Introduction

Malaria remains a major global health concern, disproportionately affecting tropical regions like the Peruvian Amazon, where environmental conditions favor the proliferation of Anopheles mosquitoes, the primary vectors of the disease [1]. Traditional measures, such as insecticide-treated bed nets [2] and indoor residual spraying [3], have proven effective but may not suffice when housing conditions are suboptimal [4]. However, challenges such as low bed nets adherence [5], rising insecticide resistance [6], and the structural unsuitability of IRS in poorly enclosed homes [7] have limited the effectiveness of these traditional tools in many high-transmission areas. Renovating homes with netting covering ceilings has the potential to further reduce Anopheles entry and resting, thereby lowering mosquito density indoors and around homes [8]. However, high-quality evidence on the effectiveness of such structural renovations in resource-limited settings is scarce.

Housing modifications have emerged as a promising avenue for malaria prevention. Improving housing infrastructure to reduce human exposure to mosquitoes has been associated with decreased malaria risk [9]. For instance, screening windows, doors, eaves, and/or ceilings can significantly limit mosquito entry into homes, thereby reducing indoor mosquito densities and subsequent malaria transmission [10]. A systematic review and meta-analysis concluded that better housing conditions are associated with a lower risk of malaria, underscoring the potential of housing improvements as a sustainable intervention [11]. Building upon this concept, the integration of long-lasting insecticide-treated netting (LLIN) into housing ceilings offers an innovative approach to vector control [12, 13]. LLINs have been extensively used as bed nets, providing both a physical barrier and an insecticidal effect against mosquitoes. Their application in housing modifications, such as installing either LLINs or mosquito nets across ceilings and open structural areas, aims to extend protection beyond sleeping spaces to the entire household environment. This strategy not only prevents mosquito entry but also targets mosquitoes that come into contact with the treated surfaces, potentially reducing both indoor and outdoor mosquito populations, with possible spillover effects on community-level mosquito abundance [9,14,15].

The stepped wedge cluster randomized trial (SW-CRT) design is particularly suited for evaluating interventions such as house renovations for malaria control. In an SW-CRT, all clusters eventually receive the intervention, but the implementation is staggered over time [16]. This design allows for the assessment of intervention effects while accounting for temporal changes and is often employed when the intervention is expected to do more good than harm, and it is logistically or ethically challenging to withhold it from any group [16]. The SW-RCT design has been effectively utilized in various public health studies, including those assessing malaria interventions [17,18]. In this context, we conducted a SW-RCT to evaluate the effectiveness of installing ceiling-mounted mosquito nets (CMMNs) in reducing Anopheles mosquito density both indoors and peridomestic in the Llanchama community, San Juan Bautista district, Loreto, Peru. This study aims to contribute to the growing body of evidence supporting housing modifications as a viable strategy for malaria control and to explore the potential of mosquito nets beyond their traditional use as bed nets.

Materials and methods

Study design

We conducted an open SW-RCT with four-steps, randomized at the household level. Using this approach, we invited the head of every eligible house to participate in the study as part of one the household clusters. The intervention was remodeling their houses to install CMMNs and the study outcomes were standard entomological indexes measured by human landing catching. Briefly, we planned the study including three clusters of Llanchama households randomly selected, which were intervened sequentially in four steps: (i) step 1 or baseline without intervention; (ii) step 2 or 3-month phase with the first cluster; (iii) step 3 or 6-month phase with the first and second cluster intervened; (iv) step 4 or 9-month phase with the tree clusters intervened (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Stepped wedge cluster randomized trial design.

https://doi.org/10.1371/journal.pone.0325089.t001

Sample size

We used a weighted sampling approach to ensure that each study sequence included house types B, C, and D. We initially excluded type A houses because of their rudimentary structure and lack of permanent residents. From our mapping, we observed that the distribution of house types A, B, C, and D was approximately 10%, 20%, 30%, and 40%, respectively. We based our sample size calculation on an 80% study power, a 5% alpha level, and an expected 50% reduction in Anopheles darlingi indoor biting. Under these assumptions, we determined that a minimum of 57 houses would be needed for a SW-RCT design with staggered groups (19 houses per sequence), including one baseline and three follow-up periods. On the basis of this estimate, we decided to enroll 23 households per sequence, to account for a < 20% loss to follow-up due to any contingency, including abandoned households or households that did not complete the renovation or do so incorrectly.

A total of 83 households were assessed for eligibility, and 14 were excluded for not meeting inclusion criteria or having been abandoned (Fig 1). The remaining 69 households were randomly assigned to three sequences of 23 households each. Although all sequences proceeded through the SW-RCT, several households did not complete renovations according to the intended timeline and were excluded from subsequent follow-ups. In the end, each sequence concluded with 19 households, mirroring the situation in which certain groups, as in the example, transferred to a later sequence or opted out of the intervention cycles (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flow chart of the households (clusters) intervention.

The figure shows that a total of 83 households were assessed for eligibility; 14 were excluded, and 69 were randomly allocated to three sequences of 23 households each. Some households did not complete the renovation process within the allotted steps, resulting in exclusions from subsequent follow-ups. The chart illustrates the progression of each sequence through the baseline and intervention steps, and the final number of households analyzed per sequence (n = 57, 19 households per sequence).

https://doi.org/10.1371/journal.pone.0325089.g001

Results

At baseline (from 03/15/24 to 03/25/24), median indoor Anopheles mosquito counts across the three clusters ranged from 36 to 62 mosquitoes per household. After the intervention, indoor mosquito counts decreased significantly, with median values dropping to 0–2 mosquitoes per household by the 9th month (Fig 2). Peridomestic mosquito counts were more variable, with baseline medians ranging from 63 to 79 mosquitoes per household, but showed no consistent reductions over time following CMMNs installation. Regarding human exposure to mosquito bites, indoor bites per person per night (BPN) were substantially reduced, from 18–31 at baseline to 0–1 by the 9th month. Peridomestic BPN, however, showed minimal change, remaining between 6 and 10 throughout the study period. Indoor bites per person per hour (BPH) followed a similar trend, with baseline values of 3–5 dropping to 0 after intervention, while peridomestic BPH remained relatively stable (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Mosquito density and biting rates during the study.

https://doi.org/10.1371/journal.pone.0325089.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Mosquito density and biting rates during the study.

The figure shows the monitoring of the mosquito density and biting rates during each study phase, including indoor mosquitoes counts (Panel 2A), peridomestic mosquitoes count (Panel 2B), indoor BPN (Panel 2C), peridomestic BPN (Panel 2D), indoor BPH (Panel 2E), and peridomestic BPH (Panel 2F). CMMN, ceiling-mounted mosquito nets; BPN, bites per person per night; BPN, bites per person per person; Gray shadow, intervention period.

https://doi.org/10.1371/journal.pone.0325089.g002

According to the multivariable mixed-effects negative binomial regression models the intervention was associated with a 55% reduction in indoor mosquito counts (IRR: 0.45, 95% CI: 0.26–0.77, p = 0.004) but did not significantly reduce peridomestic mosquito counts (IRR: 0.94, 95% CI: 0.75–1.17, p = 0.570). Similarly, indoor BPN decreased by 60% (IRR: 0.40, 95% CI: 0.22–0.73, p = 0.003), while peridomestic BPN remained unchanged (IRR: 0.93, 95% CI: 0.73–1.19, p = 0.56). Hourly bite rates (BPH) followed the same pattern, with indoor BPH decreasing by 61% (IRR: 0.39, 95% CI: 0.17–0.85, p = 0.018), whereas peridomestic BPH was not significantly affected (IRR: 0.92, 95% CI: 0.57–1.49, p = 0.740). Temporal analysis demonstrated that mosquito densities and biting rates declined sharply after the 3rd month, with greater reductions at the 6th and 9th months. By the 9th month, indoor mosquito counts were 81% lower than at baseline (IRR: 0.19, 95% CI: 0.14–0.26, p < 0.001), and indoor BPN was reduced by 88% (IRR: 0.12, 95% CI: 0.05–0.26, p < 0.001) (Table 3). Any harm was reported during the study.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Estimated Effects of CMMNs on Mosquito Density and Biting Rates.

https://doi.org/10.1371/journal.pone.0325089.t003

Discussion

This SW-CRT conducted in a malaria-endemic, riverine community of the Peruvian Amazon provides robust evidence that CMMNs are highly effective at reducing indoor Anopheles mosquito exposure. Specifically, we observed statistically significant reductions of 55% in indoor mosquito density, 60% in indoor BPN, and 61% in BPH following intervention deployment. These effects were temporally consistent and sustained across three sequential clusters, reinforcing the durability of the intervention effect. Notably, peridomestic mosquito exposure did not decline significantly, underscoring the spatial specificity of the protective benefit conferred by CMMNs.

Unlike traditional vector control tools such as long-lasting insecticide-treated nets (LLINs) or indoor residual spraying (IRS), CMMNs offer a passive, infrastructural solution that does not depend on daily behavioral adherence. As such, they represent a promising addition to the malaria prevention toolkit, particularly in settings where nightly net use is suboptimal or household structures remain highly permeable to vector entry.

Our findings are concordant with an emerging consensus that housing quality is a major determinant of malaria risk. Multiple systematic reviews and meta-analyses have affirmed the protective role of improved housing design, independent of LLIN or IRS coverage. Tusting et al. reported a 47% relative reduction in malaria infection associated with interventions that physically block mosquito entry, such as screening and improved eaves [9]. Similarly, a meta-analysis by Kua and Lee concluded that screened homes were associated with a 37% lower malaria incidence, supporting a shift toward structural, non-chemical interventions [11].

The magnitude of indoor vector suppression in our trial aligns with data from other housing intervention studies. Trials conducted in sub-Saharan Africa have shown that closing eaves and installing window screening significantly reduces indoor mosquito density and malaria incidence [9–11]. For instance, a trial in rural Gambia observed a 65% reduction in Anopheles exposure following house modifications [8]. However, few studies to date have specifically targeted ceiling areas, despite their critical role in vector entry in elevated, open-eave homes typical of Amazonian riverine environments.

Notably, our findings extend the work of Kagaya et al. and Ko et al., who evaluated the use of insecticidal ceiling nets in Kenyan children through cluster randomized trials. Although those studies focused on malaria incidence, their mechanistic premise—blocking upward mosquito entry—was conceptually validated by our entomological outcomes [12,13]. Furthermore, although our study primarily targeted Anopheles vectors, incidental reductions in other mosquito genera were observed during the study period, particularly indoors. Structural modifications such as ceiling netting likely disrupt entry pathways for both malaria and non-malaria vectors. This is supported by previous research demonstrating that improved housing design, screened ceilings, and closed eaves reduce indoor densities of Culex and Aedes mosquitoes, contributing to broader vector control benefits [21,22]. Future studies should quantify the effect of CMMNs on non-Anopheles species to assess their potential for integrated vector management.

Moreover, our study reinforces evidence from SW-CRTs evaluating vector control interventions. The SolarMal trial in Kenya, which used mass mosquito trapping, demonstrated a significant reduction in malaria prevalence, validating the SW-CRT design for evaluating community-level interventions under pragmatic conditions [17]. Likewise, trials of repellents delivered via community health workers in Myanmar showed the feasibility and operational utility of staged implementation designs in high-burden settings [18].

The observed lack of peridomestic efficacy highlights a persistent challenge in malaria elimination: residual transmission driven by exophagic and exophilic vectors [19,23]. In Amazonian settings, Anopheles darlingi exhibits predominantly endophagic behavior [24], which likely explains the intervention’s indoor-specific effect. However, residual transmission persists outdoors, especially in early evening and peri-domestic contexts [25,26], pointing to the need for complementary interventions targeting exophilic behavior—such as spatial repellents [27], insecticide-treated clothing [28], and attractive targeted sugar baits (ATSBs) [29].

Interestingly, we observed variation in intervention efficacy by housing type. Type D homes—typically more enclosed—demonstrated greater relative reductions in vector metrics compared to Types B and C. This heterogeneity suggests that baseline housing permeability may act as an effect modifier, a hypothesis supported by prior research in Uganda, where housing configuration modulated the protective effects of LLINs and IRS [4]. Such heterogeneity underscores the need to tailor structural interventions to specific housing archetypes and supports stratified analysis in future trials.

From a health systems perspective, the potential scalability of CMMNs is notable. As a one-time installation with low maintenance requirements, the intervention circumvents common barriers to behavioral compliance that undermine other malaria control strategies. Given the logistical and financial challenges of sustained LLIN and IRS coverage in remote, forested settings, CMMNs may offer a durable and acceptable complement for national programs, particularly in hard-to-reach populations [3].

This study’s principal strength lies in its pragmatic SW-CRT design, which enabled robust causal inference while ensuring equitable delivery of the intervention across all clusters. Within-cluster comparisons minimized temporal confounding, and the use of mixed-effects negative binomial models accounted for clustering, period effects, and structural heterogeneity among households. The use of human landing catches—considered the gold standard for measuring human-vector contact—strengthened outcome measurement fidelity. Nonetheless, several limitations should be considered. First, our reliance on entomological proxies—mosquito density and human biting rates—precludes direct conclusions about the intervention’s effect on malaria infection or disease, although these metrics are strongly associated with transmission risk. Second, the lack of blinding may have introduced performance or detection bias; however, the objective nature of entomological outcomes and standardized protocols likely reduced this risk. Third, contamination or spillover effects between clusters—due to geographic proximity and social interactions—may have attenuated contrasts between intervention sequences. While the average separation between clusters was 300–500 meters, and households were grouped to minimize inter-cluster proximity, it is possible that mosquitoes deflected by treated homes migrated to untreated ones in neighboring clusters. Importantly, studies in the Peruvian Amazon indicate that its typical host-seeking range is much shorter (often <500 meters) [19], and its movement is generally constrained by environmental features and host availability. Furthermore, our use of cluster-level modeling with random intercepts helped mitigate this bias, though some residual inter-cluster influence may persist. Fourth, although we adjusted for key environmental covariates, residual confounding from unmeasured or dynamic ecological factors (e.g., microclimatic variation or larval habitat changes) cannot be ruled out. Finally, clinical malaria outcomes were not assessed due to logistical and resource limitations. As a result, our evaluation focused exclusively on entomological indicators of transmission risk. Future studies should integrate parasitological outcomes, such as confirmed malaria cases or infection prevalence, alongside entomological data to more comprehensively assess the epidemiological impact of structural interventions like ceiling-mounted mosquito nets.

Following CONSORT guidelines for SW-CRTs [30], these findings should be interpreted in light of the study’s implementation context. While our results suggest entomological efficacy in a rural Amazonian community, generalizability may be limited to areas with similar housing designs, vector ecology, and health system capacities. Future studies should incorporate clinical or parasitological endpoints such as malaria incidence, PCR-confirmed parasitemia, or test positivity rates, and assess cost-effectiveness under real-world programmatic conditions. Evaluations comparing treated versus untreated netting are needed to clarify the incremental benefit of insecticidal properties in areas of evolving insecticide resistance. Additionally, operational research should explore the durability, household maintenance practices, and long-term acceptability of CMMNs, especially when integrated into comprehensive housing improvement strategies. Finally, mechanistic entomological studies, including vector behavior assessments, resting patterns, and escape responses, are warranted to refine intervention specifications and inform optimal deployment strategies for different household typologies and transmission settings.

Conclusions

This SW-CRT provides compelling evidence that CMMNs significantly reduce indoor Anopheles mosquito exposure in high-transmission, structurally vulnerable communities of the Peruvian Amazon. In an era where malaria control increasingly requires integrated, sustainable, and context-adapted strategies, structural interventions such as CMMNs deserve greater inclusion in the global vector control agenda. By offering high community acceptability, low behavioral dependency, and strong ecological compatibility, CMMNs represent a scalable and equitable complement to conventional interventions—particularly in populations underserved by current vector control tools. As the global health community strives toward malaria elimination, investments in durable, infrastructure-based solutions like CMMNs may be critical for reaching residual transmission foci and achieving lasting impact.

Supporting information

Acknowledgments

We thank the residents of Llanchama, as well as the GERESA staff, volunteers, and field workers, for their diligent efforts in documenting the study’s entomological outcomes.