Clinical and laboratory predictors of death in African children with features of severe malaria: a systematic review and meta-analysis

Background The criteria for defining severe malaria have evolved over the last 20 years. We aimed to assess the strength of association of death with features currently characterizing severe malaria through a systematic review and meta-analysis. Method Electronic databases (Medline, Embase, Cochrane Database of Systematic Reviews, Thomson Reuters Web of Knowledge) were searched to identify publications including African children with severe malaria. PRISMA guidelines were followed. Selection was based on design (epidemiological, clinical and treatment studies), setting (Africa), participants (children < 15 years old with severe malaria), outcome (survival/death rate), and prognostic indicators (clinical and laboratory features). Quality assessment was performed following the criteria of the 2011 Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2). Odds ratios (ORs) were calculated for each study and prognostic indicator, and, when a test was assessed in at least two studies, pooled estimates of ORs were computed using fixed- or random-effects meta-analysis. Results A total of 601 articles were identified and screened and 30 publications were retained. Features with the highest pooled ORs were renal failure (5.96, 95% CI 2.93–12.11), coma score (4.83, 95% CI 3.11–7.5), hypoglycemia (4.59, 95% CI 2.68–7.89), shock (4.31, 95% CI 2.15–8.64), and deep breathing (3.8, 95% CI 3.29–4.39). Only half of the criteria had an OR > 2. Features with the lowest pooled ORs were impaired consciousness (0.58, 95% CI 0.25–1.37), severe anemia (0.76, 95% CI 0.5– 1.13), and prostration (1.12, 95% CI 0.45–2.82). Conclusion The findings of this meta-analysis show that the strength of association between the criteria defining severe malaria and death is quite variable for each clinical and/or laboratory feature (OR ranging from 0.58 to 5.96). This ranking allowed the identification of features weakly associated with death, such as impaired consciousness and prostration, which could assist to improve case definition, and thus optimize antimalarial treatment. Electronic supplementary material The online version of this article (doi:10.1186/s12916-017-0906-5) contains supplementary material, which is available to authorized users.


Background
Severe malaria accounted for approximately 2 million out of 207 million estimated malaria cases in 2012 [1]. In areas with intense and stable transmission, children under the age of 5 years carry the heaviest burden, especially in the sub-Saharan region [2]. Although a correct and prompt diagnosis of severe malaria is crucial for prescribing appropriate therapy, and thus for reducing mortality, the parenteral administration of first-line treatment often remains a challenge in resource poor settings. Improved targeting of children who would benefit most from parenteral treatment rather than oral treatment would help the overall management of malaria cases.
A child is diagnosed with severe malaria when asexual P. falciparum parasitemia is detected in the peripheral blood smear or confirmed by a rapid diagnostic test, there is no other cause for its symptoms, and at least one of impaired consciousness, respiratory distress, multiple convulsions, prostration, shock, pulmonary edema, abnormal bleeding, jaundice, severe anemia, hypoglycemia, acidosis, hyperlactatemia, renal impairment, or hyperparasitemia is present. These criteria reflect the definition of severe malaria established by the World Health Organization (WHO) in 2000, according to which any child with positive blood parasitemia and at least one of abovementioned criteria is qualified to receive parenteral treatment [3].
In recent years, a decrease in the case fatality rate of malaria has been observed [4]. The reasons for this improvement are not entirely clear, but introduction of drugs with increased efficacy [5,6] and effective control programs [7] have certainly played a crucial role. A reduction in the case fatality rate of severe malaria has also been documented in controlled trials [5,6]. A potential confounder for this observed reduction may be related to a selection bias due to a shift in severe malaria case definition. In 1990, the WHO set the criteria for a strict definition of severe malaria for research and epidemiological purposes [8]. In 2000, new neurological criteria, i.e., prostration and impaired consciousness, were introduced into the definition [9], and recent works have relied on a wider pragmatic case definition. For example, the Severe Malaria in African Children (SMAC) studies included children with P. falciparum detected on blood smear and classified as "being severely ill enough to be hospitalized", without further specifications [10].
In this context, we conducted a systematic review and meta-analysis to better understand the prognostic value of clinical and laboratory findings used to diagnose severe malaria in African children. This assessment was aimed at refining the commonly employed definition of severe malaria to then explore the possibility to define 'moderately severe malaria' cases that could benefit from much more accessible oral treatment.

Search strategy and sources
We performed a systematic literature search using Medline, Embase, Cochrane Database of Systematic Reviews, and Thomson Reuters Web of Knowledge. Study selection followed the Preferred Reporting Items for Systemic Reviews and Meta-Analyses (PRISMA) guidelines [11]. The first search was undertaken in January 2014, with an update in February 2015. We searched Medline and Embase using Medical Subject Headings and subheadings used for indexing articles. We combined the following terms: "malaria/complications OR malaria/mortality" AND "treatment outcome" AND "infant, newborn OR infant OR child OR adolescent". In the Cochrane Database, we looked for the words "malaria and children" in the main title of the review. We searched the Thomson Reuters Web of Knowledge using the words "malaria child", "Africa", "mortality" and "complications". We did not put any language or time restrictions on the search and we expanded it by examining the reference list of the selected studies. Additionally, we used three landmark articles [10,12,13] on severe malaria in African children to search for citations closely related to the selected article using the PubMed option "Related citations".

Inclusion and exclusion criteria
Studies reporting clinical and laboratory variables, including at least 100 children aged < 15 years who were diagnosed with severe malaria according to the WHO definitions, and which allowed reconstructing of two-by-two tables made up of outcome (survival/death) and presence/ absence of prognostic indicator, were included in this review. Controlled trials, non-controlled trials, cohort studies, case control studies and case series, both prospective and retrospective, were considered. When necessary, authors were contacted to obtain data to construct two-by-two tables. Two independent reviewers (BG and JD) conducted this search. Two [5,10] of the included studies served as reference publications for other enclosed publications, although no direct prognostic indicators could be extracted. Three selected studies [13][14][15] considered either partial or the whole population included originally in the study comparing artesunate with quinine in severe malaria treatment in Africa (known as the AQUAMAT study). In this case, the study with a greater number of study subjects with available clinical or laboratory features associated with death was selected. Two articles [16,17] encompassed the same study population though they focused on distinct clinical or laboratory variables; thus, both of them were retained. In addition, 356 out of 2901 children enrolled in a study in The Gambia [18] also participated in the AQUAMAT study, which leads to duplication of the subjects included in these two large studies.
Finally, in view of the size of the comprised population, we also considered data from the SMAC studies [19] in our systematic review, although study inclusion criteria did not fully comply with the strict WHO definition of severe malaria. Therefore, we performed separate analyses with and without the SMAC studies.

Quality assessment
Quality of selected studies and their risk of biases were assessed by applying the 2011 revised version of the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool [20], which was adjusted to the particularity of this review following the recommendation of the Cochrane Collaboration (details in Additional file 1) [21]. When the patients' inclusion criteria differed from the WHO criteria, we reported in the methodological quality assessment that there were great concerns about the applicability of the results to the research question. Regarding prognostic indicators, clinical and laboratory features were assessed separately. Furthermore, any reported death was considered as a reference standard. Studies including less than 80% of enrolled patients were labeled as highly biased. Quality assessment was performed by one reviewer (PS) and checked by a second reviewer (BG). Any disagreements were resolved through discussion and consensus.

Data extraction
Data on clinical features among children who survived or died were extracted by one reviewer (PS) using a standardized data extraction form and checked by the second (JD), as well as on random basis by the third (BG) reviewer. Information on characteristics (design, year of publication, study country, healthcare setting), study population (size, age range, mortality, inclusion and exclusion criteria), and prognostic indicators was gathered. Any identified errors were re-examined and corrected accordingly.

Statistical analysis
A two-by-two table including crossing variables, index test (0,1) and death (0,1), was constructed for each prognostic indicator. Odds ratios (ORs) were calculated to measure the association of each prognostic indicator with death. When a prognostic indicator was assessed in at least two studies, pooled estimates of ORs were calculated. A random effects meta-analysis was performed in the case of a significant heterogeneity among studies (P < 0.05). Otherwise, the fixed-effect approach was preferred. Metan command in STATA version 12 was used to perform these meta-analyses [22]. Results for all predictors were summarized in a Forrest plot, ordering markers from the least to the most strongly associated with death. The size of each predictor's box is proportional to the global sample size of studies involved in the corresponding summary ORs. Two separate analyses were conducted; one enclosing additional findings derived from the SMAC studies and one without it, covering studies that referred strictly to the definition of WHO as diagnosis criteria. Prognostic indicators with definitive thresholds and few without single definition (acidosis, hyperparasitemia, renal failure, respiratory distress, shock) were pooled for the usage of this analysis. The combination of symptoms was not analyzed in this systematic review due to unavailability of individual records.

Results
A total of 601 studies were identified and screened in the systematic database search. Through the selection process presented in the flow diagram ( Fig. 1), 30 titles [5,10,[13][14][15][16][17][18][19] published between 1994 and 2014 were selected and used to identify predictors; 28 were finally included in the meta-analysis (no direct data could be extracted from two referral studies). Overall, 90% of eligible studies were reported in English and 10% in French. The characteristics of the studies are outlined in Table 1. The summary of quality assessment of analyzed studies, according to the QUADAS-2 tool, is presented in Table 2. The detailed analysis of each study according to the QUADAS-2 tool was captured in Additional file 2.
A total of 36 different prognostic indicators associated with death due to severe malaria were identified in 30 studies. The number of predictors of mortality evaluated per study ranged from 1 to 19 (median 6.5, interquartile range [3][4][5][6][7][8][9][10][11]. Out of 36 identified prognostic indicators, 18 corresponded with the clinical criteria of severe malaria established by the WHO. Two forest plots displaying pooled estimates of ORs with 95% confidence intervals (CI) calculated for 17 and 18 prognostic indicators included in the WHO definition of severe malaria are captured in Figs. 2 and 3, respectively. Definitions and further characteristics of the analyzed prognostic indicators are assembled in Table 3.

Discussion
The results of the meta-analysis show that there is a large variation in the strength of the association between the different WHO-defined criteria of severe malaria and death. Renal failure, coma, hypoglycemia, shock, and respiratory distress represent those with the highest prognostic value. These manifestations were also those with the highest prognostic value for death in the original paper by Marsh [12], which was supportive of the WHO definition of severe malaria. Similarly, impaired consciousness, prostration, hyperpyrexia, hyperparasitaemia, and severe anemia were weak predictors both in the present systematic review and in Marsh's paper [12]. While 5039 (35.7%) of children from the enclosed studies suffered from severe anemia, its association with death, though widely acknowledged, was insignificant. This can possibly be explained by the fact that anemic children receive blood transfusion upon admission or by the lack of other concomitant feature such as respiratory distress or neurological impairment. On the other hand, hypoglycemia, which similarly to severe anemia could be reversed if early detected, remains a significant marker of severity, which can be possibly explained by its dependency on other   Krishna et al. (1994)  The current systematic review recognizes coma (defined as Blantyre coma scale (BCS) ≤ 2) and deep breathing as robust prognostic factors of pediatric life-threatening malaria that can simply be determined and recorded by skilled observers in all types of settings. Deep breathing, as a crucial respiratory sign of severe malaria, is commonly a compensatory manifestation of underlying metabolic acidosis [44] and is more predictive than respiratory distress accompanied by signs of variable severity. These findings are nearly in line with the results from a prospective study [12] of 1844 patients in Kenya, which identified respiratory distress and impaired consciousness (defined as prostration or coma) as highly associated with death and, except for prostration, with the Lambaréné Organ Dysfunction Score, which combines coma, prostration, and deep breathing [10].
Although there is no definite consensus regarding the strongest predictors of death within the WHO clinical definition of severe malaria, the WHO distinguished three groups [1] classing clinical and laboratory features of the disease in a way to facilitate appropriate treatment. A major contrast of our results with the clinical features included in the WHO Group 1 symptoms (prostrate but conscious, prostrate with impaired consciousness, coma, mild/severe respiratory distress, shock), which are supposedly more severe and for which parenteral treatment is recommended, is that a child with prostration or impaired consciousness appears to be at a low risk of death when compared with the presence of any other listed signs and symptoms. One possible explanation for this unexpected finding is that, in some studies, the definition of impaired consciousness was less stringent than that of the WHO (BCS < 3). Interestingly, in the differentiated group of 1289 Gabonese children, Issifou et al. [36] applied a BCS Fig. 2 Pooled estimates of odds ratios (with 95% confidence intervals) of each predictor of mortality assessed in at least two studies (without the Severe Malaria in African Children studies) and number of studies by each predictor. The size of each predictor's box is proportional to the global sample size of studies involved in the corresponding summary odds ratios. *Results calculated by fixed effects Fig. 3 Pooled estimates of odds ratios (with 95% confidence intervals) of each predictor of mortality assessed in at least two studies (including the Severe Malaria in African Children studies) and number of studies by each predictor. The size of each predictor's box is proportional to the global sample size of studies involved in the corresponding summary odds ratios. *Results calculated by fixed effects The present attempt to rank clinical features according to their prognostic values was performed to potentially better distinguish children that should definitely be receiving parenteral treatment versus those that could be considered for prompt oral treatment with artemisininbased combinations. At present, the WHO recommends injectable artesunate for all children with asexual forms of P. falciparum in peripheral blood and at least one criterion of severity [45]. In the light of the very different prognostic values of the different features, Kopel et al. [46] suggested that oral treatment could be a successful alternative for patients with a detected parasitemia and a criterion considered as less severe, e.g., jaundice. Certainly, all prognostic indicators that are able to be detected at the bedside need to be searched for, and finding a lowprognostic symptom or sign does not remove the need for parenteral treatment if a high-prognostic one is present. Identifying a subset of patients with moderately severe malaria who could be safely managed with oral treatment at the primary care level would simplify the patients' management in settings where referral to hospital for injectable treatment is difficult, and allow better resources allocation. A simplified approach may be easier to implement. Already, in settings where laboratory facilities are unavailable, the laboratory tests used to define severe malaria are not considered in the classification of the disease. This new approach should be carefully assessed in a prospective multicentric clinical trial to demonstrate its safety. To our knowledge, this is the first systematic review and meta-analysis of predictors of death drawn from all relevant studies of African children with strictly defined severe malaria. Methodological quality was assessed by using a priori adjusted and defined rules of the latest version of QUADAS-2 tool, which allowed better evaluation of risk of biases in several domains. In addition, this review assessed the disease severity criteria used in the SMAC studies [19]. Indeed, this represents the largest sample size ever recruited. The fact that the results did not change much when including or not prognostic indicators from the SMAC studies increases the robustness of the findings.
The main limitation of our analysis comes from the methodological or reporting weaknesses of some studies, of which the most important one is the lack of reproducibility of reported clinical symptoms and signs. Indeed, the inter-observer (clinician) agreement on the assessment of some of the signs, such as impaired consciousness or prostration for example, can be very low. Additionally, heterogeneity between studies regarding availability of laboratory data, threshold used to define abnormality, and quality of healthcare, especially with regards to blood transfusion and management of renal failure, need to be taken into account in results interpretation. Another limitation of our review is that it did not consider combinations of clinical and laboratory features of severe malaria because of the unavailability of individual records. It has been shown that having more than one manifestation of severe malaria increases the risk of dying [13] and this has to be taken into account in a child assessment of severity, and hence in case management. Furthermore, due to lack of data in the included studies, this meta-analysis could not explore the impact of other concurrent complications that do not form part of the definition of severe malaria but are known for increasing the risk of death such as, for example, bacteremia. In addition, since all data were aggregated in each study, we were not able to analyze predictors by age group or sex. This should not alter much the relevance of our findings since approximately 80% patient population was < 5 years of age and WHO has never considered a differential definition of severe malaria for children and adults or male and female. Finally, studies reporting less than 100 cases were excluded to reduce complexity, but some of those could have brought relevant information.

Conclusion
In conclusion, the findings of this meta-analysis show that the strength of association between the criteria defining severe malaria and death is quite variable for each clinical and/or laboratory features (OR ranging from 0.58 to 5.96). Despite the heterogeneity of entry criteria, the individual studies provided concordant results. A ranking allowed the identification of features weakly associated with death, such as impaired consciousness and prostration, which could assist to refine case definition and thus optimize antimalarial treatment.