Lung-Function Trajectories Leading to Chronic Obstructive Pulmonary Disease

METHODS We stratified participants in three independent cohorts (the Framingham Offspring Cohort, the Copenhagen City Heart Study, and the Lovelace Smokers Cohort) according to lung function (FEV1 ≥80% or <80% of the predicted value) at cohort inception (mean age of patients, approximately 40 years) and the presence or absence of COPD at the last study visit. We then determined the rate of decline in FEV1 over time among the participants according to their FEV1 at cohort inception and COPD status at study end.

T h e ne w e ngl a nd jou r na l o f m e dicine C hronic obstructive pulmonary disease (COPD) is a major cause of illness and death worldwide. 1 Since the research by Fletcher and colleagues in the 1970s, 2,3 the prevailing paradigm of COPD pathogenesis has been that, in susceptible persons, exposure to particulate matter -especially tobacco smoke -leads to clinical disease through acceleration of the age-related decline in lung function, as assessed by the forced expiratory volume in 1 second (FEV 1 ). Subsequent population studies supported this paradigm and led to therapeutic trials aimed at reducing the rapid decline in FEV 1 . [4][5][6][7][8][9][10][11] Surprisingly, the observed declines in FEV 1 in these trials and in observational cohorts of patients with COPD have been variable and smaller than anticipated, particularly among persons with the most severe airflow limitation. [4][5][6][7][8][9][10][11][12][13][14][15][16] These observations question the notion that COPD always follows a trajectory of rapid decline in FEV 1 and are consistent with the hypothesis that low maximally attained lung function in early adulthood can also result in COPD later in life, even when the rate of decline in FEV 1 is within the normal range. [17][18][19][20][21] In fact, this alternative course was already suggested by Fletcher and Peto 3 and emphasized by Burrows et al. 22 but was never explored in a long-term, prospective investigation. We used data from three large cohort studies to investigate this issue.

Study Design
In this study, we used data from participants in three large and independent cohort studies, the Framingham Offspring Cohort, the Copenhagen City Heart Study, and the Lovelace Smokers Cohort. We used FEV 1 (≥80% of the predicted value [normal FEV 1 ] or <80% of the predicted value [low FEV 1 ]) at cohort inception and the presence or absence of COPD at the final cohort study visit to define four trajectories: normal FEV 1 at cohort inception and no COPD at the final visit (trajectory 1), low FEV 1 at cohort inception and no COPD at the final visit (trajectory 2), normal FEV 1 at cohort inception and COPD at the final visit (trajectory 3), and low FEV 1 at cohort inception and COPD at the final visit (trajectory 4). We then determined the extent to which the rate of decline in FEV 1 was an explanatory variable for a diagnosis of COPD at study end for a patient with a normal initial FEV 1 or a low initial FEV 1 at study inception.

Study Conduct and Oversight
The recruitment of participants and the methods used in each of the three cohorts are presented in detail in the Supplementary Appendix (available with the full text of this article at NEJM.org) and are summarized below. Written informed consent was obtained from all participants, and the studies were approved by the relevant ethics review boards. In our analyses, participants were considered to have COPD if they had grade 2 or higher COPD according to the Global Initiative for Chronic Obstructive Lung Disease (GOLD) 1 grading system; grade 2 or higher COPD according to GOLD is defined as a ratio of FEV 1 to forced vital capacity (FVC) of less than 0.70 and an FEV 1 of less than 80% of the predicted value, with the use of the prediction equations from the National Health and Nutrition Examination Survey for the Framingham Offspring Cohort and the Lovelace Smokers Cohort and local prediction equations for the Copenhagen City Heart Study. 23,24 Framingham Offspring Cohort (FOC) The FOC as it relates to this study is described in detail in the Supplementary Appendix. We identified 187 current or former smokers, 50 to 65 years of age, who had GOLD grade 2 or higher COPD at the final examination in 1996-1997 and who had undergone spirometry before 40 years of age. Six of these persons started smoking after the study began. We calculated the decline in FEV 1 by measuring the difference (in milliliters) between the first available spirometric measurement and the last measurement obtained at 50 to 65 years of age and dividing the result by the number of years between the measurements. The mean decline in FEV 1 was calculated for persons who had had an FEV 1 of at least 80% of the predicted value on baseline spirometry and for those who had had an FEV 1 of less than 80% of the predicted value on baseline spirometry. In addition, we identified 1662 persons who underwent the first spirometry before 40 years of age, regardless of whether they had COPD at the final examination, and categorized them in one of the four trajectories described above.

Copenhagen City Heart Study (CCHS)
The CCHS as it relates to this article is described in detail in the Supplementary Appendix. The annualized decline in FEV 1 was calculated as the slope between two measurements over the longest observation period. We identified 145 current or former smokers, 50 to 65 years of age, who had GOLD grade 2 or higher COPD at the final examination in 2001-2003 and who had undergone spirometry before 40 years of age. As in the FOC, we also identified 1242 participants who were 20 to 40 years of age at baseline and who underwent spirometry 25 years later, regardless of whether they had COPD at the final examination, and we categorized them in one of the four trajectories described above.

Lovelace Smokers Cohort (LSC)
The LSC as it relates to this article is described in the Supplementary Appendix. A total of 1553 persons in the LSC who had undergone at least two spirometry tests were included in the current analysis. Decline in FEV 1 was calculated as the slope defined by the first and last data points over all visits. For combined analyses with the FOC and the CCHS, we identified a subgroup of 163 persons who had similar characteristics to patients enrolled in most COPD trials with GOLD grade 2 or higher COPD and an age of 50 to 65 years at the final examination.

Statistical Analysis
We analyzed the combined rate of decline in FEV 1 among the 495 current or former smokers (187 from the FOC, 145 from the CCHS, and 163 from the LSC) who were 50 to 65 years of age and had GOLD grade 2 or higher COPD at the final examination. We report the observed decline in FEV 1 in subgroups defined according to the baseline level of FEV 1 and subsequent annualized decline in FEV 1 . The latter was based on prebronchodilator FEV 1 values and was calculated as the absolute difference in milliliters between FEV 1 values at the two examinations with the longest interval between them. We compared the proportion of persons with declines in FEV 1 of more than 1, 2, or 3 standard deviations (z scores) above the mean decline among persons who had never smoked in the FOC and the CCHS and among former smokers in the LSC. The three cohorts had different mean declines in FEV 1 , but z scores from each cohort made it possible to use a common threshold (e.g., z score >1) to define rapid decline in FEV 1 in the combined analysis. In addition to using z scores, we also used threshold values previously reported in the literature to classify the decline in FEV 1 as "rapid" if the annualized average decline was at least 40 ml per year and "normal" if the annual-ized average decline was less than 40 ml per year. 14, 25 We performed a combined analysis of 1622 participants in the FOC and 1242 participants in the CCHS who were younger than 40 years of age at baseline and who also underwent spirometry 10 to 27 years after joining their respective cohorts. In the main analysis, we defined normal or low lung function in early adulthood according to baseline FEV 1 (≥80% or <80% of the predicted value). The presence or absence of GOLD grade 2 or higher COPD at the final spirometric measurement was the second measure used to define the four trajectories. We report the distribution of participants in these trajectories and compared their characteristics and decline in FEV 1 . Because the assignment to the trajectory categories required attendance at the first and last examinations, the persons who did not attend the last examination, because of death or loss to follow-up, could not be included in the analysis. We also performed a multivariable logistic-regression analysis with incident COPD as the outcome, as described in detail in the Supplementary Appendix.
To estimate the influence of several potential confounders, we performed a number of additional analyses. First, to estimate a potential selection bias, we performed an analysis of hospital admissions for respiratory diseases, admissions for COPD, and survival in the 844 persons participating in the CCHS who attended the first examination in 1976-1978 but did not attend the final examination in 2001-2003 because of death, emigration, or loss to follow-up. These persons could not be assigned to the trajectory category because of the lack of data on final spirometry, but they could be categorized according to their FEV 1 at the baseline examination: less than 80% of the predicted value (305) or at least 80% of the predicted value (539).
Second, to estimate the way in which regression to the mean affected the observed declines in FEV 1 , we analyzed the decline in FEV 1 in the four conceptual trajectory categories using an alternative baseline FEV 1 measurement instead of the measurement from the first examination, which had been used to define these trajectories. We used data from 1091 persons in the CCHS who, in addition to attending the first examination (conducted in 1976-1978)  persons, we compared the declines between the first and fourth examinations (average interval, 25 years) with those between the second and fourth examinations (average interval, 20 years). Third, we performed a sensitivity analysis using two alternative thresholds to define normal and low maximally attained lung function in the FOC and the CCHS (FEV 1 ≥85% and <85% of the predicted value and FEV 1 ≥75% and <75% of the predicted value). This analysis investigated the effect of the dichotomization on the distribution of participants in the various trajectories.

Characteristics of the Participants in the Three Cohorts
At cohort inception, LSC participants were, on average, 20 years older than the participants in the the FOC and the CCHS (Table 1). In all the cohorts, regardless of the initial FEV 1 , the annual decline in FEV 1 was steepest among persons in whom COPD developed. The highest absolute declines were observed in the CCHS and in the LSC, but when the declines in FEV 1 were compared according to the z scores within each cohort, the persons in the CCHS and the FOC had the greatest declines, a finding that is compatible with the high prevalence of continuous smoking in these cohorts. Table 2 shows the characteristics of the participants who had ever smoked and who had GOLD grade 2 or higher COPD and were 50 to 65 years of age at the final examination, stratified according to their baseline FEV 1 (percentage of the predicted value). The baseline FEV 1 was at least 80% of the predicted value in 20% of the LSC participants, in 44% of the FOC participants, in 49% of the CCHS participants, and in 38% of the participants in the three cohorts combined. In all the cohorts, participants with a baseline FEV 1 of less than 80% of the predicted value had had a slightly higher cumulative exposure to tobacco before the baseline examination than those with an FEV 1 above that threshold, but this difference was not significant. Among persons with COPD at the final examination, the observed decline in FEV 1 and the proportion with a rapid decline in FEV 1 (defined as >2 or >3 SD [z score >2 or z score >3]) were significantly greater among participants with a baseline FEV 1 of at least 80% of the predicted value than among those with a lower baseline FEV 1 ( Table 2). The results of the multivariable logistic-regression analysis with incident COPD at the final examination as the main outcome are presented in Tables S1 and S2 in the Supplementary Appendix.  Table 3 shows the distribution in the four FEV 1 trajectory categories of the 1622 participants in the FOC and the 1242 participants in the CCHS who underwent spirometry before 40 years of age. Continuous smoking and an early age at the start of smoking were significantly more prevalent among the participants in the trajectories leading to COPD (trajectory 3 and trajectory 4) than among the participants in the trajectories not leading to COPD (trajectory 1 and trajectory 2). After an average observation time of 22 years, 332 persons -12% of the 2864 participants included in this analysis -had airflow limitation fulfilling the criterion of GOLD grade 2 or higher COPD. Of these 332 persons with COPD, 158 had had normal baseline FEV 1 , which constitutes 7% of the 2207 persons with normal FEV 1 at baseline, whereas 174 had had low baseline FEV 1 , which constitutes 26% of the 657 persons with low baseline FEV 1 (P<0.001 for the comparison between participants with an FEV 1 ≥80% of the predicted value at baseline and those with an FEV 1 <80% of the predicted value). Thus, 48% of the 332 persons with COPD at the final examination followed the trajectory from normal baseline FEV 1 to COPD, with an average decline in FEV 1 of 53 ml per year (trajectory 3 in Table 3, and Fig. S4 in the Supplementary Appendix), whereas 52% followed the FEV 1 trajectory from low maximally attained lung function (FEV 1 <80%) at baseline to COPD, with a significantly smaller decline in FEV 1 of 27 ml per year (P<0.001) (trajectory 4 in Table 3, and Fig.  S4 in the Supplementary Appendix). As seen in Figure 1, the distribution of the observed declines in FEV 1 in the four trajectory categories showed substantial variability and overlap.

Rate of Decline in FEV 1 among Persons in Whom COPD Developed
Among the 2864 participants in the FOC and the CCHS who underwent spirometry before 40 years of age, 812 reported at both the baseline examination and the final examination that they had never smoked. The average decline in FEV 1  23,24 ). † The z scores are defined according to internal controls in each cohort and describe the decline in FEV 1 1, 2, and 3 standard deviations above the observed mean decline among former smokers in the Lovelace Smokers Cohort and among those who have never smoked in the Framingham Offspring Cohort and the Copenhagen City Heart Study. T h e ne w e ngl a nd jou r na l o f m e dicine  among these participants was 18 ml per year. Only 27 of them had COPD at the final examination, which constitutes 8% of all persons with COPD. Of the persons who had never smoked and who had COPD at the last study visit, 7 (26%) followed trajectory 3, whereas 20 (74%) followed the trajectory from low maximally attained lung function (FEV 1 <80%) at baseline to COPD (trajectory 4). The average decline in FEV 1 among these persons was 37 ml per year and 23 ml per year, respectively. The analysis of persons who attended the baseline examination of the CCHS but not the final examination showed that participants with an FEV 1 of less than 80% of the predicted value had a significantly greater risk of hospital admissions for respiratory diseases than did those with an FEV 1 of at least 80% of the predicted value (25% vs. 18%, P = 0.02), as well as a significantly greater corresponding risk of admissions for COPD (9% vs. 4%, P = 0.008) and of death (32% vs. 25%, P = 0.02). Additional analyses of the decline in FEV 1 among those who did not attend the final, fourth examination of the CCHS showed a greater decline in FEV 1 in this group between the first and the second examination and between the first and the third examination than in the group of participants who attended the fourth examination (Table S3 in the Supplementary Appendix).
Analysis of the decline in FEV 1 between the second and the fourth examination of the CCHS (average interval, 20 years) that was conducted to investigate the effect of regression toward the mean showed a significantly greater decline in FEV 1 among the persons in trajectory 3 than among those in trajectory 4 (decline of 60 ml per year in trajectory 3 vs. 37 ml per year in trajectory 4, P<0.001) ( Table S4 in the Supplementary Appendix). A corresponding analysis of the decline in FEV 1 between the second and the third examination (average interval, 11 years) showed a similar pattern, with a greater decline among those in trajectory 3 than among those in trajectory 4, but the difference between these two declines was not significant (27 ml per year and 17 ml per year, respectively; P = 0.37).
Finally, we observed that the choice of the threshold value defining a normal FEV 1 level at the first examination of the FOC and the CCHS substantially affected the distribution of the participants in the two trajectory categories leading to COPD. When we used a cutoff value of 85% of the predicted FEV 1 to define a normal FEV 1 in early adulthood, we found that 232 of the 332 participants (70%) who presented with COPD at the final examination were considered to be in the trajectory 4 category (low maximally attained lung function at baseline and COPD at final examination). In contrast, when we used a cutoff value of 75% of the predicted FEV 1 to define a normal FEV 1 , only 126 of the 332 persons (38%) with COPD could be assigned to trajectory 4.

Discussion
The main new finding of this study is that the lung-function value reached in early adulthood seems to be important with respect to the diagnosis of COPD later in life. Our data suggest that approximately half the persons presenting with airflow limitation followed the paradigm that implied a rapid decline in FEV 1 from a normal level of lung function in early adulthood, whereas the other half had a rather normal decline in FEV 1 but started from a low initial value of FEV 1 .
We acknowledge that dichotomizing persons with COPD into two specific trajectories, although conceptually helpful for understanding the natural history of the disease, is artificial. Instead, there exists a wide range of individual trajectories, such that both low maximally attained lung function in early adulthood and a subsequent rapid decline may contribute to COPD. 21 In addition, it is possible that a single person over time can have periods of rapid decline in FEV 1 followed by periods when the decline is normal or almost normal. If the period of fast decline had already taken place in early adulthood, this person would have been wrongly assigned to the trajectory category of low maximally attained lung function (trajectory 4) in our study, even though a rapid decline in FEV 1 was the main mechanism behind this person's COPD. In addition, our sensitivity analyses showed that the estimation of the contribution of these two conceptual trajectories was strongly affected by the cutoff values chosen for the definition of a normal maximally attained FEV 1 . Nevertheless, even though we cannot precisely estimate the contribution of the trajectory of low maximally attained lung function (trajectory 4 in Fig. 1) to COPD using our study design, our results suggest that this contribution may be substantial T h e ne w e ngl a nd jou r na l o f m e dicine  and that populations of patients with COPD comprise persons with different rates of decline in FEV 1 (Fig. 1). This observation is in accord with previous studies and suggests that a substantial proportion of patients with COPD may not have had a rapid decline in FEV 1 , which for decades has been regarded as the hallmark of COPD. 2,3,[12][13][14] Persons with a baseline FEV 1 of less than 80% of the predicted value had a risk of COPD in midlife that was more than 3 times as high as the risk among those with a higher baseline FEV 1 (26% vs. 7%). Studies involving children and young adults have shown that the most important determinant of maximally attained lung function later in life was the lung-function value at a younger age. [26][27][28][29] Svanes et al. observed that childhood risk factors such as maternal smoking, childhood respiratory infections, and childhood asthma were strongly related to the development of COPD in young adults. 30 Nevertheless, although our results highlight the important role of maximally attained lung function in early adulthood, Shown is the distribution of observed annual decline in forced expiratory volume in 1 second (FEV 1 ) among 2864 participants in the Framingham Offspring Cohort (FOC) and the Copenhagen City Heart Study (CCHS), according to the four trajectories defined on the basis of a normal FEV 1 (≥80% of the predicted value) or low FEV 1 (<80% of the predicted value) at baseline and the presence or absence of chronic obstructive pulmonary disease (COPD) at the final examination. Participants were considered to have COPD if they had grade 2 or higher COPD according to the Global Initiative for Chronic Obstructive Lung Disease (GOLD) 1 grading system; grade 2 or higher COPD according to GOLD is defined as a ratio of FEV 1 to forced vital capacity of less than 0.70 and an FEV 1 of less than 80% of the predicted value, with the use of the prediction equations from the National Health and Nutrition Examination Survey for the FOC and local prediction equations for the CCHS. 23,24 The mean decline in FEV 1 was 24 ml per year in trajectory 1 (Panel A), 2 ml per year in trajectory 2 (Panel B), 53 ml per year in trajectory 3 (Panel C), and 27 ml per year in trajectory 4 (Panel D). The decline in FEV 1 in trajectory 3 was considered to be rapid. they also indicate that COPD does not develop in approximately 75% of persons with a low maximally attained FEV 1 . Our study has a number of strengths and limitations. Among the former is the replication of findings in large and independent cohorts. The long observation periods in the FOC and the CCHS, as compared with the shorter observation periods in other studies, allowed robust estimates of FEV 1 , because the annual decline in FEV 1 was small relative to the high measurement error in the assessment of FEV 1 . 13,31 The two major potential limitations of our study are the selection bias and the regression to the mean for the estimates of decline in FEV 1 . The long observation period in the FOC and the CCHS introduced a selection of healthy persons, which may have resulted in underestimation of lung-function decline. In line with this, we observed that the persons who did not attend the final, fourth examination of the CCHS, because of either death or loss to follow-up, had a greater decline in FEV 1 between the first and the second examination and between the first and the third examination than did the persons who attended the fourth examination (Table S3 in the Supplementary Appendix). Previous analyses of the CCHS indicated that a rapid decline in FEV 1 was associated with higher mortality. 32 However, because the persons included in the FOC and CCHS analyses were quite young at baseline, this survivor bias is likely to be smaller than that in most COPD intervention trials, which have enrolled much older patients. 12,13 The use of baseline FEV 1 measurement both to classify the participants as having a normal or low initial FEV 1 value and to estimate the subsequent decline in FEV 1 introduced potential problems with respect to regression to the mean. 33 Persons with a higher initial FEV 1 would tend to have a more rapid observed decline in FEV 1 even if there was no association between these two. In contrast, among persons with a low initial FEV 1 , regression to the mean would lead to a small decline in FEV 1 . Therefore, it would have been preferable to perform multiple FEV 1 measurements at the beginning of the studies to define the true FEV 1 baseline more precisely, as is often done in interventional trials, but we did not have such data. Although regression to the mean undoubtedly affects our findings, the anal-yses of decline in FEV 1 starting from the second rather than the first examination of the CCHS showed a consistent pattern of a greater decline among persons in trajectory 3 than among those in trajectory 4 ( Table S4 in the Supplementary Appendix). This supports our main conclusions and indicates that regression to the mean is not the only mechanism responsible for our findings.
Still another potential methodologic weakness of the FOC and the CCHS concerns the age of the participants at baseline (approximately 35 years). Although this age corresponds to the last part of the "plateau phase" of FEV 1 (i.e., the age at which maximum lung function is attained before the inevitable decline with aging begins to take place), it would have been preferable if the participants had undergone spirometry in their mid-20s, so that the potential influence of an incipient decline in FEV 1 during early adulthood could be avoided. [34][35][36][37] Finally, the appropriateness of the nomenclature used in the present article (i.e., COPD or no COPD) could be misleading, because not all people with a ratio of FEV 1 to FVC of less than 0.70 and an FEV 1 of less than 80% of the predicted value have COPD. We chose the term COPD because it is in common use, but the broader descriptive term "airflow limitation" could be considered by some to be more correct.
In conclusion, the results of this study suggest that the classic trajectory of an accelerated decline in FEV 1 from a normal level is not an obligate feature of COPD and that a substantial proportion of the persons in whom COPD develops have a low FEV 1 level in early adulthood.