Background

Chronic obstructive pulmonary disease (COPD) is a heterogeneous disease, including emphysema and large and small airway disease. The most recent update from the Global Initiative for Obstructive Lung Disease (GOLD) has addressed COPD subgroups [1], and there is a recognized need to better define COPD subtypes [2]. Several COPD subtypes have been shown to respond to specific treatments, including long-term oxygen for hypoxemic patients [3, 4], lung volume reduction surgery for upper lobe predominant emphysema [5], and medications including inhaled corticosteroids, azithromycin and roflumilast for frequent acute exacerbations [6–8].

Classic COPD subtypes include the “pink puffer” (underweight, emphysema, and normal resting oxygen saturation) and the “blue bloater” (overweight, chronic bronchitis, and hypoxemia) [9]. Though not appearing in the GOLD document and other management guidelines [1, 10], these subtypes persist in textbooks [11, 12]. The pink puffer – blue bloater distinction is vaguely defined and only applies to severe COPD patients, potentially limiting its clinical utility. However, the distinction between airway-predominant and emphysema-predominant COPD, which has some parallels to the pink puffer – blue bloater classification, may still have utility [13–17]. Historically, determinations about the extent of emphysema and airway disease could only be made on pathological specimens; however, chest computed tomography (CT) scans provide extensive anatomic information about COPD [18, 19]. As chest CT scans are becoming more frequently used for lung cancer screening and other indications [20, 21], these subtypes may be more easily identified in COPD patients.

The Genetic Epidemiology of COPD Study (COPDGene) has enrolled over 10,000 smokers with and without COPD across the United States [22]. The large sample size with extensive clinical data, including volumetric chest CT scans, allows for COPD subtyping. We aimed to create a simplified distinction between emphysema-predominant and presumed airway-predominant COPD based on the presence or absence of emphysema on chest CT scan. We hypothesized that these COPD subtypes would show associations with clinical characteristics and co-morbidities that have implications for the evaluation and management of patients with COPD. We assessed generalizability by replicating subtype associations in the Evaluation of COPD Longitudinally to Identify Predictive Surrogate End-points (ECLIPSE) Study [23].

Methods

Study subjects

COPDGene enrolled smokers with and without COPD at 21 clinical centers throughout the United States between 2007–2011 [22]. Subjects were self-classified non-Hispanic whites and non-Hispanic African Americans ages 45–80 with at least 10 pack-years of lifetime smoking. During the study visit, subjects underwent a limited physical examination, spirometry before and after inhaled bronchodilator, and a six-minute walk test to assess exercise capacity. Subjects completed questionnaires on respiratory disease, medical history and medications. The St. George’s Respiratory Questionnaire (SGRQ) measured disease-related quality of life [24]. Study protocols and questionnaires are available at http://www.copdgene.org. The ECLIPSE study is described in the Supplementary Methods (Additional file 1) [23]. COPDGene and ECLIPSE were approved by the institutional review boards at Partners Healthcare and all participating centers (Additional files 2 and 3). Subjects provided written informed consent.

Chest CT scans

All subjects underwent a volumetric chest CT scan performed at full inspiration and at relaxed expiration. All chest CT scans were subjected to a standard quality control procedure. Quantitative image analysis was performed using 3D SLICER [25, 26] and VIDA Pulmonary Workstation software (Vida Diagnostics, Coralville, Iowa). Emphysema was quantified by the percent of lung voxels with attenuation ≤ -950 Hounsfield Units (HU) on inspiratory scan [27]. Subjects were considered to have emphysema-predominant COPD if this value exceeded 10%, corresponding to 3 standard deviations above the mean in normal non-smokers from COPDGene [28]. In contrast, non-emphysematous COPD was defined by minimal to absent emphysema, specifically <5% (mean +1 SD in normal non-smokers). Airway disease was assessed by the wall area percent of segmental airways and as the square root wall area of a hypothetical airway with 10 mm internal perimeter (SRWA-Pi10) [19].

Statistical analysis

We used the COPDGene dataset version date 19-September-2012. The analysis was restricted to subjects with airflow obstruction, defined by FEV1/FVC <0.7 after bronchodilator, corresponding to GOLD stages 1–4 [1]. Based on expert opinion and previous COPDGene publications, we selected a set of clinical phenotypes and comorbidities to compare across the imaging subtypes (Table 1) [29–37]. Univariate comparisons used t-tests for continuous variables and chi-squared tests for binary variables. Logistic regression models were used to adjust for covariates including age, sex, race, body mass index, current smoking status, lifetime pack-years of smoking, and severity of airflow obstruction, assessed as post-bronchodilator FEV1 percent predicted. Stratified logistic regression models were adjusted for the same covariates excluding the quantitative of categorical variable that was used to define the strata (e.g. models in obese and non-obese subjects were not adjusted for BMI). Analyses were performed using R statistical software. Logistic regression analysis for diabetes in ECLIPSE is described in the Additional file 1.

Table 1 COPD characteristics and comorbidity definitions
Full size table

Results

COPDGene enrolled 10,300 subjects. There were 4197 current and former smokers with airflow obstruction (FEV1/FVC < 0.7 on post-bronchodilator spirometry) and inspiratory chest CT scans passing quality control included in this analysis. Figure 1 demonstrates that the classic COPD subtypes of pink puffers and blue bloaters increased in frequency in more severe COPD, yet these subjects remained infrequent even among subjects with the most severe airflow obstruction (see Table 1 for definitions). Blue bloaters were less common than pink puffers across all level of lung function impairment.

Figure 1
figure 1

Frequencies of the “pink puffer” and “blue bloater” subtypes of COPD by severity of airflow obstruction. See Methods for definitions.

Full size image

Of the 4197 COPD subjects, 1687 subjects were defined as emphysema-predominant and 1817 as non-emphysematous (Table 2, Figure 2). 693 subjects had emphysema values between 5–10% and could not be classified into one of the two categories (Additional file 4: Table S1). Emphysema-predominant subjects had more severe COPD, based on lower lung function, reduced exercise capacity on 6-minute walk test, more severe dyspnea, and reduced quality of life (higher scores on SGRQ). Non-emphysematous subjects had greater bronchodilator responsiveness expressed as the absolute change in FEV1 and as the change in FEV1 as a percent of predicted FEV1, while emphysema-predominant subjects had greater change expressed as a percent of baseline FEV1. Non-emphysematous subjects had thicker airways. The characteristics of the unclassified subjects with 5–10% emphysema generally fell between the non-emphysematous and emphysema predominant subjects (Additional file 4: Table S1).

Table 2 Demographic and clinical attributes of subjects with non-emphysematous and emphysema-predominant COPD (GOLD 1–4)
Full size table
Figure 2
figure 2

Chest CT scans from COPDGene subjects demonstrating emphysema-predominant and non-emphysematous COPD. (A) Emphysema-predominant: FEV1 55.8% predicted, 29.0% emphysema. (B) Non-emphysematous: FEV1 55.3% predicted, 4.2% emphysema.

Full size image

Table 3 shows the associations of the clinical characteristics and co-morbidities with the imaging subtypes. The COPD-asthma overlap group is more common in non-emphysematous COPD, as is bronchodilator responsiveness, when assessed using the American Thoracic Society/European Respiratory Society criteria [39]. Underweight subjects are more common among emphysema-predominant COPD. There was no difference in the prevalence of chronic bronchitis between the two imaging subtypes. Cardiovascular disease by various definitions, though statistically different between the imaging subtypes in the multivariate regression models, was not seen more frequently in non-emphysematous subjects, with the possible exception of congestive heart failure (Table 3). Both diabetes mellitus and the metabolic syndrome are significantly more common in non-emphysematous COPD (diabetes OR 2.13, p < 0.0001; metabolic syndrome OR 1.87, p < 0.0001). Using a stricter diabetes definition of self-report plus medication use (see methods in Additional file 1), there was a significant association with non-emphysematous COPD (OR 2.45, p < 0.0001). The associations with self-reported diabetes and the metabolic syndrome were similar in analyses restricted to subjects with FEV1 < 80% predicted (data not shown).

Table 3 Associations of non-emphysematous and emphysema-predominant COPD with other clinical subgroups and comorbidities
Full size table

To ensure that the associations with diabetes and the metabolic syndrome were not due to confounding, we performed logistic regression analyses stratified by various potential confounding factors (Table 4 and Additional file 4: Table S2). The increased frequency of diabetes in non-emphysematous COPD persisted in stratified analyses of obese and non-obese subjects, current and former smokers, mild-moderate (GOLD 1–2) and severe-very severe COPD (GOLD 3–4), non-Hispanic white and African American subjects, and older and younger subjects (Table 4). Odds ratios were similar or increased in the stratified analyses using the stricter diabetes definition of self-report plus medication (data not shown). The associations with metabolic syndrome persisted in stratified analyses (Table S2).

Table 4 Stratified logistic regression analyses for diabetes
Full size table

In ECLIPSE, there were 283 non-emphysema predominant and 1211 emphysema-predominant COPD subjects (Additional file 4: Table S3). Similar to COPDGene, emphysema-predominant subjects had more severe airflow obstruction. Diabetes was reported by 10.6% of non-emphysematous and 8.2% of emphysema-predominant subjects. A logistic regression model replicated the association between diabetes and non-emphysema predominant COPD (OR 1.62, 1-sided p-value = 0.034).

Discussion

In the COPDGene Study, we defined subtypes of emphysema-predominant and non-emphysematous COPD based on a distinction between high and low emphysema on chest CT scan. We found an increased prevalence of diabetes in non-emphysematous COPD subjects, which was confirmed in the ECLIPSE study. Subjects with emphysema-predominant COPD had more severe airflow obstruction, but the proportions of subjects with frequent exacerbations or exercise intolerance were similar in the two subgroups when adjusted for differences in lung function. We confirmed expected associations of asthma-COPD overlap syndrome with non-emphysematous COPD and low BMI with emphysema-predominant COPD. The classic pink puffer – blue bloater subtypes were found in low frequencies in a modern COPD population. There was no difference in the rate of chronic bronchitis across the two subtypes, highlighting another shortcoming of this classic dichotomy. The pink puffer – blue bloater distinction is unlikely to be useful and should be abandoned.

The heterogeneous nature of COPD is increasingly recognized [2]. Previous efforts to identify subtypes of COPD have relied on statistical approaches such as cluster analysis and have identified clusters with emphysema and airway-predominant disease [40–42]. Garcia-Aymerich and colleagues found a cluster of subjects with milder COPD, obesity, diabetes, and cardiovascular disease, who had less emphysema in those subjects with CT scans [43]. These statistical approaches are promising, but they often require a large number of input variables and the output may be difficult to interpret. In our analysis, the simple distinction based on the presence or absence of emphysema yielded similar results and may be easier to apply clinically than the cluster-based methods.

Diabetes and the metabolic syndrome are frequent comorbidities in COPD patients [44–47]. COPD, diabetes, and the metabolic syndrome are all related to systemic inflammation, which may explain the co-occurrence. Studies examining whether inhaled corticosteroids increase risk of diabetes have shown conflicting results [48, 49]. Regardless, any association between ICS and diabetes would not explain the high frequency of the other components of the metabolic syndrome in COPD.

In a subset of COPDGene subjects, Han and colleagues examined CT features associated with COPD exacerbations [33]. They found that diabetes was more frequent in subjects with airway disease, defined by airway wall measures. The current study using all COPD subjects in COPDGene extends those results, using a paucity of emphysema on CT scan as a surrogate for airway-predominant COPD. The concordant results using a simple definition of airway-predominant COPD strengthens the current findings. Inflammation in emphysema may be different from inflammation in airway disease [50], explaining the variable correlations between these COPD phenotypes and other diseases associated with systemic inflammation. Analyses within well-characterized cohorts of COPD patients including biomarkers, chest CT scans and comorbidities would be required to clarify these inflammatory phenotypes.

The observation that non-emphysematous COPD is associated with diabetes may have clinical implications. Current guidelines from the U.S. Preventive Services Task Force (USPTF) and other organizations recommend routine screening for hypertension and hyperlipidemia in adults [51]. However, recommendations for diabetes screening vary. The USPTF recommends diabetes screening only in adults with blood pressure greater than 135/80 mmHg [52]. The American Diabetes Association recommends screening all adults who are overweight and have one or more risk factors for diabetes, as well as all adults beginning at age 45 [53]. One might consider adding COPD, specifically non-emphysematous COPD, to the list of diabetes risk factors. With the increasing use of chest CT scans for screening and diagnosis, physicians will increasingly be able to identify COPD patients with and without emphysema. Since patients with chronic diseases may be undertreated for other medical conditions [54], our results serve as a reminder to physicians to screen for diabetes, hypertension, and lipid disorders in patients with non-emphysematous COPD. Conversely, clinicians should obtain spirometry in current and former smokers with diabetes and/or metabolic syndrome who present with respiratory symptoms.

Besides a heightened awareness of disease screening in patients with non-emphysematous COPD, our findings may potentially affect treatment. For example, the phosphodiesterase-4 inhibitor roflumilast is indicated in severe COPD patients with chronic bronchitis and frequent acute exacerbations [8]. Wouters et al. have shown that roflumilast improves glucose control in subjects with newly-diagnosed diabetes without COPD [55]. The effect of roflumilast on both COPD and diabetes outcomes in subjects with both diseases warrants further study. Since we found that chronic bronchitis was equally frequent in both imaging subtypes, non-emphysematous COPD may define a more specific subgroup that might benefit from roflumilast than does chronic bronchitis. Other anti-inflammatory agents may also have beneficial effects in COPD subtypes. Our analysis has several limitations. We divided subjects into emphysema-predominant and non-emphysematous COPD, as a surrogate for airway-predominant COPD [17]. The non-emphysematous subgroup did have large airway disease on chest CT scans. However, the major site of airflow limitation is in the small airways. Measurements of small airways (<2 mm) are limited due to the spatial resolution of CT scans. Additionally, the emphysema-predominant subtype may still include substantial airway disease since we did not account for CT airway measurements when defining the subtypes. In order to be clinically-relevant, we created a simple distinction on chest CT scans which does not require complex airway measurements. The quantitative CT emphysema categories served as a surrogate for visual reading of the presence or absence of emphysema, as would be found in a clinical radiology report. Standardized radiologist readings of chest CT scans were not available in COPDGene. However, this simplified distinction proved to be generalizable to a second study population which included COPD subjects from multiple countries.

In COPDGene, comorbidities were defined by subject self-report. Previous studies have shown subject report of diabetes to be reliable [56–58]. In addition, we confirmed our findings using a strict definition of diabetes including medication usage. However, we could not apply strict definitions of metabolic syndrome, which would require measurements of waist circumference and triglyceride, HDL, and glucose levels [59]. Using more precise definitions may strengthen the associations found using self-reported data. Additionally, we used cross-sectional data, so we cannot determine the relative timing of the onset of non-emphysematous COPD and diabetes and the metabolic syndrome. Longitudinal data would be needed to address causality. COPDGene and ECLIPSE included non-Hispanic white and non-Hispanic African American subjects; it is not known whether our findings are generalizable to other racial and ethnic groups.

Conclusions

Despite these limitations, we were able to create an imaging-based classification of emphysema-predominant and non-emphysematous COPD. The emphysema-predominant subjects had more severe COPD, based on measures of lung function, exercise capacity and symptoms. However, the non-emphysematous COPD subjects had an increased prevalence of diabetes and the metabolic syndrome, consistent with systemic inflammation. This finding may encourage clinicians to screen for diabetes and lipid abnormalities in COPD patients, specifically those without significant emphysema. With the increase in chest CT scans being performed for other indications, such as lung cancer screening, COPD patients will be more easily able to be assigned to these imaging subtypes. Conversely, patients with diabetes or metabolic syndrome who present with respiratory symptoms should undergo spirometry to evaluate for COPD. Future studies of existing and novel anti-inflammatory agents in COPD may find better results if targeted to COPD subjects without substantial emphysema.