Baseline survey results

In the baseline survey, 2559 children aged 6 months to 10 years from 1217 households were tested for malaria. The mean age of children tested was 5 years. Similar proportions of males and females were tested. The majority of children reported always using a bed net (76.8%), whilst 19% reporting never using a net. Children aged 5–10 years were more likely to report never using a net (22.6%), compared to 0–2 years (15.7%) or 2–5 years (16.5%). 73.2% (n = 1851) of children were infected by malaria (determined using RDT) at the time of the survey. Children aged 5–10 years were more likely to be RDT positive (76.9%), compared to 0–2 year olds (63.8%) and 2–5 year olds (72.7%). The results from the baseline survey were used for the restricted randomisation, to ensure that the two arms were balanced on infection prevalence (Table 1). In August 2016, the baseline mean cluster infection prevalence for the study area was 72.4 (range: 45.3–95.5) in children aged 6 months to 10 years.

Endline survey results (18 months post-installation)

For the endline survey, 2843 people from 275 compounds were included across the 40 clusters. The median age of participants was 12 years old (range: 6 months to 98 years). Net use was relatively high, with 72.5% of people reporting using a net the previous night, though this appeared lower in the intervention arm (68.0%) compared to the control arm (77.1%). The difference in net use across age groups was greater in the SET arm (ranging from 55% in 10–15 year olds to 76% in 50–100 year olds) compared to the control arm (ranging from 71% in 10–15 year olds to 84% in 0–5 year olds).

In the endline survey, mean prevalence by cluster was 43.5% (95% CI: 23.7–66.2). Infection was highest in males (47.4% vs 40.8% in females) with the burden of infection highest in children aged 5–10 years (66.8%). Infection was highest in individuals living in households classified as the lowest SES category (46.9%, compared to 43.5% and 40.1% in the middle and higher categories). Infection was lower in those who reported using a net the previous night (41.0% compared to 50.0%).

Impact of Screening and EaveTubes (SET) on infection prevalence measured by RDT

Infection prevalence was lower in the intervention arm (36.7%) compared to the control arm (50.4%), with 43% lower odds of infection in the SET arm compared to the control arm (odds ratio (OR) 0.57 (0.45–0.71), p < 0.001) (Table 2; Fig. 1).

Infection prevalence in each cluster (open circles) by intervention arm. The mean of the cluster results and 95% confidence intervals are shown in the diamonds

Full size image

In the intervention clusters, there was a reduction in malaria infection prevalence for all age groups compared to those living in control clusters. Individuals reporting sleeping under a net the night before had lower infection prevalence in both arms. There was no evidence for an interaction between intervention arm and bed net use, suggesting that the benefit of living in a SET village was additional (not multiplicative) to the protection conferred by a bed net. However, individuals living in SET villages but not using a bed net had lower prevalence (42.8%) than those using a net in control villages (47.5%) with the biggest difference between those living in control villages and not using a net (60.2%) and those living in SET villages and using a net (33.8%).

Due to the range of coverage of SET in the intervention clusters (32% to 100% of households), we assessed the impact of intervention coverage on malaria infection prevalence. Seven clusters had SET coverage of 70% or lower. At baseline, infection prevalence in the 7 lower coverage clusters was similar to infection prevalence in the 14 clusters which had > 70% coverage (infection prevalence of 72.3% (95% CI: 53.7–95.5) in clusters with coverage ≤ 70% and infection prevalence of 70.7% (95% CI: 48.4–88.5) in clusters with > 70% coverage). In the endline survey, clusters with ≤ 70% coverage had a mean infection prevalence of 43.4% (95% CI: 32.9–52.1) compared to a prevalence of 33.0% (95% CI: 23.7–56.3) in the clusters with coverage > 70%, suggesting a stronger effect of the intervention when coverage was higher. However, when comparing the lower coverage clusters to the control clusters, there was still evidence for a drop in prevalence in the lower coverage villages (OR 0.76 (95% CI: 0.60–0.95), p = 0.019). Importantly, the trial was not designed to assess the differential impact of coverage so these results should be interpreted with caution. A sensitivity analysis was performed controlling for cluster baseline prevalence and showed a similar result (OR 0.76 (95% CI: 0.59–0.98), p = 0.031).

Thirty-five (25%) of the households (297 individuals) sampled in the intervention arm did not have SET. Individuals living in these houses had comparable infection prevalence (37.4%) to those who did have the intervention (36.5%). When compared to individuals living in control villages, the impact of living in a SET cluster for those households without SET (OR 0.57 (0.40–0.83), p = 0.003) was similar to the impact for those households with SET. This suggests a community impact of the intervention, meaning that households benefited from others in their village having SET installed, regardless of whether they lived in houses with the intervention themselves.

Infection prevalence at endline compared to baseline (RDT)

Infection prevalence in children aged 6 months to 10 years for each arm was compared between baseline and endline surveys to assess whether there was a reduction in either arm, with the caveats that the two surveys took place at different times of the year and had slightly different sampling methodologies. The interaction between study arm and survey was significant (p = 0.025). Prevalence in the control arm reduced from 73.9% (95% CI: 67.7–79.3) to 69.4% (95% CI: 63.5–74.6) (p = 0.254), whilst the prevalence in the intervention arm reduced from 72.4% (95% CI 66.1–77.9) to 51.0% (95% CI: 45.8–56.1), p < 0.001, corresponding to a risk difference of 21.4% (95% CI: 13.8–29.0) (Table 2). The reduction in prevalence was not universal across all intervention clusters, with 3 SET clusters having similar prevalences between baseline and endline (Fig. 2). SET coverage in these three villages was above 60%; however, net use the previous night was below 60% in all three. The largest percentage reductions between baseline and endline were seen in clusters where baseline prevalence was highest in both arms (Fig. 2).

Percentage reduction in infection prevalence between endline and baseline surveys compared to baseline prevalence. Each circle represents a cluster

Full size image

Comparison of diagnostics: RDT and microscopy

The majority of participants in the endline survey also had a blood smear taken for microscopy (N = 2725). 36.4% (n = 1072) of participants had a microscopically detectable infection, 98% of which were Plasmodium falciparum or mixed infections. Plasmodium malariae infections were detected in 2% of slides (n = 66), whilst Plasmodium ovale infections were detected in 0.4% of slides (n = 13). Similarly, to the RDT results, the prevalence of microscopic infections were lower in participants living in intervention clusters (31.9%) compared to those living in the control clusters (41.4%) (OR 0.64 (0.44–0.93), p = 0.019).

Overall, 57.3% (707/1233) of positive RDT results were also positive by microscopy. Whilst age patterns of infection prevalence appeared similar for the two diagnostics, the agreement between the two was lowest in older participants (only 36.5% and 31.1% of positive RDT results were positive by microscopy in 25–50 year olds and 50–100 year olds respectively). Mean parasite density was also lower in infections in those age groups (< 2000 trophozoites per ml) compared to younger age groups (mean 36,384, 12,300, 3789, 6437 trophozoites per ml in 0–5, 5–10, 10–15, and 15–25 year olds, respectively).

Infection prevalence (RDT) compared to incidence

Cluster-level infection prevalence in the endline survey was associated with malaria case incidence measured in the child cohort at the end of the two years (Fig. 3). Cluster-level linear regression of case incidence on prevalence suggested that for every increase of 1 case per child per year, there was an increase of 10.9% in infection prevalence (measured by RDT) (7.3–14.5%), p < 0.001).

Association between endline prevalence (all ages) and malaria case incidence in children aged 6 months to 10 years

Full size image