Chronic obstructive pulmonary disease (COPD) is characterized by airway remodeling. Characterization of airway changes on computed tomography has been challenging due to the complexity of the ...recurring branching patterns, and this can be better measured using fractal dimensions.
We analyzed segmented airway trees of 8,135 participants enrolled in the COPDGene cohort. The fractal complexity of the segmented airway tree was measured by the Airway Fractal Dimension (AFD) using the Minkowski-Bougliand box-counting dimension. We examined associations between AFD and lung function and respiratory morbidity using multivariable regression analyses. We further estimated the extent of peribronchial emphysema (%) within 5 mm of the airway tree, as this is likely to affect AFD. We classified participants into 4 groups based on median AFD, percentage of peribronchial emphysema, and estimated survival.
AFD was significantly associated with forced expiratory volume in one second (FEV1; P < 0.001) and FEV1/forced vital capacity (FEV1/FVC; P < 0.001) after adjusting for age, race, sex, smoking status, pack-years of smoking, BMI, CT emphysema, air trapping, airway thickness, and CT scanner type. On multivariable analysis, AFD was also associated with respiratory quality of life and 6-minute walk distance, as well as exacerbations, lung function decline, and mortality on longitudinal follow-up. We identified a subset of participants with AFD below the median and peribronchial emphysema above the median who had worse survival compared with participants with high AFD and low peribronchial emphysema (adjusted hazards ratio HR: 2.72; 95% CI: 2.20-3.35; P < 0.001), a substantial number of whom were not identified by traditional spirometry severity grades.
Airway fractal dimension as a measure of airway branching complexity and remodeling in smokers is associated with respiratory morbidity and lung function change, offers prognostic information additional to traditional CT measures of airway wall thickness, and can be used to estimate mortality risk.
ClinicalTrials.gov identifier: NCT00608764.
This study was supported by NIH K23 HL133438 (SPB) and the COPDGene study (NIH Grant Numbers R01 HL089897 and R01 HL089856). The COPDGene project is also supported by the COPD Foundation through contributions made to an Industry Advisory Board comprised of AstraZeneca, Boehringer Ingelheim, Novartis, Pfizer, Siemens, Sunovion and GlaxoSmithKline.
Low fat-free mass index (FFMI) is an independent risk factor for mortality in chronic obstructive pulmonary disease (COPD) not typically measured during routine care. In the present study, we aimed ...to derive fat-free mass from the pectoralis muscle area (FFM
) and assess whether low FFMI
is associated with all-cause mortality in COPD cases. We used data from two independent COPD cohorts, ECLIPSE and COPDGene.Two equal sized groups of COPD cases (n=759) from the ECLIPSE study were used to derive and validate an equation to calculate the FFM
measured using bioelectrical impedance from PMA. We then applied the equation in COPD cases (n=3121) from the COPDGene cohort, and assessed survival. Low FFMI
was defined, using the Schols classification (FFMI <16 in men, FFMI <15 in women) and the fifth percentile normative values of FFMI from the UK Biobank.The final regression model included PMA, weight, sex and height, and had an adjusted R
of 0.92 with fat-free mass (FFM) as the outcome. In the test group, the correlation between FFM
and FFM remained high (Pearson correlation=0.97). In COPDGene, COPD cases with a low FFMI
had an increased risk of death (HR 1.6, p<0.001).We demonstrated COPD cases with a low FFMI
have an increased risk of death.
Significant heterogeneity of clinical presentation and disease progression exists within chronic obstructive pulmonary disease (COPD). Although FEV(1) inadequately describes this heterogeneity, a ...clear alternative has not emerged. The goal of phenotyping is to identify patient groups with unique prognostic or therapeutic characteristics, but significant variation and confusion surrounds use of the term "phenotype" in COPD. Phenotype classically refers to any observable characteristic of an organism, and up until now, multiple disease characteristics have been termed COPD phenotypes. We, however, propose the following variation on this definition: "a single or combination of disease attributes that describe differences between individuals with COPD as they relate to clinically meaningful outcomes (symptoms, exacerbations, response to therapy, rate of disease progression, or death)." This more focused definition allows for classification of patients into distinct prognostic and therapeutic subgroups for both clinical and research purposes. Ideally, individuals sharing a unique phenotype would also ultimately be determined to have a similar underlying biologic or physiologic mechanism(s) to guide the development of therapy where possible. It follows that any proposed phenotype, whether defined by symptoms, radiography, physiology, or cellular or molecular fingerprint will require an iterative validation process in which "candidate" phenotypes are identified before their relevance to clinical outcome is determined. Although this schema represents an ideal construct, we acknowledge any phenotype may be etiologically heterogeneous and that any one individual may manifest multiple phenotypes. We have much yet to learn, but establishing a common language for future research will facilitate our understanding and management of the complexity implicit to this disease.
Ambient air pollution accelerates lung function decline among adults, however, there are limited data about its role in the development and progression of early stages of interstitial lung disease.
...To evaluate associations of long-term exposure to traffic and ambient pollutants with odds of interstitial lung abnormalities (ILA) and progression of ILA on repeated imaging.
We ascertained ILA on chest CT obtained from 2618 Framingham participants from 2008 to 2011. Among 1846 participants who also completed a cardiac CT from 2002 to 2005, we determined interval ILA progression. We assigned distance from home address to major roadway, and the 5-year average of fine particulate matter (PM
), elemental carbon (EC, a traffic-related PM
constituent) and ozone using spatio-temporal prediction models. Logistic regression models were adjusted for age, sex, body mass index, smoking status, packyears of smoking, household tobacco exposure, neighbourhood household value, primary occupation, cohort and date.
Among 2618 participants with a chest CT, 176 (6.7%) had ILA, 1361 (52.0%) had no ILA, and the remainder were indeterminate. Among 1846 with a preceding cardiac CT, 118 (6.4%) had ILA with interval progression. In adjusted logistic regression models, an IQR difference in 5-year EC exposure of 0.14 µg/m
was associated with a 1.27 (95% CI 1.04 to 1.55) times greater odds of ILA, and a 1.33 (95% CI 1.00 to 1.76) times greater odds of ILA progression. PM
and O
were not associated with ILA or ILA progression.
Exposure to EC may increase risk of progressive ILA, however, associations with other measures of ambient pollution were inconclusive.
The aim of this study is to compare two subjective methods for the identification of changes suggestive of early interstitial lung disease (ILD) on chest computed tomographic (CT) scans.
The CT scans ...of the first 100 subjects enrolled in the COPDGene Study from a single institution were examined using a sequential reader and a group consensus interpretation scheme. CT scans were evaluated for the presence of parenchymal changes consistent with ILD using the following scoring system: 0 = normal, 1 = equivocal for the presence of ILD, 2 = highly suspicious for ILD, and 3 = classic ILD changes. A statistical comparison of patients with early ILD to normal subjects was performed.
There was a high degree of agreement between methods (kappa = 0.84; 95% confidence interval, 0.73-0.94; P < .0001 for the sequential and consensus methods). The sequential reading method had both high positive (1.0) and negative (0.97) predictive values for a consensus read despite a 58% reduction in the number of chest CT evaluations. Regardless of interpretation method, the prevalence of chest CT changes consistent with early ILD in this subset of smokers from COPDGene varied between 5% and 10%. Subjects with early ILD tended to have greater tobacco smoke exposure than subjects without early ILD (P = .053).
A sequential CT interpretation scheme is an efficient method for the visual interpretation of CT data. Further investigation is required to independently confirm our findings and further characterize early ILD in smokers.
The mean pulmonary arterial wedge pressure (mPAWP) is the critical hemodynamic factor differentiating group 1 pulmonary arterial hypertension (PAH) from group 2 pulmonary hypertension associated with ...left heart disease. Despite the discrepancy between the mPAWP upper physiologic normal and current PAH definitions, the implications of the initial mPAWP for PAH clinical trajectory are poorly understood.
To model longitudinal mPAWP trajectories in PAH over 10 years and examine the clinical and hemodynamic factors associated with trajectory membership.
Adult patients with PAH with two or more right heart catheterizations were identified from a multiinstitution healthcare system in eastern Massachusetts. mPAWP trajectories were constructed via group-based trajectory modeling. Feature selection was performed in least absolute shrinkage and selection operator regression. Logistic regression was used to assess associations between trajectory membership, baseline characteristics, and transplant-free survival.
Among 301 patients with PAH, there were two distinct mPAWP trajectories, termed "mPAWP-high" (
= 71; 23.6%) and "mPAWP-low" (
= 230; 76.4%), based on the ultimate mPAWP value. Initial mPAWP clustered around median 12 mm Hg (interquartile range IQR, 8-14 mm Hg) in the mPAWP-high and 9 mm Hg (IQR, 6-11 mm Hg) in the mPAWP-low trajectories (
< 0.001). After feature selection, initial mPAWP ⩾12 mm Hg predicted an mPAWP-high trajectory (odds ratio, 3.2; 95% confidence interval, 1.4-6.1;
= 0.0006). An mPAWP-high trajectory was associated with shorter transplant-free survival (vs. mPAWP-low, median, 7.8 vs. 11.3 yr; log-rank
= 0.017; age-adjusted
= 0.217).
Over 10 years, the mPAWP followed two distinct trajectories, with 25% evolving into group 2 pulmonary hypertension physiology. Using routine baseline data, longitudinal mPAWP trajectory could be predicted accurately, with initial mPAWP ⩾12 mm Hg as one of the strongest predictors.
COPD is a leading cause of mortality.
We hypothesized that applying machine learning to clinical and quantitative CT imaging features would improve mortality prediction in COPD.
We selected 30 ...clinical, spirometric, and imaging features as inputs for a random survival forest. We used top features in a Cox regression to create a machine learning mortality prediction (MLMP) in COPD model and also assessed the performance of other statistical and machine learning models. We trained the models in subjects with moderate to severe COPD from a subset of subjects in Genetic Epidemiology of COPD (COPDGene) and tested prediction performance in the remainder of individuals with moderate to severe COPD in COPDGene and Evaluation of COPD Longitudinally to Identify Predictive Surrogate Endpoints (ECLIPSE). We compared our model with the BMI, airflow obstruction, dyspnea, exercise capacity (BODE) index; BODE modifications; and the age, dyspnea, and airflow obstruction index.
We included 2,632 participants from COPDGene and 1,268 participants from ECLIPSE. The top predictors of mortality were 6-min walk distance, FEV1 % predicted, and age. The top imaging predictor was pulmonary artery-to-aorta ratio. The MLMP-COPD model resulted in a C index ≥ 0.7 in both COPDGene and ECLIPSE (6.4- and 7.2-year median follow-ups, respectively), significantly better than all tested mortality indexes (P < .05). The MLMP-COPD model had fewer predictors but similar performance to that of other models. The group with the highest BODE scores (7-10) had 56% mortality, whereas the highest mortality group defined by the MLMP-COPD model had 62% mortality (P = .046).
An MLMP-COPD model outperformed four existing models for predicting all-cause mortality across two COPD cohorts. Performance of machine learning was similar to that of traditional statistical methods. The model is available online at: https://cdnm.shinyapps.io/cgmortalityapp/.
Biomarker estimation methods from medical images have traditionally followed a segment-and-measure strategy. Deep-learning regression networks have changed such a paradigm, enabling the direct ...estimation of biomarkers in databases where segmentation masks are not present. While such methods achieve high performance, they operate as a black-box. In this work, we present a novel deep learning network structure that, when trained with only the value of the biomarker, can perform biomarker regression and the generation of an accurate localization mask simultaneously, thus enabling a qualitative assessment of the image locus that relates to the quantitative result. We showcase the proposed method with three different network structures and compare their performance against direct regression networks in four different problems: pectoralis muscle area (PMA), subcutaneous fat area (SFA), liver mass area in single slice computed tomography (CT), and Agatston score estimated from non-contrast thoracic CT images (CAC). Our results show that the proposed method improves the performance with respect to direct biomarker regression methods (correlation coefficient of 0.978, 0.998, and 0.950 for the proposed method in comparison to 0.971, 0.982, and 0.936 for the reference regression methods on PMA, SFA and CAC respectively) while achieving good localization (DICE coefficients of 0.875, 0.914 for PMA and SFA respectively, p <; 0.05 for all pairs). We observe the same improvement in regression results comparing the proposed method with those obtained by quantify the outputs using an U-Net segmentation network (0.989 and 0.951 respectively). We, therefore, conclude that it is possible to obtain simultaneously good biomarker regression and localization when training biomarker regression networks using only the biomarker value.