Statistics from Altmetric.com
In a recent otherwise excellent editorial in Thorax,1 Dr Seersholm indicated that our previous results on the decline in forced expiratory volume in 1 second (FEV1) based on the Copenhagen City Heart Study2 are biased. We disagree, and rather believe that our study of the general population is prone to less bias than case-control or family based studies.
Dr Seersholm argues that, because we genotyped study participants after measurement of FEV1 in 1976–8, 1981–4, and 1991–4, our results are biased.1 Certainly, if conventional risk factors are measured after development of disease, the disease might be the cause of the risk factor rather than vice versa. However, an α1-antitrypsin MZ genotype in a newborn does not change into a ZZ genotype by age. Therefore, the Pi MZ genotype preceded FEV1 outcomes in our study, even though genotypes were determined after FEV1 measurements. Using identical logic, the genotype preceded outcomes in a similar manner in other previous studies.3–6
Selection bias could potentially be a reason for discrepancies between studies on Pi MZ and COPD.1,7 In our study, where genotype distribution was in Hardy-Weinberg equilibrium, we found no evidence for selection against any α1-antitrypsin genotype.2,8 Therefore, as also pointed out by Dr Seersholm,1 selection bias is more likely in case-control and family based studies than in cohort studies of the general population.
Drs Dahl and Nordestgaard argue against selection bias if genotyping is performed after lung function tests because a newborn with a Pi MZ genotype does not change to Pi ZZ later in life. The latter is obviously true and not the issue. The reason for possible selection bias is that some people may fail to have genotype performed due to a characteristic of the lung function tests under study, and this may affect the result of a longitudinal study.
If we assume that subjects with a Pi MZ genotype have a very fast decline in forced expiratory volume in 1 second (FEV1) with premature death, a number of them may have attended the first examinations but did not live long enough to attend the last visit with genotyping. They would not therefore be included in the analysis of FEV1 decline and the result would be an underestimate of the decline in FEV1 with the possible conclusion that Pi MZ is not a risk factor for lung disease. This is usually called a “survivor effect”.
The opposite may also be true. Suppose there is a reverse relationship between pulmonary function and compliance with study visits—that is, subjects with a normal FEV1, normal FEV1 decline, and no lung symptoms may not attend the last visit because they feel well. This would tend to exaggerate any increased decline in FEV1 and the conclusion could be that Pi MZ is an important risk factor for lung disease.
These are just two of many examples of possible biases. They show that, even if the risk factor (Pi MZ) is present from birth, it is vital to postpone the collection of tests for analysis after genotyping has been performed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.