Automated Coronary Artery Calcium and Quantitative Emphysema in Lung Cancer Screening: Association With Mortality, Lung Cancer Incidence, and Airflow Obstruction : Journal of Thoracic Imaging

Secondary Logo

Journal Logo

Original Article

Automated Coronary Artery Calcium and Quantitative Emphysema in Lung Cancer Screening

Association With Mortality, Lung Cancer Incidence, and Airflow Obstruction

Balbi, Maurizio MD*,†; Sabia, Federica MSc*; Ledda, Roberta E. MD*,†; Milanese, Gianluca MD, PhD; Ruggirello, Margherita MD; Silva, Mario MD, PhD; Marchianò, Alfonso V. MD; Sverzellati, Nicola MD, PhD; Pastorino, Ugo MD*

Author Information
Journal of Thoracic Imaging ():10.1097/RTI.0000000000000698, January 20, 2023. | DOI: 10.1097/RTI.0000000000000698

Abstract

Lung cancer screening (LCS) with low-dose computed tomography (LDCT) reduces mortality in heavy smokers.1,2 Besides focusing on lung cancer (LC), LDCT enables the identification of potentially relevant respiratory and cardiovascular (CV) findings, such as emphysema and coronary artery calcium (CAC).2–4 Expanding screening for these recognized biomarkers is attractive as it may help refine risk assessment from pre-scan data and personalize follow-up intervention.5

Despite their potential clinical and prognostic implications, CAC and emphysema are not yet routinely incorporated in LCS risk prediction models. Indeed, there is still no evidence supporting any workup for these biomarkers in such a setting.6–8 A first step in assessing the benefit of reporting CAC and emphysema in LCS is to determine their prognostic value and capability to target LCS volunteers for whom preventive intervention might be beneficial. This approach could be further empowered with quantitative automated tools to reduce time analysis and improve reproducibility.9,10

The present study aimed to assess the value of quantitative emphysema and CAC provided by a commercially available artificial intelligence (AI) software for predicting all-cause, noncancer, CV, and LC mortality and LC incidence in an Italian LCS trial, namely the BioMILD trial. We also explored the correlation between these LDCT biomarkers and forced expiratory value in 1 second (FEV1) and evaluated the discriminative ability of quantitative emphysema for airflow obstruction.

MATERIALS AND METHODS

Study Participants

Details of the BioMILD trial have been described elsewhere.11 Briefly, the BioMILD trial is an ongoing prospective study performed at the Istituto Nazionale Tumori of Milan, testing the combination of plasma miRNA and LDCT to improve LCS efficacy by individual risk profiling and personalized screening intervals (clinicaltrials.gov ID: NCT02247453). Eligible participants were (i) aged 50 to 75 years and current heavy smokers of ≥30 pack-years or former smokers with the same smoking habits who stopped ≤10 years before; (ii) aged 50 to 75 years and current or former smokers of <30 pack-years with a family history of LC or a prior diagnosis of chronic obstructive pulmonary disease (COPD) or pneumonia. The exclusion criteria were the presence of neoplasms within the previous 5 years and suspected lung nodules under investigation.

All the volunteers who underwent a baseline screening round as part of the BioMILD trial were considered potentially eligible for inclusion in our retrospective study. The original Institutional Review Board approval and written informed consent allowed the use of data for future research, including the present analyses.

Demographic, Clinical, and Follow-up Data

Demographic and clinical data were collected through specific questionnaires and a direct interview with a study investigator at the baseline screening round. Information on causes of death was collected by phone calls or email contacts with the general practitioner or referring hospital and through periodic inquiries to cancer registries. The vital status was obtained through the Istituto Nazionale di Statistica (ISTAT, SIATEL 2.0 platform), which provides the exact date of death within 3 months of occurrence. Participants accumulated person-years of follow-up from the date of baseline until death or the date of the last follow-up as of November 2021.

Imaging Acquisition and Analysis

Baseline LDCT was performed on a second-generation dual-source CT scanner (Somatom Definition Flash, Siemens Medical Solutions). The whole chest volume was scanned during one deep inspiratory breath-hold with the following scanning parameters: tube voltage, 120 kVp; tube current, 30 mAs; collimation, 0.625 mm; pitch, 1.2; rotation time, 0.5 seconds. Images were reconstructed with the following parameters: thickness, 1 mm; increment, 0.7 mm; medium-sharp kernel (B50f).

The LDCT images were transferred to a dedicated graphic station (Alienware Area 51 R6 equipped with Dual NVIDIA GeForce RTX 2080 OC graphics) and analyzed using fully automated AI software (AVIEW, Coreline Soft). Emphysema was quantified using the percentage of lung volume occupied by voxels with attenuation of −950 HU or less (percentage of low attenuation areas, %LAA).12,13 CAC was measured with a 3-dimensional U-net architecture-based scoring tool previously validated in electrocardiography-gated and non–electrocardiography-gated LDCTs.14–16 The software was also tested in a subgroup of BioMILD participants to assess its longitudinal reproducibility and agreement with manual evaluation, showing consistent results (Table S1, Supplemental Digital Content 1, https://links.lww.com/JTI/A248 and Table S2, Supplemental Digital Content 1, https://links.lww.com/JTI/A248). CAC scores were stratified using prespecified Agatston score strata of 0 to 99, 100 to 399, and ≥400, in keeping with the risk categorization adopted in the low-dose cardiac CT population of the Risk Or Benefit IN Screening for CArdiovascular diseases (ROBINSCA) trial.17 %LAA and CAC were computed after convolutional neural network-based sharp to soft tissue kernel conversion18

Pulmonary Function Testing

All the participants underwent prebronchodilator spirometry at the baseline using a flow spirometer (KoKo; nSpire Health, Longmont, Colo). Respiratory function was estimated with FEV1 as a percentage of predicted and the FEV1 to forced vital capacity (FVC) ratio (FEV1/FVC). An FEV1/FVC<0.70 was used to define airflow obstruction. The severity of airflow obstruction was classified according to the Global Initiative for Obstructive Lung Disease (GOLD) stages,19 including the Preserved Ratio Impaired Spirometry (PRISm) group, where FEV1 is reduced but FEV1/FVC is preserved.20,21

Statistical Analysis

Demographic and clinical data were reported as numbers and percentages for categorical variables and as medians with interquartile ranges (IQRs) for continuous variables. Descriptive statistics of deaths and LC cases were reported as stratified by respiratory function, %LAA, and CAC. Measures of association were evaluated by the χ2 test or Fisher exact test for categorical data and by the Mann-Whitney U test for continuous variables. The presence of a trend in categorical variables was estimated by the Cochran-Armitage test.

Univariate and multivariate Cox proportional models were applied to estimate the hazard ratio (HR) and 95% CI of 6-year all-cause, noncancer, CV, and LC mortality and LC incidence. Univariate Cox models were performed for all demographic and clinical data, %LAA, and CAC. Since the distributions of %LAA were left-skewed, HR was reported per IQR. Different multivariate Cox models were performed to assess the predictive discriminations of age, sex, pack-years, %LAA, CAC, and available demographic and clinical variables that reached a statistical significance level of 0.2 in the corresponding univariate Cox models. Harrell’s Concordance Statistic (C-statistic) and time-dependent area under the curve (AUC) were reported to test the predictive accuracy of the following nested multivariate models: Modelsurvey (age, sex, and pack-years); Modelsurvey-LDCT (age, sex, pack-years, %LAA, and CAC); and Modelfinal (variables in Modelsurvey-LDCT plus demographic and clinical variables with a statistical significance level of 0.2 in the corresponding univariate Cox models). Comparisons of concordance probabilities between any 2 submodels were made by the method of Uno et al.22 To address possible bias due to the limited number of events and a number of variables, Cox regression models were performed with Firth’s penalized method. In the LC incidence analysis, all volunteers with a prevalent LC were excluded. Kaplan-Meier curves for 6-year cumulative mortality were reported in strata of CAC, and comparisons were tested by the Log-Rank test for trend.

The discriminative ability of %LAA as continuous values for airflow obstruction was investigated by the receiver operating characteristic curve (ROC) and the AUC. The sensitivity, specificity, positive predictive value, and negative predictive value of the optimal %LAA cutoff value for airflow obstruction were computed after Youden index maximization. The correlations between %LAA, CAC, and FEV1 were tested by using the Pearson correlation coefficient.

All analyses were performed using the Statistical Analysis System Software (Release SAS:9.04; SAS Institute).

RESULTS

Of the 4119 volunteers who underwent a baseline LDCT between January 2013 and March 2016 as part of the BioMILD trial, 21 were excluded due to incomplete data (Fig. 1). Detailed demographic, clinical, and pulmonary function data of the 4098 enrolled volunteers and their association with outcomes are reported in Table 1. In summary, the median age was 60 (IQR, 55 to 64; male 61%, 2491/4098), and the median number of pack-years was 42 (IQR, 35 to 52). The cause of death was ascertained for 94% (132/141) of deaths in the first 6 years since the baseline LDCT scan. Noncancer, CV, and LC death occurred in 1.3% (55/4098), 0.7% (27/4098), and 0.9% (38/4098) of cases, respectively. The LC incidence at the 6-year follow-up was 2.8% (112/4051). Higher age, male sex, number of pack-years, and self-reported COPD were significantly associated with all the mortality outcomes (P<0.05). Abnormal pulmonary function was significantly associated with all-cause, noncancer, and LC deaths (P<0.05). Higher age, a greater number of pack-years, history of cancer, self-reported COPD, and pulmonary function were associated with LC incidence (P<0.05).

F1
FIGURE 1:
Participant flow diagram.
TABLE 1 - Distribution of Demographic, Clinical, and Respiratory Function Data According to 6-Year All-cause Deaths, Cause-specific Deaths, and Lung Cancer Incidence
Total Alive 6-year all-cause deaths 6-year noncancer deaths 6-year CV deaths 6-year LC deaths 6-year LC incidence
N N (%) N (%) P N (%) P N (%) P N (%) P N (%) P
Total 4098 3957/4098 (96.6) 141 (3.4) 55 (1.3) 27 (0.7) 38 (0.9) 112 (2.8)*
5.9×1000 py 2.3×1000 py 1.1×1000 py 1.6×1000 py 4.8×1000 py
Age, y (median, IQR) 60 (55-64) 59 (55-64) 65 (60-69) <0.0001 65 (57-70) <0.0001 65 (57-70) 0.0093 65 (61-69) 0.0002 62 (57-66.5) <0.0001
Sex
 Female 1607 (39.2) 1573 (97.9) 34 (2.1) 0.0002 11 (0.7) 0.0033 5 (0.3) 0.0291 8 (0.5) 0.0212 46 (2.9) 0.6808
 Male 2491 (60.8) 2384 (95.7) 107 (4.3) 44 (1.8) 22 (0.9) 30 (1.2) 66 (2.75)
Pack-years (median, IQR) 42 (35-52) 41 (35-52) 49 (42-59) <0.0001 48 (42-62) <0.0001 47 (41-57) 0.0102 51 (45-70) <0.0001 47 (38-60.5) 0.0002
Smoking history
 Current 3247 (79.2) 3136 (96.6) 111 (3.4) 0.8792 42 (1.3) 0.5973 22 (0.7) 1.0000 31 (1.0) 0.7203 91 (2.8) 0.5861
 Former 851 (20.8) 821 (96.5) 30 (3.5) 13 (1.5) 5 (0.6) 7 (0.8) 21 (2.5)
Time since smoking cessation (median, IQR) 5 (3-8) 5 (3-8) 5 (3-8) 0.9582 6 (3-8) 0.3989 6 (3-8) 0.5233 5 (2-6) 0.3232 5 (2-6) 0.2681
Education
 Lower grade 828 (20.2) 790 (95.4) 38 (4.6) 0.5093 17 (2.1) 0.1201 8 (1.0) 0.4671 10 (1.2) 0.2297 28 (3.4) 0.5823
 HS graduate 2205 (53.8) 2138 (97.0) 67 (3.0) 24 (1.1) 13 (0.6) 21 (1.0) 56 (2.6)
 Bachelors degree 1065 (26.0) 1029 (96.6) 36 (3.4) 14 (1.3) 6 (0.6) 7 (0.7) 28 (2.75)
BMI (median, IQR) 25.4 (22.8-28.0) 25.4 (22.8-28.0) 25.9 (23.3-29.1) 0.1672 26.3 (23.4-30.9) 0.0430 27.4 (24.3-31.7) 0.0158 24.8 (22.9-28.1) 0.6221 24.9 (22.6-27.0) 0.0614
Asbestos exposure
 No 3905 (95.3) 3772 (96.6) 133 (3.4) 0.5823 54 (1.4) 0.5188 27 (0.7) 0.6364 32 (0.8) 0.0012 108 (2.8) 0.8196
 Yes 193 (4.7%) 185 (95.9) 8 (4.2) 1 (0.5) 0 6 (3.1) 4 (2.1)
Personal history of cancer
 No 3865 (94.3) 3733 (96.6) 132 (3.4) 0.7160 50 (1.35) 0.2399 25 (0.7) 0.6644 35 (0.9) 0.4750 101 (2.6) 0.0490
 Yes 233 (5.7) 224 (96.1) 9 (3.9) 5 (2.2) 2 (0.9) 3 (1.3) 11 (4.9)
Family history of lung cancer
 No 2923 (71.3) 2825 (96.7) 98 (3.4) 0.6260 37 (1.3) 0.5032 22 (0.8) 0.2909 25 (0.9) 0.4482 75 (2.6) 0.3047
 Yes 1175 (28.7) 1132 (96.3) 43 (3.7) 18 (1.5) 5 (0.4) 13 (1.1) 37 (3.2)
Asthma
 No 3884 (94.8) 3750 (96.6) 134 (3.5) 0.8888 49 (1.3) 0.0563 25 (0.6) 0.6497 38 (1.0) 0.2632 107 (2.8) 0.7027
 Yes 214 (5.2) 207 (96.7) 7 (3.3) 6 (2.8) 2 (0.9) 0 5 (2.4)
Self-reported COPD
 No 3345 (81.6) 3253 (97.3) 92 (2.8) <0.0001 31 (0.9) <0.0001 15 (0.5) 0.0004 25 (0.8) 0.0113 78 (2.4) 0.0009
 Yes 753 (18.4) 704 (93.5) 49 (6.5) 24 (3.2) 12 (1.6) 13 (1.7) 34 (4.6)
CVD§
 No 3826 (93.4) 3703 (96.8) 123 (3.2) 0.0029 42 (1.1) <0.0001 18 (0.5) <0.0001 38 (1.0) 0.1765 104 (2.8) 0.8285
 Yes 272 (6.6) 254 (93.4) 18 (6.6) 13 (4.8) 9 (3.3) 0 8 (3.0)
Hypertension
 No 2882 (70.3) 2797 (97.1) 85 (3.0) 0.0079 33 (1.2) 0.0914 13 (0.5) 0.0114 24 (0.8) 0.3311 75 (2.6) 0.4155
 Yes 1216 (29.7) 1160 (95.4) 56 (4.6) 22 (1.8) 14 (1.2) 14 (1.2) 37 (3.1)
Diabetes
 No 3778 (92.2) 3661 (96.9) 117 (3.1) <0.0001 45 (1.2) 0.0039 22 (0.6) 0.0546 33 (0.9) 0.2161 104 (2.8) 0.7997
 Yes 320 (7.8) 296 (92.5) 24 (7.5) 10 (3.1) 5 (1.6) 5 (1.6) 8 (2.5)
Statins
 No 3280 (80.0) 3177 (96.9) 103 (3.1) 0.0346 39 (1.2) 0.0881 17 (0.5) 0.0259 28 (0.9) 0.3248 87 (2.7) 0.5279
 Yes 818 (20.0) 780 (95.4) 38 (4.7) 16 (2.0) 10 (1.2) 10 (1.2) 25 (3.1)
Respiratory function
 Control 2731 (66.6) 2661 (97.4) 70 (2.6) <0.0001 30 (1.1) 0.0386 16 (0.6) 0.5457 16 (0.6) <0.0001 59 (2.2) 0.0001
 PRISm 143 (3.5) 133 (93.0) 10 (7.0) 6 (4.2) 4 (2.8) 0 1 (0.7)
 GOLD I 758 (18.5) 728 (96.0) 30 (4.0) 8 (1.1) 3 (0.4) 11 (1.5) 30 (4.0)
 GOLD II 422 (10.3) 399 (94.6) 23 (5.5) 7 (1.7) 2 (0.5) 9 (2.1) 19 (4.6)
 GOLD III/IV 44 (1.1) 36 (81.8) 8 (18.2) 4 (9.1) 2 (4.6) 2 (4.6) 3 (7.1)
Data are presented as n (%) or median (interquartile range). Bold P-values for variables significant at level 0.2 in univariate cox models and included in multivariate cox models.
*Percentage on 4051 volunteers without prevalent lung cancer.
Only for former smokers.
Either emphysema or chronic bronchitis.
§Defined as myocardial infarction, stroke, thrombosis, or angina.
BMI indicates body mass index; CVD, cardiovascular disease; GOLD, Global Initiative for Obstructive Lung Disease; PRISm, preserved ratio impaired spirometry; LC, lung cancer.

%LAA was higher in all the mortality groups than in alive participants and in subjects diagnosed with LC compared with those with no LC (P<0.05) (Supplemental Digital Content, Table S3, https://links.lww.com/JTI/A248). A statistically significant trend in CAC score was found according to all-cause, noncancer, and CV death, but no LC death nor LC incidence (Supplemental Digital Content, Table S3, https://links.lww.com/JTI/A248). A total of 709/3280 (22%) volunteers with no reported statin therapy had a CAC score of ≥100, namely above the current threshold for the initiation of preventive therapy in CV disease screening17 (Supplemental Digital Content, Table S4, https://links.lww.com/JTI/A248).

All-cause Mortality

%LAA and CAC were independently associated with a higher risk of all-cause mortality. Compared with the reference CAC category of 0 to 99, crude HRs for CAC of 100 to 399 and ≥400 were 1.89 (95% CI, 1.22-2.87; P<0.01) and 3.86 (95% CI, 2.61-5.64; P<0.01), respectively. The crude HR for an IQR increase of %LAA was 1.24 (95% CI, 1.16-1.30; P<0.01). In multivariate analyses, %LAA and CAC remained significantly associated with all-cause mortality (HRs in Modelfinal, 1.14 for an IQR increase of %LAA [95% CI, 1.05-1.23; P<0.01] and 2.13 for CAC ≥400 [95% CI, 1.36-3.28; P<0.01]) (Table 2). C-statistics for Modelsurvey, Modelsurvey-LDCT, and Modelfinal were 0.7041 (95% CI, 0.7034-0.7048), 0.7195 (95% CI, 0.7188-0.7202), and 0.7353 (95% CI, 0,7346-0,7360), respectively. The increase in concordance probabilities was statistically significant in Modelsurvey-LDCT compared with Modelsurvey (P=0.04), in Modelfinal compared with Modelsurvey (P=0.03), but not in Modelfinal compared with Modelsurvey-LDCT (P=0.22). Figure 2 shows the AUC curves over time of the three models. Kaplan-Meier curves for CAC strata are shown in Figure 3. Restricted analysis on participants with ≥30 pack-years (3834/4098, 94%) confirmed that %LAA and CAC were significantly associated with a higher risk of all-cause mortality in univariate and multivariate models (Table S5, https://links.lww.com/JTI/A248).

TABLE 2 - Univariate and Multivariate Cox Models for 6-Year All-cause Mortality
%LAA CAC, n (%)
Median (IQR) 0-99 100-399 ≥400
1.2 (0.5-2.8) 2995 (73.1) 652 (15.9) 451 (11.0)
All-cause mortality 2.0 (0.7-4.9) 72 (2.4) 29 (4.5) 40 (8.9)
HRcrude (95%CI) 1.24 (1.16-1.30) Ref 1.89 (1.22-2.87) 3.86 (2.61-5.64)
P <0.0001 0.0037 <0.0001
HRadjusted * (95% CI) 1.23 (1.15-1.30) Ref 1.78 (1.14-2.70) 3.69 (2.49-5.38)
P <0.0001 0.0086 <0.0001
HRadjusted (95% CI) 1.16 (1.08-1.24) Ref 1.27 (0.80-1.95) 2.25 (1.47-3.40)
P <0.0001 0.3008 0.0001
HRadjusted (95% CI) 1.14 (1.05-1.23) Ref 1.16 (0.73-1.79) 2.13 (1.36-3.28)
P 0.0014 0.5339 0.0008
*Multivariable model including %LAA and CAC.
Adjusted for age, sex, and pack-years (Modelsurvey-LDCT).
Adjusted for age, sex, pack-years, body mass index, self-reported chronic obstructed pulmonary disease, history of cardiovascular disease (defined as myocardial infarction, stroke, thrombosis, or angina), hypertension, diabetes, statins, and respiratory function (Modelfinal).
%LAA indicates percentage of lung volume occupied by voxels with attenuation of −950 HU or less; CAC, coronary artery calcium by pre-specified Agatston strata; Ref, reference.

F2
FIGURE 2:
Time-dependent AUC for all-cause mortality according to 3 nested multivariate models (Modelsurvey: age, sex, and pack-years; Modelsurvey-LDCT: age, sex, pack-years, %LAA, and CAC; Modelfinal: age, sex, pack-years, %LAA, CAC, body mass index, self-reported chronic obstructive pulmonary disease, history of cardiovascular disease [defined as myocardial infarction, stroke, thrombosis, or angina], hypertension, diabetes, statins, and respiratory function). %LAA, percentage of lung volume occupied by voxels with attenuation of −950 HU or less; CAC, coronary artery calcium by prespecified Agatston strata.
F3
FIGURE 3:
Kaplan-Meier mortality curves for all-cause mortality according to coronary artery calcium (CAC) by prespecified Agatston strata.

Noncancer Mortality

%LAA and CAC were independently associated with a higher risk of noncancer mortality. Compared with the reference CAC category of 0 to 99, crude HRs for CAC of 100 to 399 and ≥400 were 2.08 (95% CI, 0.96-4.19; P=0.05) and 6.61 (95% CI, 3.69-11.81; P<0.01), respectively. The crude HR for an IQR increase of %LAA was 1.28 (95% CI, 1.18-1.37; P<0.01). In multivariate analyses, the strength of associations was still significant for both LDCT parameters (HRs in Modelfinal, 1.25 for a IQR increase of %LAA [95% CI, 1.11-1.37; P<0.01] and 3.22 for CAC ≥400 [95% CI, 1.62-6.39; P<0.01]) (Table 3). C-statistics for Modelsurvey, Modelsurvey-LDCT, and Modelfinal were 0.7030 (95% CI, 0.7019-0.7041), 0.7336 (95% CI, 0.7324-0.7348) and 0.7778 (95% CI, 0.7766-0.7790), respectively. The increase in concordance probabilities was statistically significant in Modelsurvey-LDCT compared with Modelsurvey (P=0.046), in Modelfinal compared with Modelsurvey (P=0.01), but not in Modelfinal compared with Modelsurvey-LDCT (P=0.09). Time-dependent AUC curves of the three models are shown in Figure 4, whereas Kaplan-Meier curves for CAC strata in Figure 5.

TABLE 3 - Univariate and Multivariate Cox Models for 6-Year Noncancer Mortality
%LAA CAC
Median (IQR) 0-99 100-399 ≥400
1.2 (0.5-2.8) 2995 (73.1%) 652 (15.9%) 451 (11.0%)
Noncancer mortality 1.6 (0.6-5.4) 23 (0.8%) 10 (1.5%) 22 (4.9%)
HRcrude (95% CI) 1.28 (1.18-1.37) Ref 2.08 (0.96-4.19) 6.61 (3.69-11.81)
P <0.0001 0.0521 <0.0001
HRadjusted * (95% CI) 1.28 (1.17-1.38) Ref 1.93 (0.89-3.89) 6.26 (3.50-11.20)
P <0.0001 0.0804 <0.0001
HRadjusted (95% CI) 1.24 (1.11-1.34) Ref 1.47 (0.66-3.05) 4.24 (2.23-8.08)
P <0.0001 0.3239 <0.0001
HRadjusted (95% CI) 1.25 (1.11-1.37) Ref 1.24 (0.56-2.61) 3.22 (1.62-6.39)
P <0.0001 0.5833 0.0009
*Multivariable model including %LAA and CAC.
Adjusted for age, sex, and pack-years (Modelsurvey-LDCT).
Adjusted for age, sex, pack-years, education, body mass index, asthma, self-reported chronic obstructed pulmonary disease, history of cardiovascular disease (defined as myocardial infarction, stroke, thrombosis, or angina), hypertension, diabetes, statins, and respiratory function (Modelfinal).
%LAA indicates percentage of lung volume occupied by voxels with attenuation of −950 HU or less; CAC, coronary artery calcium by prespecified Agatston strata; Ref, reference.

F4
FIGURE 4:
Time-dependent AUC for noncancer mortality according to 3 nested multivariate models (Modelsurvey, age, sex, and pack-years; Modelsurvey-LDCT, age, sex, pack-years, %LAA, and CAC; Modelfinal, age, sex, pack-years, %LAA, CAC, education, body mass index, asthma, self-reported chronic obstructive pulmonary disease, history of cardiovascular disease [defined as myocardial infarction, stroke, thrombosis, or angina], hypertension, diabetes, statins, and respiratory function). %LAA, percentage of lung volume occupied by voxels with attenuation of −950 HU or less; CAC, coronary artery calcium by pre-specified Agatston strata.
F5
FIGURE 5:
Kaplan-Meier mortality curves for noncancer mortality according to coronary artery calcium (CAC) by pre-specified Agatston strata.

CV Mortality

%LAA and CAC were independently associated with a higher risk of CV mortality (Table 4). The HRs for CAC of 100 to 399 and ≥400 were 2.20 (95% CI, 0.64-6.44; P=0.18) and 10.44 (95% CI, 4.66-25.50; P<0.01), respectively. The crude HR for an IQR increase of %LAA was 1.24 (95% CI, 1.05-1.38; P<0.01). The association remained significant in multivariate analyses (HRs in Modelfinal, 1.25 for an IQR increase of %LAA [95% CI, 1.00-1.46; P=0.01] and 4.66 for CAC ≥400 [95% CI, 1.80-12.58; P<0.01]). C-statistics were 0.6831 (95% CI, 0.6816-0.6846) in Modelsurvey, 0.7608 (95% CI, 0.7592-0.7623) in Modelsurvey-LDCT, and 0.8349 (95% CI, 0.8335-0.8363) in Modelfinal. The increase in concordance probabilities was statistically significant in Modelsurvey-LDCT compared with Modelsurvey (P<0.01), in Modelfinal compared with Modelsurvey (P<0.01), but not in Modelfinal compared with Modelsurvey-LDCT (P=0.06). Figure 6 shows the AUC curves over time of the 3 models. Kaplan-Meier curves for CAC strata are illustrated in Figure 7.

TABLE 4 - Univariate and Multivariate Cox Models for 6-Year Cardiovascular Mortality
%LAA CAC
Median (IQR) 0-99 100-399 ≥400
1.2 (0.5-2.8) 2995 (73.1%) 652 (15.9%) 451 (11.0%)
Cardiovascular mortality 2.0 (0.7-5.2) 9 (0.3%) 4 (0.6%) 14 (3.1%)
HRcrude (95% CI) 1.24 (1.05-1.38) Ref 2.20 (0.64-6.44) 10.44 (4.66-25.50)
P 0.0012 0.1807 <0.0001
HRadjusted * (95% CI) 1.22 (1.02-1.38) Ref 2.07 (0.61-6.08) 8.90 (4.42-23.27)
P 0.0074 0.2160 <0.0001
HRadjusted (95% CI) 1.18 (0.98-1.35) Ref 1.69 (0.48-5.16) 7.32 (2.99-18.91)
P 0.0388 0.3860 <0.0001
HRadjusted (95% CI) 1.25 (1.00-1.46) Ref 1.32 (0.37-4.05) 4.66 (1.80-12.58)
P 0.0151 0.6515 0.0023
*Multivariable model including %LAA and CAC.
Adjusted for age, sex, and pack-years (Modelsurvey-LDCT).
Adjusted for age, sex, pack-years, body mass index, self-reported chronic obstructed pulmonary disease, history of cardiovascular disease (defined as myocardial infarction, stroke, thrombosis, or angina), hypertension, diabetes, and statins (Modelfinal).
%LAA indicates percentage of lung volume occupied by voxels with attenuation of −950 HU or less; CAC, coronary artery calcium by prespecified Agatston strata; Ref, reference.

F6
FIGURE 6:
Time-dependent AUC for cardiovascular mortality according to 3 nested multivariate models (Modelsurvey: age, sex, and pack-years; Modelsurvey-LDCT: age, sex, pack-years, %LAA, and CAC; Modelfinal: age, sex, pack-years, %LAA, CAC, body mass index, self-reported chronic obstructive pulmonary disease, history of cardiovascular disease [defined as myocardial infarction, stroke, thrombosis, or angina], hypertension, diabetes, and statins). %LAA, percentage of lung volume occupied by voxels with attenuation of −950 Hounsfield units or less; CAC, coronary artery calcium by pre-specified Agatston strata.
F7
FIGURE 7:
Kaplan-Meier mortality curves for cardiovascular mortality according to coronary artery calcium (CAC) by prespecified Agatston strata.

Lung Cancer Mortality

Increasing %LAA, but not CAC, was significantly associated with a higher risk of LC mortality in univariate analysis (HR, 1.20; [95% CI, 1.03-1.33; P<0.01]) and when adjusted for CAC (HR, 1.19; [95% CI, 1.02-1.33; P<0.01]). However, the association was no longer significant after adjustment for further confounders (Supplemental Digital Content, Table S6, https://links.lww.com/JTI/A248). C-statistics were 0.7290 (95% CI, 0.7278-0.7302) in Modelsurvey, 0.7352 (95% CI, 0.7340-0.7364) in Modelsurvey-LDCT, and 0.7826 (95% CI, 0.7816-0.7836) in Modelfinal.

Lung Cancer Incidence

%LAA was not significant associated with LC incidence nor in univariate neither in multivariate analyses (Supplemental Digital Content, Table S7, https://links.lww.com/JTI/A248). CAC of 100-399, but not ≥400, was significantly associated with LC incidence in univariate analysis (HR, 1.63; [95% CI, 1.02-2.52; P=0.04]) and when adjusted for %LAA (HR, 1.59; [95% CI, 0.99-2.47; P=0.045]). After adjustment for further confounders, the association was no longer significant (Supplemental Digital Content, Table S7, https://links.lww.com/JTI/A248). C-statistics were 0.6230 (95% CI, 0.6222-0.6238) in Modelsurvey, 0.6320 (95% CI, 0.6312-0.6328) in Modelsurvey-LDCT, and 0.6717 (95% CI, 0.6710-0.6724) in Modelfinal.

Airflow Obstruction

%LAA and CAC were linked by a significant positive association (r=0.05297, P<0.01) and negatively correlated with FEV1 (r=−0.15285 and −0.05381, respectively; P<0.01). The ROC curve in Figure 8 depicts the discriminative ability of %LAA for airflow obstruction (AUC, 0.738). The optimal cut-off value of %LAA to predict airflow obstruction by maximization of Youden Index was 1.3 (sensitivity, 72%; specificity, 62%; positive predictive value, 45%; negative predictive value, 84%). The distribution of pulmonary function test results according to outcomes and stratified by the %LAA <1.3% and %LAA ≥1.3% is reported in Supplemental Digital Content (Table S8, https://links.lww.com/JTI/A248).

F8
FIGURE 8:
Receiver-operating characteristic (ROC) curve showing the discriminative ability of %LAA as continuous values for airflow obstruction (ie, forced expiratory volume in 1 second to forced vital capacity ratio <0.70). %LAA, percentage of lung volume occupied by voxels with attenuation of −950 HU or less.

DISCUSSION

The benefit of reporting coronary artery calcification (CAC) and quantitative emphysema (%LAA) in LCS is still under debate. The present study found that %LAA and CAC assessed by automated AI software were independent predictors of 6-year all-cause, noncancer, and CV mortality in the BioMILD LCS trial, adding prognostic information to age, sex, and pack-years. Otherwise, %LAA and CAC predicted neither LC incidence nor mortality when adjusted for selected confounders. Moreover, both LDCT biomarkers negatively correlated with FEV1, with %LAA enabling the identification of airflow obstruction with moderate discriminative ability.

The prognostic value of CAC and %LAA has been explored in heterogeneous settings and following different methodologies.23–25 In LCS, there is evidence to support an independent association between these parameters and all-cause mortality.26–30 For instance, every 1% increase in the percentage of %LAA independently predicted all-cause death in the CT arm of the National Lung Screening Trial (NLST).26 In a recent meta-analysis including studies from different LCS trials, subjects in high CAC categories had a relative risk of all-cause mortality of 2.47 (95% CI, 1.75-3.47) and 2.90 (95% CI, 1.57-5.36) based on a fixed-effect model and a random-effect model, respectively.31 In keeping with these results, our findings argue in favor of %LAA and CAC to predict all-cause and noncancer mortality in LCS. Moreover, we build on previous methodologies by employing an AI-based approach, which may reduce variability while automating the process.

A CAC score of ≥400 independently predicted CV mortality. Despite differences in population composition, reference CAC categories, and endpoints, the current results align with previous studies supporting CAC for CV risk refinement in LCS.10,27–29,32–34 Jacobs et al28 observed an almost sixfold risk of coronary events in the highest risk category compared with no detectable CAC in a case-cohort study from the NELSON trial (NELSON is a Dutch acronym for “Nederlands-Leuvens Longkanker Screenings Onderzoek”). Based on the Multicenter Italian Lung Detection (MILD) trial, Sverzellati et al29 reported a fourfold increased risk of CV events and mortality in subjects with CAC >400 compared with subjects with CAC ≤400. Chiles and colleagues described similar HRs for predicting CV death in a multivariable analysis of 9045 volunteers from the NLST.32 Interestingly, we found a non-negligible proportion of patients with intermediate-to-high CAC scores not to be under treatment with statins at the study accrual. This result further highlights the potential of the extra information from CAC in optimizing CV risk factors management in LCS, notably by enabling the identification of unaware patients that may benefit from CV prevention.

There is a lack of evidence throughout the literature supporting %LAA as a predictor of CV mortality in LCS. In the MILD trial, %LAA did not show any predicting value for CV events nor correlated with CAC.29 Stemmer et al10 observed that %LAA through a fully automated algorithm did not significantly predict CV mortality when adjusted for age, sex, and pack-years in participants from the NLST. Otherwise, we found that %LAA significantly predicted CV death. Moreover, %LAA was associated with increasing CAC, and both LDCT biomarkers negatively correlated with FEV1, in accordance with previous studies demonstrating a greater burden of CAC in COPD.35,36 Although these results may suggest that the long-debated mechanisms of interaction between COPD and CAC37 had a role in the association between %LAA and CV mortality, this hypothesis remains speculative as our study was not designed to explore underlying causative links.

%LAA showed a moderate discriminative ability to identify patients with airflow obstruction. The rationale of this analysis comes from previous evidence suggesting %LAA as a biomarker to raise suspicion of COPD in LCS.26 Such potential is attractive, as it would enable targeting resources for LCS participants that would benefit from comprehensive pulmonary function testing. The optimal %LAA threshold to identify patients with airflow obstruction was the same as a recent NLST analysis, which suggested that a finding of %LAA >1% should prompt clinical providers to assess pulmonary function.26 Nevertheless, similarly to that study, we found a sizeable proportion of subjects with low %LAA but airway obstruction and subjects with a preserved pulmonary function having relatively high %LAA values. It is reasonable to hypothesize that, at least to some degree, this discrepancy might be accounted for by airway thickening.38,39 Future studies integrating CT airway metrics with %LAA are warranted to further explore the discriminative ability of CT for COPD identification in LCS.

The limited value of %LAA for predicting LC is not surprising.40 Indeed, several studies reported that automated emphysema quantification relying on densitometry is not strongly associated with LC in LCS.40–42 It is worth emphasizing that this assumption cannot be extended to visual emphysema, which has been found to refine LC risk prediction across different settings.41,43,44 Nevertheless, a comparison between visual emphysema and %LAA was beyond the scope of our analyses.

The present study has some limitations. First, because the BioMILD trial primarily targeted LC, conventional CV risk factors were mostly unknown. However, measuring risk factors such as blood pressure and lipid levels is not part of current LCS protocols and would affect the logistics and feasibility of screening practice. Second, a postbronchodilator FEV1/FVC<0.70 is required to diagnose airflow obstruction, while we used prebronchodilator spirometry. Third, we did not include data on regional quantitative emphysema, which may have the potential to further refine risk prediction models for clinically relevant outcomes.45 Last, we could not perform analyses on COPD mortality due to the small number of cases.

In conclusion, fully automated CAC and %LAA provided by AI software have shown promise to refine patient risk stratification in LCS and to identify volunteers who may benefit from preventive intervention. Validating the present findings in other LCS cohorts would enable exploring their generalizability and provide further insights for optimizing personalized recommendations for LCS volunteers.

REFERENCES

1. National Lung Screening Trial Research Team, Aberle DR, Adams AM, Berg CD, et al. Reduced lung-cancer mortality with low-dose computed tomographic screening. N Engl J Med. 2011;365:395–409.
2. Humphrey LL, Deffebach M, Pappas M, et al. Screening for lung cancer with low-dose computed tomography: a systematic review to update the US Preventive services task force recommendation. Ann Intern Med. 2013;159:411–420.
3. Yip R, Jirapatnakul A, Hu M, et al. Added benefits of early detection of other diseases on low-dose CT screening. Transl Lung Cancer Res. 2021;10:1141–1153. TLC
4. Silva M, Picozzi G, Sverzellati N, et al. Low-dose CT for lung cancer screening: position paper from the Italian college of thoracic radiology. Radiol Med. 2022;127:543–559.
5. Schreuder A, Jacobs C, Lessmann N, et al. Combining pulmonary and cardiac computed tomography biomarkers for disease-specific risk modelling in lung cancer screening. Eur Respir J. 2021;58:2003386.
6. Tsai EB, Chiles C, Carter BW, et al. Incidental findings on lung cancer screening: significance and management. Semin Ultrasound CT MR. 2018;39:273–281.
7. Pinsky PF, Lynch DA, Gierada DS. Incidental findings on low-dose CT scan lung cancer screenings and deaths from respiratory diseases. Chest. 2022;161:1092–1100.
8. Mets OM, de Jong PA, Prokop M. Computed tomographic screening for lung cancer: an opportunity to evaluate other diseases. JAMA. 2012;308:1433–1434.
9. Isgum I, Prokop M, Niemeijer M, et al. Automatic coronary calcium scoring in low-dose chest computed tomography. IEEE Trans Med Imaging. 2012;31:2322–2334.
10. Stemmer A, Shadmi R, Bregman-Amitai O, et al. Using machine learning algorithms to review computed tomography scans and assess risk for cardiovascular disease: retrospective analysis from the National Lung Screening Trial (NLST). PLoS One. 2020;15:e0236021.
11. Pastorino U, Boeri M, Sestini S, et al. Baseline computed tomography screening and blood microRNA predict lung cancer risk and define adequate intervals in the BioMILD trial. Ann Oncol. 2022;33:395–405.
12. Bak SH, Kim JH, Jin H, et al. Emphysema quantification using low-dose computed tomography with deep learning-based kernel conversion comparison. Eur Radiol. 2020;30:6779–6787.
13. Koo HJ, Lee SM, Seo JB, et al. Prediction of pulmonary function in patients with chronic obstructive pulmonary disease: correlation with quantitative CT parameters. Korean J Radiol. 2019;20:683–692.
14. Suh YJ, Kim C, Lee JG, et al. Fully automatic coronary calcium scoring in non-ECG-gated low-dose chest CT: comparison with ECG-gated cardiac CT. Eur Radiol. 2022. doi: 10.1007/s00330-022-09117-3. [Epub ahead of print].
15. Kang HW, Ahn WJ, Jeong JH, et al. Evaluation of fully automated commercial software for Agatston calcium scoring on non-ECG-gated low-dose chest CT with different slice thickness. Eur Radiol. 2022. doi:10.1007/s00330-022-09143-1. [Epub ahead of print].
16. Vonder M, Zheng S, Dorrius MD, et al. Deep learning for automatic calcium scoring in population-based cardiovascular screening. JACC Cardiovasc Imaging. 2022;15:366–367.
17. van der Aalst CM, Denissen SJAM, Vonder M, et al. Screening for cardiovascular disease risk using traditional risk factor assessment or coronary artery calcium scoring: the ROBINSCA trial. Eur Heart J Cardiovasc Imaging. 2020;21:1216–1224.
18. Lee SM, Lee JG, Lee G, et al. CT image conversion among different reconstruction kernels without a sinogram by using a convolutional neural network. Korean J Radiol. 2019;20:295–303.
19. Vogelmeier CF, Criner GJ, Martinez FJ, et al. Global strategy for the diagnosis, management, and prevention of chronic obstructive lung disease 2017 report. GOLD Executive Summary. Am J Respir Crit Care Med. 2017;195:557–582.
20. Wan ES, Castaldi PJ, Cho MH, et al. COPDGene Investigators . Epidemiology, genetics, and subtyping of preserved ratio impaired spirometry (PRISm) in COPDGene. Respir Res. 2014;15:89.
21. Wan ES, Hokanson JE, Murphy JR, et al. GOLD-unclassified smokers in the COPDGene study. Am J Respir Crit Care Med. 2011;184:57–63.
22. Uno H, Cai T, Pencina MJ, et al. On the C-statistics for evaluating overall adequacy of risk prediction procedures with censored survival data. Stat Med. 2011;30:1105–1117.
23. Bhatt SP, Kazerooni EA, Newell JD Jr, et al. Visual estimate of coronary artery calcium predicts cardiovascular disease in COPD. Chest. 2018;154:579–587.
24. Oelsner EC, Carr JJ, Enright PL, et al. Per cent emphysema is associated with respiratory and lung cancer mortality in the general population: a cohort study. Thorax. 2016;71:624–632.
25. Johannessen A, Skorge TD, Bottai M, et al. Mortality by level of emphysema and airway wall thickness. Am J Respir Crit Care Med. 2013;187:602–608.
26. Labaki WW, Xia M, Murray S, et al. Quantitative emphysema on low-dose CT imaging of the chest and risk of lung cancer and airflow obstruction: an analysis of the national lung screening trial. Chest. 2021;159:1812–1820.
27. Rasmussen T, Køber L, Abdulla J, et al. Coronary artery calcification detected in lung cancer screening predicts cardiovascular death. Scand Cardiovasc J. 2015;49:159–167.
28. Jacobs PC, Gondrie MJ, van der Graaf Y, et al. Coronary artery calcium can predict all-cause mortality and cardiovascular events on low-dose CT screening for lung cancer. AJR Am J Roentgenol. 2012;198:505–511.
29. Sverzellati N, Cademartiri F, Bravi F, et al. Relationship and prognostic value of modified coronary artery calcium score, FEV1, and emphysema in lung cancer screening population: the MILD trial. Radiology. 2012;262:460–467.
30. Schreuder A, Jacobs C, Gallardo-Estrella L, et al. Predicting all-cause and lung cancer mortality using emphysema score progression rate between baseline and follow-up chest CT images: a comparison of risk model performances. PLoS One. 2019;14:e0212756.
31. Gendarme S, Goussault H, Assié JB, et al. Impact on all-cause and cardiovascular mortality rates of coronary artery calcifications detected during organized, low-dose, computed-tomography screening for lung cancer: systematic literature review and meta-analysis. Cancers (Basel). 2021;13:1553.
32. Chiles C, Duan F, Gladish GW, et al. Association of coronary artery calcification and mortality in the national lung screening trial: a comparison of three scoring methods. Radiology. 2015;276:82–90.
33. Takx RA, Išgum I, Willemink MJ, et al. Quantification of coronary artery calcium in nongated CT to predict cardiovascular events in male lung cancer screening participants: results of the NELSON study. J Cardiovasc Comput Tomogr. 2015;9:50–57.
34. Mascalchi M, Puliti D, Romei C, et al. Moderate-severe coronary calcification predicts long-term cardiovascular death in CT lung cancer screening: The ITALUNG trial. Eur J Radiol. 2021;145:110040.
35. Chandra D, Gupta A, Strollo PJ Jr, et al. Airflow limitation and endothelial dysfunction. unrelated and independent predictors of atherosclerosis. Am J Respir Crit Care Med. 2016;194:38–47.
36. Williams MC, Murchison JT, Edwards LD, et al. Evaluation of COPD longitudinally to identify predictive surrogate endpoints (ECLIPSE) investigators. Coronary artery calcification is increased in patients with COPD and associated with increased morbidity and mortality. Thorax. 2014;69:718–723.
37. Bhatt SP, Nath HP, Kim YI, et al. Centrilobular emphysema and coronary artery calcification: mediation analysis in the SPIROMICS cohort. Respir Res. 2018;19:257.
38. Charbonnier JP, Pompe E, Moore C, et al. Airway wall thickening on CT: relation to smoking status and severity of COPD. Respir Med. 2019;146:36–41.
39. Mohamed Hoesein FA, de Jong PA, Lammers JW, et al. Airway wall thickness associated with forced expiratory volume in 1 second decline and development of airflow limitation. Eur Respir J. 2015;45:644–651.
40. Smith BM, Pinto L, Ezer N, et al. Emphysema detected on computed tomography and risk of lung cancer: a systematic review and meta-analysis. Lung Cancer. 2012;77:58–63.
41. Wilson DO, Leader JK, Fuhrman CR, et al. Quantitative computed tomography analysis, airflow obstruction, and lung cancer in the Pittsburgh lung screening study. J Thorac Oncol. 2011;6:1200–1205.
42. Maldonado F, Bartholmai BJ, Swensen SJ, et al. Are airflow obstruction and radiographic evidence of emphysema risk factors for lung cancer? A nested case-control study using quantitative emphysema analysis. Chest. 2010;138:1295–1302.
43. de Torres JP, Bastarrika G, Wisnivesky JP, et al. Assessing the relationship between lung cancer risk and emphysema detected on low-dose CT of the chest. Chest. 2007;132:1932–1938.
44. Li Y, Swensen SJ, Karabekmez LG, et al. Effect of emphysema on lung cancer risk in smokers: a computed tomography-based assessment. Cancer Prev Res (Phila). 2011;4:43–50.
45. Bae K, Jeon KN, Lee SJ, et al. Severity of pulmonary emphysema and lung cancer: analysis using quantitative lobar emphysema scoring. Medicine (Baltimore). 2016;95:e5494.
Keywords:

artificial intelligence; calcium; emphysema; lung cancer

Supplemental Digital Content

Copyright © 2023 The Author(s). Published by Wolters Kluwer Health, Inc.