Study Site and Study Population

The study was conducted at the pediatric ward of the Agogo Presbyterian Hospital (APH), a district hospital with 250 beds, situated in the Asante Akim North municipality in Ghana. The municipal area has an estimated population of 140694 inhabitants, spread over an area of 1160 km2 (2010 Census Data, Ghana Statistical Service). The region has a tropical climate and is mainly covered by secondary rain forest and cultivated land. Malaria is highly endemic in that area with seasonal peaks [20, 21].

Children ≥30 days and ≤15 years of age with a tympanic temperature ≥38°C admitted to the pediatric ward were recruited between November 2013 and April 2015. Repeated visits of participants were considered as new visits if they were at least 30 days apart or in case children were diagnosed with a new disease.

Between September 2014 and September 2015, healthy children <15 years of age with a tympanic temperature <37.5°C and without signs of infection were recruited at vaccination clinics in the surroundings of the APH. Venous blood samples were drawn for malaria microscopy, performed as described in the Supplementary Materials and Methods.

Clinical Examination and Sampling of Specimens at the Hospital

After a physical examination by a pediatrician, 2–5 mL full blood, serum, and ethylenediaminetetraacetic acid (EDTA) blood were collected. A urine sample was obtained from each child. Nasopharyngeal flocked swabs (Copan FLOQSwabs, Copan Diagnostics) were taken from all children with signs of a lower respiratory tract infection. Children with symptoms of lower respiratory tract infections and indistinct lung auscultation findings received a chest radiograph. A stool sample was collected from children presenting with diarrhea and cerebrospinal fluid samples were obtained from all children with meningeal signs or with clinical suspicion of meningoencephalitis (Supplementary Table 2). All samples were collected before drug administration and the medical treatment followed national and hospital guidelines. The diagnostic methods for each specimen and the clinical case definitions are described in the Supplementary Methods and Supplementary Table 2.

Epidemiological Analysis

Descriptive statistics were applied to summarize patient data. Categorical variables were displayed with percentages, and continuous data with the median and interquartile range (IQR). Patient diagnoses were presented stratified by age or malaria parasitemia manifestation. Data analysis was performed with Stata version 14 software (StataCorp, College Station, Texas). Details of the statistical data analysis can be found in the Supplementary Statistical Analysis Plan.

Ethical Considerations

The Committee on Human Research, Publications and Ethics, School of Medical Science, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana, and the Ethics Committee of the Ärztekammer Hamburg, Germany, approved the study design and the informed consent procedures in the underlying studies. All participants were informed about the study’s purpose and procedures. In older children, written informed consent was obtained from the patients and their parents or, in case of infants, the parents or legal guardian provided written informed consent prior to enrollment.

Hospitalized Children with Fever

From 4 November 2013 to 30 April 2015, a total of 4169 patients were admitted to the pediatric ward of the APH (Figure 1). Of these, 1238 (30%) fulfilled the study inclusion criteria and were recruited to the study. Median age of recruited children was 2 years (IQR, 1–4 years). A total of 561 (45%) patients were female and the median tympanic temperature was 39.0°C (IQR, 38.5°C–39.6°C) (Table 1). Of all children included in the study, 1078 attended the hospital once, 66 twice, 8 three times, and 1 four times during the study period.

Plasmodium Species Infection

Infection with malaria parasites was found microscopically in 728 patients (59%). Of these, 679 (93%) were P. falciparum monoinfections and 41 (6%) were mixed infections (Table 2 and Supplementary Table 3). Parasite densities varied greatly, with a median count of 74385/µL (IQR, 25508–201407/µL) (Table 2).

A total of 117 (16%) children with parasitemia were diagnosed as severe malaria, of which 17 (15%) were classified as cerebral malaria (Supplementary Table 2). Severe malaria was primarily observed in children aged 1–4 years (n = 95/740 [13%]). Nonsevere malaria increased from the first year of life (n = 58/205 [28%]) among the age groups 2–4 years (n = 269/419 [64%]) and >4 years (n = 163/270 [60%]) (Figure 2). In 370 (51%) of the parasitemic children, a co-diagnosis with one of the below-mentioned conditions was detected.

Lower Respiratory Tract Infections and Pneumonia

Respiratory tract infections were diagnosed in 410 (33%) children and were categorized as lower respiratory tract infection (LRTI; n = 290) and pneumonia (n = 120). Both LRTI (n = 62/218 in this age group [28%]) and pneumonia (n = 43/218 [20%]) were most frequently diagnosed in children <1 year of age (Figure 2). Respiratory pathogens were detected in 209 (72%) LRTI and 90 (75%) pneumonia cases. Frequently identified pathogens are listed in Table 2.

Generally, the proportion of both LRTI (Relative Risk [RR], .6; 95% confidence interval [CI], .5–.7) and pneumonia (RR, .2; 95% CI, .1–.3) was lower in children with parasitemia. LRTIs and, in particular, pneumonia were less frequently diagnosed with increasing parasite counts (Figure 3).

Urinary Tract Infection

Urinary tract infections (UTIs) were diagnosed in 218 (18%) patients. With 27% (n = 77/285), the proportion of UTI was highest in children >4 years (Figure 2), in contrast to only 15% (n = 141/953) in children <4 years. UTI was more common in girls (n = 150/561 [27%]) than in boys (n = 68/677 [10%]). Pathogens were identified in 32 (15%) children with UTI, most frequently E. coli (n = 25 [78%]) (Table 2). Plasmodium coinfections in UTI cases occurred independently of parasitemia (RR, 1.0; 95% CI, .8–1.3) (Figure 3).

Gastrointestinal and Abdominal Infections

Gastrointestinal infections were detected in 210 (17%) children. Most affected were children aged <1 year (n = 73/213 [34%]), and the frequency decreased with increasing age (Figure 2). Pathogens were found in 97 (45%) patients, with rotavirus being the most frequently detected pathogen (n = 32 [33%]) (Table 2).

The proportion of gastrointestinal infections was highest in children without malaria parasitemia (n = 110/510 [22%]); however, no clear statistical difference was observed (RR, .8; 95% CI, .7–1.1) (Figure 3).

Bloodstream Infection

Bacterial bloodstream infections were found in 62 (5%) children, of whom 43 (69%) exhibited systemic inflammatory response syndrome. The proportion of bloodstream infections increased with age and was highest in the age group >4 years (n = 17/285 [6%]) (Figure 2). Most common blood isolates were nontyphoidal Salmonella (n = 28/62 [45%]) and Salmonella Typhi (n = 19/62 [31%]) (Table 2 and Supplementary Table 1).

Bacteremia occurred predominantly in children without parasitemia (RR, .1; 95% CI, .04–.2), where 54 diagnoses occurred (11%). In children with parasite counts >10000/µL, only 4 (of 623 [1%]) presented with bacteremia, 3 caused by nontyphoidal Salmonella (Figure 3). The stratified analysis showed similar patterns among the established age groups (Figure 3).

Infection of the Central Nervous System or Meningitis

Central nervous system (CNS) infection/meningitis was observed in 11 (1%) study children across all age groups (Figure 2). In 3 (27%) cerebrospinal fluid samples, pathogens were identified (Table 2 and Supplementary Table 1). No known meningitis/encephalitis-linked viruses were detected by next-generation sequencing. Low numbers of CNS infection/meningitis patients did not allow any conclusions on associations with parasitemia (Figure 2).

Other Diagnoses

Systemic viral infections were observed in 58 (5%) patients, predominantly in children >2 years of age (n = 47 [6%]) (Table 2 and Supplementary Table 1).

In all 15 human immunodeficiency virus (HIV)–infected children, at least 1 additional clinical diagnosis could be established, namely LRTI (n = 6 [40%]), pneumonia (n = 6 [40%]), diarrhea (n = 5 [33%]), parasitemia (n = 4 [27%]), bacteremia (n = 3 [20%]), UTI (n = 2 [13%]), and tuberculosis (n = 2 [13%]). Therefore, HIV infection was not considered an acute infection potentially responsible for the current fever episode.

Other clinical diagnoses were skin-joint-bone infections (n = 82 [7%]), mastoiditis (n = 3 [0.2%]), pulmonary tuberculosis (n = 2 [0.2%]), mumps (n = 1 [0.1%]), and lymphatic filariasis (n = 1 [0.1%]).

Nineteen study children died during their hospital stay. The following diagnoses were made in these children (multiple diagnoses possible): pneumonia (8 [42%]), CNS infection/meningitis (n = 6 [32%]), LRTI (n = 4 [21%]), severe malaria (n = 4 [21%]), bacteremia (n = 4 [21%]), and HIV/tuberculosis coinfection (n = 1 [5%]).

In 109 (9%) of study participants, the etiological cause of the febrile illness remained unidentified (Table 2).

Asymptomatic Parasitemia in Afebrile Children

In total, 564 healthy children with a median age of 1 year (IQR, 0–3 years) were recruited between September 2014 and September 2015 in the study catchment area of the hospital (Table 3). In 87 (15%) participants, malaria parasites, predominantly P. falciparum (n = 82 [94%]), were detected. Median parasite count was 1696/µL (IQR, 416–6360/µL), ranging from 48 to 82520 parasites/µL.

Plasmodium Species Coinfections

Patients with high parasitemia have a lower likelihood for the additional diagnoses LRTI, pneumonia, and bloodstream infection (Figure 3). Hence, Plasmodium coinfections were observed more frequently in patients with low parasitemia.

Figure 4 shows the cumulative proportions of parasitemic children along increasing parasite counts for fever-free controls, children with both parasitemia and an alternative diagnosis and those with Plasmodium monoinfection. In children with Plasmodium coinfections, 17% (n = 60/356) and, in those with Plasmodium monoinfection, only 12% (n = 45/372) had a parasite count <10000/µL. At the parasite level of 100000/µL, 100% of the fever-free controls, 64% (n = 226/356) of the coinfected patients, and 54% (n = 192/372) of the monoinfected children had a parasitemia below that value. Thus, children with Plasmodium monoinfection had higher parasite counts compared with coinfected patients.

DISCUSSION

Using an extensive diagnostic panel, a clinical-microbiological diagnosis was made in 91% of hospitalized children with severe febrile illness. This is in line with findings from East Africa where a similar study in pediatric outpatients was able to identify possible causes of fever in 97% of children. Despite reports on decreasing malaria incidences from several sub-Saharan countries, the frequency of P. falciparum infections was still high and was found in more than half of the febrile children. Accordingly, a high number of patients belonged to 1 of 2 groups that are difficult to differentiate: (1) those with a nonmalarial disease and an asymptomatic Plasmodium parasitemia, and (2) those with a comorbidity of malaria and a nonmalarial disease.

Markedly, in the present study the likelihood of distinct nonmalarial diseases was dependent on parasite densities among coinfected patients. The frequency of respiratory tract and bloodstream infections was highest in children without malaria parasites and decreased with increasing parasite counts. In contrast, higher parasite counts were detected in children without nonmalarial comorbidity. This mutual exclusion of high parasitemia and other fever causes might be due to a selection bias (Berkson bias) where, within clinical settings, children with a severe nonmalarial disease (eg, bloodstream infection, pneumonia) have a reduced likelihood of another severe concomitant febrile infection being the reason for their fever symptoms and hence the hospital admission. As outlined above, 11% of patients without parasitemia had a bacterial bloodstream infection, while this was only diagnosed in 1% of patients with a parasite count >10000/µL. Therefore, the determination of the patients’ parasite densities is essential for the recognition of a nonmalarial disease in healthcare facilities within malaria-holoendemic areas. This information, however, is not provided by the immunochromatographic rapid test available today but rather by gold-standard malaria microscopy if performed by well-trained staff. For now, in-depth investigations for other pathogens should be initiated in case of discrepancies between disease severity and parasite count. Nevertheless, it has to be emphasized that all critically ill children with parasitemia must receive immediate antimalarial treatment, regardless of parasite counts.

To attribute severe febrile disease with P. falciparum infection to malaria, different parasite density cutoffs have been proposed or applied, most of them ranging between 5000 and 10000 parasites/µL. In the present study, 15% of healthy but parasitemic children had a parasite count >10000/µL, demonstrating that a parasite density cutoff has a weak specificity and should be adapted to the purpose (eg, patient management or clinical trial case definition). Furthermore, local malaria endemicity and age distribution has to be taken into account when deciding on a diagnostic parasite density cutoff. The crucial factor is the likelihood of asymptomatic parasitemia that is dependent on the grade of semi-immunity in a specific population. Consequently, local endemicity is an essential information for the malaria diagnosis and should be thoroughly assessed.

Acute lower respiratory tract infections were one of the most common reasons for hospitalization in the study area, which has also been reported for children <5 years worldwide. The bacterium Streptococcus pneumoniae was diagnosed in half of the children with LRTI despite the introduction of a pneumococcal vaccine in 2012 in Ghana. Although the pathogen may colonize the respiratory tract as a commensal, its presence as a risk factor for severe disease in viral coinfections has been demonstrated before.

Rotavirus and Giardia lamblia were the most common enteric pathogens in this study. Despite the introduction of a monovalent vaccine in Ghana in May 2012, the high frequency of rotavirus in children with acute diarrhea in this study may hint toward reduced vaccine effectiveness due to high rotavirus diversity in this region. The clinical relevance of G. lamblia infections has been questioned in large case-control studies [27, 28]. Similar to respiratory pathogens, it is a challenge to distinguish between intestinal carriage and true gastrointestinal infections.

Salmonella disease is the major cause of invasive bacterial febrile illness in children in sub-Saharan Africa and known to be associated with P. falciparum malaria [15, 23]. Even though in this present study, parasitemia-bacteremia coinfections are rare, they are mainly caused by nontyphoidal salmonellae.

This study is limited by the fact that the healthy control cohort was seen only once during their vaccination visit and that follow-up visits were not performed due to logistical reasons. However, a clinical examination and a medical anamnesis (data not included) provided robust insights into the current health status and we believe the selected controls were disease free. Furthermore, the healthy control group could not be used to associate other pathogens with disease symptoms. As paired sera were not available, the detection of Brucella species, Leptospira species, Borrelia species, Rickettsia species, Coxiella burnetii, and Bartonella species was performed by polymerase chain reaction (PCR) instead of the gold-standard serological methods. The frequency of these zoonotic infections may be therefore underestimated. Furthermore, no PCR methods were used to detect and quantify submicroscopic parasitemias, which may lead to an underestimation of P. falciparum infections.

CONCLUSIONS

Our study highlights P. falciparum as a major pathogen found in severely ill children admitted to a hospital in Ghana. The likelihood of comorbidity with a nonmalarial disease is dependent on the level of malaria parasites in the blood. Hence, parasite densities provide important information for patient management, in particular for antimicrobial medication. Currently available rapid malaria tests might not be sufficient for this decision, and semiquantitative rapid tests of malarial parasitemia are required as long as reliable microscopic malaria diagnosis and extensive diagnostics of nonmalarial causes of pediatric febrile illness are not available in resource-poor settings.