A model for the effect of bias for cholesterol on the population at risk.

The location of the Reference Value for an analyte within the population distribution affects the magnitude of error due to methodological bias. Using the gaussian distribution, we evaluated the effects of systematic and proportional biases of the method (positive and negative), mean value, and standard deviation on the magnitude of error. We chose four Reference Values for cholesterol as a model. For a population with a mean of 2.0 and SD of 0.36 g of cholesterol per liter, a 3% positive proportional bias causes sixfold more error at the 50th percentile than at the 97.5th. In general, the error for a given bias (proportional or systematic) is greater for a Reference Value within the body than at the tails of the distribution. Further, the magnitude of the error varies as a function of the mean and standard deviation of the population.