Lung disease severity, Coronary Artery Calcium, Coronary inflammation and Mortality in Coronavirus Disease 2019

Definition of clinically suspected COVID-19

Clinical variables considered at triage to define “clinically suspected COVID-19” were a) recent/current fever for at least 2 days, b) dry cough, c) respiratory distress during mild exercise or at rest. O₂ saturation<96% (<94% for patients with chronic pulmonary disease) using pulse oximetry was defined reduced O₂ saturation. The presence of at least one of the abovementioned clinical criteria + reduced O₂ saturation qualified the subject as clinically suspected COVID-19 and mandated fast-track HRCT, as the gatekeeper to admit patients to dedicated COVID-19 wards if showing interstitial pneumonia, even if SARS-CoV-2 viral nucleic acid laboratory results were pending. All patients were admitted to hospital after HRCT, either to COVID-19 specialized or general wards, based on the presence or absence of HRCT findings suggestive of COVID-19 pneumonia.

Definition of laboratory-confirmed COVID-19

The detection of SARS-CoV-2 viral nucleic acid in nasopharyngeal swabs was used as the only reference to adjudicate laboratory-confirmed COVID-19. This study complied with the edicts of the 1975 Declaration of Helsinki and was approved by the institutional ethics board of Parma University Hospital with expedited approval for COVID-19 studies (371/2020/OSS/AOUPR). Written informed consent was waived by the ethics commission due to the emergency state caused by COVID-19.

Data Collection

The electronic discharge letter and other available hospital electronic records were reviewed by a team of physicians working in Parma University Hospital. Clinical and laboratory information was collected on admission and during hospitalization by attending physicians and was then checked and recorded by the researchers. Patient data including demographics, medical history, comorbidities, available laboratory examinations, nasopharyngeal swabs tested with SARS-CoV-2 RT-PCR assay, or other intensive care measures and outcomes were collected.

High-resolution computed tomography

Patients were scanned with a 128-slice Somatom Definition Edge scanner (Siemens Medical Solutions, Forchheim, Germany), equipped with an integrated high-resolution circuit detector. Examinations were performed with patients in supine position, at full inspiration, without the use of contrast medium. Scan parameters were: collimation 128×0.6 mm; rotation time 0.33 ms; pitch 1.2; tube voltage 120 kVp; tube power 150 reference mAs; tube current modulation on. Scan data were reconstructed with novel model based iterative reconstruction algorithm (ADMIRE) keeping the noise at a low level and preserving intrinsic low-contrast details.

Interpretation of HRCT

Each scan was clinically evaluated by the radiologist in charge, all having more than 10-year experience, in particular using the following classification, similar to the recent Consensus Statement by the Radiological Society of North America (13). In our centre, the three following summary categories were used: a) presence of COVID-19 typical interstitial pneumonia, according to the presence of ground-glass opacities, bilateral multifocal patchy consolidation, and/or interstitial changes with a peripheral distribution, b) indetermined appearance, in case of absence of typical features and presence of multifocal, diffuse, perihilar, or unilateral ground-glass opacities with or without consolidation lacking a specific distribution and non-rounded or non-peripheral, c) negative for pneumonia. The report also focused per-protocol on the estimated semi-quantitative percentage of lung parenchyma involved (5 visually-assessed classes, <10%, 10-19%, 20-49%, 5079%, >80%).

Calcium scoring

Coronary calcium was classified using a threshold of 130 Hounsfield units (HU) involving 3 contiguous voxels for identification of a calcific lesion resulting in a minimum lesion area of 1.02 mm². The lesion score was calculated using the area density method, by multiplying the lesion area by a density factor derived from the maximal HU within the area, as described by Agatston (14). The CCS was calculated with commercially available software (CaScore; Siemens, Germany).

Peri-coronary Adipose Tissue (PCAT) Attenuation

To measure PCAT attenuation, we used a software package clinically validated for this use (AutoPlaque™ Version 2.5, Cedars-Sinai Medical Center) (15). Since in the current study we used high-quality and fast HRCT machines, but the scans were not ECG-triggered nor contrast medium used, we customized the methods otherwise typically used for PCAT attenuation measurement; we manually identified the coronary artery lumen and adjusted the vessel contours. The left anterior descending coronary artery (LAD) was the only coronary artery vessel reliably visualized in most patients, due to less motion artifacts compared with right coronary artery, in the non-contrast, non ECG-triggered HRCT datasets for the aim of PCAT attenuation measurement. Figure 1 shows the technical steps to measure PCAT attenuation. We traced the proximal 40-mm segment of the LAD and 3-dimensional layers within radial distance from the outer coronary wall equal in thickness to the average diameter of the vessel were constructed automatically, using the abovementioned semiautomated software Within the predefined volume of interest, voxels with tissue attenuation ranging from −190 up to −30 HU were considered as adipose tissue. We quantified the mean PCAT attenuation based on the attenuation histogram of peri-coronary fat within the range –190 HU to – 30 HU (10,15,16).

• Download figure
• Open in new tab

Figure 1.

Identification of the left anterior descending coronary artery from HRCT dataset, manual tracing of the vessel course and, finally, peri-coronary fat attenuation measurement.

HRCT, high-resolution computed tomography.

Outcome

The primary endpoint was incidence of in-hospital death in laboratory-confirmed COVID-19 patients. Successful clinical course leading to hospital discharge comprised relieved clinical symptoms, normal body temperature, normal O₂ saturation with no need of O₂ therapy and at least partial decrease of inflammation as shown by C-reactive protein (CRP) testing.

Statistical Analysis

No statistical sample size calculation was performed a priori, and sample size was equal to the number of patients enrolled during the study period. Categorical variables are expressed as number of patients (percentage) with 95% CIs, and continuous variables as mean□(SD) or median (interquartile range [IQR]) as appropriate. The means for continuous variables were compared using independent group χ2 tests when the data were normally distributed, otherwise, the Mann-Whitney test was used. Proportions for categorical variables were compared using the 2 test, although the Fisher exact test was used when data were limited. Stepwise multiple logistic regression was used to assess the relationship between the demographic and clinical variables, laboratory variables and, as a final step, HRCT variables and the primary end point of death. Variables that were not normally distributed such as CCS, CRP, and D-dimer were log-transformed, as is standard. We considered an increase in PCAT attenuation in steps of 1 HU and 5 HU; the results were similar and increase by 5 HU were reported for multivariable analysis. All variables with p<0.1 on univariable analysis were considered for the inclusion into multivariable logistic regression models. A 2-sided p<0.05 was considered statistically significant. All statistical analyses were performed with Stata statistical software, version 15.0 (StataCorp LLC, USA).

Suspected COVID-19 overall group (n=500)

Among the initially 500 consecutive patients with clinically-suspected COVID-19 who underwent HRCT, n=145 died (29%) during their index hospitalization, at a median of 6 (lower-upper quartile, 4-10) days after admission, while n=355 were discharged home alive after a median of 6 (lower-upper quartile, 2-13) days. n=279 (55.8%) were found positive for RT-PCR viral nucleic acid detection in the nasopharyngeal swab obtained either at presentation, or repeated in the next 2 days if the first was negative, and they received the diagnosis of laboratory-confirmed COVID-19.

In this initial cohort n=399 patients had typical findings of COVID-19 interstitial pneumonia on HRCT, n=66 were negative for COVID-19 pneumonia and n=35 were indeterminate cases. Figure 2 shows the mortality rate in the overall n=500 population, based on HRCT findings of typicality for COVID-19, but also based on laboratory detection of SARS-CoV-2 nucleic acid in nasopharyngeal swabs. HRCT findings for COVID-19 demonstrated the following prevalence of SARS-CoV-2 nucleic acid detected in patients’ nasopharyngeal swabs: 248/399 (62%) in COVID-19 typical HRCT, 19/35 (54%) in indeterminate HRCT, 12/66 (18%) in negative HRCT.

• Download figure
• Open in new tab

Figure 2.

Histograms describing mortality rate in groups classified based on HRCT results or SARS-CoV-2 PCR assay results and also histograms describing mortality rates based on visual estimate of the percentage lung parenchyma diseased, either in COVID-19 laboratory-confirmed patients or not.

HRCT, high-resolution computed tomography.

Figure 2 shows that the visual semi-quantitative estimate of diseased lung parenchyma, as reported by the radiologist in charge (here summarized in 3 classes, 0-19%, 20-49% and ≥50%) was associated with mortality, in laboratory-confirmed COVID-19 patients but also in patients with negative report for SARS-CoV-2 nucleic acid in the nasopharyngeal swabs.

Laboratory-confirmed COVID-19 group (n=279)

These patients with laboratory-confirmed COVID-19 formed the final study cohort, described and tested for univariable and multivariable association between demographic, laboratory, standard and novel variables on HRCT and in-hospital mortality.

In this cohort, n=170 (61%) patients were discharged home alive at a median of 8 days (lower-upper quartile, 5-16) after admission, while n=109 (39%) died during hospitalization, at a median of 6 days (lower-upper quartile, 4-10). Table 1 shows the baseline demographic, clinical, laboratory and HRCT characteristics in the study group and also divided based on dead or alive status at discharge. HRCT was positive for interstitial pneumonia in 248 cases (89%).

Comparing patients discharged alive with patients who died, the median (lower-upper quartiles) age was 65 y/o (56-76) vs 77 y/o (72-82), (p<0.001), n=95 (56%) vs n=74 (68%) were males, p=0.061, dyslipidemia was present in n=29 (17%) vs n=30 (28%), p=0.052, n=15 (9) vs n=26 (24), p=0.001 had prior cardiovascular disease, D-dimer median was 723 ng/ml (lower-upper quartile, 495-1119) vs 1.083 ng/ml (lower-upper quartile, 704-2248), p<0.001, CRP median 78.15 mg/L (lower-upper quartile, 36.8-132.5) vs 148.3 mg/L (lower-upper quartile, 91-209.4), p<0.001, median CCS was 17 (lower-upper quartile, 0-248) vs 188.7 (lower-upper quartile, 20.3-671.6), p<0.001 and the PCAT attenuation median was −76.4 HU (lower-upper quartile, −79.3, −76.6) vs - 68.6 HU (lower-upper quartile, −75.9, −63.3), p<0.001. Figure 3 graphically shows the distribution of age, main laboratory and imaging variables according to the in-hospital outcome. Increased heart rate (>100 beats per minute) was found during HRCT scan in 46 (27%) of the discharged-alive patients and in 22 (20%) patients who died in hospital (p=0.245).

• Download figure
• Open in new tab

Figure 3.

Distribution of age, D-dimer and CRP values, and imaging data acquired from HRCT, based on primary outcome of in-hospital death.

CRP, C-reactive protein; HRCT, high-resolution computed tomography.

Table 2 (left column), shows available demographic, clinical, laboratory and HRCT characteristics, including the CCS and PCAT attenuation, tested with simple univariable logistic regression. The following variables were significantly associated with the end point: age (OR 1.073, 95%CI 1.05-1.10, p<0.001), male sex (OR 1.67, 95%CI 1.0-2.76, p=0.046), dyslipidemia (OR 1.85, 95%CI 1.03-3.30, p=0.038), prior cardiovascular disease (OR 3.27, 95%CI 1.63-6.45, p=0.001), D-dimer (OR 1.95. 95%CI 1.45-2.63, p<0.001), CRP (OR 2.12, 95%CI 1.55-2.92, p<0.001), HRCT extent of diseased lung parenchyma (OR 2.11, 95%CI 1.60-2.78, p<0.001), CCS (OR 1.27, 95% CI 1.14-1.39, p<0.001) and PCAT attenuation (increase per 5 HU, OR 1.53, 95%CI 1.31-1.79, p<0.001). First a multivariable logistic regression model comprising clinical and laboratory variables was created (Table 2 central column), then the HRCT variables were added in a second model (Table 2 right column). This last model showed that age (OR 1.07, 95%CI 1.04-1.11, p<0.001), D-dimer (OR 1.48, 95% CI 1.02-2.16, p=0.041), CRP (OR 1.81, 95%CI 1.25-2.61, p=0.002), percentage of affected lung parenchyma (OR 1.68, 95%CI 1.20-2.35, p=0.002) and PCAT attenuation (OR 1.58, 95%CI 1.30-1.91, p<0.001) were finally significantly associated with in-hospital mortality. For the sake of completeness we also report the final multivariable model after the variable age was removed due to the moderate but significant collinearity (Spearman’s rho =0.557, p<0.001) with CCS (online appendix, Table A), instead of maintaining both variables as in Table 1.

The first model with clinical+laboratory variables showed a receiver-operating-characteristic (ROC) curve area of 0.808 (95%CI 0.756-0.859), which was significantly increased by the addition of HRCT variables to the model, with a resulting ROC area of 0.861 (95%CI 0.8170.905), p=0.004.

Figure 4 shows the comparison between two paradigmatic cases, one with lower PCAT attenuation indicating no coronary inflammation, and the other one showing higher PCAT attenuation (closer to 0 HU), indicating coronary inflammation.

• Download figure
• Open in new tab

Figure 4.

Comparison between the PCAT attenuation assessment in a COVID-19 patient with more inflamed LAD coronary artery (left panel, same patient shown in figure 1) and another COVID-19 patient with less inflamed LAD coronary artery (right panel).

PCAT, peri-coronary adipose tissue; LAD, left anterior descending.

Potential Cardiac Mechanisms Underlying Increased Mortality

Before the results of the current study we could not exclude that the prolonged hypoxemia typical of symptomatic COVID-19 represented a stress condition leading to myocardial injury, particularly in the elderly patients with asymptomatic CAD, through a demand-supply mismatch. Since the CCS is able to indirectly measure pre-existent CAD burden, and it was not independently associated with in-hospital death, we may reasonably conclude that pre-existent (known or unknown) CAD has no or limited role in worsening the outcome of COVID-19 patients. The finding that a higher coronary inflammation status in COVID-19 patients is associated with mortality is instead a novel finding and may shed some light on the possible mechanisms responsible for the frequently-found myocardial injury, associated with higher mortality in COVID-19 patients (4-7). However, the current study did not address the incidence of myocardial injury because cardiac troponins were not routinely measured in our COVID-19 cohort, so that while we can affirm that coronary inflammation is associated with in-hospital death, we cannot speculate on the relation between coronary inflammation and myocardial injury. The finding of a disproportionate increase in CRP, together with higher PCAT attenuation in patients with fatal outcome, supports the existing hypothesis that inflammation in COVID-19 is a key driver of outcome, being a possible mechanism of coronary plaque destabilization, as well as a potential trigger for cardiac arrhythmias (18). This opens the possibility that the potent anti-inflammatory drugs already in testing phase in prospective trials may help decrease mortality.

COVID-19 diagnosis by viral nucleic acid isolation or radiologically by HRCT and the Bayes theorem

Although the demonstration of SARS-CoV-2 nucleic acid in the nasopharyngeal swab remains the reference for COVID-19 diagnosis (12), it has several limitations, which may limit its diagnostic sensitivity to lower than ideal. The use of HRCT as an initial gatekeeper, as performed in the current study in presence of symptoms and signs suggestive of the disease, may prove useful during massive outbreaks, as the one the world is facing. In an outbreak setting the inherently limited specificity of HRCT – which aims at the detection of a typical interstitial pneumonia, and not specifically at the detection of the virus – is sufficient to produce a moderate-to-high positive predictive value, due to the high prevalence of disease and high pre-test probability. In turn, HRCT has a very high sensitivity and negative predictive value, which may be preferable in the gatekeeping setting of highly contagious diseases (19).

COVID-19 prognosis by HRCT

The current study also demonstrated that HRCT is capable to accurately risk-stratify patients with acute respiratory distress and laboratory-confirmed COVID-19 through simple semi-quantitative estimation of the extent of lung parenchyma affected by pneumonia, which represents the pulmonary key clinical manifestation of COVID-19. Interestingly, such risk-stratification capability, although to a lesser degree, was demonstrated also in patients in whom viral nucleic acid was not detected in at least two nasopharyngeal swabs (Figure 2). Whether this is due to RT-PCR on swabs being insufficiently sensitive or to HRCT being able to stratify also non-COVID-19 patients would require an independent reference method for COVID-19 disease, which is not available at this time.

Limitations

Our study has limitations. First, we cannot unequivocally classify those patients with HRCT findings typical of COVID-19 who tested negative to viral nucleic acid detection in nasopharyngeal swabs, since the direct virus detection remains the WHO recommended standard for COVID-19 diagnosis (12). The mounting clinical experience during the outbreak suggests that most such patients may be affected by COVID-19, since the nasopharyngeal detection of the virus is limited in sensitivity and the optimal timing to perform the swab unknown. Second, as a retrospective study, specific information regarding the prevalence of optional variables, for example the presence of myocardial injury by high-sensitivity troponins, were not available, owing to the difficult clinical conditions in the isolation wards and the urgency of containing the epidemic. Another limitation was that the presence of obesity was adjudicated as a binary variable by the caring physicians (body mass index>29 kg/m2), but individual granular data are not available.

Third, though the methods used to calculate the CCS in non-contrast and non ECG-triggered HRCT scans have been previously validated (11,20,21), the HRCT protocol is not the gold-standard to measure the CCS, and this scan protocol has never been used previously for the measurement of PCAT attenuation. To our knowledge, ours is the first study to report such PCAT attenuation measurement and its clinical results. The absolute value of fat attenuation in Hounsfield units might potentially differ between an ideal ECG-triggered and contrast computed tomography angiogram and a non-contrast, non-ECG triggered HRCT as available in the current study, although a good correlation of between PCAT measurement is now documented between precontrast and postcontrast scans (22). However, since all patients underwent the same modality of HRCT scans, the relative comparison between alive and dead patients in our study should be unbiased by this limitation, even more considering there was no difference in the number of patients with an increased heart rate (>100 beats per minute, table 1) which in theory could lead to partial volume effect of coronary artery lumen and its surroundings. Further we measured PCAT attenuation only in one coronary artery (LAD), because it was the one technically best visualized (most stable) in the given conditions of HRCT scans. Prior literature robustly supports that PCAT attenuation does not differ significantly among coronary arteries in a given individual, and right coronary artery or LAD may be used alternatively as inflammatory markers of overall coronary inflammatory status (15-17).