CT Image Acquisition
Study subjects underwent 2 volumetric CT scans according to a standardized technique.19 One scan was performed at full inspiration (total lung capacity, TLC), with the second at the end of a normal expiration (functional residual capacity, FRC). Tube potential was set at 120 kVp for all scans, and images were acquired with an effective mAs of 200 for inspiration and 50 for expiration.19 Scans were performed using 11 different scanner models from 1 of 3 manufacturers: General Electric Medical Systems (n=2511), Siemens (n=3879), and Philips (n=386). High-resolution scanning and reconstruction techniques were implemented to obtain thin-section, contiguous slices. To achieve near-isotropic voxels, scans were reconstructed with differing slice thicknesses of 0.625, 0.75, and 0.9 mm, depending on the parameters permitted by each scanner model, while the corresponding slice intervals were 0.625, 0.5, and 0.45 mm. The images were reconstructed using manufacturer-specific “smooth” convolution kernels: Standard, B31f, and B. Protocol adherence was routinely evaluated in all cases, and individual scan quality was visually assessed for the presence of motion artifacts, inclusion of all lung parts, as well as subjective measures of adequacy of inspiration/expiration. Scanners were calibrated on a weekly basis to maintain internal consistency, while monthly scans were collected on a standard phantom to track and verify the consistency of CT attenuation measurements.
CT images were analyzed using open-source 3D Slicer software,20 which provided automated lung segmentations and densitometric measures. The segmented lung volume on inspiratory scans provided a measure of CT-derived TLC (TLCCT), whereas the same measure on expiratory scans defined CT-derived FRC (FRCCT). Previously established prediction equations21 provided methods to determine predicted values for plethysmographic TLC and FRC for each subject. We compared the predicted plethysmographic values to TLCCT and FRCCT as an objective measure of adequacy of inspiration/expiration. The density mask technique was used with thresholds for lung attenuation set at −950, −910, and −856 HU. On inspiratory CT scans, low-attenuation areas (LAAs) were defined as voxels ≤−950 HU and expressed as percentage of TLCCT (%LAAI-950).22 Likewise, on expiratory scans, LAAs were defined as voxels ≤−856 HU, expressed as a percentage of FRCCT (%LAAE-856).22 Estimated tissue volume was also assessed on the basis of voxel attenuation values23 (Table 3). Data relating to the 15th percentile of lung attenuation (Perc15) and mean lung attenuation for inspiratory scans are provided in Supplemental Digital Content 1 (http://links.lww.com/IGC/A318).
Multiple linear regression was utilized to determine the effect of current smoking status on QCT-derived %LAAI-950 and %LAAE-856. The models controlled for age, sex, race, height, weight, FEV1/FVC, smoking history in years, and average daily cigarette consumption over the course of the subjects’ smoking history (cigarettes per day) by self-report, as well as TLCCT (%LAAI-950) or FRCCT (%LAAE-856). Additional models included corticosteroid use and time since quitting for former smokers. Further modeling was performed separately for the COPD cases and smoking controls. Scanner type associations and interaction terms for current smoking status with average cigarettes smoked per day and FEV1/FVC were also evaluated. A 2-sided Wald test was used to evaluate statistical significance, and a variable was considered to be statistically significant if its corresponding P value was <0.05. Diagnostics, including residual plots, were performed to verify the appropriateness of all fitted models. Statistical analyses were performed using JMP 10 (copyright © 2012 SAS Institute Inc.), the SAS/STAT software package, Version 9.2, of the SAS System for Windows XP (copyright © 2002–2008 SAS Institute Inc.), and the R software package (R Core Team, 2012).
Quantile normalization was used to align the %LAAI-950 distributions for current and former smokers so that the statistical properties were similar for both groups24; the normalizations were performed on subjects within each GOLD stage separately. Quantile normalization has its origins in high-dimensional genomic data analysis. The main goal of this procedure is to conform 2 or more technical replicates believed to have the same distribution. The procedure essentially works by making the quantiles equal, thus aligning the 2 distributions so that they are the same. The original purpose of this technique was to remove variation due to technical issues from 2 samples, thus making them similar. In this paper the goal was to align the distributions of %LAAI-950 for former and current smokers so that they have a similar center, spread, and shape: this removes the effect of smoking status on QCT measures.
A simpler adjustment method was implemented using the difference in mean %LAAI-950 between former and current smokers. Within each GOLD stage the mean difference was added to the scores for current smokers, whereas the former smokers’ scores remained unchanged. This method aligned the mean %LAAI-950 scores, but underlying differences in the distributions still persisted. This method simply shifted or scaled the distributions so that they would have the same center or spread but had no impact on conforming the shapes of the distributions to be similar.
Table 2 displays average values of %LAAI-950 and %LAAE-856 stratified by GOLD stage. Within each stage of COPD severity, as well as for the control smokers, %LAAI-950 and %LAAE-856 were both significantly lower for current smokers than for former smokers (all P<0.001). In the subset of 500 subjects who underwent visual scoring, current smokers were significantly more likely than former smokers to be scored for the presence of CN (P<0.001) (Table 4). Tables 5 and 6 display the results of multiple linear regression for %LAAI-950 and %LAAE-856, respectively. After adjusting for possible confounders, current smoking status was a statistically significant component in both models. In current smokers, %LAAI-950 was 3.5 percentage points lower than in former smokers (P<0.001), whereas %LAAE-856 was 6.0 percentage points lower (P<0.001). Inhaled corticosteroid use was significantly associated with %LAAI-950 but not with %LAAE-856, whereas time since quitting showed converse associations. Inhaled corticosteroid use and time since quitting did not contribute much or alter the other associations and thus were excluded from subsequent models. The apparent negative association between height and %LAA in the multivariate analyses is probably because CT lung volume, which correlates with height, was also included in the model. Adjustment for scanner model had minimal effect on the models shown here or the associations found with the other variables and hence were not included. Models for the visually scored subset adjusting for CN presence and extent are provided in Supplemental Digital Content 1, http://links.lww.com/IGC/A318.
The effect size of current smoking status was larger in the COPD cases than in the smoking controls for both outcomes, yet remained statistically significant (P<0.001) in all comparisons. For the cases, %LAAI-950 was 4.7 percentage points lower in current smokers, whereas it was just 1.1 percentage points lower in the control smokers (Table 7). Similarly, in the stratified models, %LAAE-856 was 6.6 percentage points lower in current smokers with COPD and 4.0 percentage points lower in control smokers (Table 7). Models for Perc15 and mean lung attenuation showed similar trends to those seen for %LAAI-950 and %LAAE-856, and are provided in Supplemental Digital Content 1 (http://links.lww.com/IGC/A318).
Figure 2 displays the distributions of %LAAI-950 from minimum to maximum values in each group. Original and normalized distributions are displayed in black for current smokers and in gray for former smokers. After quantile normalization, the distributions for former and current smokers are nearly identical, and the corresponding normalized mean %LAAI-950 values can be found in Table 2. Table 8 provides the percentile distributions of %LAAI-950 for current and former smokers before and after normalization, as well as after the mean difference adjustment. Quantile normalization successfully aligned the distributions, whereas the more basic mean difference adjustment resulted in added divergence. The normalized values of %LAAI-950 were used as the outcome variable for the models in Table 9. Upon application of this corrective factor, smoking status was not related to the normalized outcome in the combined multivariate model. Although it remained a significant component in models separated by COPD diagnosis, its direction of effect was reversed and the effect size diminished so that current smokers were associated with slightly higher values of %LAAI-950 than were former smokers.
In this study, we confirmed previous findings that density-based CT measures of emphysema and air trapping were consistently lower in current smokers than in former smokers, including those without COPD and those with varying degrees of COPD severity. We found that this current smoker effect persisted after adjustment for disease severity and other potential confounders but was not present in current smokers who did not have visible centrilobular nodules. After adjustment, the difference in %LAAI-950 between current and former smokers was 3.5%, and the difference for %LAAE-856 was 6.0%. When quantile normalization was performed to align the distributions of %LAAI-950 in current and former smokers, the estimate of the effect of current smoking was eliminated or reversed.
Prior research has consistently noted that current smokers display significantly lower QCT scores of “emphysema” than do former smokers.12–15 Although smaller, with 894 subjects, the cohort examined by Grydeland et al12 was similar in many ways (54% former smokers, 59% men, mean age 59.8 y) to the COPDGene cohort and yielded results with respect to the current smoker effect that were in line with those presented here (−4.86% LAAI-950 for current vs. former smokers). However, the authors suggested that the effect was most likely a result of survivor effect. In a study by Camiciottoli et al,13 using a threshold of 6.8% for LAAI-950 as a cutoff for the presence of emphysema, former smokers were nearly twice as likely to have emphysema than were current smokers (20.8% and 37.6%, respectively). This study was relatively small (n=266) and did not present specific values of LAAI-950 for their subjects. Furthermore, the study subjects were recruited from a lung cancer screening trial, and the authors attributed their findings to the nature of the study recruitment. The study presented here offers the opportunity to explore the current smoking effect in a large cohort that permits adequate adjustment for differences in disease severity and other potential confounders.
The finding that severity of “emphysema” increases after smoking cessation suggests that the smoking effect is a real phenomenon. Shaker et al14 followed up a group of 36 subjects who quit smoking after baseline CT scans. They defined emphysema as voxels ≤−910 HU (%LAAI-910) and collected 15th percentile of lung attenuation (Perc15) as well. Half of the group took inhaled corticosteroids during the study period, whereas the others received placebo. The authors found a statistically significant increase in %LAAI-910 of 2.6 percentage points, and Perc15 fell by 4.9 HU. Interestingly, the effect was driven by the treatment group. When evaluated alone, the placebo group did not display significant changes in lung density measures. However, the study size was too small to achieve a statistically significant difference between the groups and thus limits strong conclusions. A study by Ashraf et al15 displayed similar results in a group of 77 subjects who quit smoking for at least 2 years after baseline CT scans. The group had an average decrease in Perc15 of 6.2 HU after 1 year and a further decrease of 3.6 HU after the second year. Beyond 2 years, no further change was detected. In addition, 18 subjects relapsed during the study period and exhibited an increased Perc15 of 3.7 HU.
Cigarette smoking causes accumulation of inflammatory cells in the lung, particularly around respiratory bronchioles (respiratory bronchiolitis). Accumulation of this soft tissue attenuation material may result in an increase in CT attenuation within individual voxels, resulting in a relative decrease in the percentage of LAAs. The fact that current smoking was associated with a higher proportion of subjects with CN supports this possibility. This hypothesis is further supported by the findings by Shaker et al14 cited above, showing that former smokers who took inhaled corticosteroids had a larger decrease in lung density measures than did the untreated group. This suggests that the anti-inflammatory effect of the medication may have played a part in emphasizing the decreased QCT lung density. Our study supports the inflammatory hypothesis by showing that subjects taking inhaled corticosteroids had a relative increase in LAAs, but we found no significant difference in this effect between former and current smokers. The effect size of smoking status did not change significantly when corticosteroid use was added to the model, suggesting that the smoking effect was not affected by corticosteroid use. Although Langerhans cell histiocytosis may also cause CN, this is a relatively rare complication of smoking, and therefore we believe that most of the centrilobular nodules seen in our subjects were due to respiratory bronchiolitis.
The sample size of the current data set provided an opportunity to use the quantile normalization technique to eliminate the confounding role of smoking status in QCT measures. This technique has not previously been utilized in a data set of this type but offers a unique approach to this sort of correction. The normalization was performed separately in each GOLD stage for 2 reasons: to control for pulmonary function and because univariate assessments of %LAAI-950 (Table 2) showed differing effect sizes and variances between the current and former smokers within each GOLD stage. Upon normalization, each subject was assigned a new value for %LAAI-950. The new distributions of the current and former smoking groups became almost perfectly aligned.
To test the impact of the normalization, the multivariate models were developed again with the normalized %LAAI-950 value as the outcome of interest. In the combined model of all subjects, smoking status became a statistically insignificant component. Although the overall R2 of the model was not changed, the effect size was greatly reduced and the direction reversed. Meanwhile, the associations of the other variables remained relatively stable. In the separate case and control models, smoking status remained a significant component, but like the combined model the effect size was substantially reduced, and the direction of the effect was reversed. Quantile normalization may prove to be a relevant technique to adjust for smoking status in the evaluation of QCT lung densitometry. Alternative strategies would include adding a fixed number (eg, 3.5%) to the measured % of LAAs in current smokers, or incorporating the smoking status into a multivariate model.
The current study is not without limitations. Inclusion of subjects from multiple clinical centers introduces variations due to scanner differences. However, the scanner model did not affect the models shown here. Because of the cross-sectional nature of this study, we were unable to determine how long it takes for this smoking effect to manifest or to reverse its course upon smoking cessation. We did attempt to control for subjects’ time since quitting, but the association was statistically unimportant and made no meaningful impact on other model variables. Although we sought to control for severity of disease in the model, we cannot exclude the possibility that some of the apparent differences between current and former smokers might be due to a survivor effect. This paper did not analyze any zonal or lobar difference in lung attenuation, but this might be a topic for future research.
The quantile normalization technique used here controlled for COPD severity based on the GOLD stage but did not account for other variables that may be interrelated. Also, as the procedure requires paired data, we were obliged to exclude data for >1400 subjects who could not be paired. Those excluded subjects were disproportionately represented in the GOLD 3 and 4 categories, as there were fewer current smokers in those groups to be paired with former smokers. However, this exclusion actually helped provide a cohort that was seemingly more representative of the general COPD population. The COPDGene cohort was recruited with an intentional overabundance of GOLD 3 and 4 subjects compared with the general population. Most of the study subjects evaluated here will soon be undergoing 5-year follow-up scans. This should allow us to address the effect of smoking cessation in subjects who quit smoking before follow-up.
In summary, smoking status has a substantial effect on lung densitometry measures, and thus it is important to consider this variable in subjects being evaluated by QCT. In population studies, quantile normalization may be a helpful method to adjust lung density measures to reduce or eliminate the confounding effect of smoking status.
1. Reilly J. Using computed tomographic scanning to advance understanding of chronic obstructive pulmonary disease
. Proc Am Thorac Soc. 2006;3:450–455.
2. Coxson HO, Rogers RM. Quantitative computed tomography
of chronic obstructive pulmonary disease
. Acad Radiol. 2005;12:1457–1463.
3. Xu Y, Sonka M, McLennan G, et al.. MDCT-based 3-D texture classification of emphysema
and early smoking
related lung pathologies. IEEE Trans Med Imaging. 2006;25:464–475.
4. Nakano Y, Muro S, Sakai H, et al.. Computed tomographic measurements of airway dimensions and emphysema
in smokers. Correlation with lung function. Am J Respir Crit Care Med. 2000;162:1102–1108.
5. Lee YK, Oh YM, Lee JH, et al.. Quantitative assessment of emphysema
, air trapping
, and airway thickening on computed tomography. Lung. 2008;186:157–165.
6. Matsuoka S, Kurihara Y, Yagihashi K, et al.. Quantitative assessment of peripheral airway obstruction on paired expiratory/inspiratory thin-section computed tomography in chronic obstructive pulmonary disease
. J Comput Assist Tomogr. 2007;31:384–389.
7. Miniati M, Filippi E, Falaschi F, et al.. Radiologic evaluation of emphysema
in patients with chronic obstructive pulmonary disease
. Chest radiography versus high resolution computed tomography. Am J Respir Crit Care Med. 1995;151:1359–1367.
8. Martinez FJ, Curtis JL, Sciurba F, et al.. Sex differences in severe pulmonary emphysema
. Am J Respir Crit Care Med. 2007;176:243–252.
9. Dransfield MT, Washko GR, Foreman MG, et al.. Gender differences in the severity of CT emphysema
in COPD. Chest. 2007;132:464–470.
10. Gevenois PA, Scillia P, de Maertelaer V, et al.. The effects of age, sex, lung size, and hyperinflation on CT lung densitometry. Am J Roentgenol. 1996;167:1169–1173.
11. Camp PG, Coxson HO, Levy RD, et al.. Sex differences in emphysema
and airway disease in smokers. Chest. 2009;136:1480–1488.
12. Grydeland TB, Dirksen A, Coxson HO, et al.. Quantitative computed tomography
and airway wall thickness by sex, age and smoking
. Eur Respir J. 2009;34:858–865.
13. Camiciottoli G, Cavigli E, Grassi L, et al.. Prevalence and correlates of pulmonary emphysema
in smokers and former smokers. A densitometric study of participants in the ITALUNG trial. Eur Radiol. 2009;19:58–66.
14. Shaker SB, Stavngaard T, Laursen LC, et al.. Rapid fall in lung density following smoking
cessation in COPD. COPD. 2011;8:2–7.
15. Ashraf H, Lo P, Shaker SB, et al.. Short-term effect of changes in smoking
behaviour on emphysema
quantification by CT. Thorax. 2011;66:55–60.
16. Rabe KF, Hurd S, Anzueto A, et al.. Global strategy for the diagnosis, management, and prevention of chronic obstructive pulmonary disease
: GOLD executive summary. Am J Respir Crit Care Med. 2007;176:532–555.
17. Hankinson JL, Odencrantz JR, Fedan KB. Spirometric reference values from a sample of the general U.S. population. Am J Respir Crit Care Med. 1999;159:179–187.
18. Heyneman LE, Ward S, Lynch DA, et al.. Respiratory bronchiolitis, respiratory bronchiolitis-associated interstitial lung disease, and desquamative interstitial pneumonia: different entities or part of the spectrum of the same disease process? Am J Roentgenol. 1999;173:1617–1622.
19. Regan EA, Hokanson JE, Murphy JR, et al.. Genetic epidemiology of COPD (COPDGene) study design. COPD. 2010;7:32–43.
21. Stocks J, Quanjer PH. Reference values for residual volume, functional residual capacity and total lung capacity. ATS Workshop on Lung Volume Measurements. Official Statement of The European Respiratory Society. Eur Respir J. 1995;8:492–506.
22. Schroeder J, McKenzie A, Zach J, et al.. Relationships between airflow obstruction and quantitative CT measurements of emphysema
, air trapping
and airways in subjects with and without COPD. Am J Roentgenol. 2013;201:460–470.
23. Coxson HO, Rogers RM, Whittall KP, et al.. A quantification of the lung surface area in emphysema
using computed tomography. Am J Respir Crit Care Med. 1999;159:851–856.
24. Bolstad BM, Irizarry RA, Astrand M, et al.. A comparison of normalization methods for high density oligonucleotide array data based on variance and bias. Bioinformatics. 2003;19:185–193.
quantitative computed tomography; emphysema; air trapping; chronic obstructive pulmonary disease; smoking
Supplemental Digital Content
Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved