利用彈性模型預測細胞內水量

Flexible Modeling Approach to Predict Intracellular Water Volume

精確預測人體細胞內水量，除了可以評估營養狀態外，也具有許多臨床上的應用。本研究的目的是建構可以預測人體細胞內水量的模型，並與多頻生物電阻分析法作一比較。我們收集四十五位正常人之人體量測資料（年齡、性別、身高、體重），而多頻生物電阻分析法量測出之全身細胞內水量做為參考值。我們同時評估皮爾森公式、複迴歸線性模型與廣義累進模型預測全身細胞內水量之效能。由廣義累進模型預測之全身細胞內水量(ICW-GAM = 21.47± 0.57 L)與多頻生物電阻分析法量測出之全身細胞內水量(ICW-BIA = 21.47±0.61 L)並無統計上的差異(p = 0.96)。與皮爾森公式以及複迴歸線性模型相互比較，廣義累進模型的斯皮爾曼相關係數(0.95)最高，其均方根誤差(1.24)也最小；另外，以經驗累積分佈圖來分析，廣義累進模型的尖峰值(-0.13)最接近零誤差中心，同時雙尾間距最小(-2.13到2.79)佩斯──貝布拉克迴歸統計圖描述多頻生物電阻分析法與廣義累進模型的全身細胞內水量關係式為：ICW-BIA = 1.03 x ICW-GAM - 0.69，其斜率的95％信賴區間為0.93到1.16，截距的95％信賴區間則為-3.43到1.36，指出廣義累進模型與多頻生物電阻分析法是可以互換的方法(P > 0.10)。廣義累進模型的預測效能比傳統代數公式好，可以是一種預測正常人全身細胞內水量的方法。

In clinical practice, accurate prediction of intracellular water (ICW) volume has many constructive applications in addition to its use in the evaluation of nutritional status. Our purpose is to construct a model to predict ICW volume in comparison with the measurement by segmental multifrequency bioelectrical impedance analysis (SMF-BIA). Anthropometric data (age, gender, height, weight), and SMF-BIA were investigated in 45 healthy subjects. ICW volume measured by SMF-BIA (ICW-BIA) was the reference. Predictive performance of ICW volume based on Pierson formula, multiple linear regression (MLR) and generalized additive model (GAM) were evaluated. Predictive ICW volume derived from GAM (21.47 ± 0.57 L) were not significantly different (p=0.96) with ICW-BIA (21.47 ± 0.61 L). GAM also had higher Spearman's rank correlation coefficient (0.95) and smaller root mean square error (1.24) than Pierson formula and MLR. The GAM peak (-0.13) centered most closely to zero with the shortest tails (-2.13 to 2.79) in an empirical cumulative distribution plot. Passing-Bablok regression was described as ICW-BIA = 1.03 x ICW-GAM - 0.69, with 95% confidence interval for slope 0.93 to 1.16 and for intercept -3.43 to 1.36, indicating that GAM and SMF-BIA were interchangeable (p > 0.10). GAM outperformed conventionally algebraic equations and could serve as an excellent method for ICW volume prediction in healthy subjects.