OSA screening accuracy
Figure 3 shows the ROC curve for the predicted probabilities derived from the MOP model alongside ROC curves for the SRBD survey and BMI z-score as an objective measure of body size. The NPVs, PPVs, sensitivities and specificities are also presented in Figure 3 . The AUC was statistically significantly greater for the MOP model compared to the SRBD (P =0.007). Of note, the AUC for the BMI z-score was statistically significantly greater than that of the SRBD (P =0.038), but not statistically different compared with the AUC of the MOP model (P =0.500).
The optimal OSA screening cut-off for the MOP model was 0.428 or an OSA probability of 42.8%, while that of BMI z-score was 2.07, which corresponds to the 98th BMI percentile (Figure 4 ). Logistic regression analyses revealed 6-fold higher odds of having OSA for children with OSA probabilities ≥ 42.8% compared to those with OSA probabilities < 42.8% (OR = 5.73, 95% CI = 2.56 – 13.53, P<0.001), and 8-fold higher odds of having OSA for children with BMI z-scores ≥ 2.07 (OR = 8.35, 95% CI = 3.62 – 20.91, P<0.001) compared to those with BMI z-scores < 2.07.