The effect of age on the incidence of COVID-19 complications: a systematic review and meta-analysis

This review follows the recommendations established by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [9]; it has been registered as a protocol in the International Prospective Register of Systematic Reviews (PROSPERO; https://​www.​crd.​york.​ac.​uk) database on ID no CRD42020181539.

This review was conducted to estimate the pooled adverse complications and effects of age on each complication among COVID-19 patients. Potential studies were identified using databases PubMed/MEDLINE, Hinari, Google Scholar, and Google Search. Searches limited to English language and studies published from December 31, 2019, to April 10, 2020. Searching MeSH headings “2019-nCoV infection” OR “2019 novel coronavirus” OR COVID-19 OR “SARS-CoV-2” OR “2019-nCoV” AND “complication*” OR “Clinical characteristics” OR “Clinical course” NOT Animals were used to identify potential studies (additional Table).

All studies which have a report on the mean age of the respondents, all age groups, and all PCR-confirmed COVID-19 cases; studies having a report on clinical complications; and studies conducted worldwide were included for this review. Retrospective and prospective observational studies, case series, and cross-sectional chart review study designs were included. Case report and qualitative studies were excluded from this review.

The primary outcome of interest for this review was to estimate the pooled incidence of acute complications of COVID-19 patients. The second objective was to estimate the pooled effects of age on the incidence of COVID-19 complications. COVID-19 complications were declared as having ARDS, AKI, ACI, shock, and ALI among COVID-19 patients during hospitalization.

A total of 1247 articles were identified through the database search. After eligibility identification, 12 potential studies had been included for this qualitative and quantitative synthesis as summarized in the PRISMA flow diagram [9] (Fig. 1).

All studies were exported and managed using the Endnote™ version X9.2 (Thomson Reuters, Philadelphia, PA, USA) software. All duplicated studies were removed, and full-text articles searched using the Endnote software manually. Screening for eligibility of the individual articles was assessed independently by two reviewers (SA and MA) through review from the title, abstract, and full text. Exaggerated differences from the two reviewers narrowed through discussion and other reviewer members (ZT and DA).

The quality of the included studies assessed using the critical appraisal tool to assess the quality of cross-sectional studies (AXIS) [10]. Two reviewers (SA and MA) independently assessed the quality of the included studies. The exaggerated discrepancy between the two reviewers was managed through discussion; otherwise, a third reviewer available to arbitrate any issues that remained unresolved (DA and ZT). The quality assessment tool measures a total of twenty questions. Overall scores of the AXIS tool ranges from 0 to 20 scores [10]. The highest score was a low risk of bias whereas the lowest score was a high risk of bias among the included studies.

The data extraction for this review was summarized by two authors (SA and MA) using Microsoft Excel. The discrepancies between the two authors were managed through discussion between them and/or the other authors (DA and ZT). The main categories of data extraction were thematized on the main characteristics of the study which were author, year, journal, country, sample size, mean age, and sex; clinical characteristics (fever, cough, shortness of breath, myalgia and fatigue, headache, and diarrhea); clinical complications during hospitalization (ARDS, AKI, ACI, ALI, and shock); and imaging results.

The extracted data were imported to the STATA/MP version 16.0 software for further analysis. The descriptive analysis of the included articles was summarized using tables. The pooled estimate of each complication of COVID-19 patients was estimated by the random-effects model using DerSimonian-Laird model weight [11]. Heterogeneity in meta-analysis is mostly inevitable due to the differences in study quality, sample size, method, and different outcomes measured across studies [12, 13]. Statistical heterogeneity was checked by the Cochrane Q test and I² statistics [14]. Sensitivity analysis was also conducted to determine the effect of single studies on the pooled estimates. Univariate meta-regression was conducted for each complication by the mean age of the respondent from primary studies using the random-effects model. Egger’s statistical test was used to check publication bias [15]. Statistically significant (P value < 0.05) Egger’s test indicates that the presence of a small study effect was handled using non-parametric trim and fill analysis using the random-effects model [16].

All studies included for this systematic review and meta-analysis were conducted by Chinese scholars in China. The total sample size of 3064 COVID-19 patients was included in this review. The minimum and maximum sample sizes were 41 [17] and 1099 [18], respectively. The mean age of the patients from the included studies ranges from 47 [18] to 69 [19] years old (Table 1).

From twelve studies [17‐28], the incidence of ARDS pooled estimate was 31% (95%CI = 21.26, 41.60); significant heterogeneity was observed among studies (I² = 99.05, P value < 0.001). The highest weight among studies was observed from the studies conducted by Jin et al. and Guan et al. [18, 20] (Fig. 2). A small study effect (publication bias) was investigated using Egger’s statistical test [29]. Egger’s statistical test provides that there is no evidence for the presence of publication bias (β = −6.06, P value = 0.38).

Among ten studies [17‐19, 21‐24, 26‐28], using the random-effects model, the incidence of AKI pooled estimate was 7% (95%CI = 3.84, 10.44). The highest weight among studies was observed from studies conducted by Guan et al. [11] (Fig. 3). Egger’s statistical test evidenced that there is no publication bias among the included studies (β = −4.31, P value = 0.13).

From nine studies [17, 19, 22‐28], the incidence of AKI (18.89, 95%CI = 10.45, 27.29) had evidence of publication bias (β = −5.2, P value = 0.038). After random effects model trim and fill analysis (Fig. 4), the pooled incidence of ACI was 6.38 (95%CI = −2.84, 15.60) among hospitalized COVID-19 patients.

Using ten studies [17‐21, 23‐25, 27, 28], the incidence of shock among COVID-19 patients was 7% (95%CI = 3.84, 10.44). The highest weight among studies was observed from studies conducted by Jin et al. [20] (Fig. 5). Egger’s statistical test evidenced that there is no publication bias among the included studies (β = −4.51, P value = 0.29).

Furthermore, the pooled incidence of ALI among COVID-19 patients was estimated using six studies [20, 22, 24, 25, 27, 28]. From these studies, the incidence of ALI among hospitalized COVID-19 patients was 22.77% (95%CI = 14.05, 31.48) (Fig. 6). Egger’s statistical test revealed that there is no publication bias among the included studies (β = 5.98, P value = 0.394). The pooled incidence of acute complication of COVID-19 is summarized and shown in Table 2.

Univariate meta-regression analysis was conducted using the aggregated mean age of study level variables using the random-effects model. This meta-regression analysis evidenced that the mean age and the incidence of acute complication of 2019 novel coronavirus patients are statistically significant. As shown in the bubble plot (Fig. 7), the mean age and incidence of ARDS had a linear relationship. As the mean age increased by 1 year, the incidence of developing ARDS complications among COVID-19 patients would increase by a factor of 2.9 (β = 2.9, 95%CI 2.4–3.4) times, with a total incidence of ARDS complication of hospitalized COVID-19 patients explained by the covariate mean age of 88% (R² = 87.62). Besides, the mean age and incidence of ACI among hospitalized COVID-19 patients had a linear relationship. The incidence of ACI complications among COVID-19 patients increased by a factor of 1.6 (β = 1.63, 95%CI 1.1–2.2) times as the mean age increased by 1 year with model adjusted R² of 85%. The mean age and incidence of AKI had a statistically significant linear relationship. The likelihood of developing AKI would increase by a factor of 0.4 (β = 0.4, 95%CI 0.04–0.72) times as the mean age increased by 1 year, with an incidence of AKI explained by the mean age of 54% (R² = 54.07) (Table 3).

This study was identified as the pooled evidence of adverse clinical complications of COVID-19. Besides, the study shows the association between age and each adverse complication used to timely interventions. The first limitation of this review was few studies were included for the evaluation of adverse complications and no studies were included outside China. Second, substantial statistically significant heterogeneity was observed across studies that undermine the pooled estimate of COVID-19 complications which suggests that chance could be responsible for between-study variability. These heterogeneities could be due to different study designs, sample size variations, and age variations among the included study participants. Since all studies were from China, it will limit the generalizability to other contents.