Computer quantification of airway collapse on forced expiration to predict the presence of emphysema

Chronic Obstructive Pulmonary Disease (COPD) is characterized by airflow limitation that is not fully reversible, usually progressive and associated with an abnormal inflammatory response of the lung to noxious particles or gases, most often from cigarette smoke [1]. Being 4th leading cause of death, COPD is one of the major health challenges of the next decades, while the World Health Organization predicts that it will become the 3rd leading cause of death by 2030 [2–4]. Prevalence surveys indicate that up to almost one quarter of the adults aged 40 years and older may have mild airflow obstruction [5]. One of the challenges in such pandemic is to identify patients at risk for rapid deterioration and to develop diagnostic tools and specific treatment strategies which are directed to clinically important subgroups or particular phenotypes with poor outcome [6, 7].

Emphysema may be considered as one of the oldest phenotypes in COPD. It is characterized by the loss of lung tissue leading to breakdown of alveolar walls and subsequent airway collapse during forced expiration. It has been demonstrated that emphysema is associated with lower body mass index and bone mineral density, reduced exercise capacity, impaired quality of life and higher BODE index compared to COPD patients with similar airflow obstruction without emphysema [8–12]. Emphysema is, independently of COPD, a strong risk factor for lung cancer [13, 14] and CT screening may be most beneficial for this particular subgroup [13, 15]. Additionally, subtypes of emphysema patients may be referred for endoscopic valve placement, bullectomy or even lung volume reduction surgery. As most common variables obtained by spirometry do not accurately reflect emphysema presence, the diagnosis is currently based on CT scan of the chest, which is expensive, not routinely available and therefore not imposed every time a new diagnosis of COPD is made. Decreased variables of diffusing capacity such as D_L,CO and K_CO are often used as a surrogate marker of emphysema, for instance in the differential diagnosis with asthma, because they basically reflect impaired oxygen uptake by the loss of alveoli typically observed when the parenchyma is destroyed [16, 17]. Again, these measures need more advanced equipment restricting the analysis to respiratory physicians in secondary care. An alternative approach, however, may be found in the correct quantification of collapse during the forced expiratory phase of a flow-volume loop, the so-called ‘spirographic kink’. This kink, suddenly diminished expiratory flow, has been linked to emphysema in the past but the association has only been based on visual assessments in limited patient samples [18–20]. We therefore hypothesized that a standardized quantification of collapse may offer a more precise indication of emphysema presence, which would be of particular interest in primary care.

In the present study our objective was first to develop a mathematical model for precise computerized quantification of airway collapse on forced expiratory flow-volume loops. Secondly, we related these measures of airway collapse to CT-diagnosed emphysema in two large independent cohorts of smoking individuals.

Our study demonstrates that airway collapse observed at forced expiration can be automatically quantified by a computer model as an angle varying from 180° to 90°, correlating with the presence and severity of emphysema. Although our method was not sensitive to identify mild emphysema in the early stages of smoke-induced lung disease, an angle ≤ 131° proved to be a reliable cut-off for the positive prediction of emphysema in smoking individuals.

Emphysema is characterized by the disruption of alveolar attachments leading to loss of alveolar-airway interdependence and reduced airway tethering during breathing [29]. Upon the generation of highly positive intra-thoracic pressures during forced expiration, airway collapse occurs. Since many years it is understood that collapse represented by a spirographic kink in expiratory flow volume loop is indicative for emphysema [20]. Recent studies elegantly confirmed reduced airway diameter on CT scan with the presence of emphysema [30, 31]. To the best of our knowledge, our study is the first to validate the concept of emphysema-associated airflow collapse in a larger group of individuals comprising COPD patients of all severity stages as well as smoking controls. In our population, we found that collapse correlated well with severity of emphysema and was even better associated than measures of diffusing capacity in a multivariate approach, particularly in patients with established diagnosis of COPD based of FEV1/FVC ratio [32]. Despite the low sensitivity of our approach, the established cut-off of 131° yielded a positive predictive value of more than 95%. Revision of all 6 false positive cases revealed that 3 of them had alpha-1-antitrypsin deficiency resulting in panlobular emphysema. The latter may explain some misclassification on CT which would indicate an even higher accuracy [33, 34]. Furthermore, when combining the cut-off of collapse with a potential cut-off of 66% for Kco, sensitivity improved to 67% for a similar specificity indicating that both measures still provided additional information for the prediction of emphysema [35].

An important strength of our approach is the use of a computer model to automatically quantify collapse. Previous studies have used a beta-angle for quantification [36]. With the latter method the angular point is fixed on the flow-volume curve at 50% of FVC, with one leg through peak flow and the other leg through the X axis at the end of expiration. A major disadvantage of such technique is that early collapse after peak flow is often underestimated whereas airflow limitation at the end of expiration is overestimated. Our methodology is different by the fact that our angular point is not obligatory located on the flow-volume loop but chosen as the intersection point of best fitting regression lines representative of the data. Such approach does not only provide a more exact quantification when collapse is difficult to visualize or even absent, it also results in a quantification that closely corresponds to visual estimates when the angle is obviously present. In this context a visual cut-off of 131° is clinically useful, especially for primary care physicians who don’t have standard access to CT scan or diffusing capacity. Indeed, the identification of emphysema with spirometry in a subgroup of smoking individuals in primary care, may envisage early referral for a more extended testing and, if needed, a more rigorous follow-up. In the past, other computational methods have been used to characterise airflow limitations on forced expiration [37]. One interesting approach, is the assessment of mean transit time (MTT) as a sensitive indicator of both large and small airways obstruction [38, 39]. It is yet to be explored whether MTT is also useful in detection of emphysema at early stages and in more severe COPD [40].

As mild emphysema may also present without airflow limitation and collapse, one may hypothesize that the angle is rather representative of airway obstruction than of emphysema per se. It is obvious that the severity of emphysema is closely related to COPD severity, but the fact that within each former GOLD category patients with emphysema have significantly lower angles compared to their non-emphysematous counterparts, indicates that the angle quantifies flow limitation beyond FEV₁. From a physiological point of view, airway obstruction must present with a curvilinear decrease of flow versus volume, whereas sudden drops in flow for little volume changes are representative of collapse [41]. Although AC is sensitive to collapse and closely relates to emphysema, we do not claim that AC is a unique characteristic of emphysema, as this also not the case for decreased diffusion capacity either. Moreover, future studies with dynamic CT scans are required to differentiate with dynamic collapse of the central airways in case of tracheobronchomalacia, often occurring in severe COPD [42].

Despite the fact that we confirmed our findings in a second independent cohort, our study has some limitations. Most importantly, we used validated semi-quantitative or visual scores for the characterization of emphysema because CT scans were obtained in clinical routine with different CT equipment, different acquisitions and variable use of intravenous contrast. When using automated density measures of emphysema (defined as the percentage of voxels below −950 Hounsfield Units (HU) at inspiration), we found the expected poor relationships between AC and emphysema percentage (R² = 0.1948, p < 0.0001) [43]. Although we know that these relationships are not perfect even in larger cohorts with a uniform acquisition (e.g. in the COPD Gene study), correlations between automated scores and PFT measures usually reaches twice the value of ours [44, 45]. Surprisingly, multivariate regression still retained AC as best indicator for emphysema determined on HU, which indicate that the observed relationships with AC are independent of the methodology to measure emphysema. Another limiting factor of this study is the incapability of the algorithm to correctly compute the best fitting regression lines in all cases. Nevertheless, when taking into account badly performed maneuvers, flow fluctuation and curvilinear flow volume loops, correct computation between 180° and 90° was possible in approximately 95% of individuals. Finally, our method failed in terms of sensitivity and is therefore less useful as screening tool for the early detection of emphysema. This occurrence is inevitable due the fact that airflow limitation may be absent in patients with early emphysema, often only appearing on CT scan [46]. Whether it is clinically relevant to be diagnosed at this early stage when airflow limitation and collapse are not yet present, remains to be explored.

Taken together, our data provide strong evidence that airway collapse is one of the best lung functional correlates of visually assessed emphysema on CT scan. In primary care, detection of emphysema on spirometry may identify a population at risk for clinical deterioration and specific follow-up.