Petersen Per Hyltoft, Jørgensen Lone G M, Brandslund Ivan, De Fine Olivarius Niels, Stahl Marta
NOKLUS, Norwegian quality improvement of primary care laboratories, Division for General Practice, University of Bergen, Norway.
Scand J Clin Lab Invest Suppl. 2005;240:51-60. doi: 10.1080/00365510500236135.
To investigate the effect of composite analytical bias and imprecision in the measurements of fasting plasma-glucose (fPG) for diagnosis of diabetes mellitus and estimation of risk of development and progression of retinopathy using measurements of Haemoglobin A1C (HbA1C%).
Data on biological within-subject variation for fPG (5.7% and 4.9%) and HbA1C% (1.9%) from literature and data on fPG for a 'low-risk' population (regarding diabetes) from own investigations (ln-values of mean=1.6781 approximately geometric mean population=5.36 mmol/L and standard deviation=0.0891 approximately CV population=8.9%). Further, guidelines for diagnosis of diabetes (two consecutive measurements of fPG above 7.0 mmol/L) were obtained from literature as also the risk of development of and progression of retinopathy using measurements of HbA1C (a change in risk of 44% for a change in HbA1C% of 10%). It was assumed that each individual had values which over a short time had a Gaussian distribution about a biological set-point. Calculations of the effect of analytical bias and imprecision were performed by linear addition of bias and squared addition of imprecision to the squared error-free biological distribution. Composite variations of bias and imprecision were obtained by varying assumed imprecision and calculating the maximum acceptable bias for the stated situation.
Two diagnostic examples are described for fPG and one for risk related to HbA1C%. Firstly, the risk of diabetes as a function of set-point and bias and imprecision was investigated, using functions where the probability of two measurements above 7.0 mmol/L was plotted against biological set-points, resulting in a S-shaped curve with a 25% probability for a set-point equal to 7.0 mmol/L. Here, a maximum 5% probability of classifying an individual with a set-point of 6.4 mmol/L (upper reference limit for the 'low-risk' population) as diabetic was used to calculate the analytical quality specifications. Comparably, the 5% probability of misclassifying a diabetic with fPG of 8.0 mmol/L was investigated, and both specifications were illustrated in an imprecision-bias plot. Secondly, the percentage of 'low-risk' individuals which would be falsely diagnosed as diabetic was calculated, and this percentage was plotted as a function of bias for different assumed values of imprecision. Thirdly, the confidence intervals for a certain risk-difference for HbA1C% of 5% or 10% was used to draw an imprecision-bias plot for different assumed changes and probabilities.
Analytical quality taking the demands for bias and imprecision in account are obtainable in laboratories, but may be questionable for use of capillary blood and POCT instruments with considerable consequences for the number of individuals classified as diabetics, and thereby for the economy etc.
For clinical settings, with so clear recommendations and descriptions of risk curves as in diabetes, it is relatively easy to estimate the analytical quality specifications according to the highest level of the model hierarchy, when relevant probabilities for the events are assumed.
通过糖化血红蛋白（HbA1C%）测量，研究空腹血糖（fPG）测量中的综合分析偏倚和不精密度对糖尿病诊断以及视网膜病变发生和进展风险评估的影响。
从文献中获取fPG（5.7%和4.9%）和HbA1C%（1.9%）的生物体内变异数据，以及来自自身研究的“低风险”人群（关于糖尿病）的fPG数据（均值的自然对数值 = 1.6781，近似几何平均人群 = 5.36 mmol/L，标准差 = 0.0891，近似变异系数人群 = 8.9%）。此外，从文献中获取糖尿病诊断指南（两次连续测量fPG高于7.0 mmol/L）以及使用HbA1C测量评估视网膜病变发生和进展风险（HbA1C%变化10%时风险变化44%）。假设每个个体的值在短时间内围绕生物设定点呈高斯分布。通过将偏倚线性相加以及将不精密度平方后加到无误差生物分布的平方误差上，计算分析偏倚和不精密度的影响。通过改变假设的不精密度并计算所述情况下的最大可接受偏倚，获得偏倚和不精密度的综合变异。
描述了两个fPG诊断示例和一个与HbA1C%相关风险的示例。首先，研究糖尿病风险作为设定点、偏倚和不精密度的函数，使用将两次测量高于7.0 mmol/L的概率与生物设定点作图的函数，得到一条S形曲线，设定点等于7.0 mmol/L时概率为25%。在此，使用将设定点为6.4 mmol/L（“低风险”人群的上限参考值）的个体误诊为糖尿病的最大概率5%来计算分析质量规范。类似地，研究了将fPG为8.0 mmol/L的糖尿病患者误诊的5%概率，并在不精密度 - 偏倚图中展示了这两个规范。其次，计算了被错误诊断为糖尿病的“低风险”个体的百分比，并将该百分比作为不同假设不精密度值下偏倚的函数作图。第三，使用HbA1C%变化5%或10%的特定风险差异的置信区间，针对不同假设的变化和概率绘制不精密度 - 偏倚图。
考虑到对偏倚和不精密度要求的分析质量在实验室中是可以实现的，但对于毛细血管血和即时检验仪器的使用可能存在疑问，这对被分类为糖尿病患者的个体数量有相当大的影响，进而对经济等方面产生影响。
对于临床环境，鉴于糖尿病中有如此明确的建议和风险曲线描述，当假设事件的相关概率时，根据模型层次结构的最高级别相对容易估计分析质量规范。
