Prevalence and prognostic associations of cardiac abnormalities among hospitalized patients with COVID-19: a systematic review and meta-analysis

Although most patients recover from COVID-19, it has been linked to cardiac, pulmonary, and neurologic complications. Despite not having formal criteria for its diagnosis, COVID-19 associated cardiomyopathy has been observed in several studies through biomarkers and imaging. This study aims to estimate the proportion of COVID-19 patients with cardiac abnormalities and to determine the association between the cardiac abnormalities in COVID-19 patients and disease severity and mortality. Observational studies published from December 1, 2019 to September 30, 2020 were obtained from electronic databases (PubMed, Embase, Cochrane Library, CNKI) and preprint servers (medRxiv, bioRxiv, ChinaXiv). Studies that have data on prevalence were included in the calculation of the pooled prevalence, while studies with comparison group were included in the calculation of the odds ratio. If multiple tests were done in the same study yielding different prevalence values, the largest one was used as the measure of prevalence of that particular study. Metafor using R software package version 4.0.2 was used for the meta-analysis. A total of 400 records were retrieved from database search, with 24 articles included in the final analysis. Pooled prevalence of cardiac abnormalities in 20 studies was calculated to be 0.31 [95% Confidence Intervals (CI) of (0.23; 0.41)], with statistically significant heterogeneity (percentage of variation or I-squared statistic I2 = 97%, p < 0.01). Pooled analysis of 19 studies showed an overall odds ratio (OR) of 6.87 [95%-CI (3.92; 12.05)] for cardiac abnormalities associated with disease severity and mortality, with statistically significant heterogeneity (I2 = 85%, between-study variance or tau-squared statistic τ2 = 1.1485, p < 0.01). Due to the high uncertainty in the pooled prevalence of cardiac abnormalities and the unquantifiable magnitude of risk (although an increased risk is certain) for severity or mortality among COVID-19 patients, much more long-term prognostic studies are needed to check for the long-term complications of COVID-19 and formalize definitive criteria of “COVID-19 associated cardiomyopathy”.


Study selection.
One assessor reviewed all relevant titles and abstracts independently and selected articles for full-text review if inclusion criteria are met. Full-text review and appraisal was done by two assessors. Disagreements were resolved by discussion between the two reviewers; a third person was involved when consensus cannot be reached (Fig. 1). Data extraction. The following data from each study were extracted: first author's name, study design, population type (whether only severe and critical cases are included, or even mild and moderate cases are included), diagnostic methods for cardiac abnormalities, frequency of cardiac abnormalities, and frequencies of severity and mortality among patients with and without the aforementioned cardiac abnormalities. Studies that have data on prevalence were included in the calculation of the pooled prevalence, while studies with comparison group were included in the calculation of the odds ratio. If multiple tests were done in the same study yielding different prevalence values, the largest one was used as the measure of prevalence of that particular study.

Risk of bias (quality) assessment for prognosis studies.
Two reviewers independently assessed the quality of the included studies using the framework by Dans et al. 3 that uses the following signaling questions answerable by a "Yes" or "No": • Were all important prognostic factors considered?
• Were unbiased criteria used to detect the outcome in all patients?
• Was follow-up rate adequate?
• If clinical prediction rules are being tested, was a separate validation study done?
Results from these ratings and especially non-agreement were then the basis for discussion until final consensus is made.
Strategy for data synthesis. Meta 5 . The I 2 statistic represents the percentage of variability in effect estimates due to real dispersion among the studies 5 . I 2 of at least 50% is considered substantial heterogeneity; it means that at least half of the total variability among effect sizes is due to true heterogeneity between studies 5 . The tau-squared statistic is a function of I 2 .
Random effects models 6 were used to account for the heterogeneity of included studies. In studies with zero count for events in either the exposure or comparator group, 0.5 was automatically added in all counts 6 . Pooled prevalence of cardiac abnormalities was estimated at 95% confidence level using the Logit Transformation Method and Clopper-Pearson Intervals respectively. The Logit Transformation Method was used to estimate the pooled prevalence by log-transforming the prevalences of the individual studies 7 . The Clopper-Pearson Intervals, or more commonly known as the exact binomial test, calculates the confidence intervals based on the binomial distribution and therefore produces more conservative estimates and wide confidence intervals 8, 9 . Pooled odds ratios and 95%-CI were calculated using the DerSimonian-Laird Method. The DerSimonian-Laird Method adjusts the standard errors of the individual odds ratios to incorporate variations across different studies 10 , producing wider confidence intervals.
A funnel plot, together with Egger's Test, was used to determine potential publication bias. The results-the intercept, its confidence intervals-represent the degree of asymmetry of the funnel plot 11 . The farther it is from zero, the more asymmetric is the funnel, indicating publication bias 11 .
Post-hoc sensitivity analyses-leave-one-out analysis and Baujat diagnostics-were also done.
Analysis of subgroups or subsets. Subgroup analyses were done according to the following.

Results
Study selection. From the database search, 294 articles were retrieved and additional 120 studies were identified through Google Scholar, giving a total of 400 studies after 14 duplicates were removed. After screening articles by title, 306 articles were excluded because 281 articles are not relevant or did not satisfy the inclusion www.nature.com/scientificreports/ criteria, and 25 articles were letters, editorials, protocol summaries, or reviews, leaving only 94 articles. After screening articles by abstract, 25 articles were excluded because 19 articles were not relevant, and 6 articles were letters, editorials, protocol summaries, or reviews. 69 articles then underwent full-text screening, thus excluding 45 articles, all of which do not have the relevant population, exposure, or outcome parameters. This leaves 24 articles to be included in the final analysis. 20 articles are included in estimation of the pooled prevalence, and 19 articles are included in the meta-analysis of odds ratios.

Summary of characteristics of included studies. Among 24 papers included in the final analysis, there
are four case-control studies, two cross-sectional studies, and 18 cohort studies. All studies were done in a hospitalized setting. Deng et al. had analysis on both severity and mortality. Most of these studies involved multiple cardiac biomarkers. More details of the included studies are found in Table 1. (2)  The major weaknesses of these studies lie in their retrospective design, questionable temporality (as some are cross-sectional and case-control), different and sometimes unclear thresholds to define an "abnormal cardiac test".

Pooled prevalence calculation.
A total of 20 studies (two cross-sectional studies and 18 cohort studies) were included in the calculation of pooled prevalence, yielding a total of 4393 patients, 1040 of whom had at least one abnormal result in a cardiac test. Pooled prevalence is at 0.31 [95%-CI (0.23; 0.41)] (Fig. 2). Heterogeneity was statistically significant with I 2 = 97%, τ 2 = 0.9373, p < 0.01.

Discussion
The substantial heterogeneity among studies precludes any definitive conclusion on the magnitude of risk or odds ratio of severity or mortality associated with any abnormal cardiac finding in any given test. Nevertheless, subgroup analyses of certain cardiac biomarkers-namely, CK, Troponin T, NT-proBNP, as well as Troponin I (if the case-control study by Nie et al. is removed)-show more reliable odds ratios with their nonsignificant  www.nature.com/scientificreports/ heterogeneity. All forest plots show a clear trend towards definite increase in mortality or severity risk among COVID-19 patients exposed to a positive finding in any cardiac abnormality test.

Implications of the study findings.
In another meta-analysis of 35 studies 13 , the pooled frequency of acute cardiac injury among COVID-19 patients was at 25.3%, which is within the bounds of the 95% confidence interval estimated by this study, between 23 and 41%. Hypertension is the most common pre-existing comorbidity in these patients with a pooled frequency of 29.2% (95%-CI 24.7; 33.6%), followed by diabetes with a pooled frequency of 13.5% (95%-CI 11.5; 15.4%) 13 . Overall, fewer than one-fifth of patients had pre-existing cardiovascular diseases, at 12.6% (95%-CI 10.0; 15.2%). The risk of mortality in the presence of acute cardiac injury is increased by nearly 20 times [OR = 19.64; 95%-CI (10.28, 37.53). The heterogeneity of the studies included is also moderately to highly significant; reasons for this were not detailed in the said study. There are currently little to no studies on cardiac abnormalities on only mild and moderate COVID-19 cases as these cases are likely treated on an outpatient basis. Due to overwhelmed health systems in most countries where these studies are done, mild and moderate cases are not given enough attention nor any form of cardiac biomarker screening. In this study, a pooled odds ratio of 6.87 [95%-CI (3.92; 12.05)] means that, the COVID-19 patients with an abnormal cardiac test are 6.87 times more likely to die or have severe disease than COVID-19 patients without  With regards to severity, persistent symptoms have been reported even after recovery from COVID-19, and this has been linked to cardiac, pulmonary, and neurologic complications 14 . In a study of 143 patients who recovered from COVID-19 14 , 87.4% reported persistence of at least fatigue or dyspnea. In another study of 100 recovered COVID-19 patients 15 , high-sensitivity troponin T (hsTnt) was detectable (3 pg/mL or greater) in 71 patients (71%) and significantly elevated (13.9 pg/mL or greater) in 5 patients (5%). In the same study 15 , 78 patients (78%) had abnormal CMR findings, and endomyocardial biopsy in patients with severe findings revealed active lymphocytic inflammation. These may all contribute to an emerging picture of an emerging epidemic of "COVID-19 associated cardiomyopathy" which may affect survivors who had mild, moderate, severe, or critical COVID-19.
The findings in this meta-analysis may provide an explanation for anecdotal reports of outside-hospital sudden deaths and increasing rates of COVID-19 "recoveries" turning into "deaths", and more severe disease and more deaths from other comorbid conditions among COVID-19 survivors.
The pooled odds ratio for severity or mortality is but merely a single point estimate of a very fat-tailed risk due to the significant heterogeneity of the included studies, making it necessarily insufficient to give us any definite information for screening efforts 16 . However, there is no doubt on the precautionary principle that should be taken into consideration in implementing policies of recovery and/or follow-up. The risk for severity or mortality across all analyses presented here are asymmetrical and right-skewed. Related distribution of fatalities of pandemic outbreaks in the past 2500 years is strongly fat-tailed 17 . What we are dealing here is an "infectious" form of supposedly the most common cause of death worldwide-cardiac disease that is "infectious", so to speak.
Limitations of the study. The authors faced a major challenge in disaggregating the data of each study; hence, studies are pooled together even with different study designs. Some studies have data on prevalence but not on odds ratio. Some studies have multiple tests performed. Thus, only the maximum count of the stated outcome of the study-cardiac abnormalities found through any one of the tests-is considered in calculating the pooled prevalence and odds ratio. Consequently, subgroup analyses were performed based on the specific type of test in order to address this limitation; however, the heterogeneity did not fully disappear.
The varying tests and their varying cut-off measures for the definition of a "positive finding" in each individual study likely contributes to a significant portion in the heterogeneity even among the smaller studies. This is an understandable phenomenon given the fact that information is still evolving. Some measures may have reduced validity due to the excessive inflammation in COVID-19, which may cause spuriously high levels of serum  www.nature.com/scientificreports/ biomarkers. Therefore, we propose more studies that will eventually formalize a unified definition or diagnostic criteria for "COVID-19 cardiomyopathy".

Conclusion
Despite significant heterogeneity in most comparisons, there is a trend towards definite increase in mortality or severity risk among COVID-19 patients with any cardiac abnormality test. Due to the high uncertainty in the pooled prevalence and/or incidence of cardiac abnormalities and the unquantifiable magnitude of risk (although an increased risk is certain) for severity or mortality among COVID-19 patients, much more long-term prognostic studies are needed to check for the long-term complications of COVID-19 and formalize definitive criteria of "COVID-19 associated cardiomyopathy". By defining clear criteria, or by defining a specific test for the detection of any cardiac abnormality, the magnitude of risk can be better measured. Long-term prognostic studies using a defined criteria of "COVID-19 associated cardiomyopathy" on recovered patients should be done. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.