Skip to content
The Kids Research Institute Australia logo
Donate

Discover . Prevent . Cure .

Influence of secular trends and sample size on reference equations for lung function tests

The aim of our study was to determine the contribution of secular trends and sample size to lung function reference equations, and establish the number...

Authors:
Quanjer, P. H.; Stocks, J.; Cole, T. J.; Hall, G. L.; Stanojevic, S.

Authors notes:
European Respiratory Journal. 2011;37(3):658-64

Keywords:
Multicentre study, Pulmonary function, Reference equations, Secular trends, Spirometry, adolescent, adult, age distribution

Abstract:
The aim of our study was to determine the contribution of secular trends and sample size to lung function reference equations, and establish the number of local subjects required to validate published reference values. 30 spirometry datasets collected between 1978 and 2009 provided data on healthy, white subjects: 19,291 males and 23,741 females aged 2.5-95 yrs.

 The best fit for forced expiratory volume in 1 s (FEV1), forced vital capacity (FVC) and FEV1/FVC as functions of age, height and sex were derived from the entire dataset using GAMLSS. Mean z-scores were calculated for individual datasets to determine inter-centre differences. This was repeated by subdividing one large dataset (3,683 males and 4,759 females) into 36 smaller subsets (comprising 18-227 individuals) to preclude differences due to population/technique.

No secular trends were observed and differences between datasets comprising >1,000 subjects were small (maximum difference in FEV1 and FVC from overall mean: 0.30- -0.22 z-scores). Subdividing one large dataset into smaller subsets reproduced the above sample size-related differences and revealed that at least 150 males and 150 females would be necessary to validate reference values to avoid spurious differences due to sampling error.

Use of local controls to validate reference equations will rarely be practical due to the numbers required. Reference equations derived from large or collated datasets are recommended.