In conventional prediction equations the scatter around the predicted value is almost invariably described as constant, implying that it does not vary with the level of the predicted index. Stanojevic was the first to show that variability is age-dependent. This was corroborated in the study performed by the Global Lung Function Initiatiative (Pubmed).
One example: the coefficient of variation (CoV) of FEV1 in females is about 16% in a 3 year old, 12% in a 20 year old, and about 20% in an 80 year old healthy person. If we fix the lower limit of normal (LLN) at the 5th percentile, which is the norm in respiratory pathophysiology, then this lower limit is 1.64 times the coefficient of variation below predicted. This means that the LLN is at 75, 80 and 67 percent of predicted at ages 3, 20 and 80, respectively.
It follows that one should not use percent of predicted in evaluating test results but rather use the z-score, which in the case of the LLN is -1.64 for males, females and different ethnic groups at any age.
As shown previously, for the same age and height, there are essentially no differences in the mean FEV1/FVC ratio between ethnic groups (solid lines), signifying that differences in FEV1 and FVC between groups are proportional. However, there are differences in variability (Pubmed), due to which the LLN (dashed lines) differs.
The GOLD committee recommends to regard FEV1/FVC < 0.70 as indicating airway obstruction. The horizontal dashed red line shows that this leads to false positive findings from age 45 up. It also demonstrates that the GOLD guideline is bound to lead to ethnic differences in the prevalence rate of airways obstruction, a spurious result due the inappropriate and arbitrary lower limit of normal. This is exacerbated by the fact that e.g. Asians are smaller than whites, so that their normal FEV1/FVC ratio, and hence the LLN, is somewhat larger (the graph compares ethnic groups at the same heights).
See also: The great utility of the z-score