Pulmonary function and sputum characteristics predict computed tomography phenotype and severity of COPD

Introduction

Chronic obstructive pulmonary disease (COPD) is diagnosed in the presence of a ratio of forced expiratory volume in 1 s (FEV₁) to forced vital capacity (FVC) <70%, and assessed for severity according to the level of expiratory airflow limitation as measured by FEV₁ [1]. Clinical conditions whose pathogenetic mechanisms and degrees of severity could substantially differ are unified under the term COPD by the spirometric detection of airflow limitation. FEV₁ alone and its ratio to FVC can be considered reliable screening tools because they reflect both airway obstruction and emphysema. However, spirometry does not provide by itself an overall evaluation of the complexity and the severity of COPD [2–6]. The concept that expiratory airflow limitation, the hallmark of COPD, could be the common result of different lung pathologic changes, such as the obstruction of the conductive airways, the destruction of the lung parenchyma or various combinations of the two, has evolved from the seminal observations of the middle decades of the twentieth century [7] to the current explosion of interest in identifying the clinical phenotypes of COPD [8]. Recently, respiratory symptoms and prevalence of exacerbations have been introduced as new dimensions for the definition and grading of COPD in individuals [9, 10].

Quantitative computed tomography (CT) studies of COPD opened a new era for the understanding of the mechanisms underlying airflow limitation. CT can provide in vivo information to assess the severity of the disease and to differentiate patients with predominantly conductive airway obstruction from those with predominantly emphysematous destruction [11–17]. Recent software developments for CT data analysis have increased the reliability of the measurements of the pathological changes occurring in COPD lungs. In general, there is evidence that quantitative CT data identify the predominant phenotype [18–22] and assess the overall severity [23–27] of COPD with considerable accuracy. However, considering the high prevalence of COPD in the general population, the cost of the examination, and the risks associated with the radiation exposure necessary to obtain a CT scan [28], it is evident that quantitative CT scan data cannot, at present, be obtained in all patients affected by COPD. Further technological improvements of CT may possibly increase its use in the future in patients, such those with COPD, in whom the age-related risk of radiation exposure could be substantially low. It remains to be ascertained whether and to what extent evaluation by quantitative CT can be related to the conventional characterisation of COPD by accurate clinical history and lung function evaluation entailing, in addition to spirometry, static lung volumes and diffusing capacity.

The aim of this study was to establish a link between quantitative CT data on lung density and airway wall thickening [22, 29] and clinical and whole pulmonary function evaluation in a series of patients affected by COPD. In particular, by using a statistical approach allowing the classification of patients by large sets of variables, avoiding a priori expectations about disease characteristics, we wanted to ascertain whether the overall severity and the predominant type of the lung pathologic changes quantitatively assessed by CT could be predicted by clinical and pulmonary function data.

Results

table 1 reports anthropometric data, smoking exposure, pulmonary function and CT data of the 100 patients of the learning set (23 of whom were female) and of the 373 patients of the testing set (75 of whom were female). CT data are reported only for patients in the learning set. Noticeably, both in the learning set and the testing set the value of FEV₁/FVC was higher than that of FEV₁/vital capacity (VC). table 2 reports the data on mMRC dyspnoea score and characteristics of cough and sputum. Patients of the learning set were significantly younger, more obstructed and hyperinflated. Sputum purulence prevailed in patients of the testing set. Cardiovascular disease and diabetes prevailed in patients of the testing set, while other common comorbidities of COPD, such as osteoporosis, obesity and depression, had the same prevalence in the two sets of patients.

Figure 1a displays the relationships between the original CT measurements (AWT-Pi10 and %LAA_-950) in the 100 patients of the learning set. The two variables are associated (r = 0.25; p<0.01), but with great dispersion of the data points. This indicates that in this series of patients conductive airway wall thickening and parenchymal destruction make independent contributions to airflow limitation. Figure 1b shows the results of the application of PCA to the original CT measurements; the information contained in AWT-Pi10 and %LAA_-950 is merged in two components (CT1 and CT2). By definition (see the online supplementary material), increasing CT1 values means that parenchymal destruction (%LLA_-950 contribution) is predominant on conductive airway obstruction (AWT-Pi10 contribution), and vice versa. Increasing CT2 values means that the overall severity is increased (combined contributions of %LLA_-950 and AWT-Pi10).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1–

a) Relationships between the original computed tomography (CT) measurements (average wall thickness of an arbitrary airway with an internal perimeter of 10 mm (AWT-Pi10) and percentage of lung area with X-ray attenuation values < -950 HU (%LAA_-950)) in the 100 patients of the learning set; b) results of the application of principal components analysis to the original CT measurements; the information contained in AWT-Pi10 and %LAA_-950 is merged into two components (CT1 and CT2).

As shown in table 3, there were significant relationships between CT1 and CT2 and lung function tests. CT1 had the highest correlation with D_LCO and CT2 with FEV₁/VC.

CT1 of the patients of the learning set progressively increased (p<0.01) from a mean value of -0.25 with chronic cough to 0.13 with occasional cough and 0.52 with absent cough. CT1 progressively increased (p<0.01) from a mean value of -0.45 with chronic purulent sputum to 0.09 with chronic nonpurulent sputum and 0.42 with absent/occasional sputum. These results are compatible with a progressive prevalence of emphysema over conductive airway obstruction with increasing CT1 values. CT2 progressively increased (p<0.01) from a mean value of -0.73 with mMRC dyspnoea score zero to -0.06 with a score of 2 and 0.64 with a score of 4. Hence, increased values of CT2 are compatible with increasing severity of COPD resulting from both emphysema and conductive airway obstruction and reflected clinically by worsening of dyspnoea.

table 4 reports the multivariate models to estimate CT1 (phenotype) and CT2 (severity) by clinical and lung function variables together with their respective regression coefficients and cross-validation predicted errors. Cross-validated R-shrinkage was not significant. The variables entered in the multivariate model to predict CT1 were D_LCO %, TLC % and purulent sputum, while FEV₁/VC, functional residual capacity (FRC %), and purulent sputum entered in the model to predict CT2. The model coefficient for purulent sputum was negative in the model predicting CT1 and positive in the model predicting CT2.

As shown in table 5, there were significant relationships between the pulmonary function variables measured in the testing set and CT1 and CT2 predicted by multivariate regression models derived from the learning set. The highest correlations were found between predicted CT1 and D_LCO % and between predicted CT2 and FRC %.

Predicted CT1 increased stepward in the testing set (p<0.01) from a mean value of -0.26 with chronic cough to 0.23 with absent cough and from a mean value of -0.41 with chronic purulent sputum to 0.30 with absent/occasional sputum. Predicted CT2 increased stepward (p<0.01) from a mean value of -0.64 with mMRC dyspnoea score zero to a mean value of 0.25 with a score of 4.

figure 2 displays the distribution of the 373 patients of the testing set, using as coordinates the values of CT1 and CT2 predicted by linearly combining the coefficients of the models developed from the learning set with the clinical and functional data of each patient of the testing set. Patients' distribution is partitioned into four subsets (A, B, C and D).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2–

The distribution of the 373 patients of the testing set, using as coordinates the values of indexes predicted by multivariate regression models (CT1 (phenotype) and CT2 (severity)) computed by linearly combining the coefficients of the models developed from the learning set with the clinical and functional data of each patient of the testing set. Patient distribution is partitioned into four subsets (A, B, C and D) by two perpendicular lines drawn from the zero values of CT1 and CT2.

table 6 shows the mean values of the pulmonary function indexes that are significantly different in each subset with respect to all the others. Reported pulmonary function tests indicate that conductive airway obstruction predominates over emphysema in patients whose predicted CT1 coordinate value is below zero (subsets A and C), while emphysema predominates in the patients of the B and D subsets. Accordingly, chronic cough and purulent sputum prevailed significantly (p<0.01) in subsets A and C, while the prevalence of severe and very severe mMRC dyspnoea scores was significantly higher (p<0.01) in the B and D subsets (see online supplementary table E1). The large number of patients (120 out of 373, 32%) of the testing set whose location in the graph is close to the predicted CT1 zero line (between -0.25 and 0.25 CT1 values) may have intermediate mechanisms of airflow limitation. Irrespective of the pathogenetic mechanism (predominant conductive airway obstruction in A and C and parenchymal destruction in B and D), the severity of COPD progressively worsens with increasing values of predicted CT2.

Discussion

The main finding of this study is that pulmonary function and sputum characteristics can be used in multivariate models to classify COPD patients according to the overall severity of the disease and the prevailing mechanism of airflow limitation as assessed by quantitative CT analysis. Results have been obtained using a data-driven statistical procedure. The models have been derived from a series of patients undergoing clinical, pulmonary function and quantitative CT scan evaluation and have been cross-validated and prospectively applied to classify in a bidimensional diagram a different larger series of patients undergoing only clinical and functional evaluation. The coordinates of the diagram (predicted CT1 (phenotype) and predicted CT2 (severity)) can be estimated by measuring FEV₁/VC, TLC, FRC and D_LCO and by determining whether the patient has chronic purulent sputum. CT parenchymal destruction and conductive airway thickening progressively prevail as the main mechanism of airflow limitation with positive or negative values of CT1, respectively. CT severity of the disease progressively increases with increasing values of CT2.

D_LCO is the functional variable accounting for the greatest variance in estimating CT1. This indicates that parenchymal destruction as assessed anatomically by reduced CT lung density and functionally by reduced diffusing capacity is an important determinant of the clinical phenotype of COPD. FEV₁/VC is the functional variable accounting for the greatest variance in estimating CT2. This indicates that airflow limitation, as assessed anatomically by both reduced CT density and increased airway wall thickness and functionally by FEV₁/VC, is an important determinant of COPD severity. FEV₁/VC instead of FEV₁/FVC is entered in the model predicting COPD severity. This indicates that the use of VC at the denominator of the ratio could increase its sensitivity and accuracy in defining the severity of expiratory airflow limitation. Functional indexes of hyperinflation (TLC and FRC) subsequently enter in both models and slightly reduce the variance left after the first step of the multivariate regression process. Among the clinical variables, only chronic purulent sputum enters in both models. However, the regression coefficient connected with sputum purulence is negative for CT1 (phenotype) and positive for CT2 (severity). This indicates that presence of chronic purulent sputum, besides being an important determinant of the COPD clinical phenotype, namely predominant chronic inflammatory obstruction of the conductive airways, is also a marker of increased COPD severity.

Hence, the position assumed in the bidimensional CT1–CT2 diagram classifies each patient in relation to the prevailing mechanism of airflow limitation and the severity of the disease. The wide distribution of patients obtained in the present study is in agreement with the notion that COPD can have a continuous and wide spectrum of presentations in terms of clinical phenotype and disease severity. The strong relationships of model-predicted CT2 with indexes of both airflow obstruction and hyperinflation measured in the testing set (table 5) indicates that this parameter does reflect the overall lung dysfunction that characterises COPD. Further convincing support for the use of CT2 as an indicator of COPD severity should be obtained by comparing this variable with other markers of disease severity such as 6-min walking test, frequency of exacerbations and long-term survival.

The complexity of COPD cannot reliably be reflected by simple spirometric determinations of FEV₁ [1] or by adding symptoms scores and exacerbation frequency to FEV₁, as has been suggested [10]. The approach presented here seems to have the advantage of predicting at the same time the predominant COPD phenotype (conductive airway obstruction with negative CT1 values, intermediate with CT1 values close to zero and parenchymal destruction with positive CT1 values) and the degree of severity (ranging from mild to very severe with CT2 values progressively increasing from negative to positive). Prediction of CT1 and CT2 can be obtained objectively by the numerical values of pulmonary function evaluation and by the answer to the question about presence or absence of chronic purulent sputum. The resulting information could possibly be used to identify disease subgroups that may have different responses to treatment, and this may accelerate progress toward the development of specifically targeted therapeutic strategies, as currently recommended [18, 35].

Numerous studies have related clinical and full pulmonary function testing with quantitative CT in COPD [2, 11–14, 18, 22–27, 36]. Published relationships, however, are not readily comparable with the results obtained in this study, since, to the best of our knowledge, only the present report combines, via a data-driven statistical procedure, quantitative CT data with overlapping relationships, such as parenchymal destruction and conductive airway thickening.

The main limitation of this paper is that the series of patients from which CT1 (phenotype) and CT2 (severity) have been derived (learning set) is relatively small to be considered representative of the whole spectrum of COPD lung structural changes and, in this context, it is probably necessary to derive predictive multivariate models from larger samples of patients which could represent more accurately the general population of COPD. However, the multivariate models derived from the present learning set were able to prospectively classify a larger testing set differing in age and COPD severity. Another limitation of the paper is that the enrolment of patients on the basis of a post-bronchodilator FEV₁/FVC <70% according to GOLD guidelines [1], instead of using the lower limit of normality as suggested by the ATS/ERS guidelines [30], could have introduced a selection bias leading to overdiagnosis of COPD in male elderly nonsmokers and underdiagnosis in young smokers of both sexes [37].

Definite strengths of this work are related to the acquisition and analysis of functional and CT scan data. In fact, all data were obtained by specifically trained personnel working under standardised acquisition rules with the same pulmonary function and CT equipment and were analysed by an unbiased statistical procedure.

In summary, the data presented provide evidence that lung structural changes measured by CT scan are associated with clinical and lung function changes that may be used for an objective classification of the predominant phenotype and the severity of COPD. If these or similar multivariate models are confirmed in larger population studies, each COPD patient, despite being the individual expression of a wide and continuous spectrum of clinical presentations, after full pulmonary function testing and a short question about sputum characteristics, could be located on a diagram whose coordinates may indicate the type and severity of the structural changes underlying the mechanism of expiratory airflow limitation.