Differential impact of malaria control interventions on P. falciparum and P. vivax infections in young Papua New Guinean children

Introduction As malaria transmission declines, understanding the differential impact of intensified control on Plasmodium falciparum relative to Plasmodium vivax and identifying key drivers of ongoing transmission is essential to guide future interventions. Methods Three longitudinal child cohorts were conducted in Papua New Guinea before (2006/2007), during (2008) and after scale-up of control interventions (2013). In each cohort, children aged 1–5 years were actively monitored for infection and illness. Incidence of malaria episodes, molecular force of blood-stage infections (molFOB) and population-averaged prevalence of infections were compared across the cohorts to investigate the impact of intensified control in young children and the key risk factors for malaria infection and illness in 2013. Results Between 2006 and 2008, P. falciparum infection prevalence, molFOB, and clinical malaria episodes reduced by 47%, 59% and 69%, respectively, and a further 49%, 29% and 75% from 2008 to 2013 (prevalence 41.6% to 22.1% to 11.2%; molFOB: 3.4 to 1.4 to 1.0 clones/child/year; clinical episodes incidence rate (IR) 2.6 to 0.8 to IR 0.2 episodes/child/year). P. vivax clinical episodes declined at rates comparable to P. falciparum between 2006, 2008 and 2013 (IR 2.5 to 1.1 to 0.2), while P. vivax molFOB (2006, 9.8; 2008, 12.1) and prevalence (2006, 59.6%; 2008, 65.0%) remained high in 2008. However, in 2013, P. vivax molFOB (1.2) and prevalence (19.7%) had also substantially declined. In 2013, 89% of P. falciparum and 93% of P. vivax infections were asymptomatic, 62% and 47%, respectively, were sub-microscopic. Area of residence was the major determinant of malaria infection and illness. Conclusion Intensified vector control and routine case management had a differential impact on rates of P. falciparum and P. vivax infections but not clinical malaria episodes in young children. This suggests comparable reductions in new mosquito-derived infections but a delayed impact on P. vivax relapsing infections due to a previously acquired reservoir of hypnozoites. This demonstrates the need to strengthen implementation of P. vivax radical cure to maximise impact of control in co-endemic areas. The high heterogeneity of malaria in 2013 highlights the importance of surveillance and targeted interventions to accelerate towards elimination.


Background
Intensification of malaria control measures has been associated with marked reductions in transmission and infection and illness burden in many endemic areas [1]. In the Americas [1,2] and some parts of Asia-Pacific [3,4], these reductions have been associated with a marked shift to the predominance of Plasmodium vivax as the primary source of Plasmodium spp. infections. In parallel, the proportion of low-density, asymptomatic infections has been observed to increase [5][6][7][8] and transmission becomes more heterogeneous [9][10][11].
The reasons underlying these shifts are likely to be multifactorial. A major factor for the relative increase in P. vivax is the poor uptake and/or adherence of antihypnozoite therapy [12,13]. As a result, P. vivax hypnozoites are able to cause repeated bouts of blood-stage parasitaemia and are responsible for up to 80% of all P. vivax blood-stage infections [14]. Even in low and very low transmission settings, most P. vivax infections are asymptomatic [15,16] and often of very low density [16] but almost all carry detectable gametocytaemia [6,17,18]. These infections are thus not detected and treated by the health systems and can sustain transmission. P. vivax is also considered more easily transmissible given the rapid maturation and thus early presence of its gametocytes [19] and faster development cycle in its mosquito host [20]. Lastly, it has also been observed that P. vivax-infected mosquitoes may be younger and more likely to bite early and outdoors [21,22]. All of these factors may render P. vivax transmission less susceptible to vector control and routine case management interventions.
The highly heterogeneous nature of malaria transmission across countries, between neighbouring villages and within the same village has long been recognised [23][24][25] and is driven by an interplay of host, vector and environmental factors [23,26,27]. As transmission declines, there is a tendency for malaria infections to become increasingly clustered in high-risk populations and high-risk areas [11,28] and it becomes more important to be able to identify these clusters since they may be responsible for sustaining transmission [11]. There is growing evidence that despite achieving overall reductions in malaria transmission through improved malaria control, infections and illness burden in many hyperendemic areas remain unaltered [29][30][31] and that more targeted interventions may be necessary for elimination [11].
In the early 2000s, the overall burden of malaria in Papua New Guinea (PNG) was amongst the highest in the Asia-Pacific region, albeit with intensity of transmission geographically highly variable across the country [27,32,33]. Plasmodium falciparum and P. vivax are the two predominant species that account for most of the burden of malaria infections and illness in PNG [32,34].
Beginning in 2004, with the support of Global Fund to Fight AIDs, Tuberculosis and Malaria, PNG scaled up its malaria control interventions through scheduled 3-yearly nationwide distribution of long-lasting insecticide treated nets (LLINs), introduction of a test-and-treat approach and a switch to artemether-lumefantrine (AL) as first-line treatment [35,36]. Subsequent surveys revealed a substantial decline in the overall burden of malaria [6,33], with the nationwide infection prevalence by light microscopy (LM) declining from 11.1% in 2009 to 0.9% in 2014 [33,37]. Entomological studies also revealed a large decline in human biting rates from 83 bites/person/night to 31 bites/person/ night [37,38]. As elsewhere, these reductions in PNG have gone hand-in-hand with an increase in the proportion of asymptomatic and sub-microscopic infections [6] and a pronounced heterogeneity of residual transmission [39]. Although the prevalence of PCR-detectable P. vivax infections in community surveys has not declined to the same extent as P. falciparum infection [6], the shift towards P. vivax predominance has not yet been as pronounced as in neighbouring SE Asia-and SW Pacific countries [7].
To better understand the relationship between changing transmission and the risk profile of malaria infections and disease, it is vital to gain insight into the impact that control measures have on the two main species, P. falciparum and P. vivax. Using three consecutive longitudinal child cohorts (1-5-year-old children) conducted in the same study area, prior [40], during [41] and following 5 years of intensification (2013 cohort), we investigated the impact of improved malaria control on the breadth of metrics including clinical incidence, incidence of newly acquired infections (i.e. the molecular force of blood-stage infection, mol FOB) [42,43] and infection prevalence to better understand changing P. falciparum and P. vivax epidemiology in the context of rapid reductions in transmission. In order to guide continued reductions in transmission, we also investigated the key drivers of infection and illness in young children during the period of low transmission in 2013.

Study design and sites
Three longitudinal cohort studies of 1-5-year-old children were conducted in the same study area in the Ilahita area of Maprik District, East Sepik Province in 2006, 2008 and 2013. A detailed description of the study area is given elsewhere [40]. Briefly, the study area is located in northern PNG where malaria transmission is considered hyperendemic [34,44] and all human malaria species are endemic [40,41,45,46]. Health services are provided solely by the church-run Ilahita Health Centre with inconsistent services from a government aid post. The cohorts were conducted at three different time-points before and during the scale-up of malaria control interventions in the study area (Fig. 1).

cohort (pre-intensification)
Children aged 1-3 years were enrolled into the study and actively followed up for malaria infection and illness every 8 weeks for a total of 16 months from March 2006 to August 2007 [40,42,43]. Passive case detection at Ilahita Health Centre was maintained throughout the study for detection of clinical episodes. All rapid diagnostic test (RDT) or LM confirmed febrile illness episodes were treated with AL (Coartem®, Novartis) (if treated by study staff) or amodiaquine plus sulphadoxine-pyrimethamine as per the PNG standard treatment for common illnesses in children [47] (if receiving treatment from a non-study source). Children with P. vivax episodes were not treated with primaquine as it had not yet been introduced into PNG standard treatment guidelines [47]. Full details of the study methodology are published [40,42,43].

cohort (during early intensification)
Children 1-5 years of age were enrolled into this randomised controlled trial in April 2008, a month after the first population-wide distribution of LLIN into the study area [41]. Analysis was restricted to the control arm to allow comparability to the other two observational studies. Children were actively checked for malaria infection and illness fortnightly for the first 3 months and monthly thereafter for another 7 months. All RDT or LM confirmed febrile illness episodes were treated with AL (Coartem®, Novartis) (if treated by study staff) or Amodiaquine plus sulphadoxine-pyrimethamine as per the PNG standard treatment guidelines [47] (if receiving treatment from a non-study source). Children with P. vivax episodes were not treated with primaquine as it had not yet been introduced into PNG standard treatment guidelines [47]. Full details of the study methodology are published [41].

cohort (5 years after sustained control)
This cohort was conducted after 5 years of sustained malaria control in the study area ( Fig. 1) A total of 465 children aged 1-5 years at enrolment from 12 villages (Ilahita 1-7, Kamanokor, Sunuhu 1 and 2, Balanga and Balif) in Ilahita area were enrolled from July to September, 2013, and followed for 12 months. Of these, 45 children were excluded post hoc (11 withdrawals, 26 lost to follow-up, 8 with erratic attendance), resulting in a final sample size of 420 children (90% retention rate). All 420 children ranging in age from 0.9-6.4 years during the study period were included in the analysis investigating the key drivers of infection and illness in 2013. A subset (n = 371) aged ≤ 55 months were age-matched to earlier two cohorts to investigate the changing burden of malaria across the intervention time-points.
At enrolment, demographic and clinical data on recent illness and medications, bednet use and current state of health were recorded. Axillary temperatures were measured using an electronic digital thermometer. A 5-ml (ml) venous blood sample and two blood slides were collected. Haemoglobin level was measured using a portable HemoCue machine (HemoCue, Angholm, Sweden). The location of each child's residence was recorded using a Garmin eTrex®.
Following enrolment, children were actively followed up fortnightly for morbidity surveillance and monthly for blood sampling (250 μl finger prick sample, two blood slides and haemoglobin measurement). If a child had a febrile illness at a morbidity surveillance visit, a finger prick sample of 250 μL blood and 2 blood slides were collected. RDT for malaria was performed and, if positive, children were treated with AL (Coartem®, Novartis) and occasionally AL plus primaquine for RDT positive P. vivax, as per PNG standard treatment guidelines [48]. Over the course of the study, 9 children were documented as receiving primaquine, suggesting that  was inconsistently administered by health workers. Anaemic children with haemoglobin < 7.5 g/dL were given an anthelminthic drug (albendazole) and iron supplementations while other ailments were treated according to PNG standard treatment [48].
Plasmodium spp. infections were detected by real-time quantitative PCR assay (qPCR), as previously described [40][41][42][43]49] and LM. Briefly, parasite DNA was extracted from cell pellets (equivalent to 200 μL whole blood) using a Favorgen 96-well Genomic DNA Extraction Kit following the manufacturer's instructions and eluted in 200 μL elution buffer. The presence of P. falciparum, P. vivax, P. malariae and P. ovale infections were determined using two multiplex 2-species qPCR assays [49]. Infections with P. falciparum and P. vivax were further genotyped for Pfmsp2, Pvmsp1F3 and PvMS16 to identify individual parasite clones. All blood slides positive by first read and/or by Plasmodium screening qPCR [50], as well as 10% of the negatives, were independently examined by a second microscopist. Any discrepancies between the first and the second reads were then re-read by a third expert-level microscopist (WHO Level 1 certified). The final density was calculated by taking the geometric mean of the two concordant reads.

Statistical analysis
Analysis for this paper occurred in two parts and focussed on the two predominant species, P. falciparum and P. vivax. In the first part "Analysis of changing burden of malaria infections and illness: 2006 -2013", we aimed to compare the prevalence, mol FOB and clinical incidence across the three cohorts to determine patterns of decline for P. falciparum relative to P. vivax across the intervention time-points. In the second part, "Analysis of key determinants of malaria infection and illness during the time of low transmission 2013", the objective was to explore the full dataset of the 2013 cohort to identity factors that were key predictors of infection and illness during the period of low transmission in 2013. In both analyses, a clinical malaria episode was defined as history of febrile illness during the preceding 48 h and/ or measured temperature ≥ 37.5°C in the presence of a microscopically detectable infection of any density. The mol FOB (number of genetically unique blood-stage infections) was calculated from the number of new infections acquired during the intervals between sampling timepoints by counting all new msp2 alleles for P. falciparum and msp1F3 and MS16 alleles for P. vivax per unit time that were not present in the preceding intervals.

Analysis of changing burden of malaria infections and illness: 2006-2013
Data from each cohort were analysed separately due to the differences in the sampling schedules and the length of follow-up between the studies. However, to allow direct comparison, we used the full dataset of the 2006 cohort as the baseline while age-matched subsets of the 2008 and 2013 cohorts were used.
The population-averaged prevalence (referred to as prevalence) of P. falciparum and P. vivax infections in the three cohorts was estimated using generalised estimating equations (GEE) with a logit link and an exchangeable working correlation matrix, to account for the dependency between observations from the same child. Robust standard errors were also used to correct for working correlation matrix misspecification. Incidence rates (IR) for clinical episodes were calculated from the total number of clinical episodes experienced by each child over the study period and was modelled using negative binomial regression for the 2006 and 2013 cohorts and Poisson regression for the 2008 cohort. The relative percentage change in the prevalence and incidence was calculated using the formula: percentage change = ((current estimate − previous estimate)/previous estimate) × 100. Both the frequency of sampling and duration of blood-stage infections [51] are important factors influencing the mol FOB variable. Due to the differences in the frequency of sampling in the 2006, 2008 and 2013 cohorts, it was necessary to censor any sampling time-points that were not available across all three cohorts in order to be able to directly compare the mol FOB estimate across the cohorts. Incidence of new clones was defined as the sum of all new clones over the study period and derived using negative binomial regression, adjusting for individual time of exposure.

Analysis of key determinants of malaria infection and illness during the time of low transmission 2013
Risk factors of infection and malaria episode investigated in 2013 included the child's age (years), timing of active detection of infection visits, area of residence, bednet use in the previous night, history of febrile illness in the past 2 weeks, presence of febrile illness, which is defined as the 2-day history of fever ± axillary temperature ≥ 37. 5°C, and haemoglobin levels.
For all risk factor analyses, both univariable and multivariable regression models including all risk factors were examined. The association between prevalence of infections at monthly time-points and the risk factors was estimated using GEEs with a logit link and exchangeable working correlation matrix. Incidence of new bloodstage infections was estimated using GEE with negative binomial regression and an exchangeable working correlation matrix. Due to a very low number of clinical episodes observed in 2013, we used the total number of clinical episodes for each child across the follow-up period to assess the association between incidence of clinical infections and the risk factors. This was estimated using negative binomial regression. The risk factors were summarised across the study period for each child as follows: age at enrolment, residence (assumed not to vary across follow-up), mean haemoglobin level and mol FOB. Two multivariable models of the incidence of clinical infections, one including all aggregated risk factors and mol FOB ( mol FOB-adjusted model) and the other excluding mol FOB (base model) were examined.
Due to reduced levels of transmission in 2013, several villages had few P. falciparum or P. vivax infections detected, no clinical P. falciparum or P. vivax episodes and very few new blood-stage clones. Therefore, villages were grouped into 4 areas with geographically similar characteristics (1 = Ilahita 1, 2, 3, 4, 6 and 7; 2 = Balanga and Balif; 3 = Kamanokor and Ilahita 5; and 4 = Sunuhu 1 and 2). Due to the universally high bednet use, analyses of their association with incidence of new bloodstage infections and clinical episodes did not converge and bednet use was excluded from both analyses. The associations are expressed as odds ratio (OR) and incidence rate ratios (IRR) and were considered to be statistically significant if the Wald test p value was below the nominal level of significance of 0.05.
The analyses were conducted using Stata 12.0 (Stata-Corp, USA) and R v2. 12

Key determinants of malaria infection and illness during the time of low transmission 2013 Demographic characteristics of enrolled participants
Of the 465 children enrolled into the 2013 cohort, data from 420 were available for analyses (retention rate 90%). These children ranged in age from 0.9 to 6.4 years (mean 3.3), 53.8% were male and 93% reported sleeping under a bednet the previous night. On average, the children attended 8 out of the 10 [range 1-10] active detection of infection visits.

Prevalence of infections during follow-up
Throughout the follow-up period, 47% children had at least one P. falciparum infection and 48% had at least one P. vivax infection (detected by PCR). Overall, the averaged prevalence of P. vivax was 19.9% by PCR and 10.8% by LM, while P. falciparum prevalence was 11.0% by PCR and 4.2% by LM. Sub-microscopic infections accounted for 64% of P. falciparum and 47% of P. vivax infections.
The prevalence of PCR-detectable infections varied markedly across the different areas (Pf: range 4.5-28.8%, Pv: range 6.0-45.2%; Table 1) with significantly higher risk of infection observed amongst children living in Sunuhu 1 and 2 compared to Ilahita 1, 2, 3, 4, 6 and 7 (Pf crude OR 8.49 (CI 95 6.14-11.8) p < 0.001, Pv 12.6 (CI 95 8.11-19.6) p value < 0.001); Additional file 1). Whereas the prevalence and the risk of P. falciparum infections also varied significantly over time (range 7.1-32.2%, p < 0.0001), P. vivax prevalence and risk was more stable over time (range 17.8-23.2%, p = 0.1777; see Table 1 and Additional file 1). The risk of both P. falciparum and P. vivax infections was higher in children In multivariate analyses, area of residence, time of visit, age, haemoglobin level and the presence of a concurrent febrile illness remained independently associated with the  Table 1). Area of residence, time of visit, recent antimalarial use, age and having an episode of febrile illness in the previous 2 weeks were all associated with the risk of carrying a P. vivax infection (Table 1). Risk factors of LM-detectable infections were similar (see Additional file 2).

Molecular force of blood-stage infections in monthly intervals
Incidence of new blood-stage infections was determined for a total of 303.4 person-years of follow-up with each child at risk of acquiring new blood-stage infections for an average of 0.73 years during the cohort. The mean The rate of acquiring new P. falciparum clones was higher in Sunuhu 1 and 2 compared to Ilahita 1, 2, 3, 4, 6 and 7 (Pf IRR 3.10 (CI 95 2.08-4.63) p value < 0.001) and also in those with recent antimalarial use (IRR 10.4 (CI 95 5.92-18.2) p value < 0.001, Table 2). Age was not associated with P. falciparum mol FOB in multivariate analysis despite the significant linear association observed in the crude analysis. The P. vivax mol FOB was increased in both Sunuhu 1 and 2 and Kamanokor and Ilahita 5 compared to Ilahita 1, 2, 3, 4, 6 and 7 (IRR 8.16 (CI 95 5.38-12.4) p value < 0.001 and 6.66 (CI 95 4.24-10.5) p value < 0.001, respectively), and also increased linearly with age (IRR 1.26 (CI 95 1.13-1.40) p value < 0.001, Table 2). Both P. falciparum and P. vivax incidence varied markedly over the follow-up time period (both p < 0.0001, Table 2).

Predictors of clinical malaria episodes
Over the 10 months of follow-up, a total of 366 febrile illness episodes were observed, of which 109 (30%) were associated with microscopically confirmed infections (IR, 0.36/child/year), with 51 P. vivax (any density: IR, 0.19) and 49 P. falciparum (any density: IR, 0.18) episodes. Another 7 were P. falciparum and P. vivax mixed infections (any density: IR 0.02), 2 were P. malariae (any density: IR, 0.07). Clinical episodes with high-density parasitaemia (≥ 2500 for P. falciparum and ≥ 500 for non-falciparum infections) accounted for 63.3% (35 Pf,  Table 3). Each 1 g/dL increase in haemoglobin was associated with a 48% reduction in the incidence of clinical P. falciparum (CI 95 0.35-0.77, p value: 0.001, Table 3), and each 1-year increase in age was associated with a 38% increase in the rate of clinical P. falciparum (CI 95 1.10-1.73, p value: 0.006, Table 3). After adjustment for mol FOB, all remained associated with the rate of clinical P. falciparum episodes, and a unit increase in mol FOB (i.e. one new P. falciparum infection per child per year-at-risk) was associated with a 10% (CI 95 1.02-1.18, p value 0.008) increase in the rate of clinical P. falciparum infections ( Table 3).
The rate of clinical P. vivax episodes was also significantly higher in Kamanokor, Ilahita 5 and Sunuhu 1/2 compared to Ilahita 1, 2, 3, 4, 6 and 7 (IRR 8.01 (CI 95 3.23-19.9) p value < 0.001 and 3.71 (CI 95 1.53-8.99) p value 0.004, respectively; Table 3). Each 1 g/dL increase in haemoglobin was associated with a 69% reduction in the rate of clinical P. vivax (CI 95 0.19-0.48, p value < 0.001). After adjustment for mol FOB, only area of residence and haemoglobin remained associated with the rate of clinical P. vivax episodes (Table 3). A unit increase in mol FOB (i.e. one new P. vivax infection per child per year-at-risk) was associated with a 17% (CI 95 1.09-1.25, p value < 0.001) increase in the rate of clinical P. vivax infections. Age was not associated with the rate of clinical P. vivax episodes, either before or after adjustment for mol FOB.

Discussion
This is the first study in a P. falciparum/P. vivax coendemic area and amongst very few studies globally [52] to examine the impact of improved malaria control on the epidemiology of malaria in young children using longitudinal cohorts rather than the widely used nationwide and community household surveys and routine health information systems [6,33,37]. Longitudinal cohort studies allow for a detailed investigation into the dynamics of infection, and illness, as well as the rate of acquiring new infections ( mol FOB) and clinical illness over time.
By analysing these metrics in three consecutive longitudinal cohorts in young PNG children, we demonstrate a differential impact of control interventions on P. vivax compared to P. falciparum that may be overlooked in routine surveillance. Following the first LLIN distribution, the prevalence of P. falciparum infection and both P. falciparum and P. vivax clinical episodes declined immediately and continuously across the time period of the three cohorts. Contrastingly, the prevalence and force of P. vivax blood-stage infections did not decline, remaining initially relatively high with a substantial decline only evident in the most recent cohort that was conducted 5 years after commencement of intensified control in the area. These observations confirm that key biological differences between the two species render them differentially susceptible to standard control tools such as LLINs and case management, highlighting the need for P. vivax-focused interventions in co-endemic regions.
Notably, the relationship between transmission and mol FOB differs for P. falciparum and P. vivax. P. falciparum metrics are directly linked to blood-stage infections, which are always mosquito-derived, hence closely reflecting current levels of transmission. The reductions Multivariate negative binomial regression model-based estimates predicting risk of clinical P. falciparum and P. vivax. Backward selection approach was used to derive significant associations. Base models included all variables except mol FOB. Incidence is based on aggregated clinical data for entire 10-month study period thus precluding analysis of recent antimalarial treatment as a covariate. Bednet use was not analysed due to non-converge of data when included into models. a Overall significance level for the variable estimated using wald chi-square test; mol FOB: molecular force of blood-stage infections; b mol FOB was included as a rate; IR: Incidence rate; IRR: Incidence rate ratio. CI 95 95% confidence interval; p p value in P. falciparum mol FOB observed across these three cohorts confirm reductions in P. falciparum prevalence and EIR observed through monitoring and evaluation of the national programme [37,38]. Due to the biological ability of P. vivax to remain dormant in liver cells as hypnozoites and to serve as a continuing source of relapsing infections, P. vivax metrics are not able to differentiate between mosquito-derived and relapsing infections and therefore do not reflect active transmission as closely as P. falciparum metrics. This is particularly relevant in PNG, where P. vivax is the predominant species detectable in young children and relapses account for more than 50-80% of P. vivax infections in preschool and primary school children [14,41]. As a consequence, the P. vivax mol FOB is a composite measure reflecting the joint burden of new, mosquito-derived and relapsing infections [42,43]. This metric therefore reveals a high burden of persisting low-density relapsing infections in young children, contrasting results of nationwide surveys that showed a comparable decline in P. falciparum and P. vivax prevalence detectable by LM in both children under 5 years and the general population [37].
Given the persistence of a high burden of P. vivax infections following the initial LLIN distribution, the observation that the burden of clinical P. vivax dropped and continued to decline over the years of intensification marked a striking difference. Clinical immunity to P. vivax is acquired rapidly, even under relatively low transmission [15]. In malaria therapy patients, only few mild febrile symptoms were observed when they were reinfected with a homologous infection [53]. As relapsing infections are either genetically identical or meiotic siblings of the primary infection [54,55], it is generally thought that clinical episodes are more likely to be caused by new mosquito-bite-acquired infections. Considering that reduction in transmission results in the acquisition of fewer new mosquito-derived infections, the observation that the immediate impact of LLIN was exclusively on incidence of P. vivax clinical episodes and not on risk of infection strongly suggests that the majority of clinical episodes due to P. vivax may indeed be associated with mosquito-derived rather than relapsing infections.
The observation of a delayed impact of LLIN scale-up on P. vivax compared to P. falciparum blood-stage infections in co-endemic areas is important evidence for control programmes. It suggests that the large reservoir of hypnozoites acquired when transmission is high (prior to scale-up of control) gives rise to a sufficient burden of relapsing infections that may be transmissible, although often not symptomatic, such that minimal impact may be observed on P. vivax prevalence in the years immediately following scale-up even though transmission is being reduced. This highlights the importance of strengthening the implementation of radical cure of P. vivax in order to accelerate reduction in the burden of P. vivax [56]. Reluctance to prescribe primaquine without G6PD testing and poor adherence to the 14-day regime are major issues limiting the effectiveness of P. vivax radical cure in many settings, including PNG.
The observed impact on clinical incidence and the comparable longer-term reduction in P. vivax and P. falciparum burden of infections does however provide reassurance that vector control with LLINs can reduce the burden of P. vivax, at least in countries where malaria transmission is largely peri-domestic [57], even if coverage needs to be maintained for a longer period of time before the full effectiveness is observed. Interestingly, in many countries in Asia and the Americas where dramatic shifts to P. vivax predominance have been observed, programmes rely upon clinical case management (often with poor coverage of anti-hypnozoite therapy) as their primary malaria control strategy [2,58] and/or have highly exophilic vectors with transmission occurring mainly in forested areas where LLIN and other traditional vector control tools such as indoor-residual spraying have limited efficacy [59][60][61].
During the period of reduced transmission in 2013, the individual level of exposure to new blood-stage infections ( mol FOB) and the geographical location of the child's residence were the two key determinants of infection and illness. In the previous 2006 and 2008 cohorts, age-dependent decreases in the incidence of clinical P. vivax were observed [40,41], suggestive of rapid acquisition of clinical immunity due to high P. vivax mol FOB during those periods. Conversely, we did not observe any age association in 2013, which may be explained by the substantial decline in the force of P. vivax infection.
As documented in other settings, declining transmission leads to increasing transmission heterogeneity [60,62] and an increasing proportion of asymptomatic low-density infections [6][7][8]. In 2013, over two thirds all PCR-detected infections were sub-microscopic and the risk of clinical malaria was highly dependent on where the child lived, with higher risk of clinical illness observed in areas with higher force of infection. This pronounced spatial heterogeneity in the risk of infections and malaria illness has also been observed in the two previous cohorts [40][41][42][43] indicating that despite the declining transmission between 2006 and 2013, the high burden areas remained stable. In particular, we observed marked geographical clustering of infections and illness in two areas, Sunuhu 1/2 and Kamanokor/Ilahita 5 in 2013, the same geographical locations that were identified as highest burden areas before [40,42,43] and during scale-up of interventions [41]. The persistence of highburden areas such as these despite the ongoing implementation of control interventions is supported by observations made elsewhere [29,30] and strengthens the rationale for surveillance strategies that target interventions to these potential transmission hotspots in order to accelerate control. Such strategies will clearly need to identify the characteristics of hotspots that fuel sustained transmission and address the diagnostic challenge imposed by asymptomatic, lowdensity infections [5,[63][64][65].
A limitation of this study is the differences in the study designs, sampling schedules and the length of follow-up as well as the non-uniform structuring of the individual datasets. Consequently, each cohort was analysed separately and the calculated burden of malaria infection and disease were compared between the cohorts to determine the patterns of decline for P. falciparum and P. vivax across the intervention time-points. As such, we did not statistically test the differential patterns of decline exhibited by P. falciparum and P. vivax across the intervention time-points. However, confidence intervals of the prevalence, mol FOB and clinical incidence across the three cohorts are provided illustrating when differences are statistically significant. It should also be noted that the cohorts were conducted in the same study area with a stable population and the cohorts were agematched hence minimising variation between the cohorts.
Lastly, the impact of malaria control interventions on transmission are a function of diverse social and ecological settings leading to differences in mosquito abundance, mosquito behaviour and human-mosquito interaction. While improvements in the quality of housing have occurred over the past decade in many urban areas of PNG, housing for PNG's rural majority remains largely dependent on bush material. Higher quality of housing and socio-economic status was associated with reduced risk of malaria in previous studies in PNG [37]; however, such data is not available from the child cohorts analysed. The association between weather patterns and long-term malaria trends in PNG have also been investigated using site-specific satellite weather variables and did not explain variations in observed malaria incidence over time (Rodríguez-Rodríguez & Hetzel, unpublished data). Further work understanding the dynamic, complex and responsive ecological niches driving ongoing malaria transmission in certain areas of PNG will inform the development of targeted control and elimination efforts.

Conclusions
Scale-up of standard malaria control interventions in PNG substantially reduced the burden of malaria infection and disease in the most vulnerable 1-5-year-old age group. Data presented here suggests comparable reductions in new mosquito-derived infections for both P. falciparum and P. vivax but a delayed impact on P. vivax relapsing infections due to the previously acquired reservoir of hypnozoites. We confirm the effectiveness of sustained implementation of LLINs and case management in reducing transmission of both species in PNG but highlight the critical need to strengthen case management, radical cure, surveillance and targeted intervention strategies in order to accelerate control of malaria in co-endemic settings.
Additional file 1: Table S1. Bivariate associations between risk factors and the prevalence, mol FOB and clinical incidence. Estimates of bivariate associations calculated via generalised estimating equation (GEE) models for prevalence and molecular force of blood-stage ( mol FOB) infections and negative binomial regression model used for clinical malaria episodes.
Recent antimalarial use was not tested in the model for clinical malaria episodes due to aggregated clinical data. a Age in years at enrolment was used for clinical malaria incidence while at the start of interval was used for mol FOB. b Comparison group; For multilevel variables, comparison group estimates are presented as odds or incidence rate. PCR: Polymerase chain reaction assay; LM: light microscopy; OR: odds ratio; CI 95 : 95% confidence interval; IRR: incidence rate ratio.
Additional file 2: Table S2. Key predictors of the prevalence of infections due to P. falciparum and P. vivax as diagnosed by light microscopy. Multivariate generalised estimating equation (GEE) modelbased estimates of the risk of infection detected at each monthly active detection of infection visits via backward selection of significant risk factors. a Overall significance level for the variable estimated using wald chi-squared test. AOR: multivariate adjusted odds ratio. CI 95 : 95% confidence interval; b Excluding febrile illness at the time of visit; Data for observed positive are %.