Age and geographic patterns of Plasmodium falciparum malaria infection in a representative sample of children living in Burkitt lymphoma-endemic areas of northern Uganda

To obtain geographically and demographically representative data about malaria in northern Uganda, between January 2011 and April 2015, apparently healthy children (0–15 years of age) from the northwest and north-central regions of Uganda were enrolled as part of the EMBLEM study of BL [23]. These two regions were selected because they have historically experienced holoendemic malaria transmission [20] and they experience a correspondingly high, albeit geographically variable, BL endemicity [7]. The annual entomological inoculation rate (AEIR) varies from 397 in  the Arua District in the northwest region to 1586, the highest number recorded worldwide, in  the Apac District in the north-central region [24]. Between 2009 and 2012, the Government of Uganda responded to this high AEIR by implementing indoor residual spraying (IRS) of insecticides and distributing insecticide-treated bed nets to pregnant women and children under 5 years old in 10 districts in the north-central region, hereafter called the IRS sub-region [21]. The non-sprayed area is called the non-IRS sub-region.

A stratified multi-stage cluster sampling design was used to select a regionally representative, population-based sample of healthy children ages 0–15 years old from was selected 100 villages in the study region (Fig. 1). To select the villages, 100 census enumeration areas (EAs) were selected from a complete list of EAs in the study region obtained from the Uganda Bureau of Statistics (UBOS) [25]. This EA sample frame was stratified on ‘low-density’ versus ‘high-density’ population and ‘near water’ versus ‘far from water’ before sampling. Since the stratifying variables are thought to influence mosquito ecology and through that malaria transmission [26, 27], they were considered domains of special interest when estimating geographical patterns of malaria. Low-density EA stratum, considered a surrogate for rural areas, was defined as including those EAs having a population count of children younger than 15 years fewer than the EA mean population count (n = 2683) [25]; otherwise the EAs were categorized as belonging to a high-density EA stratum, considered a surrogate for urban areas. Near water EA stratum was defined as including those EAs with a boundary next to or within 500 m of an all-season surface water body (a swamp, river, lake), based on distances estimated from national maps incorporating geographical information metadata; otherwise the EAs were categorized as belonging to a far from water EA stratum. Because 90% of the Ugandan population resides in rural (or low-density) areas, the high-density EA stratum were oversampled by selecting low-density: high-density EAs in a ratio of 2:1. Thus, 68 EAs were sampled from low-density strata and 32 EAs from high-density with equal probability in each stratum, maintaining a ratio of 1:1 for near water versus far from water EAs within each stratum.

Because the EAs are large geographical areas that were logistically challenging for research staff to accurately survey, they were segmented into four to 17 units per EA using existing local administrative units called Local Council 1 (LC-1), also known as villages and that have distinct non-overlapping boundaries, as segments. One segment was randomly selected with equal probability per EA. Thus, one segment or village represented the entire EA. In the next stage, 22–25 households per segment were selected with equal probability from a household list constructed by study staff. In the third stage, four to 25 children were selected from a list of all children in the selected households in that segment, based on residing in the study region for at least 4 months and according to the predetermined age and sex frequency distribution of typical BL cases in the region [7].

Experienced fieldwork teams visited participants in their homes and invited them to participate in the study after obtaining informed consent. Structured questionnaires were used to collect participant information about the demographics and household characteristics, educational level of a child’s parents, a child’s exposure to malaria suppression methods (use of mosquito bed nets the night before interview and use of IRS in the house), and history of outpatient or inpatient malaria treatment (up to 6 months ago, 7–12 months ago, or 13 or more months ago). A venous blood sample for immediate malaria testing and for storage for future studies was obtained from the participant child using EDTA tubes.

Malaria infection was diagnosed by experienced laboratory technicians using thick film microscopy (TFM) to identify asexual malaria parasite forms in thick film slides stained with 10% Giemsa solution for 10 min. Malaria parasites were counted against 200 white blood cells (WBCs) and standardized to parasites/µL of blood based on the measured WBC/µL. Thin film smears were examined to identify Plasmodia species. Since ~98% were P. falciparum, this species is assumed hereafter. The technicians also used antigen–antibody capture rapid diagnostic tests (RDT) (MALARIA DUAL kits, ICT Diagnostics, Muizenberg, Cape Town, South Africa) to diagnose malaria infection. These kits detect the P. falciparum-specific malaria histidine-rich protein 2 (Pf-HRP2) and the pan-lactate dehydrogenase (pLDH) antigen shared by other Plasmodia that parasitize humans. The sensitivity and specificity of the kits for malaria in clinical samples in Uganda was reported to be 92–100% [28], which is adequate for the objectives of assessing epidemiological patterns of malaria in the current study. As only 2.1% of the children studied had fever at the time of enrolment, this number was considered negligible and unlikely to impact the overall results so these children were not excluded. The children with fever and positive RDT or TFM were treated for possible clinical malaria with artemisinin-based combination therapy (ACT) as first line drug for treatment of mild clinical malaria following national treatment guidelines. Parents of children without symptoms were advised to seek treatment from their local health centre if a new fever developed.

Questionnaires were edited in the field and computerized using DataFax and the data submitted via a secure share-portal to Information Management Services, Inc. (IMS, Calverton, MD, USA) to generate analysis files.

Descriptive analyses of pfPR, based on RDT, by geographical and demographical characteristics of participants are reported. To obtain regionally representative results, prevalence estimates were calculated as weighted probabilities of selecting the children in the study sample using standard survey statistical packages [29, 30] implemented in R (version 3.2.3). Details of the calculation of sampling weights are described in Additional file 1: Appendix 1 and 2. Thus, the reported estimates are representative for all the children in this region and the variance estimation took the weights into account and accounted for the clustering of the sample of children in the segments (villages). Unadjusted and adjusted odds ratios (ORs), standard errors [31] and Wald-type 95% CIs (95% CIs) [32] of association of pfPR with each variable were calculated. The adjusted models included only those variables with P < 0.10. Because parasite load in asymptomatic children is partly a function of cumulative acquired immunity to malaria, the overall geometric mean parasite density (GMPD) per µL were calculated and compared by sex and age using the Student’s t test statistic. As the objective of these analyses was to describe patterns to obtain granular baseline data for hypothesis generation, the results were not adjusted for multiple testing. Thus, a two-sided P < 0.05 was considered as evidence of statistical significance and values P < 0.10 as suggesting a trend. Analyses of malaria patterns were repeated separately using pfPR defined based on only TFM and on both TFM and RDT and similar results were obtained.

Overall, 2484 households were selected in 100 villages and 2467 of them participated. From these households, 1167 children were invited to participate and 1150 of them were enrolled, representing a weighted sample of 585,762 children in the region. Consistent with the design, to select children with a typical age and gender distribution typical of BL, 44.6% children participating were 6–10 years old and 52.1% were males (Table 1). Consistent with the predominantly rural population of northern Uganda, 67.6% of the children lived in low-density villages and 89.3% lived in villages near water. By region, 60.8% were from the north-central region and 39.2% from the northwest region (the location of these regions is shown in Fig. 2a). With respect to malaria suppression, 32.9% lived in the sub-region employing IRS and 67.1% in the non-IRS sub-region (as defined in Fig. 2b). To further explore malaria variation in sub-regions, the study area was sub-divided into seven sub-regions based on boundaries that correspond roughly to tribal areas (as shown in Fig. 2c). Two of these sub-regions, which contribute to the majority of BL cases in this region [7], contributed 50% of the study population. Some 28.9% of the children reportedly used a mosquito bed net on the night before interview, but 31.3% lived in houses using IRS, which is similar to the per cent living in an IRS sub-region. Regarding socio-economic status, 55.3% of mothers of participants had fewer than 5 years of formal education and 55.7% had a monthly median income below 30,000 Ugandan shillings (equivalent to ~US$10). Regarding history of malaria, 63.5% of the participants reported a history of outpatient malaria treatment, with 55.6% reporting such treatment in the past 12 months. Conversely, only 36.6% reported at least one inpatient treatment for malaria, defined as requiring some observation at a clinic or hospital, with 12.6% reporting such treatment in the past 12 months.

Malaria testing was successful for 1143 children; the seven children who failed were excluded from further analysis (Table 2). The weighted pfPR based on RDT was 54.8% (95% CI 47.3–62.4%) and it was 43.4% (36.1–50.7%) based on TFM. As expected, the sensitivity of RDT was 97.5% and the specificity was 77.8%, compared to TFM. RDT is more sensitive than TFM because it detects malaria antigens, which can be detected during early infection and for up to 35–42 days after treatment of symptomatic cases due to persistence of antigens in peripheral blood [33]. Using positivity to both RDT and TFM to define current parasitaemia, the weighted pfPR of current parasitaemia was 42.3%. Conversely, using only RDT positivity to define malaria antigenaemia alone (due to early or persistent antigen after infection has resolved), the weighted pfPR of malaria antigenaemia alone was 12.6%. The weighted prevalence of negative RDT among TFM-positive subjects (including two with parasitaemia identified only on thin film) was 1.1%. These cases (TFM-positive and RDT-negative) had lower GMPD of 178.5 parasites/µL (95% CI 115.0–277.0) than those with of current infection (RDT and TFM positive) with GMPD of 1912.3 parasites/µL (95% CI 1423.0–2570.0; P < 0.001).

Similar to prior studies from this region [20, 34, 35], there was no difference in the weighted pfPR estimate for the north-central and northwest regions (Fig. 2a). However, these similar malaria patterns masked variation in sub-regional analysis. First, pfPR was a substantially lower in IRS than non-IRS sub-regions (32.8 versus 65.7%; Fig. 2b). However, the apparent homogeneity in IRS and non-IRS sub-regions masked further, substantial variation in sub-regions in the IRS (3.2–55.3%) and the non-IRS sub-regions (55.5–75.8%; Fig. 2c). This additional variation was not explained by patterns in other covariates (Table 3). As expected, the weighted pfPR (Fig. 3a) and the GMPD (Fig. 3b) showed strong seasonal effects, both being higher during the wet season than the dry seasons, but this pattern was more striking in the high-density than in the low-density villages.

The weighted pfPR increased from 40.0% in children 1–2 years old to 61.8% in those 6–8 years old and decreased slightly to 48.9% in children 12–15 years old (Fig. 4a, b). Using logistic regression to model the weighted pfPR across five age categories that were previously defined in Emmanuel et al. (under 2, 3–5, 6–8, 9–11, 12+ years) [14], the age-specific ORs of pfPR were heterogeneous across these age categories. Compared to children under 2 years old, the ORs for pfPR peaked at 2.42 (95% CI 1.26–4.65) in children 6–8 years old and then decreased slightly to 1.43 (95% CI 0.62–3.31) in children 12–15 years old. There was a significant linear relationship between age and the weighted pfPR (Ptrend = 0.006) and a significant non-linear effect of age on pfPR (Ptrend = 0.004). These contrasting trends are consistent with the idea that the weighted pfPR increases with age as children become increasingly mobile with age and are more likely to be exposed to malaria infection. Conversely, increasing age is associated with increasing cumulative acquired malaria immunity and the chance that older children are more likely to successful resist new infection. Interestingly, these age trends were masked when pfPR across was modeled across three age categories (Table 3, Pheterogeneity = 0.185), highlighting the importance of using finer categories of age to capture rapid changes in the trends. Because BL epidemiology is characterized by male and rural predominance, the age-group specific patterns of pfPR were examined by sex and in high- and low-density villages, but no significant differences in age-specific patterns were noted by sex in the low-density (Fig. 5a) and high-density (Fig. 5b) villages.

In unadjusted analyses, pfPR was positively associated with enrolment during the wet versus dry season (OR 2.35, 95% CI 1.26–4.40), and inversely associated with living in a sub-region where IRS was implemented versus living in a sub-region where IRS was not implemented (OR 0.26, 95% CI 0.14–0.46). Similar results were obtained when exposure to IRS was assessed using self-reported questionnaire information about whether a child lived or not in a house where IRS was applied in the past year (OR 0.28, 95% CI 0.15–0.53), confirming the validity of the ecological association between IRS and pfPR. The weighted pfPR was inversely associated with a child’s mother’s income ≥30,000 versus <30,000 USHS (OR 0.68, 95% CI 0.47–1.00) and positively associated with a child reporting outpatient treatment for malaria in the past 12 months versus no treatment (OR 1.73, 95% CI 1.1.0–2.72). However, sex, proximity of village to water, population density, region, and mother’s education level, sleeping under a mosquito bed net the previous night, or history of inpatient malaria treatment were unrelated to weighted pfPR (all P > 0.10). In multivariable analyses, including only those variables with P < 0.10 in the unadjusted analyses, the weighted pfPR remained significantly heterogeneous across the seven geographical sub-regions (P = 0.001). In addition, the weighted pfPR was inversely associated with a child’s mother income (OR 0.57, 95% CI 0.37–0.87) and with living in an IRS sub-region (OR 0.37, 95% CI 0.14–1.03), but positively associated with being enrolled during the wet season calendar months (OR 1.81, 95% CI 0.95–3.46). Similar patterns were observed when these analyses were repeated using pfPR defined based on only TFM or combined positivity TFM and RDT results.