Polysomnographic predictors of blood pressure and hypertension: is one index best?

STUDY OBJECTIVES Numerous indexes derived from polysomnography are available to characterize sleep-disordered breathing, with no consensus over which measures best predict clinical outcomes. This study addresses the relative merits of using alternative polysomnography indexes by characterizing the consistency and strength of the association of each index with blood pressure and hypertension. DESIGN Cross-sectional analyses of the association of alternative polysomnography indexes with blood pressure and hypertension were performed in construction and validation data sets. Linear and logistic regression models were used to identify the best variable sets. PATIENTS Data were obtained from 6433 men and women (age 62.9 +/- 11.0 years, 52.8% women) who participated in the Sleep Heart Health Study. RESULTS In multivariable models, most indexes showed weak linear associations with systolic, with slightly stronger associations for diastolic blood pressure, and the log odds of hypertension. No single index showed consistent superiority over others. Systolic blood pressure, diastolic blood pressure, and hypertension each were associated with distinct sets of polysomnography variables. Slightly more-consistent associations were demonstrated for indexes that included hypopneas that were linked with either a 3% or 4% desaturation level than indexes that did not require hypopneas to have linked desaturation. For indexes that combined apneas and hypopneas, there was no evidence that linking obstructive apneas to desaturation or arousal altered prediction compared with counting all apneas. CONCLUSION In summary, using a rigorous cross-validation assessment, we did not identify a clear superiority of any single index for blood pressure or hypertension prediction. Detailed analyses of alternative definitions of the respiratory disturbance index support current scoring guidelines, where desaturation criteria are recommended for hypopneas but not apneas.