Risk of Lung Cancer Associated with COPD Phenotype Based on Quantitative Image Analysis

Chronic obstructive pulmonary disease (COPD) and lung cancer share a common risk factor, cigarette smoking, and COPD is associated with a 2- to 4-fold increased risk for lung cancer independent of smoking habits (1–9), even among never smokers (10). Lung cancer risk has been shown to vary with COPD phenotypes, that is, emphysema and chronic bronchitis (3, 7, 11–15). In a meta-analysis, lung cancer was associated with a previous history of COPD [OR = 2.2; 95% confidence interval (CI), 1.7–3.0], chronic bronchitis (OR = 1.5; 95% CI, 1.3–1.8), and emphysema (OR = 2.0; 95% CI, 1.7–2.4; ref. 6). Most epidemiologic studies of COPD, however, rely on self-report of COPD phenotype and are subject to both recall bias and misclassification.

Prospective studies have evaluated the association between CT evidence of emphysema and/or spirometry-defined measures of airflow obstruction and risk of lung cancer, reducing the potential for disease misclassification. Some studies report a 2- to 4-fold increased risk of lung cancer in the presence of CT evidence of emphysema, with no or lower risks associated with airflow obstruction (5, 16–18). In a limited number of studies using quantitative image analysis of CTs (qCT), lung cancer risk was not associated with a low attenuation measure of emphysema (9, 19, 20). In the National Lung Screening Trial (NLST), qCT measures of emphysema did not improve lung cancer risk prediction above that associated with self-report of COPD (21). Although lung cancer risk was not associated with airway dimensions in the NLST, risk of lung cancer has been shown to increase with decreasing forced expiratory volume in 1 second (FEV₁) even in smokers with only minimal declines in FEV₁ (15).

COPD is a heterogeneous disease, and characteristics driving the association with lung cancer are still unclear. FEV₁ alone may not fully explain the complexity of the disease (22). qCT measures of COPD can differentiate between airway-predominant and emphysema-predominant disease and may better distinguish COPD subtypes and improve estimation of lung cancer risk associated with COPD. The INHALE study evaluates alternative measures of COPD based on spirometry, quantitative image analysis more extensive than that evaluated in NLST and radiologist interpretation of low dose CT, and associations with lung cancer risk.

The INHALE study began enrollment in May 2012 and is ongoing. Lung cancer patients (cases) included were enrolled at the Karmanos Cancer Center (Detroit, MI) and Henry Ford Health System (HFHS; Detroit, MI). Institutional Review Boards at both institutions approved this study, and informed consent was obtained from each participant. Eligible cases were 21 to 89 years of age at diagnosis, enrolled within 12 months of diagnosis, able to complete the CT scan and spirometry, and never had taken amiodarone or been diagnosed with bronchiectasis or cystic fibrosis because these conditions obscure the CT images.

The analyses presented were restricted to whites and African Americans enrolled as of July 2014 and included 341 cases who had completed an interview, 292 (86% of interviewed cases) who had completed the CT scan (288 have been quantitatively analyzed), and 317 (93% of interviewed cases) who completed a spirometry or pulmonary function test.

Controls were identified through distribution of brochures/flyers in physicians' offices and throughout the community, internet and community newspaper advertisements, recruitment at community centers, health fairs and senior expos, and through current or former smoking patients aligned with HFHS. Controls were 21 to 89 years of age, able to complete the CT scan and spirometry, carried health insurance (in the event medical follow-up was required on the basis of a clinical finding on the CT or spirometry), and never had taken amiodarone, had surgical removal of any portion of either lung, or been diagnosed with lung cancer, bronchiectasis, or cystic fibrosis.

Of eligible control participants contacted through July 2014, 752 interviews were completed, 725 (96%) CTs were performed (721 have been quantitatively analyzed), and 746 (99%) completed spirometry. Approximately 11% of potential control participants contacted were ineligible, the majority due to lack of health insurance.

Interviews were conducted by phone or in-person to collect demographics, exposure and smoking history, medical history, medication use, diet, physical activity, and family health history. Smoking data include duration, number and type of cigarettes smoked, passive smoking exposure, and hookah use. Self-report of physician diagnoses of COPD (including chronic bronchitis, emphysema, and/or COPD) was included in the medical history. Only diagnoses at least one year before the lung cancer diagnosis/interview were included in the analysis. Data were also collected on prescription and over the counter medications taken for these conditions. First-degree family history of cancer (including lung cancer) or COPD and data for more distant relatives who had been diagnosed with lung cancer were enumerated.

For COPD phenotyping, participants underwent a low dose, noncontrast chest CT scan and a spirometry or pulmonary function test. Chest CTs were taken at both full inspiration and full expiration under a protocol standardized across scanners as required by the imaging software. CT images were analyzed by board-certified radiologists for abnormalities and by VIDA Diagnostics (www.vidadiagnostics.com) for the quantification of COPD phenotype. Emphysema was quantified as the percent voxels less than −950 Hounsfield units on inspiratory scan (HU_I; qCT emphysema), while air trapping was quantified as the percent voxels less than −856 Hounsfield units on expiratory scan (HU_E; qCT air trapping). Spirometry was performed by trained technicians using the EasyOne Plus Spirometer (Medical Technologies, Inc.) in accordance with ATS guidelines. FEV₁, forced vital capacity (FVC), and FEV₁/FVC were measured. Board-certified pulmonologists at both institutions reviewed the results and classified COPD status.

Age was recorded as age at diagnosis for cases and age at the time of interview for controls. Never smokers were individuals who smoked fewer than 100 cigarettes in their lifetime, whereas ever smokers included both former and current smokers. Pack-years was calculated by multiplying the number of years smoked by the average number of cigarettes smoked per day divided by 20. Family history of lung cancer was positive if the participant reported at least one first-degree relative with a diagnosis of lung cancer. Cancer staging was based on American Joint Committee on Cancer guidelines, 7th edition, and histology was defined according to clinical assessment. FEV₁/FVC < 0.70 was considered a diagnosis of COPD based on American Thoracic Society Criteria. COPD severity was classified according to Global Initiative for Chronic Obstructive Lung Disease (GOLD) standards for airflow obstruction (23). Whole-lung qCT measurements were evaluated in three ways: dichotomized on the basis of the 90th percentile threshold values among a sample of normal, nonsmoking individuals as described in the COPDGene CT Workshop Group study (qCT emphysema was dichotomized at 4.8%, while qCT air trapping was dichotomized at 19.5%; ref. 24); above and below the median in controls; and as continuous measures.

Tests of homogeneity between cases and controls were performed using χ² tests for categorical variables or t tests for continuous variables. Cohen κ was used to measure agreement between pairs of measures to determine concordance in cases and controls separately. In these analyses, the qCT COPDGene threshold values were used to dichotomize the measures. Logistic regression was used to estimate the odds of lung cancer associated with each of the COPD measures, adjusted for age, race (in the total sample), gender and pack-years, as well as inspiratory and expiratory total lung volume in models with qCT measures. The Breslow–Day test was used to evaluate homogeneity of lung cancer-COPD ORs between whites and African Americans. Backward selection was used to obtain the subset of COPD measures that maximized prediction of lung cancer risk. The full model included COPD based on spirometry (FEV₁/FVC < 0.70), and percent emphysema and percent air trapping as both continuous and dichotomized threshold values. Covariates were included regardless of statistical significance. The final backward selection model was used to generate predicted probabilities of lung cancer for variously defined COPD phenotype measures, using the mean value for all other covariates in the model. Backward regression was repeated comparing qCT measures in the “unaffected” (nontumor) lung of cancer cases with the “worst” lung among controls.

A total of 341 lung cancer cases and 752 controls were included. Cases and controls did not differ significantly according to gender or race, but cases tended to be slightly older, have a greater pack-year history, and were more likely to have a family history of lung cancer (Table 1). Although never smoking cases were enrolled, recruitment of never smoking controls has not yet been prioritized. Among cases, the majority of primary lung tumors were adenocarcinoma (53.6%), while other non–small cell histology categories comprise 31.3% of cases; 8.9% were characterized as small-cell carcinoma.

There were a significantly higher proportion of cases as compared with controls with self-reported COPD and emphysema, radiologist-assessed emphysema, and emphysema and air trapping as measured on qCT (Table 2). Cases also tended to have more severe COPD as measured by GOLD score compared with controls. Both cases and controls underreported a COPD diagnosis at interview (32.2% and 22.8%, respectively) as compared with COPD diagnosed on spirometry (50.5% and 32.3%, respectively).

In the total sample, lung cancer risk was significantly and consistently associated with each of the COPD measures (Table 3). The odds of lung cancer were increased approximately 1.4- to 3.1-fold among those with COPD compared with those without, regardless of assessment method. Associations were also consistent within and between each racial group. Among whites, the ORs indicate a significant 2- to 3.3-fold increase in lung cancer risk associated with most measures of COPD, with the exception of self-reported COPD (OR = 1.18; 95% CI, 0.78–1.77) and radiologist-evaluated emphysema (OR = 1.42; 95% CI, 0.95–2.12). Among African Americans, odds of lung cancer were increased 2- to 3.6-fold associated with COPD measures. ORs were not significantly different between whites and African Americans for any of the COPD measures.

The two clinically well-defined and consistently reported measures, COPD defined by spirometry and the COPDGene CT Workshop Group study (24) thresholds for qCT emphysema and air trapping, were used to evaluate concordance between measures. Forty-seven percent of cases and 64% of controls had no evidence of either spirometry-based COPD or qCT emphysema (percent voxels < −950 HU_I < 4.8%; Table 4). Thirty-five percent of cases and 25% of controls were discordant for the two measures, the majority of which had spirometry-defined COPD but no qCT emphysema. Nineteen percent and 11% of the cases and controls, respectively, had positive results on both measures.

Thirty-five percent of the cases and 55% of the controls had no evidence of either COPD on spirometry or qCT air trapping (percent voxels < −856 HU_E < 19.5%; Table 4). Similar percentages of cases (33%) and controls (27%) were discordant for the two measures, and proportionality tests were not significant in the two discordant groups. Thirty-two percent and 19% of the cases and controls, respectively, had positive results on both measures. Overall, 28.7% of cases were negative for these three objective COPD measures (FEV₁/FVC ≥ 0.70, percent voxels < −856 HU_E ≥ 19.5%, percent voxels < −950 HU_I ≥ 4.8%). There was minimal agreement between self-report of COPD or emphysema and these objective measures in both cases (Cohen κ: 0.18–0.28) and controls (Cohen κ: 0.13–0.21; Supplementary Fig. S1).

As concordance between objective COPD measures was low to moderate, lung cancer risk was evaluated for associations with joint measures of COPD (Table 4). Individuals with FEV₁/FVC < 0.70 but without qCT evidence of emphysema or air trapping were at 2-fold increased risk of lung cancer. Risk was slightly higher with qCT evidence of emphysema or air trapping in the absence of spirometry findings (OR = 2.99; 95% CI, 1.39–6.40 and OR = 2.42; 95% CI, 1.46–3.99, respectively). Positive findings on both spirometry and either one of the qCT measures (emphysema or air trapping) were associated with statistically significant risk estimates 3.15 and 3.95, respectively.

Using backward regression modeling to identify those COPD features most significantly associated with lung cancer, both qCT air trapping (continuous) and FEV₁/FVC < 0.70 were independent and statistically significant predictors of lung cancer risk adjusting for covariates (Table 5). Odds of lung cancer increased by 4% for each 1% increase in qCT air trapping (P < 0.001), while the odds of lung cancer were 71% higher among those with FEV₁/FVC < 0.70 as compared with the individuals with FEV₁/FVC ≥ 0.70 (P = 0.002). Predicted probabilities ranged from 0.16 (for 10% qCT air trapping and FEV₁/FVC ≥ 0.70) to 0.51 (for 40% qCT air trapping and FEV₁/FVC < 0.70; Fig. 1). Similar results were obtained when analyses were based on measures from the unaffected lung in cases compared with the “worst” lung in controls (Table 5).

Lung cancer and COPD are responsible for more than 300,000 deaths each year in the United States. These diseases share a strong risk factor in smoking and COPD increases risk of lung cancer even after adjusting for the effects of smoking. The 2- to 4-fold increased risk of lung cancer associated with a history of COPD has been consistently reported (1–9). Untangling the disease processes and identifying characteristics of COPD patients at highest risk of developing lung cancer require a more detailed evaluation of COPD phenotypes. Previous associations between COPD and lung cancer have relied on self-report, emphysema as reported by radiologist, spirometry, and more recently, emphysema and airway measures based on quantitative analysis of CT images. Each of these approaches comes with limitations, including reporting bias, observer error, operator error, and expense. We evaluated alternative measures of COPD, based on spirometry, qCT measures of emphysema and air trapping, and radiologist interpretation of low dose CT, and their associations with lung cancer risk.

Distinct groups of individuals could be identified on the basis of joint occurrence of COPD subtypes defined by FEV₁/FVC and qCT measures of emphysema and air trapping. Twenty-five percent to 30% of participants were discordant for FEV₁/FVC–defined COPD and one of the qCT measures. This type of distinction was also reported in a population of current and ex-smokers involved in a lung screening program, in which 17% of subjects had obstruction based on FEV₁/FVC and no evidence of emphysema based on radiologist evaluation of the CT, 15% had no FEV₁/FVC–defined obstruction but had evidence of emphysema based on radiologist evaluation of the CT, and 27% had findings of both (25).

COPD subtype was differentially associated with lung cancer risk; spirometry and air trapping independently predicted lung cancer risk after adjustment for age, race, sex, pack-years of smoking, and total lung volume. History of COPD was underreported on interview in this population when compared with other measures, and while providing some information with regard to risk prediction, was not included in the risk model once spirometry and air trapping measures were included.

The quantitative measure of emphysema was not independently associated with lung cancer risk. These findings are consistent with recent studies suggesting that CT measures of emphysema are not associated with lung cancer independently of other COPD measures (21). In a meta-analysis including seven studies, Smith and colleagues (9) report a 3-fold (95% CI, 2.71–4.51) increased risk of lung cancer associated with visually detected emphysema, but a nonsignificant 1.16-fold (95% CI, 0.48–2.81) increased risk of lung cancer with qCT emphysema using a 5% low attenuation volume threshold. Using a threshold of 4.8% low attenuation volume or a continuous measure, we found similar associations between qCT emphysema and lung cancer risk, which were no longer statistically significant once adjusted for FEV₁/FVC and qCT air trapping.

The use of qCT allows for an automated assessment of the presence of emphysema, limiting interobserver differences, but might miss emphysema subtypes. The threshold value used was based on COPDGene data. This threshold may not be appropriate in all populations; in our population, it appears to be a conservative measure of emphysema. The use of low attenuation measures as an aggregate across both lungs also might not fully define subtypes of emphysema that are more strongly associated with lung cancer risk. Hohberger and colleagues (26) report that malignant nodules occur more often in regions containing more severe emphysema based on radiologist scoring. A more detailed investigation into regional variation in qCT scores and alternative qCT measures is warranted.

The COPDGene study provided evidence that air trapping is strongly correlated to FEV₁/FVC and provides an independent measure of airflow obstruction (27). We could not identify a published study evaluating air trapping in relation to lung cancer risk. Using the COPDGene threshold, the median, or the continuous measure, qCT air trapping was strongly associated with lung cancer risk in our study, independent of FEV₁/FVC. This finding held when comparing both whole-lung values and unaffected/worst values between cases and controls. Mets and colleagues (28) suggest that repeat measures of air trapping based on percent voxels <−856 HU_E vary in individuals with no or mild COPD. That was a very small study and reported variation of ±10 to ±15% between scans repeated 6 weeks apart. It is difficult to determine how this type of measurement error affects estimates of lung cancer risk.

This is the first study to report an association between qCT air trapping and lung cancer independent of FEV₁/FVC; qCT emphysema did not independently predict lung cancer risk. We also show no racial variation in lung cancer–COPD phenotype associations. There are some limitations to this work. FEV₁/FVC and CT testing did not always occur on the same day. Given the case–control design of this study, cases typically had their CT around the time of diagnosis. We evaluated whether time between diagnosis and CT in the cases resulted in differing distributions of qCT measures, and no differences were detected. This suggests that measures of COPD are relatively unchanged within short time windows and that the CT accurately reflects the COPD status of the cases. We also evaluated the potential for the lung in which the cancer occurred to have altered qCT measures by analyzing data from the unaffected lung in the cases compared with the “worst” lung in controls; the results did not change.

The findings presented suggest that quantitative imaging to more specifically phenotype COPD, in addition to FEV₁/FVC, can aid in the identification of high-risk groups. It is unclear whether a combination of air trapping as identified on CT plus FEV₁/FVC < 0.70 represents a distinct clinical subtype of COPD related to lung cancer risk. We also identified a relatively large subset of lung cancer cases (29%) without positive findings on three objective measures of COPD. Understanding the shared and distinct biological pathways underlying COPD and lung cancer has implications for the identification of high-risk groups for lung cancer screening and prevention, as well as providing insight into diagnostic and treatment strategies.

No potential conflicts of interest were disclosed.

Conception and design: A.G. Schwartz, A.O. Soubani, S.M. Gadgeel

Development of methodology: A.G. Schwartz, C.M. Lusk, A.O. Soubani, G. Walworth, M.J. Flynn, D.L. Spizarny

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.G. Schwartz, S. Pandolfi, L. Mantha, A.O. Soubani, A. Wozniak, C. Neslund-Dudas, A.A. Ardisana, M.J. Flynn, T. Song, D.L. Spizarny, P.A. Kvale

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.G. Schwartz, C.M. Lusk, D. Watza, S. Pandolfi, A.O. Soubani, A.A. Ardisana, D.L. Spizarny, S.M. Gadgeel

Writing, review, and/or revision of the manuscript: A.G. Schwartz, A.S. Wenzlaff, D. Watza, M.L. Cote, A.O. Soubani, A. Wozniak, C. Neslund-Dudas, A.A. Ardisana, M.J. Flynn, P.A. Kvale, R.A. Chapman, S.M. Gadgeel

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.G. Schwartz, A.S. Wenzlaff, S. Pandolfi, L. Mantha, C. Neslund-Dudas, A.A. Ardisana

Study supervision: A.G. Schwartz, S. Pandolfi, C. Neslund-Dudas

Other (primary radiologist interpreting the CT studies at the primary location and provided input during the development stage for the CT protocol): G. Walworth

Other (quality control for CT measurements): M.J. Flynn

Other (CT image interpretation): T. Song

This research was supported by NIH grants/contracts R01CA141769 (to A.G. Schwartz, S. Pandolfi, A.O. Soubani, C. Neslund-Dudas, A.A. Ardisana, M.J. Flynn, T. Song, D.L. Spizarny, P.A. Kvale, R.A. Chapman, and S.M. Gadgeel), P30CA022453 (to A.G. Schwartz), HHSN261201300011I (to A.G. Schwartz), and the Herrick Foundation (to A.G. Schwartz).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.