Improved survival prediction from lung function data in a large population sample

Abstract

Studies relating lung function to survival commonly express lung function impairment as a percent of predicted but this retains age, height and sex bias. We have studied alternative methods of expressing forced expiratory volume in 1s (FEV(1)) for predicting all cause and airway related lung disease mortality in the Copenhagen City Heart Study data. Cox regression models were derived for survival over 25 years in 13,900 subjects. Age on entry, sex, smoking status, body mass index, previous myocardial infarction and diabetes were putative predictors together with FEV(1) either as raw data, standardised by powers of height (FEV(1)/ht(n)), as percent of predicted (FEV(1)PP) or as standardised residuals (FEV(1)SR). Quintiles of FEV(1)/ht(2) were better at predicting all cause mortality in multivariate models than FEV(1)PP and FEV(1)SR, with the hazard ratio (HR) for the worst quintiles being 2.8, 2.0 and 2.1 respectively. Cut levels of lung function were used to categorise impairment and the HR for multivariate prediction of all cause and airway related lung disease mortality were 10 and 2044 respectively for the worst category of FEV(1)/ht(2) compared to 5 and 194 respectively for the worst category of FEV(1)PP. In univariate predictions of all cause mortality the HR for FEV(1)/ht(2) categories was 2-4 times higher than those for FEV(1)PP and 3-10 times higher for airway related lung disease mortality. We conclude that FEV(1)/ht(2) is superior to FEV(1)PP for predicting survival in a general population and this method of expressing FEV(1) impairment best reflects hazard for subsequent death.