Introduction
COPD is a principal cause of death in most countries and COPD prevalence is increasing [1,2]. As none of the existing medications for COPD have been shown to modify the long-term decline in lung function, early detection and intervention is currently the best approach to reducing the burden of COPD. Among patients with COPD, those with emphysema have the lowest survival rate and the highest rate of pulmonary function decline [3]. Therefore, it is clinically important to detect emphysema during diagnostic workup of COPD patients. In patients with airflow obstruction, measurements of diffusing capacity for carbon monoxide (DLco) are regarded as reliable for distinguishing patients with emphysema from those with diseases of the airway alone (asthma or chronic bronchitis) [4]. DLco is usually decreased in patients with emphysema because of a loss of surface area of the alveolar-capillary membrane, whereas patients with asthma or chronic bronchitis generally do not have a decreased DLco[5,6,7]. Although DLco can be considered as a screening test to detect clinically unsuspected emphysema [8], the DLco test is apparently not adequately sensitive to detect emphysema because some patients with normal DLco values have shown clinically significant emphysema [9,10].
In some pulmonary laboratories, DLco levels of less than 80% of the predicted value has been defined as abnormal. However, this cutoff value has no statistical basis [11]. The American Thoracic Society (ATS) and the European Respiratory Society (ERS) have recommended that the lower 5th percentile, or 95% confidence limit, of the reference population be used as the lower limit of normal (LLN) for pulmonary function parameters including DLco [4,12]. Such cutoff values allow 5% false-positive diagnostic rates, which are considered clinically acceptable. This statistical definition of abnormality, however, is somewhat arbitrary [13]. Clinically, the lower 5th percentile may not be the best cutoff value of LLN for DLco because neither the sensitivity nor the false-negative rate, which is clinically more important in screening tests, have been considered.
We therefore attempted to validate the LLN for DLco and to determine the optimum LLN cutoff value for detecting emphysema in patients with COPD.
Materials and Methods
Study Subjects
Our study consisted of 197 COPD patients and 103 healthy adults. The COPD patients were selected from the Korean Obstructive Lung Disease (KOLD) cohort, into which patients with COPD or asthma had already been recruited from the pulmonary clinics of 11 hospitals in South Korea from August 2005 to April 2008. The inclusion criteria for the KOLD cohort have been described elsewhere [14,15]. COPD was diagnosed based on smoking history (more than 10 pack-years) and the presence of airflow limitation that was not fully reversible [post-bronchodilator forced expiratory volume in 1 s (FEV1)/forced vital capacity (FVC) <70%]. COPD patients with emphysema were defined as COPD patients in whom volumetric CT showed that the volume fraction of the lung at less than –950 Hounsfield units (HU) at full inspiration was more than 15% [16,17]. Healthy adult subjects older than 20 years were recruited. Healthy subjects were defined as those who had no smoking history, normal chest X-ray results, no previous respiratory or cardiac disease history, and no chest surgery history. This study was approved by the institutional review boards of all 11 hospitals, and each patient or subject provided written informed consent.
Pulmonary Function Tests
Spirometry was performed as recommended by the ATS/ERS using a Vmax22 instrument (SensorMedics, Yorba Linda, Calif., USA) or a PFDX machine (MedGraphics, St. Paul, Minn., USA) [18]. DLco was measured by the single-breath method using a Vmax229D instrument (SensorMedics) or a Masterlab Body(Jaeger, Würzburg, Germany), following the ATS/ERS protocol recommendations [19]. The quality control for the 2 pieces of equipment had been performed with gas-analyzer zeroing before each measurement, with a volume test every day, and also with the testing of biologic controls every month. The predicted values of DLco and DLco/alveolar volume (VA) were calculated from Park’s equation formulated using data from a healthy Korean population [20]. Estimates of the lower 5th percentile and other percentiles of DLco were calculated from predicted values on the assumption that individually measured DLco values would show a distribution close to gaussian. The ATS recommends that normal ranges should be based on calculated 5th percentiles, whereas estimates of the lower 5th percentiles based on predicted values (the –1.645σ criterion) are acceptable for parameters with distributions that are close to gaussian [12].
Computed Tomography
Volumetric CT scans were performed on all patients using 16-slice multidetector CT scanners including the Somatom Sensation 16 (Siemens Medical Systems, Erlangen, Germany), the GE Lightspeed Ultra (General Electric Healthcare, Milwaukee, Wisc., USA), and the Philips Brilliance 16 (Philips Medical Systems, Best, The Netherlands). Patients were scanned during suspended full inspiration and expiration in the supine position. CT parameters were: 16 × 0.75 mm collimation, 100 effective mA, and 140 kVp (Somatom Sensation 16); 16 × 0.625 mm, 300 mA, 140 kVp, pitch 0.938, and 0.5 sec/rotation (GE Lightspeed); and 16 × 0.75 mm, 133 mA, 140 kVp, pitch 1, and 0.75 sec/rotation (Philips 16). Each CT machine was calibrated for water using a standard phantom once per month (and after major maintenance) and for air daily. We obtained screening scans within 24 h after calibration. Acquired data were reconstructed using a standard algorithm with 0.625–0.8 mm thickness and 0.625–0.8 mm increments. Image data were stored in DICOM format. Using in-house software, images of the whole lung were extracted automatically, and the attenuation coefficient of each pixel was measured and calculated. From the CT data, the volume fraction of the lung below –950 HU (V950) was calculated automatically [14,15].
Statistical Analysis
Baseline characteristics except gender ratio were expressed as means ± SD. Univariate analysis used χ2 tests for categorical variables and one-way ANOVA for quantitative variables with Scheffe’s test as a post-hoc test for multiple comparisons. Relationships between any 2 quantitative variables were assessed using Pearson’s correlation analysis. All tests were performed employing the SPSS statistical package (SPSS version 12.0, SPSS, Chicago, Ill., USA), and values of p < 0.05 were considered statistically significant.
All measured DLco values were transformed to estimates of reference population percentiles using a standard normal table after calculating standardized residuals. Thefollowing formula was used: standardized residual = (x – µ)/σ; x = measured DLco, µ = predicted DLco, σ = residual standard deviation) [11]. To validate LLNs determined by the lower 5th percentiles of reference populations, and to compare these LLNs with LLNs obtained with reference to other percentiles, ROC curves were constructed by plotting sensitivity versus 1 – specificity (MedCalc software v. 4.2, Broekstraat, Belgium). A cutoff value providing the best combination of sensitivity and specificity was determined. Next, we compared the sensitivities and specificities of LLNs determined by several cutoff values, including the lower 5th percentile, the 80% predicted, and the best cutoff percentile, as shown by ROC curve analysis.
Results
Among the 197 COPD patients, 190 were male and 7 female. Age and FEV1 were 66.1 ± 7.2 years and 48.0 ± 15.9% predicted, respectively. We observed a moderate negative correlation between DLco (% predicted) and CT emphysema extent (V950) (fig. 1). Of the 197 COPD patients, 126 were classified as having emphysema and 71 as without emphysema. COPD patients with emphysema had significantly lower mean body mass index (BMI), FEV1, FEV1/FVC, DLco, and DLco/VA values compared with COPD patients without emphysema and with healthy subjects (table 1). There were no significant differences in mean DLco and DLco/VA values between COPD patients without emphysema and healthy subjects.
Fig. 1
Correlation between diffusing capacity (% predicted) and emphysema extent on CT (volume fraction of the lung below –950 HU at full inspiration).
Of the 126 COPD patients with emphysema, 23 had DLco values more than 80% predicted and 40 had DLco values more than the lower 5th percentile of the reference population (fig. 2).
Fig. 2
Distribution of % predicted (a) and percentile (b) values of diffusing capacity in patients with COPD (with/without emphysema on CT) and healthy subjects. Dotted lines indicate 80% predicted (a) and lower 5th percentile (b).
On ROC curve analysis, the lower 5th percentile used as the LLN for DLco had a sensitivity of 68.3% and a specificity of 84.5% to differentiate COPD patients with emphysema from COPD patients without emphysema, and a sensitivity of 68.3% and a specificity of 98.1% to differentiate COPD patients with emphysema from healthy subjects.
The lower 9th percentile was the best LLN cutoff value for DLco to differentiate COPD patients with emphysema from COPD patients without emphysema and from healthy subjects (fig. 3; table 2). The accuracy for differentiating COPD patients with emphysema from COPD patients without emphysema was 78.7% and the accuracy for differentiating COPD patients with emphysema from healthy subjects was 87.2%.
Table 2
Ability of the 2 cutoffs to differentiate COPD patients with and without emphysema and COPD patients with emphysema from healthy subjects
Fig. 3
ROC curves for percentile values of diffusing capacity used as lower limits of normal to detect COPD patients with emphysema versus COPD patients without emphysema (a: AUC = 0.84, SE = 0.03) and healthy subjects (b: AUC = 0.92, SE = 0.02).
Discussion
We have shown here that the optimum cutoff value of the LLN of DLco, used to detect emphysema among patients with COPD may not be the lower 5th percentile of the reference population. To the best of our knowledge, this is the first clinical validation of a statistically defined LLN for DLco used to detect emphysema among patients with COPD. We have previously shown that an LLN for DLco set at the lower 10th percentile was superior to an LLN defined with reference to the lower 5th percentile in detecting interstitial lung disease [21]. The results of these 2 studies indicate that statistically defining LLN for DLco as the lower 5th percentile may not be optimal for detecting emphysema or interstitial lung disease. Furthermore, these results indicate that the optimal LLN cutoff value should be determined by clinical validation, not merely statistically [22].
COPD is the only leading cause of death that is increasing in prevalence worldwide [1,2], and is widely under-diagnosed in the primary care setting [23]. Many seemingly healthy smokers were found to have emphysematous lesions on CT [24,25]. Therefore, early diagnosis of emphysema is important for clinical and epidemiologic purposes. A low DLco can detect clinically unsuspected emphysema [8,25] and is highly correlated with the degree of emphysema on lung CT scan [26,27]. DLco measurement is rapid and simple, and can be considered to be a pulmonary function test used to screen for emphysema. In recognizing the clinical importance of the DLco test, the ATS and ERS have set test performance standards [19], and have recommended that the lower 5th percentile of the reference population should be used as the LLN for DLco[4]. However, we considered that this statistically defined LLN may be of limited clinical value in detecting emphysema because this LLN does not reflect the distribution of test results in patients with emphysema. Moreover, previous studies have reported that individual patients may have significant emphysema but a normal DLco values [9,10]. Although we included only symptomatic COPD patients in the present study, the lower 5th percentile of the reference population, when used as the LLN, showed relatively low sensitivity in detecting emphysema.
We defined the emphysema group using volumetric CT, although emphysema is conventionally diagnosed using pathologic criteria such as an ‘abnormal permanent enlargement of the airspaces’ [28]. Many studies have addressed the ability of CT to accurately quantify the extent and severity of pulmonary emphysema [27,29,30]. The density mask technique has been widely used for the quantification of emphysema, with the threshold for measuring emphysematous pixels varying from under –900 HU to less than –960 HU [27]. We utilized –950 HU as the threshold for detecting emphysema because previous CT pathologic correlations have shown that the use of this criterion correlated well with diagnoses employing macroscopic/microscopic measurements [31,32]. In addition, a paired inspiratory/expiratory CT measurement study found that measurements between –900 HU and –950 HU indicated air trapping, whereas levels below –950 HU indicated emphysema [33]. Emphysema is considered present when more than 10% of pixels fall below the cutoff values of –910 or –920 HU, depending on slice thickness and the reconstruction algorithm employed [29]. We used 15% of V950, as measured by volumetric multidetector CT, as a cutoff value to distinguish between emphysema and no-emphysema patients. This cutoff value of 15% was derived from measurements of V950 in 48 healthy nonsmokers in our hospital, which found that V950 ranged from 0.15 to 13.25%, with a mean of 4.66% (unpublished data). Thus, we were confident that V950 values less than 15% were compatible with no or trivial emphysema [16,17]. We found that COPD patients with emphysema had significantly lower BMI, FEV1, FEV1/FVC, and DLco than COPD patients without emphysema and healthy adults. This was consistent with previous reports describing the characteristics of emphysematous phenotypes in COPD patients identified by high-resolution CT [16,34].
Our study has 2 limitations. Firstly, relatively few numbers of females were included. Our study result may not be generalized to female COPD patients because previous studies have suggested that gender might have an impact upon COPD manifestations [35]. Secondly, we used 3 different multidetector CT scanners produced by different manufacturers. Thus, our emphysema index values may have included some errors because different CT scanners employ distinct values of air density. However, no method correcting for differences between CT scanners has yet been established. Future studies are needed to devise a correction algorithm.
In conclusion, our findings indicate that the lower 5th percentile of the reference population may not be the best LLN cutoff value for DLco because this value had a low sensitivity for detecting COPD patients with emphysema.
Acknowledgments
The authors thank the members of the Korean Obstructive Lung Disease (KOLD) cohort study group. This study was supported by a grant (A040153) from the Korean Health 21 R&D Project, Ministry of Health, Welfare and Family Affairs, Republic of Korea and by the Asan Institute for Life Science (04-306).
Conflict of Interest Statement
J.B.S. has been an investigator in a government-sponsored study (2006–2008, Korea Science and Engineering Foundation). Y.-M.O. has been an investigator in university-sponsored studies (University of Ulsan College of Medicine) and an industry-sponsored study (AstraZeneca Korea), and has participated as a speaker in scientific meetings organized and financed by various pharmaceutical companies (GlaxoSmithKline, AstraZeneca Korea, MSD Korea, and Boehringer Ingelheim). S.-D.L. serves as a consultant to GlaxoSmithKline and has participated as a speaker in scientific meetings organized and financed by various pharmaceutical companies (GlaxoSmithKline, AstraZeneca Korea, and Boehringer Ingelheim).

Get Permission