There is still widespread use of 80% of the predicted value as the lower limit of normal in healthy subjects. This is incorrect, because the scatter does not change proportionately with the level of the predicted value. For example, as the predicted value decreases in adults, the scatter diminishes only slightly, as shown by the lower limit of normal (5th centile) relative to the predicted value. The line representing 80% of predicted moves up progressively relative to the LLN. If you held the view that values below 80% of the predicted value were compatible with obstructive lung disease, then you would be faced with a relatively large proportion of false positive test results in elderly subjects, and with some false-negative test results in younger subjects. The correct procedure, therefore, is to compare observed and predicted value, and to assess whether the difference between the two is or is not compatible with the scatter about the predicted value. Let us look at a concrete example.
The measured value is -0.66/0.50 = 1.31 times the residual standard deviation below the predicted value: the z-score = -1.31. If the data are normally distributed (this is often not the case but is true in this case), then in only 5 per cent of subjects of a reference group the z-score is smaller than -1.64, which marks the lower 5th percentile. In the present case the z-score = -1.31, well above the 5th percentile and therefore ‘within the normal range’. However, on the basis of percent of predicted (78%) you might have concluded that the FEV1 was in an abnormal range.