Methods

Subjects

Subjects in the current study were from the British Columbia (BC) Cancer Agency lung cancer screening programme, the BC-Lung Health Cohort.10 This sub-cohort is composed of smokers who had normal spirometry at baseline (ie, FEV₁ ⩾80% of predicted value; ratio of FEV₁ to forced vital capacity (FEV₁/FVC) ⩾70%); at least two spirometry measurements at least 6 months apart; and a baseline CT scan obtained using either a GE (GE Medical System, Milwaukee, Wisconsin, USA) or Siemens scanner (Siemens Medical Solutions; Erlangen, Germany).

Lung function

Spirometry was performed using American Thoracic Society guidelines without the administration of a bronchodilator.11 FEV₁ was expressed as a percentage of the predicted value (FEV₁%predicted) calculated using Crapo’s equations.12 FEV₁/FVC was calculated using actual values. The annual change in FEV₁%predicted (ΔFEV₁%predicted/year) was calculated for subjects with two visits as: (FEV₁%predicted at T1 − FEV₁%predicted at T0)/follow-up years. For subjects with more than two visits, ΔFEV₁%predicted/year was the slope of the regression line, in which all the available FEV₁%predicted measurements were plotted against age. A negative value of ΔFEV₁%predicted/year indicates worsening of the lung function.

CT technique

All CT scans were acquired in the volume scan mode at suspended full inspiration with the subject in the supine position. No intravenous contrast media were used. These CT scans were acquired using a GE scanner (Lightspeed Ultra, 120 kVp, 160 mAs, 1.25 mm slice thickness, standard reconstruction kernel) in 36 cases (25%) and using a Siemens scanner (Sensation 16, 120 kVp, 125 mAs, 1.0 mm slice thickness, B35f reconstruction kernel) in 107 cases (75%). These two image acquisition protocols have been shown to provide comparable CT densitometry measurements.13

Quantitative CT analysis

A quantitative analysis of the lung parenchyma was performed using custom software (EmphylxJ) as previously described.13 Briefly, the lung parenchyma was segmented from the chest wall and large central blood vessels in all CT images using a modified border tracing algorithm with a prior position knowledge algorithm. Total lung volume was calculated by summing the segmented pixel area in each slice and multiplying by the slice thickness. For each pixel, the mean CT attenuation (in Hounsfield Units, HU) was calculated and converted to density (g/ml) by adding 1000 to the HU number and dividing by 1000,14 and the lung inflation (ie, volume of gas/g of tissue) was calculated according to equation 1:15

where specific volume is the inverse of density. The density of the lung (tissue and gas) was measured from the CT scan, and the density of gas-free tissue was assumed to be 1.065 g/ml and constant for all subjects.16

The predicted total lung capacity (TLC) was obtained using the following equations (equations 2a and 2b) from Crapo et al:17

Women: Predicted TLC (ml)  =  59 × height (cm) − 4537 (2a)

Men: Predicted TLC (ml)  =  79.5 × height (cm) + 3.2 × age (years) − 7333 (2b)

The predicted lung weight was obtained by first calculating the predicted body weight using an equation described by Devine18 (equations 3a and 3b), and then substituting body weight into the prediction equation for lung weight modified from that originally provided by Shohl19 (see online supplement) (equation 4).

Women: Predicted body weight (g)  =  45500 + 905.5 × (height (cm) −152.4) (3a)

Men: Predicted body weight (g)  =  50000 + 905.5 × (height (cm) −152.4) (3b)

Predicted lung weight (g)  =  0.017 × ideal body weight (g) − 88.12 (g) (4)

Since TLC is the maximal volume of gas within the lung, dividing it by the predicted lung weight provides an indication of maximal lung inflation. Total overinflated lung volume was calculated by summing the pixel area with a lung inflation value greater than the predicted maximal lung inflation value in each slice and multiplying by the slice thickness. This overinflation volume was expressed as a percentage of the total lung volume (ie, %overinflation).

The “upper lung zone” was defined as the region above the carina and zonal predominance was calculated using equation 5:

The distribution was considered “upper zone predominant” if the result of equation 5 was >1 and was considered “diffuse” or “lower zone predominant” if it was ⩽1. A cluster analysis was used to estimate the size distribution of the overinflated areas.9 The inverse slope of the log-log relationship of the size of the clusters (number of contiguous voxels that are inflated beyond the predicted value for maximum lung inflation for that individual) versus the number of clusters of that size is the power-law exponent (D). Individuals with diffuse small clusters of overinflated lung will have a steeper slope (ie, greater D) than individuals with larger overinflated regions. The low attenuation lung area (LAA) with an x ray attenuation lower than −950 HU (%LAA(−950)) was calculated using the standard threshold approach and used to estimate “emphysema”.20 The zonal predominance and D were also calculated for %LAA(−950).

Airway wall dimensions were measured for all visible airways cut in cross section on each CT image using the “full-width at half-maximum method”.21 Airway dimensions included lumen area (Ai), lumen perimeter (Pi), airway wall area (Aaw) and wall area expressed as the percentage of the total airway area ((Aaw/Aaw + Ai)  =  WA%) and a normalised airway wall estimate: square root of Aaw at Pi of 10 mm (ie, √Aaw at Pi10) (see online supplement).22 A mean (SD) of 33.2 (3.1) airways were measured per subject.

Statistical analysis

Statistical analyses were performed using JMP software Version 7.0.1 (SAS Institute, Cary, North Carolina, USA). The primary outcome was ΔFEV₁%predicted/year and the explanatory variables were CT measurements of (1) overinflated lung (ie, %overinflation, zonal predominance and cluster analysis); (2) emphysema (ie, %LAA (−950 HU), zonal predominance and cluster analysis); and (3) airway dimensions (ie, Ai, %WA and √Aaw at Pi10). Covariates included age, sex, body mass index (BMI), current smoking status (ie, current or ex-smoker), pack years and baseline spirometry measurements (FEV₁%predicted and FEV₁/FVC). Three multivariate models were used to identify the CT variables associated with the primary outcome after adjusting for the covariates (overinflation, emphysema and airway dimensions were tested respectively in models 1, 2 and 3).

To illustrate the relationship between ΔFEV₁%predicted/year and baseline %overinflation, we divided the 143 subjects into quartiles according to baseline %overinflation (quartile 1 had the least %overinflation) and compared ΔFEV₁%predicted/year across four quartiles using the Wilcoxon test. A linear mixed effects model was used to evaluate the annual decline in FEV₁ (ml/year) for two groups (quartiles 1/2 and quartiles 3/4).23 Data were expressed as mean (SD) and p<0.05 was considered significant.

Results

Baseline characteristics

Descriptive characteristics of 143 subjects are shown in table 1 and baseline quantitative CT assessments are summarised in table 2.

Follow-up measurements of FEV₁

Seventy-two of 143 subjects (50.3%) were seen twice over 2.3 (0.8) years, 49 (34.3%) were seen three times over 2.3 (1.1) years and 22 (15.4%) were seen more than three times over 3.3 (1.4) years. The mean (SD) number of follow-up visits was 2.7 (0.8) (range 2–5) over 2.5 (1.1) years (range 0.5–6.4). ΔFEV₁%predicted/year observed over this time period averaged −2.3 (4.7)%/year (range −23.0 to +8.3%/year).

Risk factors associated with annual change in FEV₁%predicted

Table 3 shows the three multivariate models testing the association between CT measurements and ΔFEV₁%predicted/year. In model 1, %overinflation was inversely associated with ΔFEV₁%predicted/year whereas neither emphysema nor airway dimensions were associated with ΔFEV₁%predicted/year in models 2 and 3. In addition, in model 1, sex and baseline spirometry measurements were also associated with ΔFEV₁%predicted/year (male sex: −0.73, 95% CI −1.43 to −0.03, p = 0.04; FEV₁%predicted: −0.16, 95% CI −0.11 to −0.22, p<0.01; FEV₁/FVC: 0.19, 95% CI 0.36 to 0.01, p = 0.03).

Overinflation and annual change in FEV₁%predicted

There was a significant linear relationship between CT %overinflation and ΔFEV₁%predicted/year (see online supplement). The mean (SD) baseline %overinflation for quartiles 1–4 were 37.1 (1.5)%, 52.6 (0.8)%, 67.0 (1.1)% and 77.2 (0.6)%, respectively. The corresponding values of ΔFEV₁%predicted/year were −0.9 (0.6)%, −2.0 (0.7)%, −2.3 (0.7)% and −3.9 (1.0)%, and differed between quartiles 1/2 and 3/4 (p = 0.018, fig 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Comparison of annual change in percentage predicted forced expiratory volume in 1 s (ΔFEV₁%predicted/year) across the four quartiles of baseline CT overinflation volume expressed as a percentage of the total lung volume (%overinflation). Mean (SD) baseline %overinflation was 37.1 (1.5)%, 52.6 (0.8)%, 67.0 (1.1)% 77.2 (0.6)% for quartiles 1–4, respectively. Mean (SD) ΔFEV₁%predicted/year was −0.9 (0.6)%/year, −2.0 (0.7)%/year, −2.3 (0.7)%/year and −3.9 (1.0)%/year, respectively, and differed between quartiles 3/4 and 1/2 (p = 0.018).

The mean annual decline in FEV₁ was 0.047 l/year and 0.068 l/year for quartile 1/2 and quartile 3/4 from the linear mixed effect model (quartile 1/2: FEV₁ (l)  = 5.220 − 0.047 × age (years); quartile 3/4: FEV₁ (l)  =  6.25 − 0.068 × age (years)). The FEV₁ at age 45, 50, 55, 60, 65, 70 and 75 years was calculated using these equations for quartile 1/2 and quartile 3/4 and the values were superimposed onto the figure of Fletcher et al2 (fig 2).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

The line with surrounding grey band is the “normal range” (ie, mean±2SD) of forced expiratory volume in 1 s (FEV₁ in litres) observed in non-smokers by Fletcher et al.2 Open circles and triangles represent an extrapolation of the data from the current study. Open circles are data from quartile 1/2, in which smokers had less overinflation volume expressed as a percentage of the total lung volume (%overinflation) at baseline and an annual decline in FEV₁ of 0.047 l. Triangles are data from quartile 3/4, in which smokers had more %overinflation at baseline and an annual decline in FEV₁ of 0.068 l.

Discussion

The present results show a quantitative CT-based estimate of %overinflation using individual predicted maximal lung inflation is an independent predictor of rapid decline in lung function in smokers with normal baseline spirometry. The group with greater %overinflation at baseline exhibited a rate of decline in FEV₁ beyond the normal values predicted by Fletcher et al.2 These results suggest that, when the FEV₁ is normal, a quantitative structural assessment by CT can distinguish the smokers who will develop COPD from those who will not.

In the current study we used standard prediction equations for TLC and a prediction equation for lung weight to estimate maximal normal lung inflation for each individual. The mean (SD) value of predicted maximal normal inflation of the whole group was 5.9 (0.2) ml/g (range 5.3–6.5), of which the corresponding HU was −830 (5.3) (range −846 to −811), which is substantially different from the fixed cut-off value of −950 HU that is commonly used to define emphysema on CT scanning.20 Importantly, we do not claim that the minimally overinflated tissue identified by this procedure has undergone emphysematous destruction because we have no direct histological evidence. Based on pathology data, Leopold and Gough24 and McLean25 concluded that the dilation and destruction of the respiratory bronchioles that define centrilobular emphysema, the most common form of emphysema in smokers, is preceded by the disease in the terminal and preterminal bronchioles. We therefore strongly suspect that the minimal overinflation observed in this study may be caused by either minimal loss in the elastic recoil properties of the gas exchanging tissue and/or an increased resistance in terminal conducting airways due to inflammatory and tissue remodelling processes, both of which can occur before true emphysematous destruction.

The significance of the “overinflation” raised in our study is in agreement with that of the “hyperinflation” described by other investigators.26 27 Hyperinflation results from increased lung compliance due to emphysema and expiratory flow limitation, and patients may not perceive the negative results of it until an advanced stage because it develops slowly and insidiously over years.28 Ofir et al27 examined “hyperinflation” using lung function tests (total lung capacity, residual volume and functional residual volume) in patients with COPD GOLD stage I and healthy subjects. They found that patients with GOLD stage I COPD had more hyperinflation which increased as the intensity of dyspnoea increased. Casanova et al26 found that hyperinflation, independent of the BODE index, predicted mortality over a follow-up period of 34 months in 689 subjects with COPD. Although the findings that we have obtained with imaging tools are compatible with these studies of hyperinflation, further investigation is required to examine the relationship between these two measures of early disease.

Only a few studies have examined the relationship between quantitative CT measurements of the lung parenchyma and decline in lung function. Remy-Jardin et al29 examined 111 smokers and non-smokers and reported that subjects with emphysema visualised by radiologists at baseline had a more rapid decline in lung function than did those with normal CT scans. On the other hand, Parr et al30 and Stolk et al31 found no relationship between baseline CT emphysema and the subsequent decline in FEV₁ in subjects with COPD. These studies recruited many subjects who already had moderate COPD at baseline, which contrasts sharply with the present study which was specifically designed to determine whether CT might identify those smokers who had normal initial spirometry and subsequently developed COPD.

In contrast to the extent of overinflation, the size and location of the overinflated regions, as assessed by cluster analysis and zonal predominance, was less helpful in identifying “susceptible smokers”. Moreover, although Nakano et al22 showed that CT measurements of thickening and narrowing of the relatively large airways serve as a surrogate for the pathological changes in the small airways that are not measurable on CT scans, we failed to identify a relationship between the CT airway dimensions and ΔFEV₁%predicted. There are several possible reasons for this observation. Most important, the differences in airway dimensions that accompany a relatively small change in lung function are probably beyond the resolution of CT scans. Although many investigators have shown relationships between airway wall dimensions and airflow obstruction in cross-sectional studies, the range of lung function in those studies was much larger than in the present study.21 Second, quantitative histological studies have shown that statistically significant airway wall thickening does not become apparent until the later stages of disease (GOLD stages III and IV) with an FEV₁%predicted <50.32 Finally, our method for measuring airway dimensions may not be optimal for assessing subtle changes. Hasegawa et al33 used volumetric scanning to show that airflow limitation in COPD is more closely related to the dimensions of the distal smaller airways (ie, 5th and 6th generations) than those of proximal larger airways (ie, 3rd and 4th generations).

Fletcher et al2 were the first to show a relationship between the initial FEV₁ and its subsequent decline and referred to the phenomenon as the “horse racing” effect. However, Burrows et al34 observed an opposite relationship between ΔFEV₁ and initial FEV₁, such that the higher the initial FEV₁, the more negative the ΔFEV₁. They pointed out that this was due to “regression toward the mean”, a phenomenon in which subjects performing especially well on their first test show a greater decline because of a poorer performance on subsequent tests. Burrows and Stanescu6 34 also observed an association between initial FEV₁/FVC and ΔFEV₁%predicted, and suggested that FEV₁/FVC might provide a more reliable indicator of future loss in FEV₁. Our results confirm the findings of Burrows and Stanescu and extend their observations by showing that CT evidence of lung parenchymal overinflation is an independent predictor of decline in FEV₁.

Although the lack of an association between smoking intensity (ie, pack years) and decline in lung function might be a little surprising, this is consistent with other reports in the literature.35 We also think this lack of association between pack years and the decline in FEV₁ supports Fletcher’s concept that “non-susceptible smokers” remain unaffected regardless of their smoking history. The converse is also true—that “susceptible smokers” exhibit a decline in lung function independent of their smoking status.

A limitation of this study is that it was not originally designed as a study of COPD but was added on to a lung cancer screening cohort. Subjects were therefore not randomly chosen from the population. Follow-up spirometry was arranged at the time of follow-up CT scans which depended on the characteristics of the lung nodule(s) found on the initial CT scan. This means that different numbers of spirometric tests were performed at different frequencies between subjects. To overcome this limitation we used ΔFEV₁%predicted/year to normalise the difference in the follow-up period and the number of sampling points among subjects. ΔFEV₁%predicted/year also adjusted for the normal annual loss due to ageing and corrected for differences between women and men.

In summary, we conclude that the quantitative assessment of the lung inflated beyond individually predicted maximal lung inflation on initial CT scans may be able to identify the “susceptible minority of smokers” who eventually will develop COPD. Our working hypothesis is that the minimally overinflated lung contains the earliest forms of lesions that either increase peripheral airway resistance and/or increase lung compliance by initiating emphysematous destruction.