The effect of regression dilution bias on the nocturnal hypoxaemia V blood pressure relationship
Davies RJO., Kendrick A., Wiltshire N., Catterall J., Crosby J., Stradling JR.
Nocturnal hypoxaemia is inaccurately assessed by one night's sleep study which will artifactually reduce its relationship with blood pressure through regression dilution bias. This study aims to quantify this effect. Two data sets were analysed. 893 normal men aged 35-60, studied during a survey of OSA and mean blood pressure (MBP) (BMJ: 1990:300:75) and 204 subjects (162M, 42F, Age 49SD10.8, >4%SaO2 diprate 14SD20.3) referred with possible sleep apnoea to centre 2. The normal subjects received one night of arterial pulse oximetry (Biox 3700) at home, and the sleep clinic group two nights (Pulsox 5). The severity of nocturnal hypoxaemia was quantified as the number of >4% falls in SaO2 per recording hour (>4% diprate). To describe how regression to the mean affects nocturnal hypoxaemia, the log10>4% diprate on the first night's oximetry in the 204 sleep clinic patients was sorted into ascending order and divided into 12, 0.25 range cells with the first two cells collapsed. For the resulting 11 cells, the difference between the average log10>4% diprates on the first and second nights was calculated. Above log10>4% diprate of 1.2 (=16hr-1) regression to the mean is minor as nocturnal hypoxaemia is reproducible. Below this threshold regression to the mean changes in an approximately linear way: diff log10diprate = -0.29 x mean first night log10diprate +0.12(r2=0.81) Since only 4 of the 893 (<0.5%) normal subjects had a >4% diprate >16 hr-1 this equation was used to correct the normal data for regression dilution bias. These 893 subjects were sorted into ascending log10>4% diprate and grouped into 9, 0.20 range cells with the first two cells collapsed. The cell average MBP was plotted against the cell average log10>4% diprate and the corrected cell average log10>4% diprate and their linear relationships described: MBP = 5.4 x log10>4%diprate+103.6 (r2=0.75) MBP = 7.0 x corrected log10>4%diprate+102.8 (r2=0.83) Correction for regression dilution bias strengthens the relationship between MBP and nocturnal hypoxaemia severity. This previously unconsidered effect may have distorted apparent relationships between OSA, MBP and obesity in community studies.