Information sources

Our study conforms to the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) guidelines.[21] We retrieved literature from Ovid Medline; Embase Classic+Embase; Web of Science; EBM Reviews—Cochrane Central Register of Controlled Trials; and CAB Direct in July 2019.

Search strategy

Search strategies for each database were developed with assistance from Medical Reference Librarians at the University of Melbourne. The search terms included: “‘control*’ or ‘prevent*’, ‘cluster’ and ‘random*’, ‘rct*’, ‘mosquito*’ or ‘culicidae’ or ‘malaria*’ or ‘dengue’ or ‘arbovir*’ or ‘Buruli’ or ‘aedes’ or ‘malaria’ or ‘West Nile’ or ‘chikungunya’ OR ‘yellow fever’ or ‘filariasis’ or ‘tularemia’ or ‘dirofilariasis’ or ‘Japanese encephalitis’ or ‘Saint Louis encephalitis’ or ‘Ross River’ or ‘Barmah Forest’ or ‘la crosse’ or ‘zika’ or ‘keystone’, ‘bed net*’or ‘pyrethrum*’ or ‘pyrethroid’ or ‘pesticide*’ or ‘repellent*’ or ‘spray*’ or ‘chemical*’ or ‘retardant*’ or ‘coil*’ or ‘community mobili*’ or ‘water’ or ‘source reduction’ or ‘trap*’ or ‘biocontrol’ or ‘larvicid*’ or ‘adulticid*’ or ‘steril*’ or ‘tetracycline’ or ‘oil drip*’ or ‘DDT’ or ‘lindane’ or ‘malathion’ or ‘propoxur’. Detailed search strategies with key terms used for each database are presented in S1 Appendix.

To identify additional studies, we reviewed and hand searched reference lists of identified literature reviews and meta-analyses. Citations of studies thought potentially eligible for inclusion were extracted for screening, with duplicates removed and the remaining articles screened.

Eligibility criteria

Articles were eligible for inclusion in this review if the following criteria were met:

• Online abstract available.
• Written in English.
• Published in a peer reviewed journal at any time prior to July 2019 (not including conference proceedings).
• Described completed research (research protocols were excluded).
• cRCT design.
• The intervention activity, which was applied only to the intervention area, targeted mosquitoes without non-targeted use of insecticide.
• Quantified the occurrence of mosquito-borne disease in the study population according to control/intervention status following the intervention.

When non-targeted insecticide/s was applied in both the control and intervention arms using the same (or a very similar) approach, and a separate intervention was applied, the study was eligible for inclusion. This assumes the effects of the non-targeted insecticide were balanced across both study arms, and that any difference in outcomes between study arms is attributable to the intervention activity which was applied only in the intervention area. Where non-targeted insecticide/s was applied outside of the study protocol (such as through a government-initiated mosquito control programme), the study was eligible for inclusion and the effects of such activities were considered as possible sources of contamination bias.

Study selection

Two authors (JO and SL) screened the titles and abstracts. Articles were evaluated for inclusion according to the eligibility criteria. Where it was unclear if an article met the inclusion criteria based on the title and abstract, or where the two reviewers disagreed, the full text of the manuscript was reviewed and a decision regarding article eligibility was made by the first author (this process occurred for approximately one-quarter of screened articles).

Data abstraction

The full text of included articles was reviewed, with data abstracted independently by the two reviewers using a template specifying relevant data fields. Inconsistencies in abstracted data were resolved by re-checking the article, with discussion and consensus within the study team. Data were abstracted to a data collection template (S2 Appendix), including the study citation; disease and vector; study setting and date; inclusion/exclusion criteria; intervention(s); randomisation method, data collection and analysis methods; number, level (household, village, region etc.) and distribution of clusters; number and age distribution of participants; outcomes (entomological indicators of vector breeding and occurrence of mosquito-borne disease cases in the control and intervention arms); loss to follow-up, assessment of bias and study limitations.

Outcomes and prioritisation

Data around the study setting, design, analysis and effectiveness were used to address the first objective: to describe cRCT designs used to investigate interventions other than non-targeted insecticide aiming to reduce transmission of mosquito-borne disease in humans. Data on the effectiveness of the outcome and assessments of the studies’ strengths, limitations and bias were used to address the second objective: to comment on the strengths and weaknesses of included study designs.

Bias assessment

Both reviewers independently assessed, and commented on, the risk of bias, strengths and limitations of included studies. When assessing the risk of bias in individual studies, the Joanna Briggs Institute Critical Appraisal Checklist for Randomized Controlled Trials tool was used (http://joannabriggs.org/research/critical-appraisal-tools.html). Inconsistencies between reviewers were addressed as previously described.

Synthesis of results

Overall results are described qualitatively, with emphasis on the types of limitations associated with different study designs. Simple counts and short descriptions are presented, in particular when describing intervention types, study designs, study settings, analysis methods (Table 1), strengths and limitations (Table 2).

Download:

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

Table 1. Characteristics of included studies.

https://doi.org/10.1371/journal.pntd.0009601.t001

Download:

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

Table 2. Study strengths and potential limitations.

https://doi.org/10.1371/journal.pntd.0009601.t002

Article searches and screening

Following searches of the five medical research databases, 1,471 article citations were identified of which 1080 were duplicates and 391 articles underwent title and abstract screening (Fig 1). Twenty-one literature reviews or meta-analyses were identified; searching their reference lists identified an additional 67 articles for title and abstract screening. A total of 458 articles underwent title and abstract screening and 100 articles underwent full text screening for eligibility; eight articles met the inclusion criteria.

Download:

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

Fig 1. PRISMA Diagram.

https://doi.org/10.1371/journal.pntd.0009601.g001

The most common reasons for exclusion was that the study did not use an eligible intervention (N = 288 articles, 63% of screened articles), or that a cRCT study design was not used (N = 252, 55%). Many (N = 301, 66%) articles were excluded for more than one reason.

Cluster randomised control studies used to investigate interventions other than non-targeted insecticide for reducing mosquito-borne disease in humans

Of the eight included studies outlined in Table 1 and S3 Appendix, five aimed to reduce malaria through targeting Anopheles mosquitoes[22–26] and three aimed to reduce dengue by targeting Aedes aegypti.[8,27,28] Key characteristics of the included studies are described in Table 1, and are outlined in further detail in S3 Appendix. All eight cRCTs were conducted in low-middle income countries; three studies included urban neighbourhoods in America[8,27,28] while others included villages and outer peri-urban areas in America, South and Southeast Asia. Cluster sizes varied from single households[26] to areas with >1000 inhabitants;[23,24] the number of clusters per study ranged from 4–860.[22,26] The minimum distance between control and intervention cluster residences ranged from around 10m[22] to 3km.[23,24] One study reported outcomes only in children aged 3–9 years[8] and another in men aged 18–60 years;[22] the remaining six studies reported outcomes in all residents of the study area. Baseline prevalence data concerning the disease of interest was reported in six studies,[8,23–26] including one which ensured participants were disease-free at baseline.[22] Five studies stratified clusters at baseline prior to randomisation using a combination of mosquito density, disease incidence and population size measures.[23–25,27,28] Clusters did not appear to be matched in three studies.[8,25,26]

A variety of interventions were studied, including: mass adult mosquito trapping to reduce vector abundance;[27,28] topical repellents;[25,26] burning repellent coils at night in participants’ homes;[22] pyriproxyfen (a larvicide);[23,24] and community-led source reduction.[8] Six studies involved interventions requiring participant compliance with the intervention procedure, for example by regularly using mosquito traps, repellents, or source reduction measures.[8,22,25–28]

In three studies the intervention resulted in a statistically significant reduction in the disease of interest associated with the intervention.[8,23,24] One large study noted a 30% (95% CI: 4–55%) relative risk reduction in dengue infection among children. This study also observed reductions in entomological indicators of vector breeding (between 35% and 52%; p<0.05) [8] A smaller study noted a relative risk of 0.24 (95% CI: 0.20–0.29) for malaria in the intervention arm compared with the control arm following intervention implementation. Two entomological indicators of breeding for the target vectors also declined (between 50–84%; p<0.05), while non-statistically significant reductions in a third indicator were reported.[23] Another large study noted a relative risk of 0.30 (95% CI: 0.22–0.42) for malaria in the intervention arm compared with the control arm. Reductions (between 72–77%; p<0.05) in an indicator of target vector breeding were noted.[24] Both malaria studies observed corresponding increases in vector breeding indicators in the control arm.[23,24]

Of the five studies which did not observe disease reductions associated with the intervention; two malaria studies did not appear to consider entomological outcome measures,[25,26] and one malaria study observed a 32% reduction in the vector attack rate (p<0.05), however low vector numbers were included and other entomological indicators were not considered.[22] One dengue study did not observe any difference in entomological indicators from mass mosquito trapping[28], while a second, similar study observed a reduction (p<0.05) in vector abundance during the first of three intervention seasons only.[27]

Strengths and limitations of included study designs

A number of limitations and strengths common among the included studies were apparent (Table 2), including:

Contamination: All included studies may have been affected by contamination due to mosquito or person movement between intervention and control cluster areas, and from outside the study area. Most studies attempted to reduce mosquito contamination by ensuring a minimum distance between clusters, however space between clusters was consistently within the vector species’ known flight ranges.[23–28] Potential contamination due to the study population being highly mobile was noted in two studies.[23,24] Three studies reported non-targeted insecticide application occurred in both intervention and control clusters through government-initiated mosquito control activities. Such applications may have diluted the study intervention effects on the outcome measures.[8,24,26]

Heterogeneity: Considerable heterogeneity between study clusters was noted in two studies.[22,28] In one study, only four clusters were used and these were spread across two villages which had very different malaria rates prior to actively clearing malaria infections from participants during the recruitment phase.[22] Heterogeneity in the other study may have been related to differences in cluster ecology. These authors noted a significant vector increase in the intervention arm after initial baseline vector abundance measurements, which was hypothesised to have increased the risk of dengue.[28] Heterogeneity in most included studies was reduced by assessing vector abundance data (a proxy for disease risk) and/or disease rates in baseline clusters, then pairing clusters and randomly assigning the intervention to one cluster in each pair. Despite assessing vector baseline abundance, two studies lacked baseline seroprevalence data for the outcome disease.[27,28]

Low participant compliance with study intervention: Possible poor compliance was noted by a study observing topical repellent use for mosquito blood feeding reduction, however the extent of compliance was not well quantified.[25] False reassurance (likely due to the presence of mosquito traps, high catch rates and/or low dengue transmission during the study period) may have led to reduced participant compliance with the intervention in another study, where sub-optimal compliance was also partly attributed to participants’ concerns around the cost of electricity to run the mosquito traps and a risk of traps catching fire. Furthermore, the intervention coverage here (i.e. the length of time each household’s mosquito trap was used) was not monitored.[28]

Blinding: Full blinding was only possible where a placebo was used in the control arm. Placebo repellents were used by two studies, with participants and staff delivering the intervention blinded.[22,26] In one of these studies the outcome assessors were also blinded.[22]

Statistical under-powering: Underpowering was noted in four studies, none of which showed a statistically significant effect on the disease outcome.[25–28] Low disease transmission in the study area affected three of these studies.[26–28]

Community engagement: Two of the three studies which demonstrated statistically significant disease reductions described high community engagement with the intervention. This resulted in effectively locating and neutralising mosquito breeding sites, and comprehensive case finding.[8,23] In the dengue reduction study,[8] high community engagement with the intervention was achieved through: 1) Requesting permission to conduct the study from community leaders and discussing baseline evidence that the trial could be beneficial; 2) Conducting intervention design groups (usually with 8–10 community members per group) to discuss survey results, cost implications, and specific prevention strategies for each community; 3) Inviting local volunteers to train and work as intervention organisers and educate their communities about source reduction. Similarly, prior to a malaria reduction trial commencing, meetings were held for intervention area residents to discuss potential benefits of the trial and understand the importance of community participation for its success. A volunteer was selected from each group of 20 households. These volunteers helped field staff facilitate community discussion, and located and treated mosquito breeding sites. Community engagement strategies in the third study which demonstrated a statistically significant disease reduction were not described.[24]

Effective case finding: In two studies, two new field clinics were set up to diagnose and treat malaria cases in addition to two pre-existing clinics serving the study population. Study area residents were encouraged to attend a clinic if they developed malaria symptoms. Despite using passive surveillance for case finding, both studies showed a statistically significant malaria reduction in the intervention arm compared with the control arm. [23,24] Active case finding was used in the study which demonstrated a statistically significant dengue reduction following intervention, wherein paired saliva samples were obtained from participants and assayed for dengue antibodies to detect new infections.[8]

Bias assessment

All included studies were considered to: 1) treat study groups identically, other than by applying the intervention of interest; and 2) report completed participant follow-up or adequately describe and analyse differences between study arms. Participants were consistently analysed in the groups to which they were randomised, with outcomes measured and analysed consistently, reliably and appropriately. The trial designs were largely considered appropriate. Only three studies described the method of randomisation used.[8,27,28] Allocation to treatment groups was concealed in three studies.[8,22,26] Six studies demonstrated the study arms were similar at baseline with regards to the primary outcomes.[8,23,24,26–28]. Blinding occurred in three studies.[8,22,26] Five studies randomly assigned the intervention to one cluster in each cluster pair.[23–25,27,28] Three studies reported on participant loss to follow up.[8,22,26] Bias assessment results are displayed in S3 Appendix.