A meta-analysis on the prevalence and characteristics of severe malaria in patients with Plasmodium spp. and HIV co-infection

Co-infection with malaria and human immunodeficiency virus (HIV) increases the severity and mortality rates of both diseases. A better understanding of the effects of co-infections could help in the diagnosis, prompt treatment, prevention, and control of malarial parasites among HIV-infected patients. In this systematic review and meta-analysis, we estimated the prevalence and characteristics of severe malaria (SM) caused by co-infection with HIV. We included relevant studies that were conducted between the years 1991 and 2018 and reporting on SM. We pooled the prevalence of SM in patients with co-infection, pooled odds ratios of SM in patients with co-infection and Plasmodium mono-infection, and differences in laboratory parameters such as parasite density and leucocyte counts, between co-infected and Plasmodium mono-infected patients. The meta-analysis included 29 studies (1126 SM cases). The pooled prevalence of SM in co-infected patients using the data of 23 studies (SM = 795 cases, all co-infection cases = 2534 cases) was 43.0% (95% confidence interval [CI] 31.0–56.0%; I2, 98.0%). Overall, the odds of SM from 18 studies were pooled. The odds of SM were significantly higher in co-infected patients than in Plasmodium mono-infected patients (OR 2.41; 95% CI 1.43–4.08; I2 = 85%; P = 0.001) and also significantly higher in children (OR 9.69; 95% CI 5.14–18.3; I2, 0%; P < 0.0001; two studies) than in adults (OR 2.68; 95% CI 1.52–4.73; I2, 79.0%; P = 0.0007; 12 studies). Co-infected patients with SM had a higher parasite density than those with Plasmodium mono-infection when the data of seven studies were analysed (SMD, 1.25; 95% CI 0.14–2.36; I2, 98.0%; P = 0.03) and higher leukocyte counts when the data of four studies were analysed (MD, 1570 cells/µL; 95% CI 850–2300 cells/µL; I2, 21.0%; P < 0.0001). Thus, the prevalence of SM among patients co-infected with Plasmodium spp. and HIV is high. Because co-infections could lead to SM, patients with Plasmodium spp. and HIV co-infection should be identified and treated to reduce the prevalence of SM and the number of deaths.


Risk of bias in individual studies.
Of the 29 studies included, all included studies were judged to be of high quality (≥ 7 stars). Twenty-three studies were rated with nine stars, whereas six studies 27,31,34,35,44,45 were rated with eight stars because they did not report the information on several non-SM patients with co-infection, which was the primary outcome of the present study. Table 4 provides the data on the risk of bias of the included studies.
Pooled prevalence of SM in patients with Plasmodium spp. and HIV co-infection. The  www.nature.com/scientificreports/ analysed to estimate the pooled prevalence of SM in patients with co-infection. Overall, the pooled prevalence of SM in patients with Plasmodium spp. and HIV co-infection was 42.0%, according to 21 studies (95% CI 29.0-55.0%; I 2 , 98.2%) (Fig. 2). The highest prevalence estimate (87%) was found in the study by Hendriksen et al. 30 , whereas the lowest prevalence estimate (7%) was observed in the study by Huson et al. 32 . Prevalence estimates were stratified by the time of detection of HIV infection; the prevalence of SM among co-infected patients in whom HIV had been recently diagnosed was 45.0%, according to eight studies (95% CI 22.0-68.0%; I 2 , 99.2%); among those who received undertreatment, 44.0% according to nine studies (95% CI, 29.0%-59.0%; I 2 , 93.0%) and among those who were immunosuppressed, 21.0% according to two studies (95% CI 14.0-27.0%; I 2 , 99.1%).  Parasite density, leukocyte count, and differential counts. The differences in parasite density, leukocyte counts, and differential counts of Plasmodium spp. and HIV co-infected and malaria mono-infected patients with SM were estimated. Patients co-infected with Plasmodium spp. and HIV who had SM had a higher mean parasite density than patients with Plasmodium mono-infection, according to six studies (standardised mean difference [SMD], 1.25; 95% CI 0.14-2.36; I 2 , 97%; P = 0.03) (Fig. 4). Co-infected patients with SM had higher leukocyte counts than patients with Plasmodium mono-infection, according to four studies (mean difference [MD], 1570 cells/µL; 95% CI 850-2300 cells/µL; I 2 , 21%; P < 0.0001) (Fig. 5). The mean neutrophil counts of patients with Plasmodium and HIV co-infection and SM as well as patients with Plasmodium mono-infection did not significantly differ according to two studies (MD, 980 cells/µL; 95% CI − 1880 to 3840 cells/µL; I 2 , 81.0%; P = 0.5; Fig. 6). The lymphocyte counts in Plasmodium spp. and HIV co-infected individuals with SM and those in Plasmodium mono-infected individuals were also similar according to four studies (MD, 370 cells/µL; 95% CI − 1330 to 590 cells/µL; I 2 , 93.0%; P = 0.45; Fig. 7).

Publication bias.
There was an indication of publication bias across the included studies, as demonstrated by the asymmetrical distribution of the funnel plot (Fig. 8).
Sensitivity analysis. Because of the publication bias indicated in Fig. 8, we used the trim-and-fill method to evaluate the odds of SM caused by malaria and HIV co-infection in 18 studies. We found that the OR for the Fixed Effects model was 1.82 (P < 0.001; 95% CI 1.67-1.96), whereas the OR for the Random Effects model was 3.01 (P < 0.001; 95% CI 2.14-3.88; Fig. 9). We also used the trim-and-fill method to conduct the sensitivity analysis for the pooled prevalence of SM among co-infected patients. The pooled prevalence estimated by the Fixed Effects model was 13.5% (95% CI 12.2-14.8%), and that estimated by the Random Effects model was 16% (95% CI 2-29.9%).

Discussion
Most studies reporting on co-infection with Plasmodium spp. and HIV were performed in Sub-Saharan Africa.  50 . Previous studies suggested that co-infection can facilitate the rate of malaria transmission by the process of CD4 cell activation, up-regulation of pro-inflammatory and cytokine production, and T-cell activation resulting in a reduction in the immune response 49,51 .
The study conducted in Mozambique demonstrated the highest prevalence of SM among children co-infected with Plasmodium spp. and HIV, who were characterised by undernourishment, severe acidosis, severe anaemia, respiratory distress, and elevated blood urea nitrogen concentrations 30 . The high prevalence of SM in that study might be attributable to the fact that 896 patients suspected of having SM were enrolled. Contrarily, the study with the lowest prevalence (7.0%), that of Huson et al. 32 , was a prospective observational study of 103 patients with sepsis and 127 with malaria and 60 HIV-infected individuals as a control group.
Our meta-analysis showed a significantly increased odds of SM in patients with Plasmodium spp. and HIV co-infection compared with those with Plasmodium spp. mono-infection. Our meta-analysis the odds of developing SM in patients co-infected with Plasmodium spp. and HIV depend on age. Although the higher odds of developing SM in adults than in children had been reported 25,27,29,47,52 , our meta-analysis demonstrated that the odds of developing SM were higher in children younger than five years and in children younger than 15 years. In addition, the odds of SM among co-infected children younger than five years (OR 9.69) were higher than those among co-infected adults older than 15 years (OR 2.68). Our sensitivity analysis of odds of SM in patients with Plasmodium spp. and HIV co-infection compared with those with Plasmodium spp. mono-infection showed the odds of SM among co-infected patients were higher than those among mono-infected patients (OR, 1.82 by the Fixed Effect model; OR, 3.0 by the Random Effects model). These results suggested that the meta-analysis had the robustness of the conclusions that patients with Plasmodium spp. and HIV co-infection increased odds of SM compared with those with Plasmodium spp. mono-infection. The development of SM among adults could be reflected by a failure to acquire immunity, which resulted in a higher parasite density among patients co-infected with Plasmodium spp. and HIV 47 . Conversely, Plasmodium spp. and HIV co-infection in children was associated with the rapid onset of cerebral malaria mediated by defects in macrophage phagocytosis 34 . This was supported by a previous study demonstrating lower absolute counts of CD4+ T cells, B cells, and NK cells in co-infected children who developed cerebral malaria 35 . That previous study demonstrated that HIV-positive patients are prone to additional opportunistic infections and febrile illnesses, which may be difficult to clinically distinguish from malaria 23 . Co-infection with Plasmodium spp. and HIV has been associated with a reduction in anticoagulant protein S and markers of endothelial activation, resulting in increased morbidity among co-infected patients 32 .
Our meta-analysis found that Plasmodium spp. and HIV co-infected patients with SM had a higher parasite density than Plasmodium spp. mono-infected patients with SM. We found that children younger than 5 years 25,34 and children younger than 15 years 31,43 who were co-infected with Plasmodium spp. and HIV and had SM had higher parasite densities than children with Plasmodium mono-infection. However, the study of adults aged 15-49 years that was conducted in Zambia 28 demonstrated no difference in the mean parasite densities, whereas the study of both co-infected children younger than 15 years and adults older than 15 years that was conducted in Mozambique demonstrated that the SMD of parasite density was higher in children and lower in adults 30 . In patients with Plasmodium spp. and HIV co-infection, it was reported that malaria caused an increase in transitory HIV viral load 53 and that HIV infection caused an increased susceptibility to malaria infection 53 as well as induced more severe parasitaemia and higher rates of treatment failure 13 . These likely effects of HIV infection lead to impairment of the immune system, resulting in reduced control of parasite multiplication 50 .
Only a few studies have reported on the effects of co-infection on haematological parameters such as leukocytes, platelet counts, and haemoglobin levels. Our meta-analysis showed that Plasmodium spp. and HIV coinfected patients with SM had higher leukocyte counts than patients with Plasmodium spp. mono-infection. The leukocyte counts, particularly the neutrophil count, were significantly higher in patients with high parasitaemia compared with those with low and moderate parasitaemia, whereas lymphocyte counts were significantly lower in patients with high parasitaemia 54 Table 2. Parasitemia level and leukocyte differential counts in co-infections and Plasmodium monoinfections. NA not applicable.

No.
Author, year www.nature.com/scientificreports/ Only these two studies, however, contained information on neutrophil counts. Therefore, the difference in the leucocyte counts should be investigated further.

Plasmodium mono-infection Co-infection
Our meta-analysis of lymphocyte counts showed lower lymphocyte counts in two studies conducted in Malawi during the periods of 2005-2006 35 and 1996-2010 31 but higher lymphocyte counts in the study conducted in Malawi during 2016-2017 37 . The heterogeneity of lymphocyte counts among the three studies might be explained by the fact that two of these studies included patients who have recently been diagnosed with HIV 31,35 , and the other study included HIV-infected patients who were undertreated 37 . These findings were in agreement with that of a previous study that demonstrated that a lower lymphocyte count in HIV-infected patients was associated with a more clinically advanced disease 55 . For other haematological changes in patients with Plasmodium spp. and HIV co-infection, such as red blood cell parameters, another previous study demonstrated that severe anaemia was caused by a reduction in erythropoiesis 29 .
Previous studies have shown that the mortality risk among individuals with Plasmodium spp. and HIV coinfection was twice as high as those with HIV mono-infection 38,42,56 . The mortality caused by the Plasmodium spp. and HIV co-infection was reported to be 282% higher in children and 64% higher in adults with SM compared to HIV-negative patients 30 . A previous study suggested that the severity and mortality of immunosuppression by HIV might be associated with hypoglycaemia and hypotension 23 .
Our study had several limitations. First, we excluded full clinical drug trials because our objective was to investigate the odds of SM in co-infected patients who did not receive any malaria treatment. A further metaanalytic study of the risk of SM in full clinical trials should be conducted. Second, patients with HIV status who rejected malaria testing or were not tested for malaria may have resulted in the underreporting of HIV and malaria co-infection, because HIV patients may present with atypical signs and symptoms of malaria 57 . Third, the difference in the CD4 cell count between patients with co-infection and those with Plasmodium mono-infection could not be meta-analysed as the CD4 data reported by some included studies were insufficient. Clinicians in the regions where both Plasmodium spp. and HIV are endemic should carefully consider co-infection as a differential diagnosis to prevent SM. Moreover, an early evaluation of HIV patients with suspected malaria may help reduce disease severity and mortality. Further longitudinal studies should focus on the impact of HIV on malaria infection to inform the management of co-infected individuals living with HIV/AIDS. In conclusion, our systematic review and meta-analysis demonstrated that Plasmodium spp. and HIV co-infection could lead to SM. As patients with Plasmodium spp. and HIV co-infection had a greater risk of developing SM than those with Plasmodium spp. mono-infection, it is necessary to diagnose and treat patients with Plasmodium spp. and HIV co-infection to reduce the number of cases of SM and death from co-morbidities.

Methods
Data sources and search strategy. The present systematic review and meta-analysis adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) 58 . The searches were performed systematically in three databases, including PubMed, Scopus, and the Web of Science. The search terms: '(malaria or Plasmodium) AND HIV AND (coinfection OR co-infection)' were used for the searches, applying search strategies relevant to each of the individual databases. Table S1 describes the details of the search strategy for all research databases. The end date for the search was 5 May 2020. All relevant articles (no limitation in the year of publication but limited to the English language) reporting on SM in patients with Plasmodium spp. and HIV co-infection were screened for eligibility. The reference lists of included studies and review articles were examined for additional studies. Searches in other sources, including Google Scholar, were also performed to maximise the number of included studies.  (3) studies involving human samples. Any reports of a small number of cases (fewer than five), such as case reports, case series, commentaries, letters to editors, short reports, and research notes, were excluded from this study. As we aim to investigate the pooled prevalence of severe malaria in patients with malaria and HIV co-infection patients who did not receive any malaria treatment, rather than the incidence www.nature.com/scientificreports/      was assessed independently by two authors (MK and FRM) using the Newcastle-Ottawa Scale for assessing the quality of nonrandomised studies in meta-analyses 59 . All included studies were judged based on three broad parameters, namely the selection of the study groups, the comparability of the groups, and the ascertainment of the outcome of interest 59 . A star system was developed for rating the quality of each included study with a ranging system from 1 to 9. The risk of bias was high if the study was rated < 7 stars, and the risk of bias was low if the study was rated ≥ 7 stars.  www.nature.com/scientificreports/ Statistical analysis. The primary outcome of the present study was to estimate the pooled prevalence of SM among patients with Plasmodium spp. and HIV co-infection. The pooled prevalence of SM among patients with Plasmodium spp. and HIV co-infection was estimated using the Random Effects model (method of Der-Simonian and Laird) 60 . The results were demonstrated as the pooled prevalence estimate and 95% confidence intervals (CIs) using a forest plot. The meta-analysis of pooled prevalence was performed using Stata version 12.1 (StataCorp LP, College Station, TX, USA). As mentioned above, the secondary aim of the present study was to determine whether Plasmodium spp. and HIV co-infection is associated with higher odds of SM when compared with Plasmodium spp. mono-infection. The pooled odds ratio (OR) and 95% CI was estimated using (1) the number of patients with SM in the presence of Plasmodium spp. and HIV co-infection and those with Plasmodium spp. mono-infection; (2) the total number of patients with Plasmodium spp. and HIV co-infection and those with Plasmodium spp. mono-infections. The pooled mean differences (MDs) and 95% CI between laboratory parameters, including parasite density, and leukocyte and differential counts were estimated based on the means and standard deviations (SDs) between the two groups. Medians and ranges/interquartile ranges reported by included studies were transformed to means and SDs as described elsewhere 61 . Meta-analyses of the pooled ORs and MDs were performed using Review Manager (RevMan) 5.3 software (Version 5.3, London, UK). The heterogeneity among included studies was tested and quantified by the Cochrane chi-square, and I 2 statistics were presented in the forest plots. If the I 2 statistic was higher than 50%, indicating substantial heterogeneity 62 , the Random Effects model was used in the meta-analysis. A subgroup analysis of age groups and locations of participants was also performed to identify any difference in the odds of SM among subgroups.
Publication bias. Publication bias was evaluated by visual inspection of funnel plot asymmetry. Generally, if symmetry is observed, this indicates no publication bias, whereas asymmetry suggests publication bias across the included studies. If the results indicated a publication bias, we revised the estimate of the prevalence and the odds ratio after correcting for such publication bias in the sensitivity analysis using the trim-and-fill method 26 utilizing Stata ver. 14 (Stata Corporation, College Station, TX, USA).