The Dunedin Study investigates health and behaviour in a population-based cohort born in Dunedin in 1972/1973.12 13 The cohort was formed at age 3 years when 1037 individuals (52% male; 91% of eligible births) attended the first follow-up. The cohort has been further assessed at ages 5, 7, 9, 11, 13, 15, 18, 21, 26, 32, 38 and 45 years (see online supplement for additional methods).
MeasurementsHeight and weight were measured at each assessment. Spirometry was performed at each assessment since age 9.14 15 At age 45, spirometry was repeated after inhaling 200 µg salbutamol via a large-volume spacer.16 Predicted values for FEV1, FVC and FEV1/FVC at each age were generated using multiple linear regression models for non-smoking, non-asthmatic, non-pregnant men and women adjusting for height and height squared. Airflow obstruction was defined as a postbronchodilator FEV1/FVC ratio <0.7 at age 45.17
Definition of variablesChildhood asthma was defined as a parent/self-reported diagnosis with compatible symptoms or asthma medication use within the previous year at ages 9, 11 or 13 years.18 19 Childhood airway hyper-responsiveness (AHR) was defined as a positive response to methacholine/salbutamol challenge at ages 9, 11 or 13 years.20
Participants were considered to be breastfed if breastfeeding continued for at least 4 weeks.21 Preschool attendance was reported at age 3. Household overcrowding at age 3 was defined as fewer than two rooms (excluding kitchen and bathroom) per child.22 Childhood cat and dog ownership was defined as having lived with cat and dog by age 9.22 Parental smoking was ascertained at ages 7, 9, 11 and 13 years.22 At 3 years, parents were asked about the number of coughs and colds the participant had had in the previous year.
Childhood socioeconomic status was assessed throughout childhood based on the education and income associated with their parents’ occupation.18 Parental history of asthma or hay fever was ascertained when participants were aged 7 and 18.
Atopy was assessed at ages 13, 21 and 32 years by skin prick tests for common aeroallergens. Atopy was defined as a weal diameter 2 mm greater than the negative control to one or more allergens.23
Cumulative smoking was calculated from the reported number of cigarettes smoked up to age 18 years and the number smoked between each subsequent assessment (one pack-year=20 cigarettes/day for 1 year).16
Statistical analysisWe analysed FEV1 at each age (from 9 to 45 years) using latent profile analysis (LPA) to identify distinct subgroups of study participants whose FEV1 measurements followed a similar pattern over time. LPA is described extensively elsewhere.24 Briefly, it identifies unobserved subgroups of individuals who share similar characteristics, based on a series of continuous variables. We used estimated posterior probabilities to assign individuals to the trajectory that they were most likely to belong to.
Based on previous studies,7 8 we hypothesised at least four FEV1 trajectories. LPA models with 4–12 trajectories were fitted to the data. After excluding models with trajectories with <2% of the cohort, the model with the lowest Bayesian information criterion (BIC) value was selected.25 Descriptive labels were assigned to the identified trajectories. The primary analysis included study members with at least six spirometry measures between ages 9 and 45, including at least one measure from each of childhood (ages 9–15), early adulthood (ages 18–32) and mid-adulthood (ages 38–45). Baseline/early childhood characteristics were compared between those included and not included in these models. Missing data were assumed to be mostly missing completely at random. A sensitivity analysis included those with at least two spirometry measures at any ages from 9 to 45 years.
Associations between childhood and adult factors (exposures) and trajectories were examined using multivariable binary logistic regression where the outcome belonged to that particular trajectory vs belonging to any other trajectory. Further analyses compared each trajectory with the ‘average’ trajectory. The choice of plausible exposures was informed by a directed acyclic graph (online supplemental figure S1). Initially, all exposures were included in each model, followed by univariable models for exploratory purposes. Firth logistic regression was used when the size of the trajectory of interest was below 50.26 Statistical significance was determined using the Bonferroni-Holm method (per variable, separately for univariable and multivariable models). This controls the family-wise error rate without requiring assumptions about the (in)dependence of the hypotheses being tested.27 This analysis was repeated for each identified FEV1 trajectory.
The same approach was used for FVC and FEV1/FVC. One participant with developmental abnormalities affecting lung function was excluded from all analyses. Measurements were excluded at ages when pregnancy was reported. Two-sided p values <0.05 were considered statistically significant. Analyses used Stata V.15.0 (StataCorp).
DiscussionUsing up to 10 assessments of spirometry in a large prospective population-based cohort followed from ages 9 to 45 years, we identified distinct trajectories of FEV1, FVC and FEV1/FVC. A striking observation is that most, but not all, of the identified trajectories were approximately parallel, demonstrating that, for most people, their lung function trajectory into mid-adult life was already established before adolescence. Notable exceptions included childhood AHR-associated persistently low trajectories, which started low and diverged further from the others over time, and accelerated-decline trajectories associated with smoking and higher adult BMI. We also identified catch-up trajectories for FEV1 and FVC, which, for FVC, were associated with lower adult BMI values. Importantly, we found that the three FEV1 and three FEV1/FVC trajectories leading to the lowest values at age 45 contributed most of the airflow obstruction at age 45 despite representing small proportions of the cohort.
The FEV1 trajectories that we identified are broadly consistent with those reported in other cohorts. The only other study to cover a similar age range from childhood to mid-adulthood (ages 7–53), identified 6 distinct FEV1 trajectories among 2438 Australian participants.7 Using a similar statistical approach to ours, a study of two British population-based cohorts identified four FEV1 trajectories based on three or four spirometry measures from ages 5 to 24.8 Another British cohort observed high and low FEV1 trajectories from ages 10 to 26.9 Among asthmatics, the Childhood Asthma Management Programme classified 684 participants into four FEV1 trajectories based on spirometry measures from ages 7 to 26.28 Looking at changes after peak adult lung function, an American study identified five FEV1 trajectories from ages 25 to 55.10
We identified more distinct FEV1 trajectories than all these studies.7–10 28 This is likely because we selected the model with the best empirical fit (lowest BIC), whereas other studies either limited the maximum number of trajectories or merged similar-looking trajectories. However, even if we were to merge similar trajectories, the combined evidence suggests that there are at least five FEV1 trajectories in the general population: persistently high, average, below average, accelerated decline and persistently low. In addition, an early below average with catch-up during adolescence FEV1 trajectory was observed in both our study (8%) and the TAHS (8%).7 None of the other cohorts had sufficient data to identify this. For comparison with earlier studies, graphs with fewer trajectory classes are shown in online supplemental figure S3.
We also identified 10 distinct FEV1/FVC trajectories. Two studies found normal/high and persistently low trajectories from ages 11 to 32 and 10 to 26 years.6 9 Only the recent TAHS study investigated FEV1/FVC trajectories beyond early adulthood, reporting six trajectories.11 Although they used a different statistical approach and had fewer ages of assessment with few lung function measurements in early adulthood, the patterns of trajectories were broadly consistent with our findings. In particular, the adverse trajectories leading to airflow obstruction in mid-adult life (‘early low-rapid decline’, ‘early normal-rapid decline’, ‘early low-normal decline’) were similar to the persistently low, accelerated-decline and always-well-below-average trajectories that we have identified. Graphs with fewer FEV1/FVC classes are shown in online supplemental figure S4.
In our study FEV1/FVC trajectories and FEV1 trajectories had similar associations with AHR: for both parameters, persistently low (both 3% of the cohort) and always-well-below-average (7% and 8%, respectively) trajectories were associated with childhood AHR, whereas those in the always-well-above-average FEV1/FVC trajectory (15%) and always-above-average FEV1 (14%) trajectories were less likely to have childhood AHR. Other studies have found that asthma is associated with persistently low FEV1 trajectories.7 8 AHR is a hallmark of asthma and these findings indicate that childhood asthma plays a key role in the impaired development of FEV1 and airflow obstruction. Although parental asthma has been reported to be associated with impaired lung function among offspring,6 29 we did not find any direct association between parental asthma and lung function trajectories, after accounting for the potential mediator of childhood asthma in the participants. The TAHS study found an association between parental asthma and low FEV1/FVC (but not FEV1) trajectories but did not report whether this association persisted after adjusting for childhood asthma.7 11 All participants in the persistently low FEV1/FVC trajectory were atopic, although this was not a statistically significant independent predictor of this trajectory once adjusted for the other variables including AHR.
Unsurprisingly, accelerated-decline-II FEV1 and accelerated-decline and above-average-but-with-decline FEV1/FVC trajectories were all associated with cumulative smoking. Although smoking was not a statistically significant predictor of the accelerated decline FEV1 trajectory in TAHS, this trajectory had the highest prevalence of smoking,7 and smoking was associated with the FEV1/FVC trajectories leading to adult airflow obstruction.11
We identified nine distinct trajectories for FVC. To our knowledge, only two previous studies have investigated FVC trajectories. A high and a low FVC trajectories were identified in the Isle of Wight cohort, based on three measures at ages 10, 18 and 26 years.9 The TAHS study identified five FVC trajectories.11 Although fewer in number, several of these trajectories appear consistent with the trajectories that we identified, with persistently low, persistently high and catch-up trajectories. However, they did not identify an accelerated decline FVC trajectory. Graphs with fewer FVC classes are shown in online supplemental figure S5.
We found a high degree of overlap between identified FEV1 and FVC trajectories (figure 4), for example, 68% of participants in the persistently low FEV1 trajectory were assigned to either the persistently low or always-well-below-average FVC trajectory, and 87% of those in the persistently high FEV1 trajectory were assigned to either the persistently high or always-well-above-average FVC trajectory. We also found considerable overlap between the FEV1 and FEV1/FVC ratio trajectories but there was less overlap between the FVC and FEV1/FVC trajectories (online supplemental figures S6 and S7). The extent to which FVC and FEV1/FVC trajectories provide additional health information to FEV1 trajectories remains unclear and needs further investigation: although TAHS found that trajectories leading to obstructive, restrictive and mixed patterns of lung function disorder had different predictors and health associations, these classifications were based on the lung function outcomes at age 53 rather than the trajectories themselves.11
Heatmap showing overlap between trajectories of FEV1 (rows) and FVC (columns) with darker shades for higher counts. Numbers indicate number of participants. FEV1, forced expiratory volume in 1 s; FVC, forced vital capacity.
We identified few associations between our potential exposures and the better than average lung function trajectories. BMI at age 9 was positively associated with the persistently high FEV1 trajectory and negatively associated with the always-below-average trajectory, suggesting that early childhood growth may have a long-term protective effect on adult FEV1, although there was also evidence that overweight children may be more likely to have always-below-average trajectories. By contrast, BMI at age 45 was associated with accelerated decline FEV1 and FVC trajectories, and a lower likelihood of the catch-up FVC trajectory, confirming the detrimental impact of increasing body weight in adolescence and adulthood on lung function.
Collectively, the three trajectories leading to the lowest FEV1 values (persistently low, always-well-below-average and accelerated-decline-II) represented 19% of the cohort and 55% of postbronchodilator airflow obstruction (meeting GOLD spirometric criteria for COPD) at age 45. Moreover, the three FEV1 trajectories that did not reach the normal peak in early adulthood (persistently low, always-well-below-average and always-below-average) represented 22% of the cohort and 44% of airflow obstruction at age 45. This is consistent with an analysis of three cohorts in which half of the participants with spirometric COPD had lower lung function in early adulthood, followed by a normal rate of lung function decline.3 It is also supported by our observations for the FEV1/FVC trajectories: the persistently low and always-well-below-average trajectories represented only 11% of study cohort, but contributed 60% of postbronchodilator airflow obstruction at age 45, indicating that over half of mid-adulthood spirometric COPD has childhood origins.
Our finding that most of the identified trajectories were approximately parallel suggests that, for most people, these trajectories are already established by adolescence. This raises important issues. First, it remains unclear how to maximise lung growth in early life to set people on beneficial trajectories. Second, it is unclear whether interventions involving better management of childhood asthma/AHR, avoiding smoking and maintaining healthy body weight could ‘correct’ the adverse lung function trajectories, or at what age(s) these interventions could be effective. Third, the underlying mechanisms of persistently high and catch-up lung function trajectories are unknown: understanding these would provide important insights into promotion of healthy lung function development in the general population. Finally, the long-term consequences of different lung function trajectories for health and longevity are not yet established, for example, we do not know whether low lung function in late adulthood from persistently low trajectories has similar consequences to low lung function due to accelerated decline.
To our knowledge, these are the most complete longitudinal data on lung function from childhood to mid-adulthood currently available, and this is only the second study to characterise lung function trajectories in terms of all three main spirometric parameters. Strengths of the study include the prospective population-based design, large sample size, high retention rates during follow-up, assessments of lung function at up to 10 time points and an objective approach to model selection. Although 17% of the original cohort was excluded from the main analysis due to insufficient lung function measures, findings were very similar in the sensitivity analysis which included participants with at least two spirometry measures (n=1008; 97% of the original cohort; online supplemental figure S2). Furthermore, those included in and excluded from the main analysis had similar baseline characteristics (online supplemental table S1). We chose to analyse the data using LPA because this allows each individual to be assigned to the most likely trajectory. In our study, the posterior probabilities of belonging to the trajectories were very high (online supplemental table S12), indicating a good model fit. The trajectories for each individual participant associated with the identified FEV1, FVC and FEV1/FVC trajectory classes are shown in online supplemental figures S8–S10. The LPA approach has been used by several previous studies30 and is conceptually similar, but mathematically different, to group-based trajectory modelling that has been used by some others.7 11
Several limitations should be acknowledged. We do not have lung function data before age 9, when the trajectories of lung function may be largely determined,31 32 and we do not yet have lung function measures in later adulthood to complete the trajectories across the life course. Our definition of postbronchodilator airflow obstruction at age 45 does not necessarily indicate clinical COPD. However, in our cohort at this age, the lower limit of normal for FEV1/FVC ratio (0.69) is very similar to the GOLD criterion of FEV1/FVC ratio <0.7 that we used. Although we have reports of coughs and colds at age 3, we may have missed early life exposures such as severe lower respiratory tract infections.33 34 We also compared exposures for each trajectory to all other trajectories combined, whereas previous reports have tended to use the ‘average’ trajectory as the reference group. Our approach risks diluting risk factors that might be shared by more than one trajectory. On the other hand, the ‘average’ group may simply have a mix of beneficial and adverse factors and may not necessarily represent ideal lung health. To facilitate comparisons with previous studies, multivariable analyses using the ‘average’ trajectories as the reference groups are shown in online supplemental tables S13–S15. Overall, there were few differences with the main analyses, although some of the associations changed statistical significance—mostly becoming non-significant, which is likely because of the lower power due to the smaller size of the reference category. Finally, the precision of our estimated associations for some exposure–trajectory combinations is limited by the size of the less common trajectories along with the collinearity between some exposures, such as AHR, asthma and atopy.
留言 (0)