Evaluation of the global lung function initiative 2012 reference values for spirometry in a Swedish population sample

The study sample was recruited from the Obstructive Lung Disease in Northern Sweden (OLIN) Studies population-based cohorts. In 2006, a randomly selected cohort in ages 20–69 years was invited to a postal questionnaire survey (n = 7,997) together with a follow-up of a previously recruited randomly selected cohort aged 30–84 years in 2006 (n = 7,004). Of the responders (n = 12,055, 80.4%), 1016 subjects were randomly selected after stratification reflecting the age and sex distribution of the general population in the study area. They were invited to examinations including structured interviews and lung function tests in which 726 subjects (71.5%) participated. In order to obtain a sufficient number of healthy non-smokers, an additional sample of 738 healthy non-smokers according to the 2006 questionnaire survey were also invited to the examinations and 448 (60.7%) participated. All examinations were performed during 2008–2013, and, in total, 501 Caucasians (49% women) were identified as healthy non-smokers with acceptable spirometry quality and constitute the reference population.

Healthy non-smokers [ 15] were defined as subjects without a history of any airway or lung disease, breathlessness, cough, wheeze, ischemic heart disease, rheumatic disorders or a previous life-time exposure of > one pack-year of smoking. Their characteristics are described in Table  1. Eligibility criteria are presented in an appendix [see Additional file 1]. The age was calculated by one decimal point as the difference between date of birth and date of examination. Height was measured in stocking feet with an accurate stadiometer with 0.5 cm precision. Weight was measured with 0.5 kg precision without jacket and shoes and with empty pockets. Written informed consent for participation in the study was obtained from the participants and the study was approved by the Regional Ethical Review Board at Umeå University, Sweden.

Spirometric measurements included FEV ₁, FVC and SVC performed on two Jaeger Masterscope spirometers (JLAB version 5.21, CareFusion, Würzburg, Germany). The spirometers were calibrated each morning, and a minimum of three and a maximum of eight measurements were performed per subject. The procedures were performed without use of bronchodilators and following the ATS/ERS recommendations [ 16] but with a repeatability criterion of ≤5% deviation from the second highest value [ 17]. Once the data had been exported from the spirometers to an electronic data file, data validity controls were performed. The repeatability criterion was not met for 2% of the measurements and each such measurement was thoroughly examined by post-hoc ocular control of flow-volume charts, resulting in the exclusion of two subjects.

The GLI reference values are based on pre-bronchodilator values, and only pre-bronchodilator values from the reference population were collected and analysed. Using the Excel macro for GLI [ 18], reference values, lower limit of normal (LLN), Z-scores and percentiles for FEV ₁, FVC and the FEV ₁/FVC ratio were calculated for each subject in the reference population. The GLI Z-score is a standardized measure of the positioning of an observed measurement in the distribution of the population from which the GLI reference values are derived and takes both between-subject and age- and height-related variability into account. If the agreement between the observed values in the reference population and the GLI reference values is perfect, the mean Z-scores should ideally be zero, and the standard deviation (SD) should be one [ 19]. Mean values and standard deviations were calculated, and Q-Q plots were scrutinized to determine if the Z-scores were normally distributed. Possible relationships between Z-scores and age, height, weight and sex were examined by multiple linear regression models. If the GLI reference values are applicable, no such relationship should exist. LLN was defined as the lower fifth percentile in the distribution from which the GLI reference values are derived, as calculated by the GLI Excel macro, if not explicitly stated otherwise. The 90% limits of normality, which are expected to include 90% of the observations if the agreement is perfect, were defined as observations with GLI Z-scores within the −1.645 to +1.645 ranges. To identify the lower 2.5th percentile, the Z-score threshold of −1.96 was used.

For comparison, a number of other commonly used reference values were also applied on the reference population, i.e. ECSC [ 1] and Hankinson [ 20], and from Scandinavia, Berglund [ 2], Hedenström [ 3, 4], Langhammer [ 21], Gulsvik [ 22] and Viljanen [ 23]. Mean observed measurements, expressed as percent of predicted, were calculated across the different sets of reference values.

When applying the GLI reference values [ 5] on the reference population, the Z-scores for FEV ₁, FVC and FEV ₁/FVC were reasonably, although not perfectly, normally distributed, but not centred around zero. The mean Z-score was 0.21 (SD 0.91) for FEV ₁, 0.35 (SD 0.92) for FVC and −0.25 (SD 0.85) for the FEV ₁/FVC ratio and differed significantly from zero for all three measures (p < 0.001). Both FEV ₁ and FVC exceeded the predicted values across all ages except for FEV ₁ among 22–29 year olds. The pattern was the opposite for the FEV ₁/FVC ratio, where Z-scores generally were below zero across all ages, in particular among women. Independent samples T-tests showed a significant difference in mean Z-scores between women and men for the FEV ₁/FVC ratio, but not for FEV ₁ or FVC [Table  2]. Mean Z-scores deviated significantly from zero in the same age groups as where mean percent of predicted significantly deviated from 100% as displayed in Figures  1 and 2.

For FEV ₁, 93.6% (94.3% among women, 93.0% among men) of the 501 observations were within the 90% limits of normality. For FVC, the corresponding proportion was 90.2% (89.3% among women, 91.1% among men), and for the FEV ₁/FVC ratio, the proportion was 92.2% (88.5% among women, 95.7% among men). However, only 2.0% of the women and 1.6% of the men had FEV ₁/FVC values above the 95th percentile. The pattern of Z-scores was the opposite for FEV ₁ and FVC. For FEV ₁, 0.4% of the women and 1.2% of the men had values below the 5th percentile, and 5.3% of the women and 5.8% of the men above. For FVC, 0.8% of the women and 1.2% of the men had values below the fifth percentile, and 9.8% of the women and 7.8% of the men above. The GLI percentile frequency distributions for FEV ₁, FVC and the FEV ₁/FVC ratio were shifted, particularly among women, towards higher percentiles for FEV ₁ and FVC, and towards lower percentiles for FEV ₁/FVC [Figure  3].

When analysing the Z-scores for FEV ₁, FVC and the FEV ₁/FVC ratio in relation to age, height, weight and sex, small but statistically significant associations yielding R-square values of 1-2% were found. Sex had a significant impact on the Z-score for the FEV ₁/FVC ratio (Beta-coefficient 0.256 (p = 0.001)), sex and height had a significant impact on the Z-score for FEV ₁ (Beta-coefficient 0.276 (p = 0.023) for sex, −0.016 (p = 0.005) for height), and weight had a significant impact on the FVC Z-score (Beta-coefficient −0.009 (p = 0.001)).

The prevalence of spirometrically defined airway obstruction in the reference population according to LLN for the FEV ₁/FVC ratio was 9.4% (95% CI 5.7%-13.1%) among women and 2.7% (95% CI 0.7%-4.7%) among men (p-value = 0.002). When applying the lower 2.5th GLI percentile as LLN; the prevalence of obstruction in the reference population was 3.4% (95% CI 1.1%-5.7%) among women and 1.5% (95% CI 0.0%-3.0%) among men. Figure  4. illustrates that subjects defined as having obstruction according to the LLN criterion tended to have Z-scores for FEV ₁ below rather than above zero and Z-scores for FVC above rather than below zero.

When other sets of reference values were applied to the reference population, the mean FEV ₁ and FVC as percent of predicted significantly exceed 100% for both sexes according to ECSC, Hankinson and Berglund. Mean FEV ₁ and FVC as percent of predicted for reference values Hedenström (Sweden), Langhammer (Norway), Gulsvik (Norway) and Viljanen (Finland) were closer to 100%. In general, the FVC percent of predicted values were more overestimated than the FEV ₁ values for all sets of reference values, and particularly so among women. Consequently, FEV ₁/FVC and FEV ₁/VC as percent of predicted were generally below 100%, and more pronounced so among women compared to men [Figure  5].