Misclassification

Misclassification based on spirometric findings may occur because:

1. There is uncertainty about the reference interval in a population. We estimate the reference interval from samples of the population. The smaller the sample, the greater the potential error in an arbitrarily selected cut-off value, such as a 95th or 5th percentile. The reference group from which such cut-off limits are derived should therefore preferably be large.
2. Within-person variability is responsible for the fact that on one occasion measured values fall within the ‘normal’ range, yet on another occasion outside that range. The greater the ratio between variability within and between subjects, the larger the proportion of misclassifications is going to be.

An example (ref. 1). The mean FEV1 of a group of subjects is 3500 mL, SD 350 mL. The within subject variability (SD) in this example is either 105 or 140 mL. We classify an observation as ‘compatible with airway obstruction’ if it falls below the 2½ percentile. In a healthy population such a verdict is to be expected in 2½ per cent of the population. However, taking into account variability within individuals, so that actual observations may be above or below their representative mean level, then 26% and 35%, respectively, of ‘suspect values’ is not recognized.

In clinical practice the prevalence of disease is higher than in a normal population; in that case the proportion of observations ‘compatible with airway obstruction’ will be greater than in a normal population. Even in that case the ratio of within person and between person variability matters. The table below illustrates that if the prevalence rate of ‘suspect’ observations is 10% or 30%, 2.6 to 10.5% will be erroneously classified, respectively.

Ref. 1 - Example from: Quanjer PhH, Stocks J, Polgar G, Wise M, Karlberg J, Borsboom G. Compilation of reference values for lung function measurements. In: Standardization of lung function tests in paediatrics. Eur Respir J 1989; 2 suppl 4: 184s-261s.