The current analysis of three studies performed in South, Central and North Germany confirmed the integrity of the basic determinants of lung function that are usually taken into account in prediction equations. This demonstrates that the data sets used were comparable to data found in the literature. Our analysis revealed that in addition to commonly used predictors such as gender, age and height, several other factors can also be used. These include: the packyears of cigarette smoking, environmental tobacco smoke exposure, the level of education, asthma, body weight, obesity, diabetes, hypertension, and medication. Although there were differences in effect size between the three study populations, these differences were not statistically significant.
Both the European Coal and Steel Community (ECSC)  and the American Thoracic Society (ATS)  have published comprehensive lists of reference equations for spirometry, and a number of more novel reference equations for different ethical groups and age ranges have been discussed [7–10]. Moreover, the ATS and the European Respiratory Society (ERS) have recently recommended a revision of the reference equations . However, all these reference equations were derived from healthy populations and include only gender, age and height as predictors.
Here, we analyzed a broad panel of potential predictors of lung function in three population-based studies. Using this approach we were not only able to assess whether these factors showed significant associations with lung function, but also whether their contribution to the overall predictive power was significant. On the one hand, this could be relevant to explain differences between prediction equations in different populations, as the additional factors might vary between these populations. On the other hand, a limited size of their additional effect could be the prerequisite for generalizing reference equations from one subpopulation to another or to the general population.
In line with the commonly used reference equations we found that gender, age and height are the major determinants of FEV1, FVC and PEF. Additionally, we found that body weight, and the presence of obesity were associated with changes in lung function. Weight itself had only limited effect but regarding obesity the SHIP data demonstrated a considerable reduction of FEV1 and FVC. One reason for the strong negative effect might be that in the SHIP study the rate of obesity was twice as high as in the other two studies. Our data are in line previous findings describing an impairment of lung function in subjects that are extremely overweight [12, 20, 21].
Cigarette smoking is the major risk factor for accelerated lung function decline in adults , and it has been demonstrated that the number of cigarettes smoked per day is linearly associated with the rate of decline of lung function . It is also an established result that exposure to ETS has negative effects on respiratory health . ETS exposure has been linked to several diseases, including asthma and COPD, and was demonstrated to be associated with reduced lung function both in children and in adults . In a similar manner our data showed associations between smoking, ETS exposure and lung function impairment. However, the effect of ETS was only detected in the SHIP data, which also showed the highest percentage of ETS exposure.
Socioeconomic status (SES) is another important determinant of lung function and pulmonary diseases . Associations between low SES and lung function, primarily FEV1 and FVC, are well known. Smoking contributes to the negative effect of poverty, but other factors are also involved, such as specific environmental and occupational exposures, increased indoor air pollution, low birth weight and increased frequency of respiratory tract infections in childhood [12, 25]. Moreover, our analysis revealed a negative correlation between education level and lung function in all three data sets.
Clearly, a number of disorders are also related to impaired lung function. Dyspnoea on exertion, bronchial hyperresponsiveness, asthma and COPD are examples of such conditions . In our study we confirmed asthma to be a strong predictor of impaired lung function, which was statistically significant for almost all lung function measures in the three studies.
Prospective population studies have identified a relationship between low levels of ventilatory function and an increased risk for cardiovascular diseases . FVC and FEV1 were inversely related to myocardial infarction , and PEF negatively to ischemic heart diseases and stroke . Impaired values of FVC and FEV1 were also found in patients with hypertension . Our data are in line with these results, as we confirmed the negative association between hypertension and lung function in the ECRHS study. Cross-sectional studies have also reported negative associations between markers of glucose intolerance and ventilatory function, and FEV1 and FVC were inversely correlated to insulin resistance and the prevalence of Type 2 diabetes mellitus . Also, subjects not having the diagnosis of diabetes showed a link between lung function and a raised plasma glucose level [31, 32]. Furthermore, we noticed the impairment of lung function in subjects with diabetes in our analysis but this effect became statistically significant only in elderly subjects of the SHIP study.
The multitude of factors potentially influencing lung function is responsible for some of the difficulties in establishing adequate and reliable prediction equations for lung function measures. Taking into account many of these factors, we were able to establish regression models showing a good correlation between predicted and measured lung function values. However, as expected, the major effect was attributable to gender, height and age, which accounted for more than 95% of the variability explained by our regression models. This was true except for in the case of the ratio of FEV1 to FVC, where our regression models could explain only a very small part of the variability. The explanation for this is that the ratio is self-adjusted. This means that the most important determinants for FEV1 and FVC, which are gender, age and height, are cancelled out. This again suggests that simple prediction equations with gender, age and height explain a fair substantial part of lung function variance for FEV1 and FVC. Additionally, the comparison of prediction equations between the three studies pointed out, that each equation was capable of predicting lung function of all three populations with similar accuracy. This suggests that for different subpopulations of Germany a single prediction equation for each of the lung function measures can still be used. However, it should be noted that at the extreme ends of lung function the accuracy of prediction was no longer guaranteed and systematic deviations appeared. One explanation for this lack of agreement might be that beyond the predictors included in our analysis there are additional factors influencing lung function at the extreme ends. There might be specific anthropometric characteristics or exposure to environmental and occupational pollution, including ozone, nitrogen dioxide, sulfur dioxide, dust, chemicals and gases, all of which are known to have adverse effects on lung function . Furthermore, genome-wide studies have revealed that genetic factors have an influence on pulmonary function [34, 35], and specific regions on various chromosomes have been identified to be significantly associated with lung function and the occurrence of COPD in smokers [36, 37]. However, it has to be considered that these genetic factors only account for a very small proportion (less then 0.15%) of the variance in lung function parameters and that they do not substantially add to clinical variables in predicting the onset of COPD. Another factor related to lung function might be diet. There is a positive association between lung function and the intake of fatty acids or antioxidant vitamins such as vitamin C and E [38, 39]. As we did not have information on these factors, further studies taking these into account might be of value.