Early Serial Echocardiographic and Ultrasonographic Findings in Critically Ill Patients With COVID-19

Background Cardiac function of critically ill patients with COVID-19 generally has been reported from clinically obtained data. Echocardiographic deformation imaging can identify ventricular dysfunction missed by traditional echocardiographic assessment. Research Question What is the prevalence of ventricular dysfunction and what are its implications for the natural history of critical COVID-19? Study Design and Methods This is a multicenter prospective cohort of critically ill patients with COVID-19. We performed serial echocardiography and lower extremity vascular ultrasound on hospitalization days 1, 3, and 8. We defined left ventricular (LV) dysfunction as the absolute value of longitudinal strain of < 17% or left ventricle ejection fraction (LVEF) of < 50%. Primary clinical outcome was inpatient survival. Results We enrolled 110 patients. Thirty-nine (35.5%) died before hospital discharge. LV dysfunction was present at admission in 38 patients (34.5%) and in 21 patients (36.2%) on day 8 (P = .59). Median baseline LVEF was 62% (interquartile range [IQR], 52%-69%), whereas median absolute value of baseline LV strain was 16% (IQR, 14%-19%). Survivors and nonsurvivors did not differ statistically significantly with respect to day 1 LV strain (17.9% vs 14.4%; P = .12) or day 1 LVEF (60.5% vs 65%; P = .06). Nonsurvivors showed worse day 1 right ventricle (RV) strain than survivors (16.3% vs 21.2%; P = .04). Interpretation Among patients with critical COVID-19, LV and RV dysfunction is common, frequently identified only through deformation imaging, and early (day 1) RV dysfunction may be associated with clinical outcome.

The ongoing COVID-19 pandemic is the most lethal in recent history, with > 250 million people infected and > 5 million deaths. 1 Cardiac complications reportedly are common among patients with COVID-19. The first case series of critically ill patients with COVID-19 in the United States documented probable cardiomyopathy in one-third of patients, consistent with early reports from China. 2,3 Whether this cardiac dysfunction is the result of cytokine storm, viral myocarditis, or acute cor pulmonale resulting from elevated right ventricle (RV) afterload is unclear. Complicating interpretation of these data are reports in patients with COVID-19 of a high frequency of VTE, which also may affect RV afterload. 4 A number of unanswered questions remain. Among critically ill patients with , what are the frequency and implication of myocardial dysfunction identified with the more sensitive and specific method of deformation imaging? 5,6 What is the time course of echocardiographic abnormalities? What is the frequency of DVT and the possible clinical role of screening for it? To address these questions, we performed a prospective, multicenter study of critically ill patients with COVID-19 in which serial echocardiographic and vascular ultrasound imaging were performed during the first week of ICU admission.

Study Design
This was a multicenter, prospective, observational cohort study conducted within the Influenza and Other Viruses in the Acutely Ill Network. We enrolled patients at eight acute care medical centers in eight cities on hospitalization day 1 between September 2020 and January 2021. The protocol was approved by the Institutional Review Board (Identifier: 200973) and the Institutional Review Board (Identifier: 21-073). All participants (or their surrogates) provided informed consent.

Patients
We used a convenience sample of adult patients (aged $ 18 years) admitted with a confirmed diagnosis of COVID-19 who were critically ill, defined as receiving any of the following therapies: highflow nasal oxygen, noninvasive ventilation for indication other than sleep apnea, invasive mechanical ventilation, continuous renal replacement therapy, or IV vasoactive medications (either vasopressor or inotrope). We excluded patients who were expected to die within 24 h or who had a contraindication to clinicianperformed point-of-care ultrasound evaluation.

Transthoracic Echocardiography and Point-of-care Ultrasound
Patients underwent limited transthoracic echocardiography on study days 1, 3, and 8 while they remained hospitalized. This imaging protocol has been applied successfully in critically ill patients in prior studies (e-Appendix 1). 7 Where capability existed, complete transthoracic echocardiography was performed. Imaging was performed by a cardiologist, cardiac sonographer, or intensivist or emergency physician trained in critical care echocardiography and point-of-care ultrasound. Lower extremity vascular ultrasound consisted of two-point compression ultrasonography, that is, imaging bilateral femoral and popliteal veins. All imaging protocols were in adherence to guidelines on imaging patients with COVID-19. 8 If the patient was in prone position, assessment was performed when the patient was supine, if possible. Study images were anonymized before analysis.

Ultrasound Interpretation
All imaging data were interpreted by clinicians masked to clinical status. Cardiac images were interpreted offline by an advanced cardiac sonographer (T. D. O.) with physician overread (M. J. L.) or by direct physician read (S. P. D.). Vascular ultrasound images were interpreted directly by physicians (M. J. L. and S. P. D.). Both physician interpreters were board certified in critical care echocardiography and ultrasonography. Deformation imaging using speckle tracking to measure ventricular longitudinal strain (which assesses the extent to which adjacent portions of the myocardium move closer to each other during systole) was performed with Image Arena (TomTec). In cases where limited image quality prevented Take-home Points Study Question: How does ventricular function change over time during the first week of critically ill COVID-19, and is it associated with clinical outcomes? Results: Deformation imaging (ventricular strain) was abnormal during the first week of illness, while traditional assessments like ejection fraction were normal. One-third of patients with left ventricular dysfunction improved over the first week. Nonsurvivors had worse right ventricular strain. Interpretation: Ventricular dysfunction is common in critically ill patients with COVID-19, and is often only identified using deformation imaging. Ventricular function often changes during the first week of COVID-19 illness. measurement of global ventricular longitudinal strain of all 16 segments of the heart, we substituted longitudinal strain from the apical four-chamber only, consistent with prior studies. 9, 10 We did not use strain assessments in echocardiograms demonstrating arrhythmia-inadequate image quality. RV free-wall strain was calculated from the RV-focused view, if available, and from standard apical four-chamber view if not. We defined left ventricle (LV) systolic dysfunction as either left ventricular ejection fraction (LVEF) of < 50% or absolute value of LV longitudinal strain of < 17%. 11 We defined RV systolic dysfunction as RV fractional area change (FAC) of < 35%, tricuspid annulus systolic planar excursion (TAPSE) of < 1.7 cm, or absolute value of RV free-wall strain of < 22%. We selected RV FAC and TAPSE based on ease of measurement and ubiquity in clinical practice and chose thresholds based on published guidelines or reference studies. [12][13][14] Vascular ultrasound was scored as thrombus absent or present depending on whether a thrombus was visualized.

Additional Clinical Data
We collected demographic data and presence of medical comorbidities. At the time of each ultrasound assessment, we recorded vital signs; ventilator settings; receipt of fluid and vasoactive medications; and adjunctive therapies for ARDS, such as prone positioning, neuromuscular blockade, or extracorporeal life support. We converted vasopressor infusion rates into norepinephrine-equivalent dosing, per previously described methods. 15 We recorded clinically obtained arterial blood gas and serum lactate levels. We calculated Sequential Organ Failure Assessment scores. 16 Patients were assessed daily for clinical evidence of atrial fibrillation, atrioventricular conduction block, or other arrhythmia. Patients underwent research laboratory assessments of troponin-I and B-type natriuretic peptide (BNP) on days 1, 3, and 8. We recorded in-hospital death and in-hospital clinical diagnosis of VTE.

Statistical Analysis
We compared clinical characteristics of survivors and nonsurvivors using c 2 tests for categorical variables and Wilcoxon tests for continuous variables. We compared patients with and without LV dysfunction on day 1 using a similar approach. Day 1 measurements were illustrated graphically using violin plots and scatterplots. We characterized echocardiographic changes over time using descriptive statistics and longitudinal spaghetti plots. We calculated the proportion of patients with thrombosis identified by research ultrasound examinations. The relationship between LV strain and LVEF was shown graphically using a scatterplot. We did not stratify analyses by treatments, such as fluid or vasoactive therapy, because of heterogeneity in treatments received and a moderate sample size that limited statistical power for subgroup comparisons. For longitudinal comparisons, we accounted for within-patient correlation by using the Friedman test (nonparametric repeated measure analysis of variance) and Wald c 2 test (generalized linear mixed-effects model) depending on the nature of the variable.
In an exploratory analysis, we predicted LV strain from LVEF by performing a linear regression. Residuals were compared using the Wilcoxon rank-sum test, stratifying by in-hospital mortality. In this analysis, a higher residual indicated a greater-thananticipated difference between a patient's actual strain and what the model predicted from ejection fraction data. For ease of graphic representation and reporting, LV strain and RV free-wall strain were converted to absolute values; troponin-I and BNP values were log transformed for graphic representation. P values of < .05 were considered statistically significant; no correction for multiple comparisons was used. The statistical analysis was conducted using R version 4.0.4 software (R Foundation for Statistical Computing).

Frequency and Implications of Systolic Dysfunction
LV dysfunction was present at admission in 38 of 110 patients (34.5%). Nineteen of those patients (50.0%) with LV dysfunction received a diagnosis only via deformation imaging. Median LVEF was 62% (IQR, 52.4%-69.2%) at baseline, whereas median absolute value of LV strain was 16.0% (IQR, 13.7%-19.1%) at baseline. Day 1 LV function could not be determined in 20 patients (18.2%) because of limited image quality. In 22 patients, day 1 LV strain was measured using only the apical four-chamber view. Day 1 RV free-wall strain was measured in 56 patients. We noted no significant difference between survivors and nonsurvivors with respect to day 1 LV strain ( Patients with LV dysfunction on day 1 were significantly more likely to have pre-existing pulmonary disease (31.6% vs 9.6%; P ¼ .009), and rates of pre-existing cardiovascular disease were similar (63.2% vs 55.8%; P ¼ .48). We observed no other differences between patients with and without LV dysfunction in frequency of arrhythmia, peak BNP or troponin values, or in-hospital mortality ( Table 2). Additional comparisons between normal and abnormal LV function are displayed in Table 2. chestcc.org We noted that LV strain, when measurable, identified LV dysfunction in 41% of patients with normal LVEF (Fig 1). Patients with normal LVEF and normal LV strain showed a mortality of 27.3% (Fig 1,

Serial Echocardiographic Assessments
Characteristics of cardiac function over time are presented in Table 3 and Figures 2 and 3. Median absolute value of LV strain was abnormal across study assessments on days 1, 3, and 8 (16.0%, 16.0%, and 18.2%, respectively), whereas median LVEF was normal (62%, 62%, and 65%, respectively) ( Table 3, Fig 2). Additionally, median RV free-wall strain was mildly abnormal on days 1, 3, and 8 (19.2%, 18.3%, and 20.8%, respectively), whereas RV FAC (39%, 40%, and 43%, respectively) and TAPSE (2.0 cm, 2.0 cm, and 2.0 cm, respectively) consistently were normal. The RV to LV end-diastolic area consistently was 0.9 in all three assessments (Table 3). Of patients who demonstrated normal LV function on day 1, 11.5% progressed to LV dysfunction on the day 8 echocardiogram, whereas 34.2% of those with LV dysfunction improved to normal on the day 8 echocardiogram. Figure 3 demonstrates changes in individual participants over time. Although we noted improvements in in LVEF (P < .001) and changes in RV to LV ratio over time (P ¼ .04), we observed no significant temporal trends in LV strain or in RV strain among the study cohort over the study period ( Table 3).

Discussion
This was a prospective, multicenter, observational study of critically ill patients with COVID-19 using serial echocardiographic evaluations over the first week of hospitalization for severe illness. We identified a high proportion of ventricular dysfunction, much of which was identified only through deformation imaging. This proportion is comparable with other observations of similar cohorts that used clinically obtained echocardiograms. 17,18 Deformation imaging revealed a much higher frequency of abnormal ventricular function compared with traditional imaging techniques such as LVEF, TAPSE, or RV FAC. This discrepancy was not characterized previously in a prior study that evaluated LVEF and strain. 17 We did not observe an association between day 1 LV dysfunction and mortality, although we noted survivors showed better day 1 RV free-wall strain values than nonsurvivors. This association between RV dysfunction and mortality expands on prior work that used traditional RV imaging techniques. 17,19 We also characterized the course of cardiac dysfunction over the first week of critical illness. Our data suggest limited usefulness of routine prospective screening for lower extremity DVT in this population in the first week of critical illness.
We found that ventricular strain more often was abnormal than traditional echocardiographic measurements in critically ill patients with COVID-19. LV strain is more sensitive at detecting myocardial injury compared with ejection fraction, because ejection fraction is more dependent on loading conditions than longitudinal strain. 20 We observed that around 25% of interpretable echocardiograms showed ejection fractions of > 70% (Table 3), which can occur with reduced LV preload or afterload. RV free-wall longitudinal strain also similarly was abnormal more often compared with FAC and TAPSE, which generally were normal. The disparity between deformation imaging and traditional echocardiographic markers may reflect early cardiomyopathy, higher circulating catecholamine levels, lower ventricular preload or afterload, or shock severity. Patients with COVID-19 and respiratory failure frequently demonstrate shock consistent with this pattern. 21,22 We observed that patients who died were more likely to have worse strain than would have been expected from ejection fraction values. We speculate that these patients have ventricular dysfunction with falsely reassuring ejection fraction values because of low loading conditions. Patients with hyperdynamic ejection fractions previously were reported with higher mortality rates in the ICU setting. 23 Critically ill patients with low preload are more likely to show worse strain that improves on fluid resuscitation. 24 The high mortality observed in patients with normal LVEF and abnormal strain is both intriguing and supported by these physiologic rationales, but inferences are limited because of smaller numbers.  We observed that some patients with LV and RV dysfunction improved over the 8-day study period, although we observed no significant change in the proportion of patients with LV dysfunction in LVEF, LV strain, or RV strain during this period. Additionally, we observed no significant trends in deformation imaging or traditional echocardiographic measures of cardiac function over the first week. Additional inferences regarding trends within individuals were limited because of small numbers. We did not observe a change in median values for troponin-I or BNP over the study period.
We observed that very few patients received a diagnosis of VTE during hospitalization, and even fewer during the first week. These rates seem substantially lower than those reported in other COVID-19 cohorts. 25 Possible explanations for this discrepancy include the possibility that VTE may not develop until after the first week of critical illness or that these cohorts were different than those described early in the pandemic, where patients might have been more thrombophilic as a result of clinical management, differences in mobilization practices, or underlying disease process. It is also possible that a lower extremity clot might have been missed by two-point compression venous ultrasound, although evidence suggests that it is comparable with whole-leg ultrasound when performed by experienced clinicians. 26 The diagnosis of VTE during hospitalization was made according to routine hospital care, and it is possible that routine hospital care may not have captured all instances of VTE.
Strengths of this study include its multicenter recruitment, its use of deformation imaging, and its incorporation of prospective serial imaging over the early course of disease. Prior convenience samples relying on clinically obtained biomarkers and traditional imaging have suggested that cardiac dysfunction is common among hospitalized patients with COVID-19. 18,27 Abnormal strain also is common among noncritically ill hospitalized patients with COVID-19 and is associated with mortality and increased inflammatory cytokines. [28][29][30][31] Inferences regarding the early ICU stay from prior studies are limited, because most patients were not critically ill, and echocardiography was performed late in the hospital course, sometimes weeks after admission. One singlecenter study addressed the limitations of prior retrospective studies. 32  cardiac injury was common and typically occurred within the first week of illness. Our study significantly expands on those findings by incorporating deformation imaging, more than doubling the sample size, and the improved generalizability based on multicenter enrollment.
Our study has limitations. Although performed by clinicians skilled in performing and interpreting pointof-care ultrasound in critically ill patients, imaging quality sometimes was limited in this cohort, which may have led to an underestimation of the incidence of cardiac dysfunction, including images that used strain from the apical four-chamber view only in lieu of global LV strain. This study may be susceptible to selection bias. The number of enrolled patients is much lower than the number of critically ill patients with COVID admitted to the study hospitals, many of whom could not provide written informed consent. As such, this study is probably classified more appropriately as a convenience sample, although we note a somewhat higher rate of enrollment than the one patient, site, or month typical of enrollment in critical care trials. 33,34 Because many patients did not undergo prior imaging, the baseline cardiac function of these patients is not known. Because of limitations of patient exposures during the COVID-19 pandemic, interobserver variability was not assessable. Because of heterogeneity, this study did not account for treatments received over time at the point of serial assessments. Many of the study comparisons are univariate and do not account for other patient variables. Patients enrolled at participating centers may not be representative of US patients with severe COVID-19. Patients were enrolled between September 2020 and January 2021, which omits the newer Omicron variants of SARS-2-CoV. In addition, we may have missed cardiac injuries that occurred after the eighth ICU day. Specific thresholds for strain, including the lower limit of normal, are not well established, which could affect patient assignment into normal or abnormal function categories. Similarly, we chose a threshold of 50% for ejection fraction based on simplicity, although intersocietal guidelines use thresholds of 52% for male patients and 54% for female patients. 13 We also acknowledge the risk of type I statistical error, with potential for spurious associations to be found by chance because of multiple comparisons.

Interpretation
Using sensitive methods and prospective ascertainment, we confirmed that cardiac dysfunction is common among critically ill patients with COVID-19. The most common abnormalities were abnormalities in RV freewall strain and LV longitudinal strain. Nonsurvivors tended to have worse LV strain than expected for LVEF values and tended to have worse RV strain. We did not observe significant changes in cardiac function over the first week of critical illness, although inferences on temporal changes are limited. Early lower extremity VTE screening seems to have limited usefulness in this patient population.