INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The association between low weight for height and advanced disease in patients with chronic obstructive pulmonary disease (COPD) is a common clinical observation. Even though the prognostic value of nutritional status in these patients has been a subject of interest for decades, it is not clear whether the effect is independent of other prognostic factors. Studies with mortality as outcome are limited, and have been done with either small or selected populations (i.e., based on hospitalized patients or outpatients) (1). In addition, the data from some studies are insufficiently adjusted for important prognostic factors such as FEV₁ (1) and smoking (1, 2), or are of limited value because of a short follow-up (2, 3). We, therefore found it appropriate to investigate whether the findings in these studies are applicable to a large nonselected population. The aim of the present study was thus to prospectively examine whether body mass index (BMI) in subjects with COPD is a significant predictor of mortality after adjustment for other known prognostic factors. Analyses were based on a large random sample of the general population followed for an average period of 17 yr.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Population and Design

We used data from the Copenhagen City Heart Study, which has previously been described in detail (5). In summary, the population was derived from a random, age-stratified sample of 19,698 individuals aged 20 yr or more, recruited from among 90,000 people living in a defined area of Copenhagen. A total of 14,223 subjects underwent a first examination during a 25-mo period from 1976 to 1978. Exclusion of subjects with self-reported asthma or missing data for crucial variables (FEV₁, FVC, or BMI) left a total of 13,589 subjects. Among these, 2,132 subjects (15.7%) with COPD defined as FEV₁/FVC < 0.7 were identified and included in the analyses.

Variables of Interest

The variables of main interest were BMI and ventilatory function. BMI was calculated as weight/height² (kg/m²), and was categorized into four groups: underweight (less than 20 kg/m²), normal weight (20 to 24.9 kg/m²), overweight (25 to 29.9 kg/m²), and obese (30 kg/m² and above). FEV₁ and FVC were obtained by spirometry (Model N 403; Monaghan, Littleton, CO), with the spirometer calibrated daily. As a criterion for correct performance, we required at least two measurements differing by less than 5% from each other. The largest volume was used in the analyses. Predicted values of FEV₁ were based on internally derived values, generated from a subgroup of lifetime nonsmokers without diabetes mellitus, bronchial asthma, heart disease, or pulmonary symptoms, and with a daily consumption of alcohol of less than five drinks. Linear regression of age and height on FEV₁ was done separately for each gender. For each participant, the observed FEV₁ as % predicted (FEV₁%pred) was calculated. Subjects were classified into three categories according to severity of ventilatory impairment: (1) severe COPD (FEV₁%pred less than 50); (2) moderate COPD (FEV₁%pred 50 to 69); and (3) mild COPD (FEV₁%pred 70 or more). Cutoff limits were chosen according to the European Respiratory Society guidelines on COPD (6).

In addition to BMI and ventilatory function, we included the following potential risk factors: age, gender, tobacco consumption, and chronic mucus hypersecretion. The participants gave information about their smoking status as never-smokers, ex-smokers, or current smokers. Duration of smoking (years) and the amount of current tobacco consumption (grams per day) were also reported; one cigarette was considered equal to 1 g of tobacco, one cheroot to 3 g of tobacco, and one cigar to 5 g of tobacco. Smokers reported about whether or not they inhaled tobacco smoke. Three categories of current smokers were defined: (1) light smokers (1 to 14 g/d); (2) medium smokers (15 to 24 g/d); and (3) heavy smokers (more than 24 g/d). Besides being treated as a continuous variable, duration of smoking was also separated into decennials and treated as a categorical variable. Chronic mucus hypersecretion was defined as the presence of cough and sputum for at least 3 mo per year for more than 1 yr.

Follow-Up

Notification of deaths and cause of death were obtained from the Danish National Board of Health. COPD deaths were defined as deaths in which the immediate or contributory cause of death was registered through codes 490-493 of the International Classification of Diseases, 8th Revision (ICD-8). All subjects were followed until January 1, 1995 for all-cause mortality, and until January 1, 1994 for COPD mortality.

Statistical Analyses

To assess the independent contribution of BMI to mortality in COPD, we used the Cox proportional hazards model (7) with age as an underlying time-scale and delayed entry as appropriate. Analyses were adjusted for FEV₁%pred, current smoking status, inhalation, duration of smoking, and chronic mucus hypersecretion. FEV₁%pred was entered as a continuous covariate after ensuring that a linear association between FEV₁%pred and outcome did not violate the data. Using age as an underlying time-scale ensured optimal adjustment for age. To establish the influence of BMI in various stages of COPD, we conducted subsequent analyses in different strata of ventilatory function. Initial analyses were done separately by gender, but as the association between BMI and both all-cause and COPD mortality did not differ between men and women, our final analyses were done on the entire sample stratified by gender, thus assuming the same effect of covariates in men and women but allowing for different baseline hazards. We tested the validity of this assumption by including interaction terms between gender and each individual covariate, using the likelihood ratio test.

Data were analyzed with the STATA software system (Stata Corporation, College Station, TX) (8). The results of regression analyses are given in terms of estimated relative risks (RRs) (hazard ratios), with corresponding 95% confidence intervals (CIs).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Baseline characteristics of the 2,132 subjects with COPD (FEV₁/ FVC < 0.7) at study entry are shown in Table 1. At the initial examination, the men were significantly older and had a higher mean BMI than the women. A considerably larger proportion of the women were never-smokers, and among current smokers, the men had higher exposure in terms of quantity smoked daily and duration of smoking.

All-Cause Mortality

The results of the initial regression analysis of all-cause mortality are given in the upper half of Table 2. There was an independent effect of BMI on survival, with significantly higher mortality seen in underweight subjects than in those of normal weight. This was present independently of sex, with RRs of 1.64 (95% CI: 1.20 to 2.23) in men and 1.42 (95% CI: 1.07 to 1.89) in women. Parameter estimates were similar in men and women, and subsequent analysis was performed on the pooled data. In this analysis, association between BMI and mortality was tested in the three different strata of ventilatory function, as shown in the upper half of Table 3. An interaction term between BMI and these three strata of FEV₁%pred was found to be significant, indicating that the effect of BMI on mortality depended upon stage of COPD. A significant effect of BMI on all-cause mortality was present only in subjects with severe COPD (FEV₁%pred < 50), in whom mortality was lowest in the obese and increased with decreasing BMI (test for trend: p < 0.001). In subjects with mild or moderate COPD, the associations between BMI and mortality did not reach significance, but the relation tended to be U-shaped. The minimum point of risk in moderate COPD seemed to be broader than in mild COPD, with tolerance of overweight, since the risk for these subjects (BMI: 25 to 29.9 kg/m²) equalled the risk for subjects of normal weight. Interaction between sex and each covariate was tested separately. The only interaction found was for tobacco smoke inhalation, for which women had a higher RR than men (RR = 2.00 ([range: 1.48 to 2.70] and 1.21 [range: 0.98 to 1.48], respectively). Graphic presentation of the relations is shown in Figure 1, in which differences in mortality risk for different stages of COPD are also taken into account in order to allow a direct comparison of the risks. Subjects of normal weight with mild COPD were used as a reference (RR = 1).

View larger version (17K): [in this window] [in a new window]   Figure 1.   All-cause mortality by BMI in subjects with mild, moderate, and severe COPD. Normal-weight subjects (BMI 20 to 24.9 kg/m2) with mild COPD were used for reference.

COPD Mortality

The estimated RRs for death from COPD are listed in the lower part of Table 2. Mortality was highest in underweight subjects, and decreased with increasing BMI in both men and women (tests for trend: p < 0.001). The impact of BMI on COPD mortality was stronger than that on all-cause mortality, with RRs between the lowest and highest BMI of 5.56 (range: 2.47 to 12.54) and 7.17 (range: 2.45 to 21.00) in men and women, respectively. Again, associations between BMI and COPD mortality were similar in men and women, with results from the pooled analysis shown in the lower half of Table 3. As in the analysis of all-cause mortality, interaction between BMI and the three strata of FEV₁%pred was significant. In all three stages of COPD the highest mortality was found in underweight subjects. In subjects with severe COPD mortality continued to decrease with increasing BMI, with an RR of 7.11 (range: 2.97 to 17.05) in underweight as compared with obese subjects. A similar but weaker association was found in subjects with mild and moderate COPD (test for trend: p = 0.06 and p = 0.05 in moderate and mild COPD, respectively). However, as indicated in Table 2, the number of deaths from COPD was limited. The associations between BMI, stage of COPD, and subsequent COPD mortality are depicted in Figure 2.

View larger version (19K): [in this window] [in a new window]   Figure 2.   COPD-related mortality by BMI in subjects with mild, moderate and severe COPD. Normal-weight subjects (BMI 20 to 24.9 kg/m2) with mild COPD were used for reference.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

This is the first study to examine the relation between BMI and mortality in subjects with COPD in a large, randomly selected population. Our main finding was an independent effect of BMI on both all-cause mortality and COPD mortality in both men and women; poor prognosis was related to underweight. Furthermore, the association between BMI and survival differed according to stage of ventilatory impairment; in subjects with severe COPD, mortality continued to decrease with increasing BMI.

A significant proportion (20.9%) of women with COPD were never-smokers. Self-reported asthmatic individuals were excluded from the study, but some of the remaining never-smokers could have had unrecognized asthma and been incorrectly classified as COPD patients. If this were the case, and if these subjects were also underweight and had a higher mortality than other subjects with similar lung function, our results could be biased. However, there were only six deaths with asthma as a contributory cause of death, and although asthmatic individuals have a higher mortality from other respiratory disease than the background population, the inclusion of subjects with unrecognized asthma is unlikely to have biased our results, on the basis of the small number of asthma-related deaths.

Several studies have been done of the relation between BMI and all-cause mortality in general population samples (9- 12). The shape of the mortality curves reported has not been quite consistent, probably in part because of selection of the populations and insufficient control of confounders. In general, however, the optimal BMI seems to be between 21 and 25 kg/m², but with considerable tolerance to overweight. In our population this seemed fitting for subjects with mild to moderate COPD, whereas our data strongly suggested that optimal BMI increased with increasing severity of COPD. Some of the differences between our study and the studies of others could have originated from the expected greater proportion of COPD deaths in our population subset with COPD than in the general population. With regard to mortality from COPD, the association between BMI and mortality was very clear, as risk of death from COPD was dramatically increased in underweight subjects. The prognostic value of BMI was particularly convincing in subjects with severe COPD, and was present across the full scale of BMI.

Previous studies of mortality in patients with COPD have yielded results compatible with ours (1). Wilson and colleagues (2), in a study of 779 men with moderate to severe COPD, described body weight for height and all-cause mortality as inversely related in three different strata of pulmonary function after adjustment for age, FEV₁, TLC, exercise capacity, and heart rate, but not for smoking. Gray-Donald and coworkers (4), in a study of 348 patients with severe COPD, found low BMI to be a strong predictor of death from respiratory causes after adjustment for age, gender, current smoking status, FEV₁, and use of home oxygen, although BMI was not found to be a significant predictor of all-cause mortality in a subgroup of nonhospitalized patients, probably because of a small sample size. When adjusted for age, gender, pulmonary function, arterial blood gases, hematocrit, use of steroids, and use of oxygen, BMI was a significant predictor of survival in 135 nonsmoking patients with severe COPD and moderate hypoxemia in a recent study by Górecka and colleagues (3). In all of these studies, patients were recruited from clinical centers or hospitals where the diagnosis of COPD had been established, whereas our population was randomly selected and presumably included subjects in whom at the initial examination COPD may have been undiagnosed. This suggests that subjects with early disease, who might not have consulted a doctor for respiratory symptoms, could have the same negative effect of low BMI as patients in pulmonary clinics or hospitals. A corollary of this observation is the need for earlier identification of subjects at risk. Prevention of weight loss in such a less disabled target population could produce a favorable cost-effectiveness of intervention, as suggested in a recent paper by Schols and associates (13). In our study a similarly strong effect of low body weight on mortality was found, and the negative effect was somewhat reversed by appropriate therapy.

Although the present study has the advantage of being based on a large, nonselected, prospectively examined population with a long duration of follow-up, it also has shortcomings. Underestimation of COPD as a primary cause of death is plausible. Potentially, this error could have been unevenly distributed among the different weight groups. If deaths in underweight individuals are more likely to be classified as COPD deaths because of the known prevalence of underweight in this disease, the result would be overestimation of the effect of BMI on deaths from COPD. Confounding by other diseases affecting both BMI and mortality is also possible. Diseases leading to changes in both pulmonary function and BMI, such as congestive heart failure with pulmonary congestion and fluid retention, could especially have been responsible for the association between BMI and mortality in the upper scale of BMI. Additionally, residual confounding by smoking cannot be excluded, although we made a strong effort to take the full effects of smoking into consideration.

The role of low BMI as a determinant of poor survival in these patients could have been due to several factors, such as respiratory muscle weakness (14), impaired gas exchange (17), and impaired immune response (20), all of which have been related to malnutrition in COPD patients. It is still possible, however, that the decline in BMI in patients with COPD is a marker of advanced disease, corresponding to a currently unknown factor or factors that are also responsible for the decline in pulmonary function and progression of the disease. Hypoxia and hypercapnia are related to the severity of COPD and have been linked to malnutrition (19). Similar associations have been found for low diffusion capacity and increased TLC (2), suggesting pathogenic factors for emphysema similar to those for weight loss. We have no information on TLC and diffusion capacity in this population, and therefore no data on the prevalence of emphysema among the study participants.

We expected to find a higher mortality rate among underweight subjects but not that overweight and even obesity should confer protection against mortality from COPD. Since the majority of deaths in subjects with severe COPD are indeed COPD related, we found a strong association with all- cause mortality in this group, whereas in subjects with mild to moderate COPD other causes of death, related negatively to overweight (i.e., cardiovascular disease), would obscure the association. However, causality should be inferred with extreme caution. There is no known pathogenic mechanism that could explain why obesity should protect against mortality in subjects with COPD. Conversely, obesity in itself contributes to low FEV₁, indicating that obese subjects classified as having severe COPD may not in fact have experienced a severe decline in ventilatory function, and may therefore have had a survival corresponding to that of individuals with less severe disease.

The independent prognostic value that we found for low BMI on mortality in COPD may not establish a genuine causal relation. To resolve this important question, studies with different designs are needed. The current knowledge from intervention studies with nutritional supplements and high-calorie diets administrated during both in- and outpatient regimens has not been sufficient to clarify the relation of BMI to mortality in COPD (16, 21). Anabolic steroids and growth hormone have been used in other intervention designs in this decade (26), but so far only one study has indicated an effect on mortality (13). Because intervention studies are difficult and expensive to conduct, the effect of weight changes over time on prognosis in COPD could be further investigated in population studies, and could thereby contribute to identifying the extent to which a low BMI contributes to increased mortality in COPD, and whether the relation is causal.

On the basis of the present study of subjects with COPD, we conclude that irrespective of stage of disease, underweight is an important independent risk factor for mortality. In mild to moderate COPD the best prognosis is found in normal-weight or overweight subjects, whereas in severe COPD overweight, and even obesity, is associated with a better survival.