Screening for Obstructive Sleep Apnea: Bayes Weighs in

A fundamental challenge associated with screening tests is recognition of the impact of disease prevalence upon the predictive value of the result. For example, in the common circumstance of screening for low prevalence dis- eases, even good tests may have unacceptably high false positive rates. The converse situation, screening in high preva- lence populations, is less common but occurs with obstructive sleep apnea (OSA): even ostensibly good screening tests may have unacceptably high false negative rates. The challenge of recognizing false negative OSA screening results has important implications as screens are increasingly implemented in high risk populations. This raises two clinically impor- tant questions: 1) How sensitive and specific should a screening test be to minimize false negative results across a spec- trum of baseline prevalence; and 2) Given a screening test with known sensitivity and specificity, in what range of disease prevalence may the test be reasonably applied? Simple graphics are presented that incorporate acceptable risk thresholds and illustrate combinations of prevalence, sensitivity, and specificity in which disease probability remains high despite a negative test result. Adopting a Bayesian approach, together with acceptable risk thresholds, may help to avoid potential pitfalls of false negative screening results. Although the baseline prevalence in the United States of obstructive sleep apnea (OSA) is estimated in the range of 3- 28% (1), in certain populations the prevalence is much higher. For example, higher prevalence has been reported for patients with refractory epilepsy (33%) (2), recent stroke (58%) (3), refractory hypertension (63%) (4), heart failure (35%) (5), polycystic ovary syndrome (65%) (6), anterior ischemic optic neuropathy (89%) (7), children with Down's Syndrome (63%) (8), and those undergoing bariatric surgery (80%) (9-10). Treatment of OSA may be associated with improved outcomes in some settings (11). One particular area of interest involves OSA in the peri-operative setting, which may be associated with complications and/or longer hospital stay (12-13). Given the limited resources for labora- tory polysomnogram (PSG) as a screening tool, screening questionnaires may assist in OSA risk stratification. A screening tool with the acronym "STOP-Bang" was recently developed to provide dichotomous risk stratification (low versus high OSA risk) in general surgical populations (12-13). In this population, adverse outcomes involving res- piratory compromise were increased in the peri-operative period in those with OSA, defined as an apnea-hypopnea index (AHI) >5 by standard PSG. The STOP-Bang screen involves four yes/no patient questions (snoring, daytime tiredness, observed nocturnal apnea, high blood pressure), combined with four yes/no clinical features (BMI >35, age >50, neck circumference >40 cm, male gender). This simple tool was validated in a surgical population (177 patients, including general, gynecologic, orthopedic, urologic, plastic, ophthalmologic, and neurosurgery cases) (13), with a posi- tive result defined as >3 "yes" responses. Within the study limitations (including high refusal and no-show rates for PSG), the sensitivity for OSA varied with OSA severity