These analyses use cross-sectional data obtained from a broad, representative healthy population sample from Austria to investigate the fit of the GLI lung volumes reference equations. As the GLI equations failed to demonstrate a good fit with our population-based data in normal subjects, a new set of sex-specific reference values was created for lung volumes.

Reference values are indispensable when interpreting lung volumes in clinical practice, using the LLN with TLC and ULN with RV for defining restrictive impairment and hyperinflation respectively [13]. Until recently, assessments in Austria and Europe relied mostly on the ECSC reference equations for adults, despite several studies having demonstrated inconsistencies between these reference equations, so the update by GLI was highly anticipated [14,15,16].

When using the GLI spirometry equations in our population a good fit was observed. We therefore considered our cohort comparable to the Caucasian cohorts used by GLI to create equations for spirometry and lung volumes. While small differences exist especially for females, we consider the equations sufficient for the detection of obstructive anomalies in our cohort [17]. This is consistent with previous analyses reporting a good fit with the GLI spirometry equations for other European cohorts [7, 18]. While some authors still report significant differences [19], the GLI equations, at least for Caucasian populations, offer consistent cut-offs and improved comparability between cohorts. The large amount of collated data, smoothing out small differences between populations, seems one of the main advantages. Additionally, even ethnic-specific equations created by GLI are available for spirometry. But the accuracy of these compared to globally merged equations was questioned lately [20].

However, GLI lung volume reference values did not fit well within our cohort. Large differences were observed, with mean Z-scores > 0,5 for TLC, RV and RV/TLC. Also, the percentage under the LLN and over the ULN was lower and higher respectively than expected. The difference was even more pronounced in females including significant differences for IC and ERV. These deviations could lead to an under-detection of restrictive disorders and overdiagnosis of hyperinflation in the Austrian population.

So far there is few data about the performance of the new GLI equations in European cohorts. The number of observations for lung volumes was much lower than for spirometry, and no equations are available for different ethnic backgrounds than Caucasian. A recent study from Belgium found similar results, with the GLI equations underestimating especially the values for RV [21]. Furthermore, the percentage under the LLN was lower than the expected 5% for TLC. A study in Algerian adults also reported, despite good fitting GLI spirometry values, similar results for RV, RV/TLC and TLC [22].

One potential explanation for the poor fit of the GLI lung volume equations is that our data were collected recently (starting 2011). Longitudinal studies have shown that populations are getting taller and healthier [23], with average population lung function increasing [24,25,26,27], potentially influenced by socioeconomic factors, or reduced occupational or environmental exposure [25, 28]. While in literature the impact of these developments in lung function is still discussed, the large size of our cohort might especially contribute to visible differences [29].

There were less obese and overweight individuals in our cohort compared to GLI. As the significance of weight as predictor of static lung volumes is not yet conclusively understood [6], we used weight as an predictive variable in an early version of the equations. This only minimally altered the coefficients, and so wasn’t used further (data not shown). While weight seems to have only a small impact on overall lung volume reference equations, the effect of body composition could be more important and may explain some of the differences between cohorts.

Future analyses could investigate and include the effect of body compartments on lung volumes.

Other factors contributing to the need to revisit equations could be changes in methods and equipment. Various studies in patients with obstructive lung diseases have demonstrated significant differences between lung volumes measured by gas dilution methods versus plethysmography, although the situation in healthy individuals is less clear [30]. Indeed, GLI found statistically significant differences between these two methods in their cohort, but regarded the differences as not clinically relevant, although the majority of their data were derived from plethysmography [6]. In addition, use of different body plethysmography devices and software could potentially impact the results. For example, in GLI devices manufactured by JAEGER (which we used in our study) measured somewhat higher values than those from other manufacturers, especially for RV [6]. Recently, authors from COSYCONET demonstrated differences in FRC up to 0.67 L between two manufacturers [31].

So, while the simplicity of one equation spanning different techniques, equipment, and populations is one argument for the use of the GLI equations, this might not appropriately represent all different populations and methods. It is to be expected that reference values derived directly from the specific examined population would fit that population better than standardised equations – although it is important that for such population-based equations to be useful, the examined population has to be representative of the broad population, as has been shown to be the case with the LEAD cohort [8] Still, adding more data to the GLI equations, may in the future improve the generalizability and render population based equations obsolete.

In this study the population derived reference equations from LEAD demonstrated a superior fit for all lung volume indices compared to the GLI equations. Lung volumes in our cohort were influenced by sex, age and height. Some studies have included weight as predictive variable for lung function [15, 16, 32], but as with GLI we found only a small influence of weight [6], and our equations therefore do not need to include this parameter. Importantly, we included obese individuals in our analyses, since reference values should be generalisable to the intended population [17]. Our newly derived equations might be usable in other European countries with similar population characteristics and equipment. This will have to be analysed in future studies.

Our analyses were conducted according to the ERS/ATS workshop report requirements [2]. While these published already over 20 years ago, they are still the most recent criteria available. We used strict selection criteria for our healthy cohort, only including never smokers, and excluding those reporting any respiratory symptoms. In addition, the population was distributed over all age groups, although with an overrepresentation of children, adolescents and of females, potentially due to the exclusion of those with a smoking history. We used standardised methods for the measurement of lung volumes, with strict quality control [8], and to create the reference equations we used the same statistical models as GLI. In particular, the LMS model allows the equations to cover the entire age range, avoiding discrepancies when entering the adult age [5].

The main limitation of our analyses is the single centre aspect of our lung function testing. The comparison of measurements done in another site showed only very small, not clinically relevant differences. Still a systemic bias can’t be ruled out, as only the device and software of one manufacturer was used. This also limits generalisability to other equipment and software. Furthermore, our cohort included no individuals aged < 6 years and > 80 years, so we recommend the use of our equations only between the ages of 6 and 80 years. Ethnicity wasn’t documented, as participants of the LEAD study, corresponding to the Austrian population, were predominantly of European ancestry. The Austrian population is known to consist just a very minor part of subjects different than Caucasian ancestry, so ethnicity wasn’t considered in the initial study design. Therefore the reference values are only applicable to similar Caucasian populations. We used strict exclusion criteria, but still subjects with physiologically abnormal lung function measurements or undiagnosed respiratory disease could have been present in the analysed cohort.