Longitudinal changes of body mass index, spirometry and diffusion in a general population

The body mass index (BMI), body weight (kg) to squared height (m) ratio, is a well known index that is receiving increasing attention to evaluate the effects of overall obesity on ventilatory function.

Besides age and height, BMI has recently been considered as an additional independent variable in models for deriving spirometric prediction equations 1, 2. In particular, the present authors have previously observed that BMI improved the precision of predictions for both volumes and flows, regardless of sex 2.

Furthermore, BMI or body weight gains have been shown to be related to longitudinal decline of forced expiratory volume in one second (FEV₁) and forced vital capacity (FVC) in adults, both in occupational cohorts 3, 4 and in general population samples 5–7. This effect of BMI on lung function has been shown to be independent of age 4, 5. In studies with both males and females, significantly higher effects have been found in males 5–7.

Little is known about the effect of longitudinal changes of BMI on variations of the carbon monoxide diffusing capacity of the lung (D_L,CO) in large general population samples. Body weight, but not BMI, change was included among predictors of longitudinal change of D_L,CO in adults from the Tucson Epidemiological Study of Obstructive Lung Disease, although its effect was not specifically addressed 8. By applying a statistical model analogous to that applied in the above mentioned paper, the present authors have previously found that weight at baseline and change in weight were significant predictors of D_L,CO longitudinal increases over an 8-yr period, both in adult males and females from the Po river delta prospective epidemiological study in Italy 9. With regard to clinical studies, large BMI values have been shown to be an independent determinant of increased diffusion indexes in groups of patients with chronic obstructive pulmonary disease (COPD) 10 and obstructive sleep-apnoea 11. In a small study on massively obese but otherwise normal, nonsmoking, young adults, the single-breath D_L,CO increased with the degree of obesity 12.

The aim of this study was to evaluate the effects of changes of BMI, over an 8-yr follow-up, on longitudinal changes of slow vital capacity (VC), FVC, FEV₁, and D_L,CO in a general population sample of North Italy.

Methods

Results

Baseline and follow-up samples characteristics were compared and the results are shown, separately by sex, in tables 1⇓ and 2⇓. Medians, 5th and 95th percentiles were computed to describe anthropometric measures and respiratory function. Age variations approximately corresponded to the lengths of follow-up time. Forty-two per cent of subjects were interviewed 8 yrs apart and over 90% between 7 and 9 yrs apart. Lung function indexes showed negative variations in the overall medians mainly due to aging. No major differences were observed between males and females: variations over time were almost identical, as well as baseline and follow-up values of age and BMI; baseline and follow-up values of weight, height, and lung function measurements were less in females than in males, as expected.

Among the variables considered, only age and FEV₁ evaluated at baseline were significantly different (p<0.05 by Wilcoxon rank-sum test) between those who had complete longitudinal data and those who had not (43 versus 46 yrs mean age, and 3.06 versus 2.93 L mean FEV₁).

It was observed that most of those who actually lost weight improved their lung function and those who gained weight reduced their lung function. In the sample, as shown in table 3⇓, there seemed to be a linear relationship in the trend by which the lung function changes (dVC, dFVC, and dFEV₁) decreased within increasing dBMI quartiles. The trend was clearer in males than in females.

Tables 4⇓ and 5⇓ report the regression estimates for the final models for the five lung function and the two diffusing capacity indexes separately by sex. The only significant interaction was observed in both sexes between baseline BMI and dBMI in the model for dVC, suggesting that baseline BMI represents an effect modifier of the relationship between dBMI and dVC. In both sexes, BMI increases (dBMI >0) over time caused reductions in dFVC (dFVC <0) and dFEV₁ (dFEV₁ <0). For example, as shown in table 4⇓, −20 and −16 mL were the median values of dFEV₁ and −3 and −11 mL were those of dFVC, corresponding to a dBMI unit change in males and females, respectively. The effect of dBMI on dVC can be seen in the models considering the joint effect of the terms dBMI and the interaction BMI*dBMI: the overall outcome is negative. This effect will be further exemplified below as a comment to figure 1⇓. Among females, dBMI caused statistically significant decreases of dFEV₁/VC and nonsignificant decreases of dFEV₁/FVC. In contrast, in males, dBMI caused nonsignificant increases of dFEV₁/VC and statistically significant increases of dFEV₁/FVC. These relationships may be explained considering that in males VC and FVC reductions after BMI increases were greater than reductions of FEV₁, causing positive changes of the ratios. BMI increases seemed to provoke a slight positive variation in median D_L,CO, although not significant. A strong and significant association with baseline BMI was shown by dD_L,CO whereas it wasn't by dK_CO. All lung function indices decreased with aging.

Although some of the risk factors considered were associated with changes of spirometric indices and D_L,CO, none of them appeared to be a potential confounder except for baseline BMI that remained in the final models. As an example, table 6⇓ shows the effects on dFEV₁ of the risk factors that were not included in the final regression models shown in tables 4 and 5 for males and females separately. For both sexes, low socioeconomic status and persistent smoking showed negative effects on FEV₁ changes. Occupational exposure was significantly associated with FEV₁ decrease in males only.

To graphically represent the effect modifications of baseline BMI values, figure 1⇓ shows predicted values of dVC (by regression estimates in table 2⇓) for two males and two females at 40 yrs with baseline BMI of 20 and 30 kg·m⁻², respectively. Those who increased their BMI reduced their lung function. Males appear to have larger reductions of VC than females as they get fatter. For a unit increase in BMI, 40-yr-old males with 20 and 30 kg·m⁻² of BMI at baseline lose 0.021 and 0.053 L of VC, respectively. Whereas in 40-yr-old females with 20 and 30 kg·m⁻² of BMI at baseline, a unit increase in BMI corresponds to a VC loss of 0.003 and 0.026 L of VC, respectively. VC losses after BMI increases are greater for more obese individuals. Intersections with the vertical axes show the median decline for those who reported no BMI changes over time (−0.04 and −0.12 L for the males, −0.02 and −0.16 L for the females with a BMI of 20 and 30 kg·m⁻², respectively). Intersections with the horizontal axes represent the reduction of BMI needed to achieve no VC change (−1.9 and −2.2 kg·m⁻² for the males, and −6.5 and −6.2 kg·m⁻² for the females with a BMI of 20 and 30 kg·m⁻², respectively). However, predictions beyond the sample range of BMI changes, reported in table 1⇓, should be carefully interpreted.

Regression models were also built including changes in weight instead of changes in BMI (data not reported). The models included the same covariates as in tables 3 and 4⇓⇓ (replacing BMI with weight). There was a significant interaction between weight changes and baseline values of height. The models based on weight changes gave predictions of lung function changes that were similar to those obtained by models based on BMI changes.

Discussion

In a general population sample of adults surveyed in northern Italy, it was observed that gains of BMI over an 8-yr follow-up time (i.e. dBMI >0) induce VC, FVC and FEV₁ decreases (i.e. dVC < 0, dFVC < 0 and dFEV₁ <0). The detrimental effect of gaining weight might be reversible for many adults, as it was observed that most of those who reduced their BMI values also increased their lung function. The extent of lung function loss tends to be higher among those who, at baseline, report greater BMI values. Males appear to experience larger losses than females. Conversely, it seems that longitudinal changes of BMI cause a slight and nonsignificant increase in D_L,CO values in both males and females.

In order to exclude weight changes that could be related to physical growth, as usually happens in youth 25, the present authors chose to study adults aged ≥25 yrs. In fact, in a previous study by the current authors 2, on cross-sectional and longitudinal observations from subjects 8–64 yrs old, they observed that lung function increased up to a certain value of BMI and then fell according to a parabolic trend. However, when subjects under 25 yrs were excluded, the lung function was steadily decreasing as observed in the present study.

The authors considered whether weight changes might have been a better choice than BMI changes at predicting lung function longitudinal variations. Indeed BMI changes and changes in weight were highly correlated (R-squared=0.9165). However, the effects of weight changes on lung function were also dependent upon baseline values of height. That is the taller the subject the lesser the effect of weight changes on lung function. On the contrary, BMI changes were independent of baseline values of height. Since it was observed that selected individuals in the present study showed height changes over the follow-up time, the authors chose to use changes in BMI rather than changes in weight.

With regard to the effects of weight changes on spirometric indexes, the present results are comparable with those found by other authors both in occupational cohorts and in general population samples: Carey et al. 7 (96 mL FEV₁ loss for males and 51 mL for females per a 10 kg increase in weight over 7 yrs of follow-up, in a population-based study on adults aged 18–73 yrs); Wise et al. 6 (11.1 mL for males and 5.6 mL for females per a 1 kg increase in weight over 5 yrs, in smokers aged 30–60 yrs who had quit smoking); Chinn et al. 4 (17.6 mL per a 1 kg increase in weight over ∼7 yrs in male shipyard workers aged 45–75 yrs). In agreement with other studies, the present authors have observed that the effect of gaining weight on decline of pulmonary function was independent of age, smoking habit, and occupational exposure 3–5, 7.

Also, in accord with the results observed in population-based studies 5–7, the present authors observed that the effect of weight gain on lung function was greater in males than in females. The results reported here support the hypothesis that this effect is likely due to sex-related differences in weight distribution. Males tend to deposit fat centrally (increasing the circumference of the abdomen), while in females the deposition is typically peripheral (increasing the circumference of the hip). Thus, a mechanical effect on the diaphragm in men, impeding expansion of lungs during inspiration, could justify their higher impairment of ventilatory function. Indeed, it has been recently observed that abdominal obesity, as measured by waist-to-hip ratio, is associated with significantly greater reductions of FVC in males compared to females and with significant reduction of FEV₁ in males but not in females 26. Since in the present sample body measurements, such as thickness of the skin-folds, girth of the abdomen, or breadth of the hips were not recorded, it was not possible to take the independent effect on lung function of different patterns of body fat deposition into account 27, 28.

In this sample it was observed that in males BMI increases over time provoked decreases in lung function that were greater for VC and FVC than for FEV₁. Therefore, the ratios FEV₁/VC and FEV₁/FVC increased as BMI increased, although VC, FVC and FEV₁ all decreased. In females, however, the ratio decreased mainly due to the fact that females did not report as large reductions of FVC as males.

It was also observed that the magnitude of VC reductions following BMI increases depended upon the baseline BMI values, whereas FVC reductions did not. In particular, the more obese at baseline the greater VC reductions over time. For 40-yr-old males with BMI values of 20 and 30, VC reductions were 21 and 53 mL, respectively, whereas FVC reductions were 34 mL for both. Such a difference could be explained considering that the slow manoeuvre is apt to yield larger volumes than the forced manoeuvre, and therefore might be more sensitive to mechanical constraints to diaphragm expansion caused by obesity.

In the final models several risk factors were not included that, although possibly significant, appeared not to confound the relationship of main interest between lung function and BMI. Among them, smoking habit is potentially modifiable over time as well as BMI. Quitting smoking compared with keeping smoking improves FEV₁ change of about 11 mL in males and 49 mL in females. Conversely, reducing BMI by one unit improves FEV₁ change of about 20 mL in males and 16 mL in females.

With regard to D_L,CO, far fewer papers have analysed its relationship with BMI changes, other than in small clinical samples. In a case-controlled study, higher D_L,CO and K_CO values were found in healthy nonsmoking obese adults compared to nonobese matched controls, while the correlation analysis showed that BMI was a significant determinant for K_CO 29. In nonsmoking massively obese adults otherwise healthy, % predicted D_L,CO has been shown to increase with increasing obesity over the range of weight/height ratios between 0.6 kg·cm⁻¹ and 1.2 kg·cm⁻¹ 12. In the same clinical sample, it was also shown that the D_L,CO decreased after remarkable weight loss 12. A significant correlation between BMI and D_L,CO, and between BMI and K_CO was found in a group of outpatients with stable COPD 10 and in a group of never smoker obese patients with moderate-to-severe obstructive sleep apnoea 11, respectively. Data from the different samples above consistently show a positive correlation between BMI and D_L,CO, which is independent from concomitant comorbidity, respiratory disease and smoking habit.

In a previous paper 9 the present authors have observed, by using a different approach of statistical analysis, that weight at baseline and change in weight were significant predictors of D_L,CO increases over an 8-yr period, both in adult males and females >40 yrs 9. In the present paper, BMI was used instead of weight, as stated above. However, an analogous positive effect of BMI changes and BMI values at baseline (significant in males) on D_L,CO changes in both sexes was found. For the K_CO index, baseline BMI was not a significant contributor, possibly because alveolar volume, which correlated with BMI, is already accounted for in the computation formula (i.e. K_CO=D_L,CO/V_A).

However, the highest D_L,CO values with weight gain have usually been observed in obese individuals. This relationship has been explained by the increased capillary blood volume in the lungs, which increases with increasing BMI 11, 12. In the present sample from a general population, an increase in D_L,CO values was observed along with changes of BMI that was not statistically significant. It may be attributed to the fact that the majority of the subjects were not overweight (BMI <25 kg·m⁻²) and only a few were obese (BMI >30 kg·m⁻²), according to the classification of the American Society for Clinical Nutrition 30.

As mainly healthy subjects from a general population living in a rural area with low pollution were evaluated 15, 16, the lack of correction for haemoglobin levels should not have affected D_L,CO data. Indeed, it has been demonstrated that the correction for haemoglobin levels does not significantly change the uncorrected D_L,CO values in a large series of hospital patients in usual clinical conditions 31.

In the analyses a median regression was applied to quantify the magnitude of the effects of BMI changes on lung function variations, after adjusting for other risk factors. Assumptions for making inference by using median regression are less restrictive than those for linear regressions that are also more sensitive to outlying values (e.g. few excessively large changes).

Selecting only those with complete data may have introduced some bias into the results. As expected, those who were older and reported lesser FEV₁ values at baseline were more likely to drop out. Baseline age and FEV₁ mean differences between those who had complete longitudinal data (compliers) and those who had not (drop-outs) were of ∼3 yrs and 126 mL, respectively, and statistically significant. However, making the assumption that the drop-outs who were older might have had a slight increase of BMI and a large decline of FEV₁, as was observed in the compliers, the bias would be in the conservative direction.

The results presented here from a large general population sample point out that the increase in body mass index value leads to a worsening of ventilatory function, especially in males. Furthermore, it was observed that most subjects who reduced their body mass index values also improved their ventilatory function. Therefore, in agreement with the study of Morgan and Reger 32 on male workers, the current authors hypothesise that the detrimental effect of gaining weight might be reversible. This may have important clinical implications, as, for instance, overweight patients with ventilatory impairment could be routinely encouraged to lose weight to improve their lung function.

Appendix

Median regression

In usual linear regression, the expected (mean) value of the dependent variable (Y) is a linear function of a set of covariates (X): where β is the coefficient vector to be estimated. Regression coefficients are estimated by minimising the sum of the squared residuals, which are the differences between the observed values Y and expected values Xβ (least-squares estimates). To obtain estimates for coefficients and standard errors normality and heteroschedasticity of the distributions of the residuals is usually assumed.

Instead in the median linear regression, the median value of the dependent variable is a linear function of a set of covariates: Coefficients are estimated by minimising the sum of the absolute values of the residuals (L-1 estimates). In order to obtain estimates and standard errors assumptions do not have to be made as for usual regression. Indeed no parametric assumption is required. The estimates for the standard errors are obtained according to the asymptotic methods proposed by Koenker and Bassett 33. The estimates of the standard errors were also validated by bootstrap sampling procedures 34.

An example

The current authors present a simple example to show what is stated in the Methods section, that median regression techniques provide answers similar to the least-squares regression when the data are linear with normally distributed errors, but differ significantly from the least-squares fit when the errors do not satisfy the normality conditions or when the data contain significant outliers. Suppose there are six observations taken on two random variables (X, Y) as follows: (1, 10), (2, 20), (3, 40), (4, 70), (5, 80), (6, 110). In figure 2a⇑), the six observations (scatter dots), the median regression line (····) and the least-squares regression line (–) are shown. The two regression lines overlap (thus the latter is not visible in the figure), meaning that least-squares and median regressions give very similar answers. Figure 2b⇑) and c) show how the regression lines are affected by the presence of an outlier in two possible scenarios.

Suppose that the last observation was mistakenly recorded as (6, 510) instead of the correct (6, 110). The median regression is almost the same as in panel a), as if the data were correctly recorded, whereas the least-squares regression is highly changed by the presence of the outlier. Figure 2c⇑) is similar to panel b), except that the fourth observation was mistakenly recorded as (4, 510) instead of the correct (4, 70). Again the least-squares regression line is highly affected by the presence of an outlier whereas the median regression is not.