Chan Chung-hong, Ng Daniel K
Department of Paediatrics, Kwong Wah Hospital, Kowloon, Hong Kong.
Pediatr Pulmonol. 2007 Aug;42(8):711-5. doi: 10.1002/ppul.20642.
It is common for apnea-hypopnea index (AHI) to be used as an outcome variable in ordinary least squares linear regression. However, the distribution of AHI is not tested. The assumption of ordinary least squares linear regression may be violated.
The distribution of AHI from a pediatric sleep laboratory was assessed by Kolomgorov-Smirnov test. Transformation of AHI was attempted. In addition, we fitted an ordinary linear regression model (OLSM) and negative binomial regression model (NBRM) of the relationship between body mass index and the rate of apnea and hypopnea events. OLSM and NBRM were evaluated by residuals analysis.
AHI from the studied population deviated significantly from normal distribution. Commonly used transformation algorithm could not transform AHI to normal distribution. In addition, OLSM violated the underlying statistical assumptions of homogeneity of variance and normality of error. NBRM, on the other hand, was not restricted by these assumptions.
The current study suggested AHI is not likely to be normally distributed and its distribution cannot be transformed to normal. Negative binomial regression of the total number of apnea and hypopnea with an offset of log TST should be used in data analysis.
在普通最小二乘线性回归中，将呼吸暂停低通气指数（AHI）用作结果变量很常见。然而，AHI的分布未经过检验。普通最小二乘线性回归的假设可能会被违反。
通过柯尔莫哥洛夫-斯米尔诺夫检验评估儿科睡眠实验室中AHI的分布。尝试对AHI进行变换。此外，我们拟合了体重指数与呼吸暂停和低通气事件发生率之间关系的普通线性回归模型（OLSM）和负二项回归模型（NBRM）。通过残差分析对OLSM和NBRM进行评估。
研究人群的AHI显著偏离正态分布。常用的变换算法无法将AHI变换为正态分布。此外，OLSM违反了方差齐性和误差正态性的基本统计假设。另一方面，NBRM不受这些假设的限制。
当前研究表明AHI不太可能呈正态分布，且其分布无法变换为正态分布。在数据分析中应使用以对数总睡眠时间为偏移量的呼吸暂停和低通气总数的负二项回归。
