Predicting sleep disordered breathing acutely after stroke: comparing eight sleep questionnaires and a logistic regression model

Introduction: Sleep disordered breathing (SDB) is common in stroke patients and negatively affects its outcome. Diagnosis in acute stroke subjects is challenging. We aimed to evaluate the performance of existing questionnaires to predict SDB in this setting and to assess the additional predictive power of stroke specific factors. Methods: Within a prospective multi-centre cohort of acute stroke patients, we studied 395 subjects who underwent early limited-channel cardiorespiratory sleep apnea testing. The association between anthropometric, demographic and stroke specific parameters, and SDB was investigated with univariate and multivariate logistic regression models. We compared the power to predict SDB of the resulting model to the Berlin Questionnaire (BQ), STOP-BANG, NoSAS, SACS, NoApnea, SOS, ESS and SLEEP-IN at different AHI cut-offs. Results: Multivariate logistic regression analysis identified a positive association of two stroke-related factors with SDB: cardio-embolic stroke aetiology (for AHI ≥15/h, p =0.003) and NIHSS at hospital discharge (for AHI ≥30/h, p=0.037). The best multivariate model showed an acceptable performance (AUC 0.76). Still, with the exception of STOP-BANG (AUC 0.69, p = 0.110) and SACS (AUC 0.66, p=0.054), it was significantly better to predict an AHI ≥ 15/h than the questionnaires. Conclusion: Neither the questionnaires nor the multivariate model is sufficiently powerful to diagnose SBD in acute stroke patients. Nonetheless, STOP-BANG and SACS are useful to triage acute stroke patients with a higher likelihood for SDB to testing with limited channel sleep studies for early diagnosis and treatment in the acute phase after stroke.