The Global Lung Function Initiative 2012 Equations Are as Well-Suited as Local Population Derived Equations to a Sample of Healthy Professional Firefighters

Background and Objective We aimed to assess the validity of using the Global Lung Function Initiative's (GLI) 2012 equations to interpret lung function data in a healthy workforce of South Australian Metropolitan Fire Service (SAMFS) personnel. Methods Spirometry data from 212 healthy, nonsmoking SAMFS firefighters were collected and predicted normal values were calculated using both the GLI and local population derived (Gore) equations for forced expiratory volume in one second (FEV1), forced vital capacity (FVC), and FEV1/FVC. Two-tailed paired sample Student's t-tests, Bland-Altman assessments of agreement, and z-scores were used to compare the two prediction methods. Results The equations showed good agreement for mean predicted FEV1, FVC, and FEV1/FVC. Mean z-scores were similar for FEV1 and FVC, although not FEV1/FVC, but greater than 0.5. Differences between the calculated lower limits of normal (LLN) were significant (p < 0.01), clinically meaningful, and resulted in an 8% difference in classification of abnormality using the FEV1/FVC ratio. Conclusions The GLI equations predicted similar lung function as population-specific equations and resulted in a lower incidence of obstruction in this sample of healthy SAMFS firefighters. Further, interpretation of spirometry data as abnormal should be based on both an FEV1 and FEV1/FVC ratio < LLN.


Introduction
Firefighters' risk of developing chronic respiratory diseases is well known and no better exemplified than by the marked deterioration in lung function of first responders to the 9/11 disaster in New York [1]. The US-based National Fire Protection Association (NFPA) now recommends that spirometry be performed on an annual basis [2].
The interpretation of the results of spirometry rely, in part, on their comparison to a reference standard derived from normative data obtained in a healthy population. The 2005 American Thoracic Society and European Respiratory Society (ATS/ERS) statement on spirometry recommended the use of population-specific predicted normal equations [3], which should be updated approximately every ten years to reflect the changes that are likely to occur in anthropometric and ethnic characteristics (e.g., changes in mean heights for a given age over time). This is logistically difficult to do in every population. The Global Lung Function Initiative (GLI) 2012 therefore developed a new set of multiethnic predicted normal equations [4,5] using the pooled resources of 26 countries and data from more than 74000 subjects. They have been evaluated and shown to be well matched to some adult populations (in Australasia [6] and Europe [7,8]) but not others, such as Finland [9] and Sweden [10], where local population-specific reference values may be more relevant. Therefore, care needs to be taken when recommending whether GLI equations should be implemented in any particular population or laboratory.

Canadian Respiratory Journal
This issue must be considered when interpreting the lung function of professional firefighters. We have previously shown larger forced expiratory volumes in one second (FEV 1 ) and, in particular forced vital capacities (FVC), in South Australian Metropolitan firefighters compared to agematched controls, in both the entire sample and the majority who have no history of doctor-diagnosed lung disease [11]. Larger values may be attributable to a "healthy worker effect" [12], as well as the relatively high standard of physical fitness required for entry into the fire service. Selection of the most appropriate reference equations relative to this cohort is important in both the surveillance of serving firefighters and the assessment of potential recruits.
The purpose of this paper is to determine if the GLI equations are well matched to this healthy workforce, in light of their relatively large FVCs, compared to reference equations that were derived from the local population [13] (and not included in the GLI pool of data). We hypothesised that there would be no difference in the number of fire fighters who would be classified as having abnormal results (lung function less than the lower limit of normal (LLN)) between the two equations.

Study Participants and Data Collection.
We used spirometric data from full-time South Australian Metropolitan Fire Service (SAMFS) firefighters collected between June 2014 and April 2015 using a pneumotachograph-based spirometer (Masterscreen6 PFT system, CareFusion, Yorba Linda, CA). Spirometric data were collected as part of the ongoing longitudinal surveillance of lung function and respiratory health in the SAMFS, which commenced in 2007. All spirometry was performed prebronchodilator and in accordance with ATS/ERS guidelines [14]. Age was calculated to at least one decimal point as the difference between date of birth and date of examination. Participants provided information on medical and smoking history by written questionnaire following spirometry. Only never-smokers and firefighters with no history of doctor-diagnosed asthma or lung disease, based on questionnaire responses, were included in this analysis. Further details of procedures and equipment used in data collection have been previously described [11,15,16]. Calibration was performed on a daily basis using a threelitre syringe while zero flow was set immediately before each measurement. Data collection was funded by the SAMFS and ethical approval was obtained from the University of South Australia Human Research Ethics Committee (0000032662).

Reference Equations.
Predicted normal values for FEV 1 , FVC, and the FEV 1 /FVC ratio were calculated for each subject using two different sets of equations: firstly, using prediction equations derived from a random sample of the South Australian population that also used a pneumotachographbased spirometer (Gore) [13] and, secondly, using prediction equations from the Global Lung Function Initiative (GLI) [4], following the specific instructions. Individual -scores were calculated by subtracting the predicted value from the measured value and dividing by the standard deviation. The individual LLN was statistically defined by the lower fifth percentile (i.e., -score = −1.645).

Data Analysis.
Data were checked for normal distribution and two-tailed paired samples Student's t-tests or Wilcoxon Signed Ranks Tests were used to compare the means of predicted lung function and the LLNs, as well as mean -scores. Independent samples Student's -tests were used to compare included firefighters to those excluded based on medical history. The -score is a standardised measure of the position of a measurement within the distribution of the population from which the reference values are derived and takes into account age and height-related variability. We follow Hall and colleagues in defining the minimum physiologically relevant difference to be 0.5 -scores [6]. A significance level of = 0.01 was set for all tests to allow for multiple testing. Bland-Altman's 95% limits of agreement analysis (LoA) were used to quantify the difference and random error between the two equations. Bland-Altman plots provide information on how the difference between the two equations changes as the scale increases/decreases [17]. Limits of agreement were defined as mean difference ± 1.96 SD. Good agreement was defined by the LoA being less than ATS/ERS standard of acceptable repeatability (0.15 L) [14]. Data were analysed using SPSS5, version 22.0.0 for Windows, PC (IBM, Chicago, IL, USA).

Characteristics of Study Population.
From spirometry measures collected in 409 full-time firefighters, 212 participants were included in this analysis. The five full-time female firefighters were excluded from the analyses as well as a further two males (due to incomplete spirometry data). Twenty-one firefighters were excluded for having incomplete information on smoking status, along with 13 current smokers and 87 former smokers. A further 69 firefighters were excluded based on having a history of doctor-confirmed asthma or respiratory disease. The mean age (SD) of the included participants was 46.4 (8.7) years, mean height was 181.1 (6.2) cm, and mean body mass was 89.6 (12.6) kg. Measured spirometric values, predicted values, and LLNs calculated using both reference equations are shown in Table 1. Comparing included and excluded firefighters (see Supplementary Table 1 in Supplementary Material available online at https://doi.org/10.1155/2017/6327180), measured FEV 1 and FVC and mean -scores using Gore for FEV 1 and FVC were significantly lower in the excluded firefighters ( < 0.01).
Amongst all firefighters, there was a statistically significant ( < 0.01) difference between mean -scores produced by each equation for the FEV 1 /FVC ratio, but not FEV 1 or FVC (Table 2). When categorised by age, younger firefighters tended to have higher FEV 1 and FVC -scores with Gore relative to GLI, with this pattern reversing as age increased. Mean GLI FEV 1 /FVC ratio -scores were generally closer to zero amongst all age categories than those produced with Gore.

Discussion
This analysis demonstrated that the GLI equations are as well-suited to a sample of healthy professional firefighters, who typically have above-average lung function, as the population-specific Gore equations. The two equations in our study showed good agreement for mean predicted FEV 1 and FEV 1 /FVC, but not for FVC,  which was of clinical importance. There was also a significant difference between mean -scores for FEV 1 /FVC, but not FEV 1 and FVC. Hall and colleagues previously determined that GLI equations are well matched to Australasian spirometry [6], reporting mean -scores (SD) of 0.23 (1.00) for FEV 1 , 0.23 (1.00) for FVC, and −0.03 (0.87) for FEV 1 /FVC using the GLI equations. Observed FEV 1 and FVC -score means in our sample were both greater than those observed by Hall and colleagues, as well as the minimum physiologically relevant difference of 0.5 -scores. These higher values may be partly attributable to a healthy worker effect or to the preemployment selection process. Potential recruits with low lung function may be excluded directly as part of their prehire mandatory medical evaluations, while the intense prehire physical fitness evaluations of simulated firefighting tasks may naturally select those with above-average lung function. A possible explanation for the low FEV 1 /FVC ratios ( -score means <−0.7 for both equations) in the presence of above-average FEV 1 may involve the concept of airway/parenchymal dysanapsis, whereby an individual may have comparatively large lungs (which determines FVC) without a correspondingly large airway diameter (which determines FEV 1 ) [18,19], although why this phenomenon would feature so prominently in this population is unclear.
These analyses showed considerable differences between the subsequent LLNs, which were clinically meaningful, given their recommended use in detecting abnormality [3]. The impact of switching reference equation on the incidence of airflow obstruction has been investigated, with both Quanjer et al. and Brazzale et al. observing minimal differences when comparing the GLI to the European Community of Steel and Coal (ECSC) and The Third National Health and Nutrition Examination Survey (NHANES III) equations [5,20]. Hulo et al., however, observed more considerable differences when comparing the GLI to the ECSC equations [8]. By definition, five per cent of a healthy population sample would be expected to be below the LLN (lower 5th percentile). In this firefighter cohort, rates of FEV 1 /FVC less than the LLN (indicative of obstruction) were higher than this, as well as those reported by Backman et al. [ [5], when using the GLI equations. However, when interpretations are made using clinically important airflow limitation (when both the FEV 1 /FVC and the FEV 1 are below their LLNs) the rates of abnormality were greatly reduced to ≤2% for both equations. While some organisations such as the British Thoracic Society and the Global Initiative for Chronic Obstructive Lung Disease advocate the use of evaluating FEV 1 with the FEV 1 /FVC ratio to grade the severity of obstruction [21,22], it is of particular importance in selected (healthy) populations with large FVCs, to reduce the likelihood of misclassification. Such misclassification has important practical implications for firefighters, beyond the obvious detection of disease or abnormality, given the ongoing recommendation from firefighting organisations that prehire medicals determine abnormal spirometry based on a fixed cut-off of the FEV 1 /FVC ratio alone [2,23].
The NFPA recommends annual spirometric assessment of firefighter lung function, with interpretations based on expressing lung function as a percentage of predicted normal, adjusted for age, height, gender, and ethnicity [2]. Such interpretations may systematically misclassify diseased firefighters whose lung function was greatly above normal in the first instance. A more valid means of examining lung function in a population like this is to examine the annual rate of change for each individual and compare this to an established limit of normal longitudinal decline [24]. This is the intention of our surveillance program, and preliminary results have previously been reported [15]. Longitudinal surveillance may also reduce the misclassification of those whose lung function lies close to the LLN or upper limit of normal, given that such classifications can change over follow-up [25].

Limitations of This Study.
A limitation of this analysis is that it was not known whether any of the firefighters truly had clinically diagnosed or undiagnosed obstructive lung disease, given that this information was self-reported. Although FEV 1 was normal, disease may still have been present if participants had abnormally large FEV 1 at the beginning of their careers.
The present study used and discussed the validity of the FEV 1 /FVC ratio and its implications for assessing obstruction in firefighters. The ATS/ERS however define obstruction as a reduced FEV 1 to vital capacity (VC) ratio, below the 5th percentile of the predicted value [3]. As slow VC is expected to be greater than FVC [26], use of the FEV 1 /VC ratio could potentially increase the likelihood of misclassified obstruction in a population with proportionally large lungs relative to airway diameter.
At the time of the study, the SAMFS maintained a workforce of 861 full-time firefighters, 409 of whom voluntarily participated (47.5%). The main reason for nonparticipation was for logistical reasons, as a large portion of nonparticipating firefighters were either in nonmetropolitan areas or not present during scheduled lung function testing at a given station; many SAMFS firefighters hold positions unattached to a particular station and frequently move between locations. While privacy and anonymity were ensured, some firefighters with respiratory symptoms or asthma or who smoked may have chosen not to participate, possibly contributing to the above-average lung function observed in this study. Those who did participate may also have denied certain information.
A further limitation of the study was the relatively narrow age range of the men. The LLN for the Gore equations was calculated by subtracting the measured value from the predicted value while GLI equations use the lambda-musigma method (to account for the larger variation seen in older adults). The differences in the subsequent LLNs are accentuated when many older adults are included in the sample. However, firefighters are usually less than the age of 60 and so the results are no less valid for this population.

Conclusions
The GLI equations predicted similar lung function as population-specific equations and resulted in a lower incidence of obstruction in this sample of healthy SAMFS firefighters. Identification of abnormal spirometry should rely on interpretation of both the FEV 1 /FVC ratio and the FEV 1 value in relation to the LLN.

Disclosure
An earlier version of this work was presented as an abstract at the Thoracic Society of Australia & New Zealand and the Australian & New Zealand Society of Respiratory Science, Annual Scientific Meeting 2015.

Conflicts of Interest
The authors declare that they have no conflicts of interest.