Prospective validation of a brief questionnaire for predicting the severity of pediatric obstructive sleep apnea

Introduction: Pediatric obstructive sleep apnea (OSA) is diagnosed and stratified by polysomnography. However, due to cost and inaccessibility, up to 90% of children undergo tonsillectomy and adenoidectomy (T&A) solely based on clinical criteria. We previously developed a data-driven brief screening questionnaire ('Selected Features,' SF) that predicted OSA severity than alternatives. The SF asks the parent whether a child: (i) has had breath-holding spells at night over the past 4 weeks, (ii) is a mouth-breather during the day, (iii) has stopped growing at a normal rate any time since birth, and (iv) is overweight. This study sought prospectively validate the SF questionnaire. Methods: We conducted a prospective assessment of the predictive accuracy of SF compared to the Pediatric Sleep Questionnaire-Sleep Related Breathing Disorder (PSQ-SRBD) scale in otherwise healthy children with sleep disordered breathing referred for T&A. We compared the model fits of PSQ-SRDB and SF for (i) a linear regression model for the prediction of OSA, and (ii) a logistic regression model for severe OSA, defined as apnea hypopnea index (AHI) > 10. P < 0.05 was significant. Results: A total of 124 patients were included. The average age was 7.3 years (95% confidence interval, 6.6-8.0) and 66 (54%) were male. The racial composition was 54 (44%) black, 41 (33%) white, and 28 (23%) other. The median AHI was 4.8 (interquartile range 12) and 43 (35%) of patients had severe OSA. In linear and logistic regression models, SF outperformed the PSQ-SRBD and null models as measured by Akaike Information Criteria. The overall accuracy in predicting AHI >10 for PSQ-SRBD was 0.65 (0.56-0.73, P = 0.54) compared to 0.73 (0.64-0.80, P = 0.04) for SF. Conclusion: By eliminating redundancy, we have developed a questionnaire with improved prediction of OSA and its severity, in children with high pre-test probability of the condition. While multi-site validation is necessary, SF demonstrates value in screening children prior to T&A in resource-limited environments.