Investigation of the Obesity Paradox in Chronic Obstructive Pulmonary Disease, According to Smoking Status, in the United States

Abstract

An obesity paradox in chronic obstructive pulmonary disease (COPD), whereby overweight/obese individuals have improved survival, has been well-described. These studies have generally included smokers. It is unknown whether the paradox exists in individuals with COPD arising from factors other than smoking. Nonsmoking COPD is understudied yet represents some 25%–45% of the disease worldwide. To determine whether the obesity paradox differs between ever- and never-smokers with COPD, 1,723 adult participants with this condition were examined from 2 iterations of the National Health and Nutrition Examination Survey (1988–1994, 2007–2010), with mortality outcomes followed through December 2011. Using Cox proportional hazards models, adjusted for sociodemographic factors, lung function, and survey cycle, ever/never-smoking was found to modify the association between body mass index and hazard of death. Compared with normal-weight participants, overweight/obese participants had lower hazard of death among ever-smokers (for overweight, adjusted hazard ratio (aHR) = 0.56, 95% confidence interval (CI): 0.43, 0.74; for obesity, aHR = 0.66, 95% CI: 0.48, 0.92), but never-smokers did not (overweight, aHR = 1.41, 95% CI: 0.66, 3.03; obesity, aHR = 1.29, 95% CI: 0.48, 3.48). An obesity paradox appeared to be absent among never-smokers with COPD. This, to our knowledge, novel finding might be explained by pathophysiological differences between smoking-related and nonsmoking COPD or by smoking-associated methodological biases.

An obesity paradox has been described in chronic obstructive pulmonary disease (COPD). Many studies have reported an inverse association between body mass index (BMI) and all-cause mortality in COPD; overweight and obese individuals have improved survival compared with their normal-weight counterparts (1, 2). Such a finding is considered paradoxical, because elevated BMI is associated with increased mortality among the general population (3).

Of the studies that have demonstrated an obesity paradox in COPD, all have enrolled current or former smokers. This is unsurprising—smoking is a common risk factor for COPD. However, COPD can arise in never-smokers, who might account for 25%–45% of individuals with COPD worldwide (4, 5).

COPD that is not related to smoking is relatively understudied (6). Treatment and management recommendations are derived predominantly from clinical trials that have explicitly excluded these individuals. Pathophysiological differences between smoking-associated and nonsmoking COPD are plausible but are poorly characterized.

Additionally, smoking is strongly associated with weight loss and the incidence of many diseases that increase mortality. Although studies reporting an obesity paradox have generally controlled for smoking through regression adjustment, its strong effects have led to recommendations within obesity epidemiology to restrict analyses to never-smokers to eliminate biases that arise by including smokers (7, 8).

We sought to examine whether the obesity paradox is present in nonsmoking COPD. If never/ever-smoking modifies the association between elevated BMI and decreased mortality among individuals with COPD, this would 1) identify a potential phenotypic difference between smoking-associated and nonsmoking COPD and/or 2) suggest that biases introduced through smoking could partially explain an obesity paradox in COPD.

To address this question, we analyzed data from 2 National Health and Nutrition Epidemiologic Survey (NHANES) studies, large population-based epidemiologic surveys conducted by the US National Center for Health Statistics.

METHODS

Study design

Analyses were performed among a pooled cohort assembled from 2 NHANES studies, NHANES III (1988–1994) and NHANES Continuous (2007–2010) (9, 10). Participants in these studies underwent physical examination, answered detailed questionnaires, and were followed for vital status by linkage to the National Death Index (11).

Study population

Participants with COPD were included. In this study, we defined COPD as a prebronchodilator ratio of forced expiratory volume in the first second (FEV1) to forced vital capacity (FVC) below the lower limit of normal, as described by Hankinson, among individuals ≥40 years old (12, 13). A sensitivity analysis was performed using the alternate Global Initiative for Chronic Lung Disease definition of a fixed ratio below 0.70 (14). Pre- rather than postbronchodilator spirometry values were used due to absence of bronchodilator testing in NHANES III.

Safety exclusions for spirometry differed between the NHANES studies and are listed in Web Appendix 1 (available at https://academic.oup.com/aje). Participants were excluded if no spirometry was available, spirometry did not meet quality standards for interpretation (classified as not reproducible and reliable in NHANES III or American Thoracic Society grade D or below in NHANES Continuous), or the participant responded “yes” to having current asthma or pregnancy.

Exposure and covariate definitions

Categories of BMI (weight (kg)/height(m)²) were defined as follows: underweight (<18.5), normal weight (18.5–25.0), overweight (25.0–29.9), and obese (≥30.0). Additionally, BMI was parameterized as a restricted cubic spline, with 3 knots placed at default locations specified by Harrell (15). Age in years was categorized into deciles from 40 to 80, with an additional category for 80 years or older. Educational attainment was categorized as less than high school, some high school, high-school graduate, or more than high school. Percent-predicted FEV1 and FVC were derived from the equations of Hankinson et al. (16). As per previous NHANES analyses, Hispanic ethnicity was reclassified as Mexican, and an adjustment factor of 0.88 was applied to those with “other” ethnicity (17, 18). COPD severity categories were defined by Global Initiative for Chronic Obstructive Lung Disease criteria (14); moderate to severe COPD was defined by an FEV1 of <80% predicted.

Among ever-smokers, individuals were classified as current or former smokers by self-report, and groups were defined according to years of smoking, divided into tertiles. Additional covariates were extracted, including 1) presence of cardiovascular disease (yes/no), ascertained by self-reported history of heart attack, congestive heart failure, or stroke, and additionally in NHANES Continuous (additional questions available) by self-report of peripheral vascular disease, angina, and coronary artery disease; 2) history of diabetes, defined if a participant reported a physician diagnosis of diabetes outside of pregnancy; and 3) history of cancer, defined if a participant reported being told they had cancer (in NHANES III, this question was phrased as cancer other than skin cancer).

Outcome assessment

All-cause mortality was ascertained by linkage of each participant to the National Death Index, with data available to December 31, 2011, through the Linked Mortality File. Linkage was performed by the National Center for Health Statistics using a probabilistic match on the basis of multiple self-reported participant identifiers; methodological details published previously (11).

Statistical analyses

Variable distributions were examined and summary statistics calculated according to smoking status. Comparisons of participant characteristics between ever- and never-smokers were made by χ² and t test, as appropriate. Age-adjusted mortality rates across BMI and ever/never-smoking categories were standardized to the 2000 US Census by deciles of age (19).

The relative hazard of death according to BMI category was estimated via Cox proportional hazard models, with normal weight as the reference category, and BMI represented as a spline, with 22 as the reference value (representing the approximate midpoint of normal weight). The time metric was months from NHANES examination. Participants exited on occurrence of death or were administratively censored on December 31, 2011. The full model controlled for age, sex, ethnicity, highest attained educational level, FEV1 percent predicted, and NHANES cycle. Effect modification by never- or ever-smoking was examined by a Wald test on the multiplicative term for interaction between ever-smoking and BMI category.

We performed 5 sensitivity analyses. We excluded those with Hispanic or “other” ethnicity, because spirometric predictive equations were not expressly created for these groups, and we isolated the analytical population to those with moderate to severe COPD, to ensure that obstructed nonsmokers likely had physiologically significant disease. Among those participants from NHANES Continuous who completed postbronchodilator testing (n = 336), we excluded individuals whose postbronchodilator FEV1/FVC ratio was no longer below lower limit of normal. We used a fixed ratio instead of lower limit of normal to define spirometric obstruction, and we stratified ever-smokers into current and former smokers, to assess biases related to smoking cessation.

A subsidiary analysis was performed to assess for residual confounding among smokers as a potential explanation for an obesity paradox. Successive covariates plausibly associated with smoking, body weight, and mortality were introduced into the model. Additional covariates included 1) 6 strata of smoking intensity defined by current/former smoking and tertiles of smoking duration, 2) history of cardiovascular disease, 3) history of diabetes, and 4) history of cancer.

Survey weights were applied to produce population-based estimates. All analyses were performed in Stata, version 13 (StataCorp LP, College Station, Texas). Statistical significance was accepted at a 2-sided P < 0.05. Because NHANES data and the Linked Mortality File are fully deidentified and publicly available, this study was not subject to institutional review.

RESULTS

A total of 1,723 individuals with COPD, over the age of 40 years, who had complete covariate information and mortality outcomes were identified (Table 1, Web Figure 1). In total, 389 individuals were never-smokers and 1,334 were ever-smokers. There were statistically significant differences in baseline covariates between never-smokers and ever-smokers with COPD, with exception of ethnicity and BMI. More specifically, never-smokers were overall younger, more likely to be female, attained higher education, and had milder COPD.

Table 1.

Characteristics of Participants in a Study of the Obesity Paradox Among Persons With Chronic Obstructive Pulmonary Disease, National Health and Nutrition Examination Survey III (1988–1994) and Continuous (2007–2010), United States^a

Characteristic	Never-Smoker (n = 389)		Ever-Smoker (n = 1,334)		P Value
	%	Mean (SE)	%	Mean (SE)
Age, years		56 (0.8)		59 (0.4)	<0.01
Age category, years
40–49	41		25		<0.01
50–59	31		31
60–69	14		26
70–79	11		17
≥80	3		1
Male sex	45		57		0.04
Ethnicity
Non-Hispanic white	80		86		0.06
Non-Hispanic black	9		8
Mexican and/or Hispanic	7		3
Other	4		3
Educational level
Less than high school	6		9		<0.01
Some high school	7		16
High-school graduate	22		34
More than high school	65		41
Smoking intensity
Former			45
Current			55
Years smoked				33 (0.7)
BMIb		27 (0.4)		26 (0.2)	0.10
BMI category
Underweight	3		4		0.73
Normal weight	36		40
Overweight	42		37
Obese	19		20
COPD severity
Mild	66		44		<0.01
Moderate	32		47
Severe/very severe	2		9
Spirometry
FEV1, % predicted		85 (1.1)		77 (1.0)	<0.01
FVC, % predicted		101 (1.2)		97 (0.9)	<0.01
FEV1/FVC		65 (0.3)		61 (0.3)	<0.01

Abbreviations: BMI, body mass index; COPD, chronic obstructive pulmonary disease; FEV1, forced expiratory volume in 1 second; FVC, forced vital capacity; SE, standard error.

^a Percentages and point estimates are weighted.

^b Weight (kg)/height (m)². Categories of BMI: underweight (<18.5), normal weight (18.5–25.0), overweight (25.0–29.9), and obese (≥30.0).

Over 16,384 person-years of follow-up, 710 individuals died, corresponding to a crude mortality rate of 43 deaths per 1,000 person-years. Across all BMI categories, age-standardized mortality rates were higher for ever-smokers compared with never-smokers who had COPD (Table 2). The age-standardized mortality rate was 38 deaths per 1,000 person-years (95% confidence interval (CI): 35, 41).

Table 2.

Age-Adjusted Mortality Rates (per 1,000 Person-Years) Among Persons With Chronic Obstructive Pulmonary Disease, National Health and Nutrition Examination Survey III (1988–1994) and Continuous (2007–2010), United States^a

BMIb Category	Never-Smoker		Ever-Smoker
	MR	95% CI	MR	95% CI
Underweight	32.3	10.2, 54.4	68.3	45.6, 91.0
Normal weight	24.8	16.6, 33.0	44.2	38.3, 50.1
Overweight	31.6	22.8, 40.3	34.4	29.1, 39.7
Obese	24.9	9.4, 40.4	37.1	30.7, 43.4
Overall	28.5	23.1, 33.9	39.3	36.6, 42.0

Abbreviations: BMI, body mass index; CI, confidence interval; MR, mortality rate.

^a Age-adjusted to the 2000 US Census (19).

^b Weight (kg)/height (m)². Categories of BMI: underweight (<18.5), normal weight (18.5–25.0), overweight (25.0–29.9), and obese (≥30.0).

In Cox models, ever/never-smoking modified the association between BMI and hazard of death (Table 3). In the full model, compared with those of normal weight, overweight and obese individuals had a decreased hazard of death among ever-smokers with COPD (for overweight, adjusted hazard ratio (aHR) = 0.56, 95% CI: 0.43, 0.74; for obesity, aHR = 0.66, 95% CI: 0.48, 0.92), but never-smokers with COPD did not (for overweight, aHR = 1.41, 95% CI: 0.66, 3.03; for obesity, aHR = 1.29, 95% CI: 0.48, 3.48). Similarly, underweight participants had a higher hazard of death among ever-smokers with COPD (aHR = 1.73, 95% CI: 1.07, 2.82), but never-smokers did not (aHR = 1.08, 95% CI: 0.42, 2.78). The P value for interaction (Wald) was 0.01. Results were consistent with BMI parameterized as a spline, with statistical evidence of increased hazard of death at BMI of >32 among never-smokers with COPD and with no threshold identified among ever-smokers (Figure 1).

Table 3.

Adjusted Hazard Ratios for Mortality Among Persons With Chronic Obstructive Pulmonary Disease, National Health and Nutrition Examination Survey III (1988–1994) and Continuous (2007–2010), United States^a

BMIb Category	Never-Smoker			Ever-Smoker
	No.	aHR	95% CI	No.	aHR	95% CI
Underweight	11	1.08	0.42, 2.78	52	1.73	1.07, 2.82
Normal weight	143	1.00	Referent	549	1.00	Referent
Overweight	155	1.41	0.66, 3.03	483	0.56	0.43, 0.74
Obese	80	1.29	0.48, 3.48	250	0.66	0.48, 0.92

Abbreviations: aHR, adjusted hazard ratio; BMI, body mass index; CI, confidence interval; FEV1, forced expiratory volume in 1 second.

^a Adjusted for age, sex, ethnicity, highest attained educational level, FEV1 percent predicted, and National Health and Nutrition Examination Survey cycle; P for interaction (Wald) = 0.01.

^b Weight (kg)/height (m)². Categories of BMI: underweight (<18.5), normal weight (18.5–25.0), overweight (25.0–29.9), and obese (≥30.0).

Figure 1.

Hazard ratios for mortality among persons with chronic obstructive pulmonary disease among never-smokers (A) and ever-smokers (B), National Health and Nutrition Examination Survey (NHANES) III (1988–1994) and NHANES Continuous (2007–2010), United States. Body mass index (calculated as weight (kg)/height(m)²) was parameterized as a restricted cubic spline; hazard ratios for mortality with 95% confidence intervals (dashed lines) are from a Cox proportional hazards model (log scale). Analyses were adjusted for age, sex, ethnicity, highest attained educational level, forced expiratory volume in 1 second (FEV1) percent predicted, and NHANES cycle.

Sensitivity analyses that excluded individuals with “other” or Hispanic ethnicity (Web Table 1), limiting the analytical population to those with moderate to severe COPD (Web Table 2), excluding those from NHANES Continuous who were no longer obstructed postbronchodilator (Web Table 3, n = 138), and using a fixed ratio to define obstruction (Web Table 4) did not change inferences. An obesity paradox was also identified in both current and former smokers with COPD (Web Table 5).

In a subsidiary analysis among ever-smokers, the inverse association between BMI and mortality was somewhat attenuated with successive introduction of potential confounders (Web Table 6). However, even with the addition of these covariates, overweight individuals with smoking-related COPD continued to have a lower hazard of death compared with normal-weight individuals. There was loss of statistical significance for lower hazard of death for obese individuals after additional adjustment, although the hazard ratio remained consistently below one.

DISCUSSION

In a population-based US cohort of individuals with COPD, ever/never-smoking modified the association between body mass index and all-cause mortality. These results, which were robust to multiple sensitivity analyses, suggest that the obesity paradox is absent in never-smokers with COPD. Importantly, inferences were maintained when limiting analysis to individuals with moderate to severe COPD, suggesting that the finding was not driven by systematic differences in disease severity that might exist between these subtypes.

Proposed explanations for the obesity paradox in COPD cluster within 2 themes (20). In the first, normal-weight individuals are hypothesized to have a greater burden of conditions that increase mortality (e.g., “unhealthy normal”), such as a higher prevalence of peripheral arterial disease and emphysema, lower fat-free mass, poorer nutritional characteristics, and worse exercise tolerance (21–24). In the second, overweight/obese individuals are proposed to have physiological advantages that might improve survival (e.g., “healthy obese”), such as less hyperinflation, preserved fitness, and greater metabolic reserve (25, 26).

Studies of the obesity paradox in COPD have generally enrolled a substantial number of current or former smokers (1, 2). However, COPD can arise in never-smokers; population estimates suggest that these individuals might account for approximately 25%–45% of COPD worldwide (4, 5). Nonsmoking COPD is a heterogeneous disease entity with numerous risk factors, including occupational exposures, indoor air pollution, and low early-life lung function. In contrast to smoking-associated COPD, nonsmoking COPD is relatively understudied and might have differing disease trajectories and responsiveness to treatment (6). Our results suggest that there might be pathophysiological differences between smoking-associated and nonsmoking COPD that warrant further investigation.

Alternatively, the absence of an obesity paradox in nonsmoking COPD raises the possibility that the paradox could be explained by factors associated with smoking rather than COPD per se. Because smoking is associated with weight loss and with numerous diseases that increase mortality (diseases that could in turn further alter weight), biases through confounding, misclassification, and selection are likely (27). Partial attenuation of point estimates in our subsidiary analysis accounting for smoking intensity, cardiovascular disease, and diabetes suggests that residual confounding is at least somewhat operative in the current analysis. Among smokers with COPD, individuals with normal weight might be enriched for smoking-related illnesses that lead to weight loss (such as undetected malignancy, poorly controlled diabetes, or end-stage heart disease), increasing mortality compared with overweight/obese groups. As a corollary, possible differences in the paradox among former smokers compared with current smokers might reflect higher BMI as a marker of improving health (28). Further, because obese smokers in the general population are at higher risk of death (29), survival bias is introduced by studying mortality in those with prevalent COPD—obese smokers with COPD might be systematically less likely to be selected into prevalence studies because of antecedent death, thereby biasing observed mortality rates downward (30). Because both BMI and smoking are associated with COPD incidence and morbidity, relationships could become exaggerated by collider stratification (31–34). These possibilities are bolstered by parallel findings of an obesity paradox among individuals with diabetes and cardiovascular disease that disappears when restricting analyses to never-smokers with these conditions (7, 35). Mortality differences between never/ever-smokers with COPD within the same BMI strata are most pronounced among those at lower BMI, suggesting that a future focus on normal and underweight individuals according to smoking strata could help to unravel the mechanisms underlying the paradox.

As a final point of interest, the separation in the BMI-mortality association between ever- and never-smokers was increased in a sensitivity analysis restricted to those with moderate to severe obstruction. This is consistent with prior results demonstrating a stronger magnitude of the obesity paradox among individuals with more severe COPD (36). Differential misclassification of COPD severity as a result of spirometric restriction among obese individuals (restriction leading to lower FEV1 and overestimation of obstruction severity) has been proposed as an explanation (37). However, this would be expected to also attenuate the association between being overweight/obese and mortality among never-smokers with COPD when analysis is restricted to moderate to severe disease, which was not seen.

Our study has a number of strengths. Participant characterization was rigorous and standardized. Assessment of anthropometry and lung function was standardized and performed by trained technicians, and the same spirometers were used across the 2 NHANES studies. Results were unchanged using either a lower limit of normal or fixed ratio to define the study population. Pertinent comorbidities were recorded in the data set, providing the capacity to perform informative sensitivity analyses. Mortality was ascertained through the National Death Index, a nationwide, federally managed database that is highly accurate for vital status determination. Finally, analyses within a nationally representative cohort provide meaningful outcomes for the US population with COPD. Our is one of the few studies to contrast smoking-related and nonsmoking COPD from a population sample, and these results provide important insights into epidemiologic risk factors for death in these subgroups.

A few limitations are noted. Because NHANES III did not perform bronchodilator testing, some individuals within our population might represent cases of undiagnosed and uncontrolled asthma. However, there was little quantitative change in results with exclusion of those from NHANES Continuous who had reversible obstruction, and those with reported asthma were intentionally excluded. Additionally, individuals with certain medical conditions were excluded from spirometry for safety reasons; our results might therefore not be applicable for these individuals. Because of the short time duration between the newer iteration of NHANES and date of censoring, we were insufficiently powered to examine differences between these cycles. We were similarly underpowered to examine differences across age strata. Finally, we did not have information on weight at the time of COPD diagnosis, which has relevance for disease risk (33, 34). An ideal design would necessitate measurement of BMI before incident COPD, incorporate detailed assessment of nonsmoking risk factors for disease development, and include other measures of anthropometry and adiposity, such as fat-free mass index or mid-thigh muscle cross-sectional area, as better representations of body composition (38).

In summary, within a nationally representative cohort, overweight and obesity was associated with lower mortality rates when compared with normal weight among ever-smokers with COPD. This association, the obesity paradox, was not present among never-smokers with COPD. This might be explained by physiological differences between smoking-associated and nonsmoking COPD or by distortion of the BMI-mortality association through smoking-associated biases. These findings call for further investigation of causality among ever-smokers as well as additional evaluation of the pathologic differences between ever- and never-smokers with COPD.

Abbreviations

aHR

adjusted hazard ratio

BMI

body mass index

CI

confidence interval

COPD

chronic obstructive pulmonary disease

FEV1

forced expiratory volume in 1 second

FVC

forced vital capacity

NHANES

National Health and Nutrition Examination Survey.