Dengue Virus Infection and Associated Risk Factors in Africa: A Systematic Review and Meta-Analysis

Dengue contributes a significant burden on global public health and economies. In Africa, the burden of dengue virus (DENV) infection is not well described. This review was undertaken to determine the prevalence of dengue and associated risk factors. A literature search was done on PubMed/MEDLINE, Scopus, Embase, and Google Scholar databases to identify articles published between 1960 and 2020. Meta-analysis was performed using a random-effect model at a 95% confidence interval, followed by subgroup meta-analysis to determine the overall prevalence. Between 1960 and 2020, 45 outbreaks were identified, of which 17 and 16 occurred in East and West Africa, respectively. Dengue virus serotype 1 (DENV-1) and DENV-2 were the dominant serotypes contributing to 60% of the epidemics. Of 2211 cases reported between 2009 and 2020; 1954 (88.4%) were reported during outbreaks. Overall, the prevalence of dengue was 29% (95% CI: 20–39%) and 3% (95% CI: 1–5%) during the outbreak and non-outbreak periods, respectively. Old age (6/21 studies), lack of mosquito control (6/21), urban residence (4/21), climate change (3/21), and recent history of travel (3/21) were the leading risk factors. This review reports a high burden of dengue and increased risk of severe disease in Africa. Our findings provide useful information for clinical practice and health policy decisions to implement effective interventions.


Introduction
Dengue is an important arboviral disease, with the highest incidence in tropical and subtropical regions, with a potential to spread into other geographical areas. In the past four decades, dengue has caused a significant impact on human health and national economies [1,2]. Approximately 390 million people are infected with dengue virus (DENV) annually. Of these, 96 million develop clinical manifestations that lead to 500,000 hospitalizations and 25,000 deaths, annually [3]. Dengue is caused by an RNA virus of the family Flaviviridae that is transmitted to humans through a bite of infected Aedes mosquitoes.

Quality Assessment
The methodological quality of selected prevalence studies was evaluated by two reviewers using a quality assessment checklist adapted from Hoy and others [13]. Risk of bias was assessed using nine domains: target population, sampling frame, sample selection method, likelihood of non-response bias, data source, case definition, study instrument that measured the parameter of interest, mode of data collection, and numerator and denominator of the parameter of interest. The risk of bias levels was low (score = 0) or high (score = 1), and the overall risk of bias was defined as low (score 0-3), moderate Viruses 2021, 13, 536 3 of 17 (4)(5)(6), and high (7)(8)(9) (Table S1). Any discrepancy was resolved through discussion with a third reviewer.

Statistical Analysis
The extracted data were pooled using MetaXL version 5.3 software (EpiGear International Pty Ltd., Queensland, QLD, Australia). A random effect model was used to estimate the overall prevalence and 95% confidence intervals (CI), and results were presented in forest plots. The percentage of heterogeneity between studies was quantified using I 2 and chi-square tests, and I 2 ≥ 50% was considered significant. Sensitivity analysis to test the effect of each study on summary prevalence, by excluding each study step by step, was used to evaluate the robustness of overall prevalence. A funnel plot and Egger's regression test were used to detect publication bias. All results with p-values < 0.05 were considered statistically significant. Descriptive statistics, narrative synthesis, and relevant figures were used to summarize the information where statistical pooling was not possible.

Search Results and Characteristics of Selected Studies
A total of 2170 records were retrieved from database searches. After duplicates removal and screening, 43 studies were finally included in the review ( Figure 1). The methodological quality of studies ranged from low (0-3 score, 37 studies) to moderate (4-6 score, 6 studies). No study had a high risk of bias, six (13.9%) studies had moderate risk, and 37 (86.1%) had low risk (Table S2). Out of 43 studies, 34 were prospective crosssectional, six were retrospective cross-sectional, two were prospective cohort, and one was a case-control study (Table 1). denominator of the parameter of interest. The risk of bias levels was low (score = 0) or high (score = 1), and the overall risk of bias was defined as low (score 0−3), moderate (4−6), and high (7−9) (Table S1). Any discrepancy was resolved through discussion with a third reviewer.

Statistical Analysis
The extracted data were pooled using MetaXL version 5.3 software (EpiGear International Pty Ltd., Queensland, QLD, Australia). A random effect model was used to estimate the overall prevalence and 95% confidence intervals (CI), and results were presented in forest plots. The percentage of heterogeneity between studies was quantified using I 2 and chi-square tests, and I 2 ≥ 50% was considered significant. Sensitivity analysis to test the effect of each study on summary prevalence, by excluding each study step by step, was used to evaluate the robustness of overall prevalence. A funnel plot and Egger's regression test were used to detect publication bias. All results with p-values < 0.05 were considered statistically significant. Descriptive statistics, narrative synthesis, and relevant figures were used to summarize the information where statistical pooling was not possible.

Search Results and Characteristics of Selected Studies
A total of 2170 records were retrieved from database searches. After duplicates removal and screening, 43 studies were finally included in the review ( Figure 1). The methodological quality of studies ranged from low (0−3 score, 37 studies) to moderate (4−6 score, 6 studies). No study had a high risk of bias, six (13.9%) studies had moderate risk, and 37 (86.1%) had low risk (Table S2). Out of 43 studies, 34 were prospective cross-sectional, six were retrospective cross-sectional, two were prospective cohort, and one was a case-control study (Table 1).

Dengue Virus Outbreaks and Serotype Distribution
Since 1964, 45 dengue outbreaks were reported in 14 countries ( Table 2). Most of the outbreaks occurred in East (17/45) and West (16/45) Africa. DENV-1 and DENV-2 were dominant serotypes in most of the outbreaks ( Figure 2). During the past decade (2010-2020), there was an expansion of multiple DENV serotypes occurrence in Africa ( Figure 3).

Risk Factors
Evidence from 21 reports published between 2007 and 2020 showed that old age, lack of mosquito control, living in urban areas, climate change, and history of recent travel were the leading risk factors of dengue. Other risk factors were type of occupation, lack of education, low income, and known diabetes mellitus status (Table 4).

Discussion
This systematic review reports the distribution of outbreaks and the prevalence of dengue in Africa during the outbreak and non-outbreak periods. Our results show that dengue has been reported in 24 of 54 countries and has become endemic, with repeated outbreaks in most of them. Since 1960, all four DENV serotypes (DENV-1-4) caused epidemics in all African sub-regions, with DENV-1 and DENV-2 dominating. Laboratory confirmed outbreaks were reported in 13 African countries, with the East Africa region contributing over 50% of the epidemics. These observations support evidence previously documented [5,7]. After 2010, severe dengue cases have been increasingly reported in different countries, including Burkina Faso [79], Côte d'Ivoire [80], Djibouti [30], the Republic of Sudan [81], Senegal [22], and Tanzania [31]. The previous report shows that these countries have experienced continuous active DENV transmission in the past decade [5].
Our analysis revealed an increased occurrence of multiple DENV serotypes in Africa during the past decade (2010-2020) (Figure 3), with a greater proportion of serotypes reported in East and West Africa (Figure 2). Concurrent infections with multiple serotypes may pose a risk of severe dengue because lifelong immunity against primary infection by one serotype does not cross-protect subsequent infections by a different serotype. In secondary infection, antibody-dependent enhancement facilitates viral multiplication in the host cells, resulting in severe disease [82]. Expansion of multiple DENV serotypes in Africa may be caused by several factors. International travel of infected people from epidemic and endemic countries has been associated with the introduction of DENV-1 and DENV-3 serotypes in several African countries [83,84]. Spill-over of sylvatic DENV-2 strains from forest Aedes mosquitoes into a human transmission cycle possibly facilitates the spread into urban or new geographical areas with the potential to cause epi-demics [85,86]. Further, increasing recognition of DENV as the cause of undifferentiated febrile syndromes [87], and the availability of more sensitive and specific molecular-based laboratory tests in the past decade, may have contributed to more detection and reporting of DENV serotypes [51,56,88].
Meta-analysis results show that the overall prevalence of dengue virus in Africa is 14%. This prevalence is relatively higher than the 7% reported in previous meta-analysis [36]. The discrepancy could be due to differences in a number of prevalence studies included in the meta-analysis. More studies included in this review possibly contributed to an increased number of dengue positive cases. In addition, our review included studies conducted during outbreaks, thus, large studies with a higher proportion of dengue positive cases were expected. During an outbreak, the prevalence of dengue virus was 29%. Our results agree with the 30% prevalence reported in the previous meta-analysis that included prospective cross-sectional studies conducted during epidemics or following recognized epidemics in the Republic of Sudan [89]. These observations indicate a high burden of dengue of up to 39% in febrile patients during an outbreak, highlighting the need for routine laboratory dengue diagnosis in tropical Africa.
Low dengue virus prevalence of 3% in febrile patients was found during the nonoutbreak period ( Figure 6). In comparison, our results were relatively lower than values reported in a previous meta-analysis by Simo et al. (2019) involving febrile patients from studies conducted during the non-outbreak period [36]. This difference could be due to the selection and epidemiological contexts of included studies. For instance, inclusion of studies conducted during ongoing epidemics or following epidemics are likely to contribute to a higher number of dengue positive cases [32]. As a result, the overall prevalence could have been overestimated. Despite the low prevalence observed during the non-outbreak period, a burden of up to 5% in febrile patients is still of a public health concern that needs appropriate interventions. Persistent occurrence of sporadic dengue cases may indicate endemicity, therefore, routine laboratory dengue diagnosis and enhanced mosquito surveillance could help to detect cases early and identify hotspots, respectively. Despite substantial variability (I 2 = 98.89) between prospective cross-sectional studies, more than 90% of the studies had good precision on the overall dengue prevalence (Table 3) and low risk of methodological quality bias (Table S2). The presence of publication bias (Egger's test, p = 0.0022) in the outbreak studies ( Figure S1), could be due to small studies with non-significant results not being published.
Co-existing unrecognized co-morbidities can complicate dengue diagnosis and patient management. The findings from previous studies [90,91] show that co-morbidities increase the risk of severe disease and fatal outcomes among dengue patients. Our results show that in the past decade (2010-2020), malaria and dengue co-infections were the most prevalent, followed by dengue and chikungunya co-infection [24,31,40,43,53,56,79,[92][93][94][95]. A similar occurrence pattern of malaria and dengue co-infection dominance followed by dengue and chikungunya co-infection was previously reported [96]. In Africa, co-morbidities of dengue are not usually diagnosed due to a lack of diagnostic capacity to differentiate dengue from other mosquito-borne acute febrile illnesses such as chikungunya, Zika, and yellow fever. The diseases develop similar non-specific clinical signs and can be co-transmitted with dengue [97]. These findings underscore the need to enhance differential diagnosis of non-malaria febrile illnesses in Africa.
Results from this review show that increasing age, lack of mosquito control, living in an urban area, climate change, and recent history of travel were the leading risk factors of dengue in Africa (Table 4). A high risk of contracting DENV in the old age group may be explained by continuous exposure to Flaviviruses [54,69,72]. The presence and abundance of Aedes mosquito vectors are known to increase risk of dengue exposure [98,99]. Evidence from some studies shows that people living in areas surrounding waste dump sites, opening windows at night, presence of stagnant water at home, households with indoor bathrooms, and living with open water containers were associated with a high risk of dengue [27,45]. Other potential risk factors included occupation type, lack of education, low income, and known diabetes mellitus status [48,77]. These findings disclose gaps in individual and environmental practices that could limit Aedes mosquito abundance and spread in African settings.
This review had some limitations. First, we could not establish a meta-analysis of DENV NS1 prevalence due to an inadequate number of studies reporting NS1 prevalence alone. Most studies had overlap data between NS1 and RT-PCR test. Second, a small number of studies (n < 10) limited subgroup meta-analysis of dengue prevalence based on setting (community versus healthcare facility), geographical sub-regions, and the design other than prospective cross-sectional. Third, it is possible that some individuals were asymptomatic and could not be detected in the included studies, thus, the number of dengue positive cases may be higher than reported in this review. Despite the limitations, we are confident that our findings provide useful information for clinical practice and public health policy decisions.

Conclusions
In conclusion, this review reveals a high burden of dengue infection and highlights an increased risk of severe disease in Africa due to the increasing circulation of multiple dengue virus serotypes. We advocate for the need of routine laboratory dengue diagnosis in Africa to facilitate early detection of cases, provision for appropriate patient care, identification of serotypes/genotypes, and outbreak preparedness. It is important to implement effective mosquito surveillance to identify hotspots, and control through the promotion of education on individual behaviours and environmental management practices that can limit the spread of dengue infection in Africa.