Risk factors for critical outcome of COVID-19 differ according to location: a systematic review and meta-analysis


 Background

The mortality rates of the coronavirus disease 2019 (COVID-19) differ across the globe. While some risk factors for poor prognosis of the disease are known, regional differences are suspected. We reviewed the risk factors for critical outcomes in extensive number of studies according to the study location.
Methods

We searched the PubMed, Embase, Cochrane Library, and Web of Science literature databases from January 1, 2020 to June 8, 2020. We defined the critical outcome as death, admission to the intensive care unit, or critical type of COVID-19. Candidate variables to predict the critical outcome included patient demographics, underlying medical condition, symptoms, and laboratory findings. Pooled relative risks (RRs) and standardized mean differences were calculated for each variable and were also determined according to the study’s continent.
Results

A total of 80 studies were included from Asia (n = 48), Europe (n = 22), and North America (n = 10). The risk factors for the critical outcome in the overall population included male sex, age, and all inspected underlying medical conditions. Symptoms of dyspnea, anorexia, dizziness, fatigue, and certain laboratory findings were also indicators of the critical outcome. Subgroup analysis was performed according to study location, and we found several discrepancies. Underlying respiratory disease was associated higher risk of the critical outcome in studies from Asia (pooled RR 2.16 [1.60–2.92] and Europe (pooled RR 1.50 [1.32–1.69]), but not North America. Underlying hepatic disease was associated with a higher risk of the critical outcome from Europe (pooled RR 1.34 [1.15–1.56]), but not from Asia and North America. Symptoms of vomiting (pooled RR 2.43 [1.60–3.69]), anorexia (pooled RR 2.38 [1.45–3.91]), dizziness (pooled RR 2.23 [1.51–3.28]), and fatigue (pooled RR 1.92 [1.23–3.02]) were significantly associated with the critical outcome in studies from Asia, but not from Europe and North America. Hemoglobin and platelet count affected patients differently in Asia compared to those in Europe and North America.
Conclusions

There are several discrepancies among risk factors for critical outcomes among patients with COVID-19 according to the location of the infected patient.


Study selection
Studies were selected by following the PRISMA ow diagram [13]. After removing duplicates, the titles and abstracts were screened to identify eligible studies for full-text review. When different outcome data were found in the same study population with similar study periods, the data with the larger population were selected. Studies with ≤ 5 patients with critical outcomes were excluded because the calculations of mean and standard deviations (SDs) were considered unreliable in these studies. When data were not presented according to the critical outcome, the authors were contacted to provide organized results.
Studies performed in the ICU, or those including patients with negative COVID-19 polymerase chain reaction results were also excluded.

Data extraction
From each study, we collected article information including the authors, study design, location, period, restriction in patient selection, and study outcome. Patient characteristics were collected including sex, age, ethnicity, underlying medical condition, symptoms, and laboratory ndings. The underlying medical condition included smoking history, hypertension, diabetes, cardiac disease, renal disease, respiratory disease, hepatic disease, cerebral disease, malignancy. The comorbidities were de ned differently in each study, as shown in Additional le 3: Table E1. The symptoms included fever, fatigue, myalgia, dizziness, headache, dyspnea, chest tightness, cough, sputum, sore throat, rhinorrhea, anorexia, nausea, vomiting, abdominal pain, and diarrhea. The laboratory ndings included white blood cell count, neutrophil count, lymphocyte count, monocyte count, hemoglobin, platelet count, creatinine, blood urea nitrogen, aspartate transaminase (AST), alanine transaminase (ALT), total bilirubin, creatine kinase, lactate dehydrogenase, and prothrombin time.
The characteristics were organized according to the critical outcome de ned in each study. For categorical variables, the input variables were organized as a two-by-two table. For continuous variables, means and SDs were organized as recommended by the Cochrane handbook [15]. Google Translate was used to translate the articles published in other languages to English.

Statistical considerations and assessment of bias
Forest plots with a random-effects model were used to explore the baseline characteristics and the impact of each variable on the critical outcome. I 2 statistics were used to assess the heterogeneity [16].
Pooled relative risks (RRs) were calculated for categorical variables. For continuous variables, standardized mean differences (SMDs) were calculated for most variables because of the differences in scale, except for age for which weighted mean difference (WMD) was calculated. The 95% con dence intervals were calculated for each pooled value and are presented in square brackets throughout the manuscript.
Quality assessment of each study was performed according to the recommended six areas of potential study biases: study participation, study attrition, outcome measurement, confounding measurement and account, and analysis [17]. Egger's regression tests were performed to assess publication bias [18].
Analyses were performed in the overall population and in subgroups according to the continent. The impact of ethnicity on the critical outcome was inspected separately with studies specifying the race according to the four categories: non-Hispanic white, non-Hispanic black, Hispanic, and Asian. To reduce the heterogeneity of the results, sensitivity analyses were performed among studies without any restriction in patient selection, critical outcome con ned to death, and at least partly achieving every standard of the six areas of potential study biases.
The process of study screening, data extraction, and assessment of quality and risk of bias were performed by two independent reviewers, and an agreement was reached through group discussion. All statistical analyses were performed using Stata version 16

Search ndings and study characteristics
The initial search revealed 3,071 studies, which narrowed to 2,151 studies after duplicate removal. After screening, 1,578 studies were removed and 573 articles were assessed with full-text review. After removing 493 non-relevant studies, our systematic review included a total of 80 studies (Additional le 3: Figure E1). The full list of the included studies is available in Additional le 4 .

Assessment of study quality and publication bias
While most studies at least partly met the quality standards of each area, several studies did not. The two studies did not represent the population of interest (study participation), two studies did not adequately measure the prognostic factor of interest (prognostic factor measurement), and nine studies did not account for important potential confounders (confounding measurement and account). (Additional le 3: Table E2).

Sensitivity analysis
A sensitivity analysis was performed in 17 studies without any restriction in patient selection, outcome con ned to death, and at least partly achieving every standard of the six areas of potential study biases. , I 2 = 5.8%) were associated with higher risk of death. The extent of heterogeneity in these categorical variables was largely reduced when further analyses were performed according to each continent. The association between laboratory ndings and death was consistent with the main analyses, except that lower monocyte counts were associated with a higher risk of death (pooled SMD − 0.33 [-0.57--0.09], I 2 = 92.2%). The results of the sensitivity analysis are summarized in Additional le 3: Figure E5, and the details are described in Additional le 3: Table E4.

Discussion
This is the rst study to summarize the risk factors for the critical outcomes (death, admission to the ICU, or critical type of COVID-19) of COVID-19 according to the study location. It is also the largest updated systematic review regarding risk factors for the poor prognosis of patients with COVID-19. While most risk factors were largely similar across the three continents, several differences were noted. The presence of respiratory disease was associated with a higher risk of the critical outcome in Asia and Europe, but not North America. The presence of hepatic disease was associated with a higher risk of the critical outcome in Europe, but not in Asia and North America. Symptoms of vomiting, anorexia, dizziness, and fatigue were signi cantly associated with the critical outcome in Asia, but not Europe and North America. While platelet count was inversely associated with the critical outcome in Asia, it was not in Europe and North America. In contrast, lower hemoglobin levels were associated with the poor outcome in Europe and North America, but not in Asia.
Our ndings of the overall population are consistent with those of previous reviews. Male sex, older age, underlying comorbidities, and several laboratory parameters have been repeatedly emphasized as risk factors for poor outcomes in patients with COVID-19 [9,10,[99][100][101][102][103]. This disease is well-known for malesex predominant deterioration. A nationwide study from Denmark reported that male sex was an independent risk factor for death even after adjusting for age and comorbidities [104]. The underlying mechanism for this observation has not yet been elucidated but may be explained by the immune regulatory genes encoded by the X chromosome, which makes men more susceptible to viral infections as compared to women [105]. In addition, sex hormones may act directly in innate immune cells to regulate their function, and indirectly via non-immune cells resulting in immune cell actions [106]. Older age was also a risk factor for grave prognosis among COVID-19 patients in previous systematic reviews [9,10]. This is easily understandable as old age is also a well-known risk factor for death among patients with community-acquired pneumonia and in uenza [107][108][109]. Among many symptoms, dyspnea was the only symptom that was consistently associated with a higher risk of the critical outcome in all three continents, a nding concordant with those in previous reviews [9,102]. Dyspnea is relatively uncommon among COVID-19 patients despite typical lung involvement [2,110]. Therefore, the presence of dyspnea could imply extensive involvement of the lung and lead to poor outcomes.
The results of our study not only con rm previous knowledge regarding the risk factors for the deterioration of COVID-19 patients but also reveal some novel ndings. First, some risk factors revealed inter-continental differences. Recognizing such differences can inform the development of proper guidelines for the management of patients according to their region and ethnicity. As noted, underlying respiratory disease was associated with the critical outcomes in Asia and Europe, but not in North America. Although the exact reason for this disparity is beyond the scope of our review, it may be partly explained by the differences in therapies for the treatment of these chronic respiratory diseases [111]. In China, only about 56% of patients with chronic obstructive pulmonary disease receive treatments that are standard in Western countries, while 23% receive Chinese traditional treatments [111]. Considering the protective effect of corticosteroids in the treatment of COVID-19 [112], such a gap in the treatment of chronic respiratory disease may have led to different outcomes among continents. Among the symptoms of COVID-19, vomiting, anorexia, dizziness, and fatigue were risk factors in Asia, but not in Europe and North America. These symptoms can be associated with weight loss and poor nutritional status during the course of the disease, while body mass index is mostly higher for individuals living in Europe and North America, compared to those in Eastern Asian countries [113]. Although our meta-analysis suggested that Hispanic patients may have better prognosis compared to non-Hispanic white, the impact of ethnicity on the prognosis of COVID-19 is yet to be explained. While a regional study from the United States reported that ethnicity may be a factor for diverse outcomes [114], other studies denied these ndings after adjusting for risk factors [115,116]. A recent meta-analysis of has suggested that, after adjusting patient characteristics, ethnicity may not be an independent prognostic factor [117].
Second, with enough pooled analysis, various comorbidities are proven to be risk factors. Because viral infections can cause a systemic in ammatory response and localized vascular in ammation [118], studies have focused on diseases associated with cardiovascular outcomes as risk factors. In previous systematic reviews including 13,16,25, and 36 studies [9,99,101,119], underlying cardiovascular disease, hypertension, diabetes, congestive heart failure, cerebrovascular disease, chronic kidney disease, respiratory disease, and cancer were identi ed as risk factors for poor patient outcome. The reviews did not nd or mention any signi cant impact of underlying liver disease. However, several studies have inferred the impact of liver disease on the prognosis of COVID-19. For example, laboratory abnormalities associated with hepatic dysfunction were frequently observed in patients with COVID-19, and were more common in severe forms of COVID-19 [120]. Furthermore, a pooled analysis showed a higher incidence of acute hepatic injury in severe COVID-19 compared to that in non-severe disease [121].
To correctly acknowledge our study ndings, several limitations should be noted. First, most of the included studies had retrospective design. This was inevitable because COVID-19 is a novel disease that caused a sudden pandemic. Second, residual heterogeneity was observed in the analyses of continuous variables. The residual extent of heterogeneity may be partially explained by differences in the reported forms of the variables (i.e., mean and SD, median and range, median, and interquartile range).

Conclusions
Our extensive systematic review summarized the risk factors associated with the critical outcome (death, admission to the ICU, and critical type of COVID-19) of COVID-19 patients according to location of infected patients (Asia, Europe, and North America). Although the risk factors were mostly consistent across the three continents, underlying diseases, patient symptoms, and laboratory ndings posed different impact on patient prognosis in each location. Future studies are required to understand the reasons for such discrepancy. Availability of data and materials The datasets used and/or analyzed during the current study are available from the corresponding author upon a reasonable request.

Competing interests
The authors declare that they have no competing interests.

Funding
This study was not funded by any organization.
Authors' contributions HJK, JJY, and JL contributed to the conception and design of the work. HJK and H Hwang contributed to the data acquisition. HJK, H Hwang, H Hong, JJY, and JL contributed to the analysis and interpretation of the data for the work. HJK drafted the article. H Hwang, H Hong, JJY, and JL revised it critically for important intellectual content. All authors approved the nal version of the manuscript to be published, and agreed to be accountable for all aspects of the work, in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.