Predicting an accelerated lung function decline in smokers: is there a proper threshold?

To the Editor:

We read, with very much interest, the publication by Akkermans et al. [1] in the European Respiratory Journal. In this study they reanalysed data from the first Lung Health Study by forming four groups, based on the forced expiratory volume in 1 s (FEV₁)/forced vital capacity (FVC) <0.70 or <5th percentile, according to the LMS (lambda, mu, sigma) approach. However, we do have some remarks concerning the validity of their analysis and propose a means of reinterpretation of their outcome.

Akkermans et al. [1] concluded that the decline of the post bronchodilator FEV₁ in the group fixed⁺/LMS^- (FEV₁/FVC <0.70 and >5th percentile) was significantly lower compared to the group fixed⁺/LMS⁺ (FEV₁/FVC <0.70 and <5th percentile). The mean±sd decline was 43.8±50.0 mL·year⁻¹ versus 53.5±51.5 mL·year⁻¹, respectively, and thus one recommended the use of the 5th percentile threshold to define the presence of chronic obstructive pulmonary disease (COPD).

Interestingly, this conclusion was reached after they excluded, in total, 1842 subjects from the initial 5887 subjects. Therefore, the question arises: what is the consequence of such a vast post hoc exclusion on the estimates of the FEV₁ decline in the two groups?

The majority of subjects (n=1276 ) were excluded as they showed “an unstable classification” and one can ask oneself what the FEV₁ decline was in those excluded subjects. Some were reported to have an improved lung function over time and some were reported to have become more obstructive. Data, on whether this exclusion was “randomised” over the four groups formed, is lacking. The text of the publication does not give detailed information on the reasons as to why the authors chose to follow this path. Anyway, one must realise that it is highly likely that the groups, thus formed, suffer from selection bias and the outcome of the analysis is, therefore, also subject to some degree of bias. This selection bias is illustrated in table 1 of the article [1] where it is shown that the excluded subjects had a better FEV₁ and FEV₁/FVC, despite the same amounts of pack-years.

Of the remaining 566 excluded subjects, a number were excluded based on “missing at least one follow-up spirometry”. Now, the authors used the classical statistical approach for longitudinal studies, a random intercept and random slope analysis (mixed procedure in SAS (SAS software 9.2; SAS Institute Inc., Cary, NC, USA)). This type of analysis is characterised by the fact that such missing data is not detrimental to the validity of the outcome. The number of observations within a subject does not appear to be equal for all. It seems that the authors excluded too many subjects than was strictly necessary.

Even when the exclusion of subjects does not lead to bias in the outcome, the conclusion that the FEV₁/FVC <5th percentile threshold is to be favoured over the FEV₁/FVC <0.70 has to be interpreted with caution. If indeed, the decline in <5th percentile group is steeper, it does not mean that in the FEV₁/FVC <0.70 but <5th percentile group decline is absent. The mean decline differed by a small 10 mL·year⁻¹, but in both groups is it steeper than expected in healthy subjects (estimated to be ∼27 mL·year⁻¹ for the FEV₁, according to the European Respiratory Society reference equations [2]). When the FEV₁ decline in the FEV₁/FVC <0.70 but >5th percentile group is normally distributed with a mean±sd of 43.8±50.0 mL·year⁻¹, it is evident that in this group some very steeply declining subjects are present.

Akkermans et al. [1] correctly state that a rapid lung function decline is pivotal [3]. When a rapidly declining subject still shows a post bronchodilator FEV₁/FVC ≥5th percentile, that subject receives a non- COPD label and has to wait for that crossing of the 5th percentile threshold before the COPD label is imposed. This means a loss of time, while it is clear that such a rapid-declining subject is truly diseased, those subjects with an initial (very) high FEV₁ (e.g. a starting value at the 95th percentile) must “wait” a long time before they receive a proper label, i.e. while their FEV₁ deteriorates.

In conclusion, in our view the discussion on which threshold to follow/choose should be replaced by a debate on how to locate rapidly declining subjects with a whether or not they hover still above, at, or already below whatever threshold.

• Received May 21, 2013.
• Accepted May 24, 2013.

• ©ERS 2014