Based on a general population sample of the second phase of the Norwegian Hordaland County Cohort Study, we identified four delimitation levels defining four increasingly healthier and narrower pulmonary reference population types. We derived and compared prediction equations for all four population types and found that self‐reported obstructive lung diseases, smoking history, breathlessness, cough and wheeze were optimal exclusion criteria when defining a pulmonary reference population. As expected, predicted lung function values derived from the parent population (population type A) were lower and had a much steeper age‐related decline than lung function values derived from the other increasingly healthier population types. Some minor differences were also found between predicted values for the population without the cardinal respiratory symptoms breathlessness, cough and wheeze (population type B) and predicted values for the never‐smoking population without the cardinal respiratory symptoms (population type C). The never‐smoking population without cardinal respiratory symptoms and the never‐smoking population without any respiratory symptoms did not differ from each other with regard to lung function equations.
This is the first study to use analysis of the associations between self‐reported respiratory symptoms and lung function as a tool to identify simple exclusion criteria for a pulmonary reference population. It is also the first study to compare various definitions of pulmonary reference populations with a general population sample. Such comparison enabled us to examine associations between lung function and the predictors age, height and sex, both in a general population and in increasingly respiratory healthier and narrower reference populations.
A limitation of this study is the age range. The study did not include children and there were only a limited number of elderly subjects, especially men. Furthermore, the study population consisted of Caucasian subjects from an affluent Western country. In clinical practice it is important that reference values are derived from a population that shares such basic characteristics with the patient population.33
There is a need for other studies, preferably with a larger proportion of elderly subjects, to assess whether the exclusion criteria identified here are also optimal in other types of populations.
It is possible that the study population was healthier than a general population since all analyses were based on a follow‐up study rather than a cross‐sectional study. However, a previous report from the Hordaland County Cohort Study showed overall small differences between responders and non‐responders in the follow‐up survey in 1996–7, and there were no differences in the associations between risk factors such as age and smoking, and respiratory disorders.34
Several factors potentially affecting lung function were not examined in the present study—for example, abnormal chest radiographs, diabetes, cardiovascular disease, a history of tuberculosis and malnutrition. It is possible that the presence of such characteristics and diseases might have influenced lung function values in the reference population types. However, a previous study of the same population cohort in 1987 assessed that only 15 of 540 subjects (3%) had diseases that might affect pulmonary function,14
suggesting that such factors are not widespread in a healthy reference population.
The lack of detail in the international recommendations has led to differences in reference population exclusion criteria between studies. The number of exclusion criteria in various published studies presenting spirometric prediction equations varies from 3 to 20 and heavily influences the relative size and health status of the reference populations.5,8,10,16,35,36
Furthermore, the wording of questions has been shown to influence prevalence rates.26
In the present study we have used a questionnaire with questions originating from three previously published and validated respiratory questionnaires.24,26,27
The NRQ has been used extensively for more than 35 years in Norway.37,38
The other two questionnaires have been mostly used in Europe, but parts of them have also been applied in US studies.5,39
To enable comparable lung function prediction equations across studies, it is important to promote uniform exclusion criteria for reference population, with standardised wording of questions. To ensure standardised wording across languages, the methodology of translation and back‐translation is recommended to enhance questionnaire reliability and validity.
Prediction equations for LLN lung function are used in clinical practice to determine abnormal spirometric rates. If airways obstruction is defined as a FEV1/FVC ratio below the LLN, the prevalence would be 5.0% in the parent population based on the type A reference equations. When implementing LLN reference equations from population types B, C and D, however, the percentage of subjects with airway obstruction in the parent population increased to 9.5%, 11.4% and 10.1%, respectively. Following a stricter definition of airway obstruction as the presence of both FEV1/FVC and FEV1 below the LLN, the prevalence would be 2.4% with type A reference equations, 5.3% with type B equations, 6.7% with type C equations and 6.1% with type D equations (results not shown). In any case, the most important difference between abnormal spirometric rates occurs when the cardinal respiratory symptoms of breathlessness, cough and wheeze were excluded from the general population. Some difference was also observed when never‐smokers were excluded. However, results from the present study suggest that maintaining reference populations without any respiratory symptoms at all would have no further clinical implications.
In the present study, exclusion of ever‐smokers increased observed lung function in the male reference population but decreased lung function in the female reference population. This is probably due to a sex difference in the association between smoking habits and age. While exclusion of ever‐smokers made the male reference population considerably younger, it made the female reference population older. More young women than elderly women were smokers or ex‐smokers. Even with the absence of respiratory symptoms, the ageing of the female population when excluding ever‐smokers led to a natural decrease in lung function. When adjusting for age in the prediction models, expected lung function was higher after exclusion of never‐smokers among both men and women. Furthermore, a previous study from the same population has shown that smoking was a strong predictor for the incidence of chronic obstructive pulmonary disease during a 9 year period for subjects with normal lung function at baseline,40
rendering support to the notion that ever‐smokers should be excluded from reference populations even if they do not report respiratory symptoms. On the other hand, one could argue that exclusion of ever‐smokers from reference populations is not possible in all parts of the world. In some developing countries the proportion of adult never‐smoking men may be so small that it would be necessary also to include healthy ever‐smokers in reference populations in order to estimate predicted lung function. Future studies should further explore similarities and differences between healthy ever‐smokers and healthy never‐smokers with regard to respiratory status.
Reference population criteria in the present study entailed self‐reported respiratory symptoms rather than information on respiratory symptoms obtained by a physician in a clinical examination. Self‐administered questionnaires have several advantages over physician‐administered interviews. Observational bias affecting answers is not a problem, and the reproducibility for research purposes is better in the absence of a physician's subjective interpretation.41,42
A disadvantage with the use of self‐administered questionnaires, however, is that it depends on a literate study population.
Never‐smokers without cardinal respiratory symptoms (population type C) and never‐smokers without any respiratory symptoms (population type D) constituted 23% and 14% of the parent population (population type A), respectively. Whether we used population type C or D to derive prediction equations had no implications on the resulting reference values. There are important methodological advantages to keeping the reference population as large as possible, relatively speaking. Narrow reference populations will result in higher statistical uncertainty concerning the reference value estimates than will broader reference populations owing to fewer observations, especially in the older age groups. Rigid exclusion criteria will lead to selection bias and skewed age distributions. More elderly subjects than younger subjects suffer from respiratory symptoms, so older age groups will be under‐represented in a reference population based on rigid exclusion criteria. This, in turn, may influence the association between age and lung function as observed in the present study where the association between age and FEV1
/FVC did not reach statistical significance in the healthiest population types. There were only 13 never‐smoking men in the 70–82 year age group without cardinal symptoms (population type C) in the present study. A larger proportion of elderly subjects in the reference population would perhaps lead to a significant reduction in FEV1
/FVC with age, more in line with what has been observed in other studies.1
A similar selection bias was also observed with occupational exposure to dust or gas. After exclusion of cardinal respiratory symptoms, additional sex‐stratified analyses showed no difference in lung function between persons who had been occupationally exposed to dust or gas and those who had never been subjected to such exposure (p>0.05, results not shown). However, the size of the reference populations would have decreased by 57% for men and 27% for women if all occupationally exposed subjects were excluded.
Subjects with respiratory symptoms were excluded in a stepwise order based on both existing clinical and methodological considerations.19,20,21
The symptom whose exclusion led to the largest increase in mean lung function was excluded first. It could be argued that the order of exclusion influenced the resulting reference population exclusion criteria. However, additional analyses in which the exclusion order was changed among the seven cardinal respiratory symptoms (breathlessness grades 1–4, morning cough, chronic cough and wheeze) gave the same overall results regarding statistical significance and change in lung function. Breathlessness grades 1–4 and morning cough also remained significant if ever‐smokers were excluded first (p<0.05, results not shown).
Exclusion of all ever‐smokers with obstructive lung diseases or any respiratory symptoms in the present study left only 14% of the total population. However, excluding only ever‐smokers with obstructive lung diseases or the cardinal respiratory symptoms of breathlessness (grades 1–4), cough (morning cough and chronic cough) and wheeze resulted in a substantially larger and valid reference population. We believe that these exclusion criteria will be feasible and sufficient to enable derivation of prediction equations comparable across studies. We recommend testing the exclusion criteria with identical wording in both existing and future larger populations for further validation. The results from the present study should be included in the future discussion of a more detailed and standardised international definition of pulmonary reference populations.