Chest CT–derived pulmonary artery enlargement at the admission predicts overall survival in COVID-19 patients: insight from 1461 consecutive patients in Italy

Clinical data were collected by each center according to a centralized electronic case report form (CRF). Clinical data at the admission consisted of the following: demographic characteristics (sex and age), comorbidities (hypertension, diabetes, chronic lung disease, cardiovascular disease), and laboratory tests: white blood cell count (WBC), creatinine, C-reactive protein (CRP), lactate dehydrogenase (LDH), troponin I, interleukin-6, and D-dimer. Outcome data were oro-tracheal intubation and death.

The derivation cohort consisted of patients from the three hospitals (OSR, PCZ, BOL) that completed the CRF and shared chest CT images for centralized analysis within May 20, 2020, established as the first deadline for data collection. Patients from the remaining four participating hospitals (BGE, CRE, SGM, CCM), who provided data and CT images within the end of June, were used for the external validation. Prior to the analysis, data were cross-checked with medical charts and verified by data managers and clinicians for accuracy.

A total of 1469 consecutive patients fulfilled the inclusion criteria: 68 of the 829 patients of the derivation cohort and 7 of the 640 patients of the validation cohort were excluded for lacking data on comorbidities or follow-up (Fig. 1). The median follow-up was 51 days for the derivation cohort and 63 days for the validation cohort.

Complete demographic and clinical features of both cohorts are presented in Table 1.

The primary outcome was overall survival (OS) measured from admission to the emergency department. Secondary outcomes were in-hospital mortality and time-to-orotracheal intubation from admission to the emergency department.

All chest CT examinations were performed on multidetector scanners with at least 16 detector rows. The list of scanners and acquisition protocols are reported in Supplementary Materials (ESM 1). Briefly, all volumetric chest scans were sent to the core-lab where they were reformatted at a 2.5-mm slice thickness without gap or overlap. Images were reconstructed with sharp kernel for lung parenchyma evaluation and a soft kernel for mediastinum evaluation, and they were visualized using a standard window (lung: width 1400 HU, center − 450 HU; mediastinum: width 350 HU, center 40 HU).

The analysis was performed by a radiologist with 9 years of experience in cardiothoracic imaging, blinded to all clinical data in order to reduce the possible bias in the analysis and to increase homogeneity and reliability of data. Parameters of lung involvement (residual respiratory lung reserve, pneumonia extension, and features) and pulmonary artery metrics were extracted (Fig. 2).

Residual respiratory lung reserve was measured as well-aerated lung volume, quantified using automatic software (IntelliSpace Portal 8.0, Philips Medical Systems) based on a previously defined HU threshold value [17]. Pneumonia extension was scored in 0% (absent, score 0), 1–25% (minimal, score 1); 26–50% (mild, score 2); 51–75% (moderate, score 3); and > 75% (severe, score 4). Qualitative features of pneumonia were scored as follows: score 0 for absent pneumonia, score 1 for prevalent ground-glass opacities (GGOs), score 2 for prevalent consolidation, and score 3 for GGOs and consolidation equally represented. The metrics of pulmonary arteries included the measurement of the MPAD at the level of its bifurcation, of the left (LPAD) and of the right pulmonary artery diameter (RPAD) at their origin. MPAD was classified according to the severity classification system for the diagnosis and prognosis of PH [13] in four classes: normal (≤ 27 mm for females and ≤ 29 mm for males, according to the Framingham sex-specific normative values [16]); mild enlargement (from > 27 to < 31 mm for females and > 29 to < 31 mm for males); moderate enlargement (from ≥ 31 to 34 mm for both sexes), and severe enlargement (> 34 mm for both sexes). To assess inter-observer reliability, a second reader with 5 years of experience in cardiothoracic imaging, blinded to the measurements of the first reader, assessed MPAD and pneumonia score.

Cox’s regression analysis was employed for evaluating the role of MPAD classes [13, 16] in predicting the OS, un-adjusting and adjusting for demographics and comorbidities (age, sex, hypertension, diabetes, history of cardiovascular disease, and chronic lung disease). Multivariable Cox’s regression analysis was employed for evaluating the simultaneous role of CT parameters (MPAD, pneumonia extension, well-aerated lung volume), laboratory tests (WBC, CRP, creatinine), demographics, and comorbidities in predicting OS. The final model was obtained with a backward variable selection with a removal significant level of 0.05. Model calibration was performed as detailed in Supplementary Materials (ESM 1). Model discrimination was evaluated with the optimism corrected c-index. To assess the role of gender in categorized MPAD with respect to the outcome, the final Cox model was estimated adding an interaction term between the two variables.

The adjusted Cox regression analysis, the calibration curve, and the c-index of the final multivariable Cox’s model were computed on the validation cohort using CT data from two independent readers.

Inter-observer reliability was assessed with Rothery’s non-parametric intra-class correlation coefficient (ICC) [18] for MPAD and Cohen’s kappa coefficient for the pneumonia score.

Missing data were not imputed; thus, each analysis considered only complete cases for the variables used in the analysis. All tests were 2-sided and the significance level was set at 0.05. Confidence intervals (CIs) were computed at a 95% level. All statistical analyses were performed using R 3.5.0 (http://​www.​R-project.​org/​). All details are reported in Supplementary Materials (ESM 1).

The median age was 69 (interquartile range IQR, 58–77) years and 537 (70.6%) patients were male (Table 1). Hypertension was the most frequent comorbidity (484 patients, 63.6%) and median oxygen saturation was 92% (IQR, 88–95%).

At chest CT, 752 patients (98.8%) had pneumonia, with prevalent GGOs (432 patients, 56.8%). Pneumonia extension greater than 50% was recorded in 139 cases (18.3%).

Median MPAD was 27 [25–29] mm. Twenty-one (2.8%) patients were not hospitalized after emergency department evaluation, while 459 (60.3%) patients were discharged within 40 days from admission with a median hospitalization length of 11 (IQR, 7–17) days. The cumulative incidence of orotracheal intubation at 40 days was 21.28%. At 40 days, non-survivors were 182/761 patients (23.9%), with death occurring at a median time from an admission of 7 (IQR, 4–12) days. The characteristics of survivors and non-survivors are reported in Table 2. OS at 40 days was 76.08% (95% CI: 73.11–79.17%).

Out of 761 patients, 86 (11.3%) had mild enlargement, 98 (12.9%) moderate enlargement, and 36 (4.7%) had severe MPAD enlargement.

The mortality rate was 25.6% (22/86), 38.8% (38/98), and 47.2% (17/36) in patients with mild, moderate, and severe MPAD enlargement, respectively, while it was 19.4% (105/541) in patients with normal MPAD (global Fisher’s exact test p < 0.001). Indeed, the classes of MPAD were associated with a different risk of death (40-day overall survival [95% CI]: for normal 80.56% [77.32–83.99%], for mild enlargement 74.40%[65.72–84.23%], for moderate enlargement 61.22% [52.30–71.67%], and for severe enlargement 52.78% [38.75–71.89%]; log-rank test p < 0.001) (Fig. 3a). At Cox’s regression analysis, moderate to severe MPAD enlargement (≥ 31 mm) resulted to predict a significantly reduced overall survival with respect to normal MPAD (hazard ratio, HR [95% CI]: mild enlargement vs normal 1.387 [0.876–2.196], p = 0.163; moderate enlargement vs normal 2.311 [1.594–3.350], p < 0.001; severe enlargement vs normal 3.001 [1.797–5.010], p < 0.001). Similar results were obtained also at unadjusted Cox’s regression analysis (Table S1). Therefore, in the subsequent analyses, the MPAD was dichotomized into moderate-severe MPAD enlargement (≥ 31 mm) vs normal-mild MPAD enlargement (< 31 mm). The percentage of moderate-severe MPAD resulted to be similar between genders (M: 18.6% vs F: 15.2%, p = 1) (Table S2). The new categorization achieved HR = 1.741 (95% CI: 1.253–2.418, p < 0.001) at adjusted Cox’s regression (at unadjusted analysis: HR = 2.367 [1.725; 3.249], p < 0.001) (Fig. 3b).

Moderate-severe MPAD enlargement remained associated with a higher risk of death at multiple Cox’s regression analysis also when pneumonia extension, well-aerated lung volume, laboratory tests, demographics, and comorbidities were considered (HR [95% CI]: 1.592 [1.154–2.196], p = 0.005) (Table 3). Figure 4 reports two exemplifying cases with normally and moderately enlarged MPAD, respectively. Besides MPAD, the other variables associated with higher risk of mortality in the model were old age (HR [95% CI]: 1.063 [1.052–1.075], p < 0.001), male gender (1.586 [1.079–2.329], p = 0.019), higher creatinine (2.468 [1.824–3.339], p < 0.001), and greater pneumonia score (1.397 [1.084–1.799], p = 0.010), while greater well-aerated lung volume was a protective factor (0.699 [0.581–0.841], p < 0.001). The final model showed a good calibration (Figure S1) and discrimination (c-index [95% CI]: 0.826 [95% CI: 0.796–0.851]).

In the validation cohort, the median age was 69 years (IQR, 59–77 years) and 421 (66.5%) patients were males (Table 1). Hypertension was the most frequent comorbidity (285 patients, 45%) and median oxygen saturation was 92% (IQR, 88–96%).

Most of the patients (626, 98.9%) had pneumonia at chest CT, with prevalent GGOs (328 patients, 51.8%). Thirty-one percent of cases (196) had lung involvement greater than 50%.

Mild, moderate, and severe MPAD enlargement was found in 58 (9.2%), 101 (15.9%), and 33 (5.2%) patients, respectively.

Inter-reader agreement was excellent for both MPAD (ICC [95% CI]: 0.937 [0.931–1]) and pneumonia score (Cohen’s kappa [95% CI]: 0.886 [0.856–0.916]).

The mortality rate of the validation cohort was 22.1% (140/633 patients). OS at 40 days was 77.69% (95% CI = 74.50–81.03%).