Longitudinal association of body mass index with lung function: The CARDIA Study

The data used in these analyses were collected in the Coronary Artery Risk Development In Young Adults (CARDIA) study, a multi-center cohort study occurring in the US. The cohorts were recruited from the general population, mostly by telephone, randomly sampled from a prepaid health plan in Oakland, CA and from populations in Birmingham, AL, Chicago, IL, and Minneapolis, MN. The response rate was approximately 50%, which was considered acceptable given the required long term commitment to the study. The detailed methods, instruments and quality control procedures are described in other published reports [26, 27]. In 1985–86 (year 0), 5,115 black and white men and women were recruited for the year 0 examination; 4,624 were reexamined in 1987–88 (year 2); 4,352 in 1990–91 (year 5); 4,086 in 1992–93 (year 7); and 3,950 in 1995–96 (year 10). At year 0, CARDIA included approximately equal numbers of participants who were black and white, men and women, aged 18–24 and 25–30, and had more than or less than or equal to high school education [26, 27]. We excluded 58 participants who were outside the 18 through 30 age range at year 0, 7 women who were pregnant at baseline, and anyone missing baseline lung function, BMI, physical activity, alcohol intake, or smoking, leaving 4,734 participants for analysis. Of these, 4,277 attended year 2, 4,043 attended year 5, and 3,668 attended year 10. We excluded 147 observations in women who were pregnant at followup measurement of lung function, since pregnancy might influence both BMI and lung function, but included observations in those same women when not pregnant.

Clinic attendance was somewhat higher at the year 10 exam among whites (82%) than among blacks (73%). The participants lost to follow-up after years 0, 2, or 5 did not differ significantly in most of their year 0 characteristics when compared with those observed at year 10. Specifically, both mean FVC and FEV₁ at year 0 did not differ significantly across those whose last examination attended was year 0 (n = 203), 2 (n = 232), 5 (n = 221), 7 (n = 410), or 10 (n = 3668).

Body weight was measured in light clothing to the nearest 0.1 kg with a calibrated balance beam scale, height without shoes was measured to the nearest 0.5 cm using a vertical ruler, and BMI (kg/m²) computed.

Demographic characteristics, lifestyle habits, and medical history were collected by self-report using a questionnaire. Physical activity was measured using an interviewer-administered questionnaire [28] concerning the frequency of participation in 13 different activities during the past 12 months. Because participants were not asked specifically about duration of physical activity, exact energy expenditure cannot be estimated and the activity is expressed approximately in "Exercise Units" (EU). A score of 100 EU is roughly equivalent to participation in activities such as a vigorous exercise class or bicycling faster than 10 miles per hour, two or three hours a week for six months of the year. Average weekly alcohol intake was determined separately for beer, wine, and liquor. Smoking status was categorized into four groups: never smokers, ex-smokers, current smokers of ≤ 15 cigs/day, and current smokers of >15 cigs/day. Asthma diagnosis [29] was made at a given examination if the subject was taking asthma medication (usually based on examination of medicine containers) or self-report of a medical diagnosis of asthma (not asked at year 5). The asthma variable had three categories: asthma diagnosed before the beginning of the study, asthma diagnosed during the study and people that never had asthma diagnosed either before or during the study.

Lung function was measured using a Collins Survey 8-liter water sealed spirometer and an Eagle II Microprocessor (Warren E. Collins, Inc., Braintree, MA). Standard procedures of the American Thoracic Society [30] were followed at all examinations. Daily checks for leaks, volume calibration with a 3-liter syringe and weekly calibration in the 4–7 liter range were undertaken to minimize methodological artifacts between exams. We analyzed FVC and FEV₁ as the maximum of five satisfactory maneuvers and represented as percent of predicted [12, 31‐34]. In almost all cases, the maximum and second highest maneuvers agreed to within 150 ml.

Year 0 (baseline) BMI, divided into quartiles, was the primary predictor variable. The use of standard NHLBI BMI based adiposity categories to categorize the distribution of BMI in this population resulted in unequal distribution of the population in each category and prevented the detailed evaluation of lung function in thin participants at baseline (Table 1); furthermore, participants changed categories during follow-up. Hence study specific year 0 (baseline) BMI quartile cutpoints were used as the primary predictor variable. However, the percentage of people progressing to different obesity categories (as defined by standard NHLBI cutoffs: normal, < 25 kg/m²; overweight, ≥ 25 kg/m² – < 30 kg/m²; and obese, ≥ 30 kg/m²) within each baseline BMI category over a 10 year period was calculated [35]. Change in BMI was evaluated as an additional predictor variable. Participants were divided into three age groups: 18–21 years, 22–26 years, and 27–30 years based on their year 0 age. The effect of BMI on lung function was evaluated in these 3 age groups separately.

We considered that methodological differences might exist between examinations, reflecting small changes in spirometry procedures that occurred by using different technicians across examinations, despite the formal procedures remaining the same. As the first analytic step, we estimated such methodological differences adjusting for race, gender, age, age², height, and height² by subtracting the mean lung function value for participants of a given age at a later exam from the mean lung function value for other participants of the same age at an earlier examination, then averaging over ages (age-matched calendar time differences) using the method of Jacobs et al. [36]. Relative to year 10 measurements, we added 53 ml, 54 ml, and 16 ml to the predicted FVC at year 0, 2, and 5, respectively; added 6 and 21 ml to the predicted FEV₁ at years 0 and 2, and subtracted 25 ml from the predicted FEV₁ at year 5; and subtracted 0.94, 0.55, and 0.91 units from 100* the predicted FEV₁/FVC ratio at the respective years.

Analyses of lung function and BMI relationships in three narrow age ranges allowed us to separate the cross-sectional and longitudinal relationships as people went through different phases of lung development, plateau, and decline [9, 37‐39]. Longitudinal changes in lung function over 10 years, as estimated by FVC, FEV₁, and FEV₁/FVC at years 0, 2, 5, and 10 were estimated within each age group across different baseline BMI quartiles. Using the lung values corrected for methodological differences, a repeated measures regression model (SAS PROC MIXED) adjusted for current age, time, race, sex, height, age group category, smoking status, physical activity, and alcohol intake (all at baseline) and baseline prevalence and incidence of asthma [9, 36, 38, 39] was used to estimate the association of baseline BMI with lung function. The covariates were selected a priori based on their associations with the variables of interest. Linearity assumptions and goodness of fit were verified by examining the sequence of mean dependent variable values at each age within each BMI category in reference to the fitted lines. Goodness of fit was adequate (data not shown). Serial correlation was modeled as compound symmetry. We estimated the effect of concurrent change in BMI from year 0 to year 10 on lung function using a repeated measures regression with change in the lung parameter as the dependent variable (3 repeats: year 2 – year 0, year 5 – year 0, and year 10 – year 0), and baseline BMI quartiles, change in BMI in 5 categories, and their interaction as the independent variables of interest. The 4 categories of least change in BMI stratified a large number of people, while the highest category allowed a closer evaluation of change in lung function among those who gained a considerable amount of weight (≥ 6 kg/m²). In analyses evaluating the association between change in BMI and change in lung function, we added as covariates change in smoking status, change in physical activity, and change in alcohol intake. People who were heavier at baseline tended to gain more weight over 10 years than did people who were lighter. Therefore this additional model evaluated how much of the lung function and BMI relationship persisted after accounting for subsequent weight change. It also evaluated the relationship of lung function to change in BMI itself. For comparison with the model of change since baseline, we also examined a transition model [40] in which the repeated changes were for year 2 – year 0, year 5 – year 2, and year 10 – year 5. Here a BMI increase ≥ 6 kg/m² was rare given that the maximum time interval between examinations was 5 years, so the highest BMI change category was ≥ 2.5 kg/m².

The study sample at year 0 was aged 24.9 ± 3.6 years (Table 2). There were 1017 in the 18–21 year old birth cohort (mean age 19.6 years), 1842 in the 22–26 year old cohort (mean age 24.1), and 1875 in the 27–30 year old cohort (mean age 28.5). By design, the participants were evenly distributed among race-sex groups. Thirty nine percent had no education past high school. 2225 had never smoked and never had asthma either prior to year 0 or during the 10 years of study. Quartile cut points for year 0 BMI, computed before exclusion for missing covariates, were: 25th percentile 21.2 kg/m², median 23.4 kg/m², and 75th percentile 26.4 kg/m². A higher percentage of participants were black in the highest BMI quartile as compared to lower BMI quartiles (64% vs. 46%). Progression in NHLBI BMI-based adiposity categories [35] is depicted in Figure 1. Overall, the mean increase in BMI over 10 years was 3.0 ± 3.5 (SD) kg/m². Category cut points 0, 1, 2.5, and 6 kg/m² for 5 categories of change in BMI were selected to represent weight loss and gradations of weight gain. The increase in BMI tended to be larger, the higher the initial BMI (Table 3).

The change in FVC over a 10 year period differed across baseline BMI quartiles (p < 0.0001). Average 10 year FVC change was 71, 19, -72, and -185 ml in the lowest through highest BMI quartile (Table 4), with the increase in the low BMI participants being more pronounced in the youngest birth cohort (Figure 2). The estimated mean FVC generally increased for 5 years, then plateaued in all birth cohorts in the thinnest people at baseline (Figure 2). A pattern of increase and plateau is seen from mean age 19.6 years, with no decrease in FVC through mean age 38.5 years (Figure 2, oldest birth cohort, year 10). In contrast, for those in the highest BMI quartile FVC decreased continuously over the same time period in all birth cohorts. People in the second quartile of baseline BMI displayed a tendency to increase FVC over 10 years, but less so than in quartile 1 (data not shown) and people in the third quartile of baseline BMI displayed a tendency to decrease FVC, but less so than in quartile 4 (data not shown).

The change in FEV₁ over a 10 year period also differed across baseline BMI quartiles (p < 0.0001). The FEV₁ change was 60, 18, -28, and -64 ml in the lowest through highest BMI quartiles (Table 4), respectively, with the increase in the low BMI participants being more pronounced in the youngest birth cohort and no suggestion of a decline in FEV₁ through age 38 in the lowest BMI quartile (Figure 3).

In contrast to the FVC and FEV₁, estimated mean FEV₁/FVC tended to decrease in the thinnest participants for the first 5 years and then increase over the next 5 year period as compared to a continuous increase in the participants in the highest baseline BMI quartile (Table 4, Figure 4); p for changes in the ratio across BMI categories was < 0.0001 within each age group and did not vary significantly by age group. Age-adjusted mean change in FEV₁/FVC (averaged across the 3 birth cohorts) was -0.07, 0.29, 1.00 and 2.03 in the lowest to highest BMI quartiles, respectively (Table 4), with the initial decrease in the low BMI participants being more pronounced in the youngest cohort.

Findings were similar among the 2225 participants at year 0 who were never smokers and did not have asthma at any time during the study (Table 4). Race and gender did not significantly modify the association between year 0 BMI and any of the lung function variables (data not shown).

Change in BMI was a significant predictor of FVC, FEV₁, and FEV₁/FVC over the 10 year study period. The direction of change in lung function according to change in BMI was dependent on year 0 BMI, but not on birth cohort (p < 0.0001 for the interaction of change in BMI and baseline BMI, Table 5). Averaging across the 3 birth cohorts, FVC increased over the study period in the lowest BMI quartile across all categories of change in BMI, although the increase was least in those who lost weight (5 ml) or who gained > 6 kg/m² (5 ml) as compared to a increase in FVC in those who gained 1–5.9 kg/m² (15–65 ml) (p for difference for those who lost weight as compared to those who gained 2.5–5.9 kg/m² < 0.0001) (Table 5). In contrast, in individuals in the highest baseline BMI quartile FVC increased in those who lost weight (22 ml) or those who gained 0.1–0.9 kg/m² (15 ml), but decreased progressively as weight gain increased, reaching a loss of 264 ml in those who gained > 6 kg/m² (p for difference <0.0001). Within each category of change in BMI, baseline BMI remained a significant predictor of FVC (p value < 0.0001).

Averaging across all 3 birth cohorts, FEV₁ decreased in all baseline BMI quartiles across all categories of change in BMI. In the lowest baseline BMI quartile, the decrease was lower in those who lost weight (-38 ml) or gained minimal weight (-31 ml) as compared to those who gained > 6 kg/m² (-110 ml) during the same period (Table 5) (p for difference for >6 kg/m² as compared to those who lost weight = 0.001). Individuals in the highest baseline BMI quartile also lost increasing amounts of FEV₁ with increasing change in BMI (Table 5) (p for difference for > 6 kg/m² as compared to those who lost weight < 0.0001). The magnitude of loss of FEV₁ was higher in the highest baseline BMI quartile as compared to the lowest baseline BMI quartile.

The FEV₁/FVC decreased among all participants. The decrease in FEV₁/FVC among the lowest baseline BMI quartile was higher with increasing weight gain (p for difference for > 6 kg/m² as compared to those who lost weight < 0.0001) as compared to the decrease in FEV₁/FVC among the highest baseline BMI quartile with increasing weight gain (p for difference = 0.01 for those who gained > 6 kg/m² as compared to those who lost weight) (Table 5).

The transition model using year 2 – year 0, year 5 – year 2, and year 10 – year 5 as repeats but otherwise parallel to the model shown in Table 5 showed a similar pattern for each lung function measure to that shown in Table 5, but lung function changes for each BMI change category were generally smaller than in Table 5, consistent with the shorter exposure intervals being modeled (2, 3, and 5 year intervals in the transition model, compared to 2, 5, and 10 year intervals in Table 5).

Restricting to participants who were never smokers and never had asthma during the study, the direction and magnitude of association between change in BMI and FVC, FEV₁, and FEV₁/FVC was similar to that seen in the entire population (data not shown). This similarity included that change in BMI was a significant predictor of FVC, FEV₁ and FEV₁/FVC within each baseline BMI category (p <0.0001 for the interaction of change in BMI and baseline BMI).