Conclusion(s): We have developed a clinical risk score for having moderate-severe OSA that has very good discrimination properties and correctly classifies 81% of children referred for PSG.