A cross-sectional study of asymptomatic Plasmodium falciparum infection burden and risk factors in general population children in 12 villages in northern Uganda

Between October 2011 and February 2014, a cross-sectional survey of children in 12 random villages in 22 districts Northwest and North-central regions of Uganda (Fig. 1). The children were enrolled as pilot population controls in a case–control study entitled the Epidemiology of Burrkitt Lymphoma in East African children and Minors (EMBLEM) study [13, 18], which is part of a larger study investigating risk factors of eBL in Uganda, Tanzania, and Kenya [19]. The target regions span from 621 to 2000 m above sea level and include areas covered by savannah, slow flowing rivers, swamps, lakes, which are separated into two regions (Northwest and North-central) by a gorge of the East African Rift valley. Both regions experience warm (20–30 °C) and wet climate (1250–1500 mm of rainfall per year [20] in two rainy seasons: April to June and September to December), which is conducive for mosquito breeding and perennial malaria transmission [12]. The malaria prevalence in these regions is high, but geographically heterogenous (prevalence ranging from 3.2 to 75.8% [13]). As noted in the introduction, IRS and LLIN were implemented in 10 districts in the study area between 2008 and 2015 [21], and universal LLIN has been implemented since 2017.

The target population of children (0–< 16 years old) in northern Uganda was estimated to be 3,031,494 in 2015, based on the national population census data [20]. This population was sampled using a stratified multi-stage cluster sampling design (Fig. 2) [13]. In the first sampling stage, 12 census enumeration areas (EAs) were randomly selected from 4 strata, defined by ‘low population-density’, ‘high population-density’, ‘near water’, and ‘far from water’, from a list of all EAs obtained from the Uganda Bureau of Statistics (UBOS) [20]. These strata are expected to be associated with malaria transmission [22, 23]. Population density of an EA was categorized as low if it was below the median population count (n = 2683), and high if it was equal or higher than the median population count [13]. Proximity of an EA to water was defined as near surface water when the EA boundary was next to or within 500 m of an all-season swamp, river, or lake, based on distances estimated from national maps incorporating geographical information metadata; otherwise, it was defined as being far from water. Four EAs were sampled from the low-density near water stratum and the low density far from water stratum, and 2 EAs from the high density near water stratum and the high density far from water stratum. In the second stage, one village per EA was randomly selected and a household survey conducted, and eligible children in the household invited to participate. Children were eligible if they were aged 0–< 16 years old, were usual residents of the household, and were apparently healthy, i.e., not having symptoms requiring hospital care. Children were not eligible of they had a cancer diagnosis or symptoms requiring hospital treatment.

An experienced field team consisting of laboratory technicians, interviewers, and a local guide, led by a local medical doctor, visited consecutive households in each village to enroll eligible children. For practical reasons, enrollment in each village continued until a sample size of at least 70 children was obtained. The sample size per village was based on a target sample size of 840–1080 children from 12 villages, which was needed to enable precise estimation of population of pfPR in healthy children in the 12 villages as well as in each village. This calculation assumed overall malaria prevalence of 50%, that each village would have at least 30 households with an average household size of 2–3 eligible children, yielding a sample size of 60–90 children per village. Enrollment in each village lasted 2–3 weeks per village, and between October 2011 to February 2014 for the 12 villages. The period of enrollment was prolonged because the study design stipulated that about half of the villages should be enrolled during the dry season and the other half during the wet season. This necessitated phased implementation to enable the single team would be able to mobilize and safely enroll children in all the villages, which were remote, located far away from one another and from the EMBLEM field laboratories (Fig. 1). Written informed permission for the child to take part was obtained from the child’s guardian (usually one or both parents), and assent from children aged 8 years or older. Structured questionnaires were administered to record each child’s age, sex and household characteristics, parental educational level, malaria prevention methods, including ownership and use of LLIN the night before interview and application of IRS in the house, and a history of outpatient or inpatient malaria treatment (≤ 6, 7–12, or ≥ 13 months ago).

Venous blood for research (10 mL) and clinical (4 mL) tests was drawn in EDTA tubes from each child. The research blood samples were transported in cold boxes to local EMBLEM field laboratories within 2 h from sampling, and centrifuged for 15 min at 1300g to separate plasma, buffy coat, and red cell fractions that were stored in barcoded cryovials at − 80 °C. Clinical samples were immediately tested for malaria parasites by experienced local technicians using thick film microscopy (TFM) to visualize asexual malaria parasite forms and commercial rapid diagnostic tests (RDT) (MALARIA DUAL kits, ICT Diagnostics, Muizenberg, Cape Town, South Africa) to detect malaria antigens [13]. TFM was performed on slides stained with 10% Giemsa solution for 10 min, and the visualized parasites were counted against 200 white blood cells (WBCs) and the results expressed as parasites/µL of blood, based on the measured WBC count. The RDT kits used in the current study detect the P. falciparum-specific malaria histidine-rich protein 2 (Pf-HRP2) and the pan-lactate dehydrogenase (pLDH) antigen and have a reported sensitivity and specificity of 92–100% in Uganda [24]. Because malaria antigens may appear in blood before asexual parasites and remain in blood for several weeks after asexual parasites have been cleared from blood, RDTs were used as the primary outcome measure to capture malaria infection defined by both visualized parasites and circulating antigens [25]. Although RDT tests can yield false-negative results, particularly for parasites that have deleted pfhrp2 or pfhrp3 genes that code for the RDT antigens [26], the frequency of such parasites is low (< 1.1) and was considered unlikely to distort the general malaria epidemiological patterns. Thin film smears were examined to identify Plasmodium species. Because ~ 98% of the parasites were P. falciparum, this species is assumed hereafter.

Questionnaires and result forms were reviewed for completeness and accuracy, and computerized by DataFax or computer data entry. DataFax uses intelligent character recognition built-in capabilities to perform data entry and reduce data entry errors. Logical consistency checks were performed on the electronic data, and data queries arising were corrected before creating analysis files.

The response rate was calculated as the proportion of enrolled children out of all eligible children in the selected households that were enrolled. The outcome variable for this analysis was P. falciparum prevalence (pfPR), based on RDT positivity, which captures infection defined by visualized asexual parasites and circulating falciparum antigens [25]. Although polymerase chain reaction (PCR) would provide a more accurate result for antigenemia. One child who reported a fever on their questionnaire and had a high parasite count on TFM (> 2500 parasites/uL) was considered to have clinical malaria excluded from analysis. Children who reported a fever on their questionnaire but were malaria negative (n = 6) or had low parasitemia (n = 11, parasite load ranging from scanty to 1200 parasites/uL) were considered to have incidental malaria and not excluded from analysis. The geometric mean parasite density (GMPD) per µL was assessed overall and by age group to confirm low-level parasitemia among children with visualized parasites, as would be expected in high-endemic areas [9].

The results were weighted by the inverse probability of being sampled into the study, and are weighted back to the general population of children aged 0–< 16 years old in Northwest and North-central Uganda in 2015 (projected to be 3,138,360). Weights were trimmed by replacing the value of the weights in the highest 3% of the weight distribution with the value of the weight at the 97th percentile to minimize the impact of outlier weights. Stratification and clustering at the village and household levels were accommodated in the variance computations of estimates. The reported results are weighted estimates, unless stated otherwise.

The analyses were performed using the “survey” package [27, 28] (v. 3.32-1) in R (version 3.4.3, r-project.org) following methods previously described [13]. Descriptive analyses of weighted pfPR are reported overall, by stratum, by village, and by demographical and geographical variables as in [13]. Associations with pfPR were assessed for: (a) household variables reflecting crowding (number of other children in the household), parental characteristics (mother’s income, parents’ occupation), malaria infection in the enrolled household members (any child with infection, infection in a child below 5 years old, infection in a younger sibling); (b) individual characteristics, including history of fever due to malaria or non-malaria conditions (at enrollment, in the past 6 months, in the past 12 months); and (c), use of herbs for treatment of skin or gum disease and lifetime number of admissions. When considering the number of other children in the household or the number of other children in the household below 5 years old, this variable was coded as 0 if there was no other child in the household or no child below 5 years in the household, otherwise it was coded as the number of other children or the number of children below 5 years in the household. Associations between pfPR and keeping different types of animals inside the house or nearby were evaluated because mosquito blood-feeding preferences (anthropophily/zoophily) could influence the risk of malaria positivity [29]. Unadjusted and adjusted odds ratios (ORs), standard errors [30] and Wald-type 95% CIs (95% CIs) [31] of association of pfPR with each variable were calculated. Forward stepwise logistic regression was used to construct adjusted models, using the covariates that in unadjusted models had associations with prevalence with P < 0.05. Covariates with several levels were coded with dummy variables for the categories in the univariate analysis and using trend coding in the adjusted analyses. As the objective of these analyses were descriptive and for hypothesis generation, statistical tests were not adjusted for multiple testing. A two-sided P < 0.05 was considered statistically significant.

As shown in Fig. 2, 1007 (86.2%) of 1168 eligible children in 354 (79.5%) of 445 households randomly selected from 1121 eligible households in 12 villages were enrolled. These children represent a trimmed weighted sample of 2,829,988 children aged 0–15 years (Additional file 1: Table S1 and Additional file 2: Fig. S1). The demographic distribution was 45.4% being males and 54.6% females; 42.7% were children aged less than 5 years old, 34.4% aged 5–9 years old, and 23.9% aged 10–15 years old. The size and age distribution of the weighted population were comparable to the corresponding population in the study area in 2015, based on data from UBOS [20].

Of 1007 enrolled children, 942 (93.5%) were successfully tested for malaria. One subject who reported a fever at enrollment and had parasite load > 2500 parasites/uL was excluded from analysis. In the remaining set, the weighted pfPR was 52.4% (95% CI 19.9–84.8%) by RDT read by experienced local technicians and 32.7% (9.3–56.2%) by TFM performed by experienced local technicians (Table 1). The sensitivity of RDT to detect asexual parasites was 97.5% (95% CI 95.8–99.7%) and the specificity was 69.7% (95% CI 43.4–96.1%), compared to TFM by experienced local technicians.

The overall weighted GMPD was low 733.3 parasites/µL. GMPD was significantly higher in children who were both RDT and TFM positive than those who were TFM-positive but RDT negative (768 versus 107 parasites/µL, P = 0.002). GMPD did not vary by sex (P = 0.167), and was not significantly different in children aged below 5 years versus older (955 versus 615 parasites/µL, P_difference = 0.201). GMPDs showed modest variation across the villages (189–724 parasites/µL, except for two villages with an average of 1357 and 1807 parasites/µL).

The weighted pfPR in villages varied from 10.6% in village 19 in the North-central region to 86.2% in village 40 in the northwest region (Table 2). pfPR was lower in IRS villages and higher in non-IRS villages (18.4% versus 75.2%, P < 0.0001), but it varied both within IRS (10.6–34.8%) and non-IRS villages (63.6–86.2%). No overall differences were noted between wet versus dry seasons (66.9% versus 46.2%, P = 0.40). However, within IRS villages, pfPR was lower in villages enrolled in the dry season compared to those enrolled in the wet season (10.6–17.6% versus 20.1–34.8%). No differences in pfPR by season were observed within non-IRS villages (dry: 63.6–75.2% versus wet: 74.8–86.2%; Table 2).

The stratum-specific weighted pfPR by season are shown in Fig. 3. Within the low-population density stratum, pfPR in the wet and dry season was similar (67.3% versus 70.8%, P = 0.802). Conversely, within the high-population density stratum, pfPR was higher in the wet season than in the dry season, but not statistically significant (62.1% versus 37.9%; P_{heterogeneity} = 0.589; Fig. 3a). GMPD did not vary by season in the low- or high-population density strata (Fig. 3b), but the GMPD tended to be higher in the dry season than the wet season in the high-density stratum.

The age-specific pfPR was higher in non-IRS versus IRS villages, but slight differences in the actual peaks. pfPR peaked between 6 and 8 years old in IRS villages and slightly later between 9 and 11 years old in non-IRS villages (Fig. 4a). Among children with visualized parasites, the age-specific-GMPD was higher in non-IRS villages (Fig. 4b), in children up to 6–8 years of age and then decreased rapidly in older children. The GMPD in children in IRS villages was very low for all ages, although a spike was observed in children aged 6–8 years old (Fig. 4b), probably due to random fluctuation.

The associations between pfPR and children’s demographical, malaria prevention, illness history, lifetime malaria treatment, environmental, and household characteristics are shown in Table 3. No differences in pfPR were noted between females and males (53.3% versus 51.4%; P = 0.768), by distance of the home from the source of drinking water being ≥ 1 or < 1 km (55.7% versus 49.4%; P = 0.723). pfPRwas associated with keeping goats inside or near the house (49.8% versus 60.5%, OR 0.65, 95% CI 0.48–0.87; P = 0.021).

Compared to living in a non-IRS village, living in an IRS village was associated with decreased pfPR (18.5% versus 75.2%, OR 0.07, 95% CI 0.05–0.11). Consistent with this result, pfPR was decreased in children for whom their questionnaire reported that IRS was applied to their house, as compared to no application (23.7% versus 59.0%, OR 0.22, 95% CI 0.05–0.86). Consistent with previous results [13], sleeping under an LLIN on the night before enrollment, versus not, was not associated with decreased pfPR (64.6% versus 48.8%, P = 0.395). Compared to having no other child in the household, having one or more other children in the household, as was associated with decreased pfPR (P_{heterogeneity} < 0.0001; Table 3). The results were similar when the number of children below 5 years old was considered (P_{heterogeneity} = 0.0002).

Reporting a history of non-malaria related fever in the past 6 months, versus not, was unrelated to pfPR (60.6% versus 51.4%, P = 0.384). However, reporting four or more malaria-related fever episodes in the past 6 months, versus none, was significantly associated with elevated pfPR (OR 2.80, 95% CI 1.11–7.05, P_{heterogeneity} = 0.0003). Compared to children reporting a history of inpatient malaria treatment in the past 12 months, pfPR was lower in those reporting a history of inpatient malaria more than 12 months ago (OR 0.72) and those reporting no such a history ever (OR = 0.47, P_trend = 0.026; Table 3). In contrast, compared to children reporting a history of outpatient malaria treatment in the past 12 months, the pfPR was elevated in those reporting a history of outpatient treatment of malaria more than 12 months ago (OR 1.83, 95% CI 1.15–2.91), but decreased in those reporting no such history ever (OR 0.57, 95% CI 0.41–0.80, Table 3).

A few children (n = 19) who reported a fever on their questionnaire were deemed to have incidental malaria at the time of enrollment. Compared to those who did not report a fever, reporting a fever was associated with pfPR (OR 5.38, 95% CI 2.01–14.39). pfPR was associated with having at least one sibling who was RDT positive (versus none: OR 4.24 95% CI 1.61–11.15) and with having a younger sibling who was RDT positive (versus none: OR 2.20 95% CI 1.40–3.46). Season, parental education and occupation, herbal treatment, and keeping other animals (other than goat) inside or near the house was not associated with pfPR (Table 4).

Only 6 variables of 14 evaluated using stepwise logistic regression remained significantly associated with pfPR in the multivariable model. Specifically, pfPR was inversely associated with living in an IRS village (adjusted OR 0.06, 95% CI 0.04–0.07), with having two or more other children in the household (versus none, P_trend < 0.0001), or 2–4 children in the household aged below 5 years of age (versus none, P_trend = 0.014), and keeping a goat inside or near the house (aOR 0.42, 95% CI 0.29–0.62). Conversely, pfPR was positively associated with reporting a fever at enrollment (aOR 5.02, 95% CI 1.95–12.92) and having at least one sibling who was RDT positive (aOR 5.39, 95% CI 2.94–9.90).