Larviciding to prevent malaria transmission.

BACKGROUND: Larviciding refers to the regular application of chemical or microbial insecticides to water bodies or water containers to kill the aquatic immature forms of the mosquito (the larvae and pupae).
OBJECTIVES: To summarize research evidence evaluating whether larviciding with chemical or microbial insecticides prevents malaria transmission. SEARCH
METHODS: We searched the Cochrane Infectious Diseases Group Specialized Register; the Cochrane Central Register of Controlled Trials (CENTRAL), published in the Cochrane Library; MEDLINE; Embase; CAB Abstracts; LILACS; the World Health Organization International Clinical Trials Registry Platform (WHO ICTRP); ClinicalTrials.gov; and the ISRCTN registry up to 6 June 2019. SELECTION CRITERIA: We included cluster-randomized controlled trials (cRCTs), interrupted time series (ITS), randomized cross-over studies, non-randomized cross-over studies, and controlled before-and-after studies (CBAs) that compared larviciding with no larviciding. DATA COLLECTION AND ANALYSIS: We independently assessed trials for eligibility and risk of bias, and extracted data. We assessed the certainty of evidence using the GRADE approach. MAIN
RESULTS: Four studies (one cRCT, two CBAs, and one non-randomized cross-over design) met the inclusion criteria. All used ground application of larvicides (people hand-delivering larvicides); one evaluated chemical and three evaluated microbial agents. Studies were carried out in The Gambia, Tanzania, Kenya, and Sri Lanka. Three studies were conducted in areas where mosquito aquatic habitats were less extensive (< 1 km²), and one where habitats were more extensive (> 1 km²; a cross-over study from The Gambia).For aquatic habitats of less than 1 km², one cRCT randomized eight villages in Sri Lanka to evaluate chemical larviciding using insect growth regulator; and two CBA studies undertaken in Kenya and Tanzania evaluated microbial larvicides. In the cRCT, larviciding across all villages was associated with lower malaria incidence (rate ratio 0.24, 4649 participants, low-certainty evidence) and parasite prevalence (risk ratio (RR) 0.26, 5897 participants, low-certainty evidence) compared to no larviciding. The two CBA studies reported lower malaria prevalence during the intervention period (parasite prevalence RR 0.79, 95% confidence interval (CI) 0.71 to 0.89; 70,902 participants; low-certainty evidence). The Kenyan study also reported a reduction in the incidence of new malaria cases (RR 0.62, 95% CI 0.38 to 1.01; 720 participants; very low-certainty evidence).For aquatic habitats of more than 1 km², the non-randomized cross-over trial using microbial larvicides did not detect an effect for malaria incidence (RR 1.58, 95% CI 0.94 to 2.65; 4226 participants), or parasite prevalence (RR 1.15, 95% CI 0.41 to 3.20; 3547 participants); both were very low-certainty evidence. The Gambia trial also reported the mean haemoglobin level, and there was no difference across the four comparisons (mean difference -0.13, 95% CI -0.40 to 0.13; 3586 participants).We were unable to summarize or pool entomological outcomes due to unreported and missing data. AUTHORS'
CONCLUSIONS: Most controlled studies on larviciding have been performed with microbial agents. Ground larviciding for non-extensive larval habitats may have an effect on malaria transmission, and we do not know if there is an effect in large-scale aquatic habitats. We found no studies using larviciding application techniques that could cover large aquatic habitats, such as aerial spraying using aircraft.

Background

Description of the condition

Malaria is caused by the Plasmodium parasite, which is transmitted by female Anopheles mosquitoes. There are five Plasmodium species that cause disease in humans; however, the most important species in terms of disease burden are Plasmodium falciparum, which is prevalent in sub‐Saharan Africa, and Plasmodium vivax, which is more common in Asia and South America. There were an estimated 219 million malaria cases and 435,000 deaths worldwide due to malaria in 2017 (WHO 2018). Sub‐Saharan Africa carries a disproportionately high share of the malaria burden, with 92% of cases and 93% of malaria deaths in 2017 (WHO 2018). As well as direct effects on health, malaria is a major cause of poverty and underdevelopment in many countries, due to household and health system costs, absenteeism from school or work, reduced productivity, and premature death (Chima 2008). Malaria‐endemic countries are, on average, poorer by more than five‐fold and have lower rates of economic growth than non‐malaria endemic countries, with a mean growth of per‐capita gross domestic product (GDP) of 0.4% per year versus 2.3% between 1965 and 1990 (Sachs 2002).

Vector control tools, such as long‐lasting insecticidal nets (LLINs) and indoor residual spraying (IRS) of insecticides, play a major role in malaria control, alongside diagnosis and effective treatment of malaria cases, and chemoprevention in some population groups. Scale‐up of vector control, diagnosis, and treatment averted 663 million clinical cases of malaria between 2000 and 2015 (Bhatt 2015). However, progress against malaria is stalling and a high burden of morbidity and mortality still remains (WHO 2017; WHO 2018). The World Health Organization (WHO) set out ambitious targets in the Global Technical Strategy to eliminate malaria in at least 35 countries by 2030 (WHO 2015a).

Description of the intervention

Larviciding refers to the regular application of microbial or chemical insecticides to water bodies or water containers to kill the aquatic immature forms of the mosquito (the larvae and pupae) (Tusting 2013).

Malaria vectors lay their eggs in standing water and the eggs develop through a series of life stages (larvae and pupae) into adults. The type of standing water selected by ovipositing females depends on the species in question and can be natural or man‐made, temporary or permanent (Bruce‐Chwatt 1985). For example, Anopheles stephensi prefers containers such as water tanks, some species prefer brackish habitats (Anopheles aquasalis in Latin America), while others prefer riceland habitats (Anopheles arabiensis).

There are several different types of larvicide, including chemical larvicides (using conventional insecticides, such as temephos, or insect growth regulators, such as pyriproxyfen, methoprene, and diflubenzuron), microbial larvicides (such as Bacillus thuringiensis israeliensis (Bti) and Bacillus sphaericus (Bs)) and oils. Larvivorous fish have also been used as a form of malaria control. Larvicides have varying modes of action. For example surface films, such as mineral oils and alcohol‐based surface products, suffocate the mosquito larvae and pupae by covering the surface of a water body. This is different from synthetic organic chemicals, such as organophosphates, which inhibit cholinesterase and affect the central nervous system of the mosquito. Insect growth regulators interfere with insect metamorphosis and prevent adult emergence from the pupal stage. Microbial larvicides function by bacterial proteins binding to the larval gut, which cause the larvae to stop eating and die (WHO 2013).

How the intervention might work

Larviciding aims to reduce malaria transmission by targeting the immature stages (larvae and pupae) of the anopheline mosquito, to reduce the number of mosquitoes that reach adulthood. By reducing adult vector populations in this way, larviciding is expected to reduce the transmission of Plasmodium species by anopheline mosquitoes, and reduce morbidity and mortality from malaria (Figure 1).

1

Logic model of the proposed effect of larviciding on various entomological and epidemiological outcomes. EIR: entomological inoculation rate.

Many of the principles behind vector control come from the theory of vectorial capacity developed by George Macdonald in the 1950s (Macdonald 1957). Vectorial capacity describes the total number of potentially infectious bites that would eventually arise from all the mosquitoes biting a single perfectly infectious (i.e. all mosquito bites result in infection) human on a single day. Vectorial capacity can be linked to the basic reproduction ratio of a disease which is the estimated number of secondary infections potentially transmitted by a single infected individual in a totally susceptible population (Black 1968). The basic reproduction number represents the theoretical estimate of the intensity of transmission. The George‐Macdonald model shows that vectorial capacity is most sensitive to changes in adult mosquito survival, which led to the prioritization of IRS and LLINs as vector control tools in the 1950s. However, the vectorial capacity model does not adequately consider the aquatic stages of the vector and so the potential of larviciding is likely to have been underestimated (Brady 2016). Models show that larval source management (LSM) reduces mosquito population density linearly with coverage if adult mosquitoes avoid laying eggs in treated habitats, but quadratically if eggs are laid in treated habitats and the effort is therefore wasted (Smith 2013). This would mean that if the most productive habitats are targeted, larviciding could be highly effective even without extensive coverage. Larviciding may also operate against both indoor and outdoor (e.g. An arabiensis) biting and resting mosquitoes, unlike LLINs and IRS. This is beneficial, since in some settings anthropophillic vectors are able to sustain transmission even with high coverage of LLINs or IRS, or both (Bayoh 2010; Russell 2010; Lwetoijera 2014), and several studies have also shown evidence of behavioural adaptation of vectors towards early evening biting which may reduce the effectiveness of indoor interventions (Gatton 2013). Thus larviciding may be effective against ‘residual malaria transmission', which is generally defined as transmission that exists despite universal coverage of LLINs or IRS to which vector populations are fully susceptible (Durnez 2013; Killeen 2014).

Why it is important to do this review

There is a need for new tools in malaria vector control if the goals set by the WHO Global Technical Strategy are to be achieved (WHO 2018). Malaria vector control currently relies largely on LLINs and IRS. Although the WHO recommends the use of LSM (including larviciding) as a supplementary control measure (WHO 2013), larviciding is not widely used by malaria control programmes. This is despite historical and contemporary successes with the use of larviciding for vector control. Programmatic application of Paris Green, an arsenic‐based compound toxic to larvae, contributed to the elimination of species belonging to the Anopheles gambiae complex in Egypt and Brazil (Soper 1943; Shousha 1948). Larviciding is routinely practiced by mosquito control programmes in the USA and Europe (Becker 1997; Floore 2006). Larviciding has also been hugely successful against other vector‐borne diseases; for example, Bti and temephos were used to control species of the Simulium damnosum complex – vectors of onchocerciasis – in Brazil and the continent of Africa as a supplement to mass drug administration (MDA) (Sékétéli 2002; Gustavsen 2011).

Larviciding has the potential to overcome several challenges currently facing malaria vector control. First, larviciding is able to target outdoor resting and biting mosquitoes that are less affected by LLINs and IRS. Second, it could be used to tackle residual foci of malaria where high coverage of LLINs and IRS is not sufficient to eliminate malaria. Last, larviciding could be used together with other interventions as part of an insecticide resistance management strategy. Insecticide resistance has been reported in all major malaria vectors and involves all classes of insecticide (but particularly pyrethroids) and may threaten the effectiveness of insecticide‐based vector control (WHO 2012a). The distribution and intensity of insecticide resistance has been increasing over time. Of 80 malaria‐endemic countries reporting insecticide resistance monitoring data since 2010, 68 reported resistance to at least one insecticide class and 57 reported resistance to two or more insecticide classes (WHO 2018). The WHO Global Plan for Insecticide Resistance Management recommends the use of insecticide‐based and non‐insecticide‐based interventions targeting both immature and adult mosquitoes as an insecticide resistance management strategy (WHO 2012a). This is also aligned with Integrated Vector Management (IVM), an adaptive, evidence‐based, and multi‐sectorial approach to vector control, which is recommended by the WHO for more effective, sustainable, and ecologically sound vector control (WHO 2008).

A Cochrane Review of LSM for controlling malaria was published in 2013 (Tusting 2013). This contributed to WHO deliberations that led to the recommendation of LSM as a supplementary malaria vector control intervention, and a WHO operational manual on LSM (WHO 2012b; WHO 2013). Although all LSM interventions have the aim of reducing mosquito larvae, the ways they are carried out are very different and effectiveness is likely to differ. For example, habitat modification (a permanent alteration to the environment such as drainage of aquatic habitats) is different to regular application of chemical or microbial larvicides to a water body. Due to the diversity of forms of LSM, a new assessment of larviciding alone is justified, thus splitting the original Cochrane Review on LSM (Tusting 2013).

Objectives

To summarize and appraise experimental and quasi‐experimental studies evaluating the effect of larviciding with chemical or microbial insecticides on malaria transmission.

Methods

Criteria for considering studies for this review

Types of studies

Cluster‐randomized controlled trials (cRCTs) with: the unit of randomization being a cluster, and at least two clusters per arm. As larvicides are distributed at a community level, we did not expect to find trials with individual randomization.

Randomized and non‐randomized cross‐over trials with: the unit of randomization being a cluster, at least two clusters per arm, and a suitable washout period during which malaria or entomological indices have returned to baseline levels. As larvicides are distributed at a community level, we did not expect to find trials with individual randomization.

Controlled before‐and‐after studies (CBAs) with: a contemporaneous control group, and at least two sites per arm.

Interrupted time series (ITS) studies with: a clearly defined point in time when the intervention occurred, and at least three data points before and three during or after cessation of larviciding.

We excluded studies if:

the intervention was applied for less than one year in sites with perennial (year‐round) transmission (as reported by the study authors); or less than one transmission season (defined as the period from the onset of rains until one month afterwards) in sites with seasonal transmission (as reported by the study authors);

the follow‐up periods for the intervention and control periods were not identical.

Types of participants

All people living in a rural or urban malarious area that is at any level of endemicity, including both stable and unstable transmission.

We planned to include and analyse studies specific to special groups, such as refugees and soldiers, separately from other studies but none were identified.

Types of interventions

Intervention

Larviciding using chemicals (insecticides and insect growth regulators), microbial agents, or oils. We excluded plant products, because formulations have not been standardized and studies are thus not comparable. We also excluded biological larviciding using larvivorous fish, covered in a separate Cochrane Review (Walshe 2017).

Control

Not receiving larviciding interventions as described above. Any co‐interventions such as LLINs, IRS, topical repellents, spatial repellents, environmental manipulation, environmental modification, MDA, and case management must have been received in both control and intervention arms.

Types of outcome measures

Studies must have reported at least one primary outcome for inclusion.

Primary outcomes

Clinical malaria incidence: we used site‐specific definitions, provided they include: demonstration of malaria parasites by blood smear or a rapid diagnostic test (RDT), or both; and clinical symptoms including fever or history of fever, detected passively or actively.

Malaria parasitaemia incidence: measured as a count per person unit time of infections or new infections, both defined as parasitaemia confirmed by blood smear microscopy or RDT. New infections were defined as either infection in participants who were negative for parasites at an earlier survey or infection in participants who were cleared of parasites using drug treatment at an earlier survey.

Malaria parasite prevalence: proportion of surveyed people with confirmed parasitaemia.

Secondary outcomes

Entomological

Adult mosquito density measured by a technique previously shown to be appropriate for the vector (measured using human baits, light traps, knock‐down catches, baited huts, or other methods). Adult mosquito density would most likely have been reported as bites/person/night for human landing catches and mosquitoes/traps/night for trap catches or pyrethrum spray catches.

Sporozoite rate measured as the number of caught adult mosquitoes positive for malaria sporozoites. Sporozoites can be detected through molecular or immunological methods.

Entomological inoculation rate (EIR): the estimated number of bites by infectious mosquitoes per person per unit time. This was measured using the human biting rate (the number of mosquitoes biting a person over a stated time period measured directly using human baits or indirectly using light traps, knock‐down catches, baited huts, or other methods of biting rate determination) multiplied by the sporozoite rate.

Epidemiological

Incidence of severe malaria: we used site‐specific definitions, provided they include (a) and either (b) or (c):

(a) demonstration of parasitaemia by blood smear;

(b) symptoms of cerebral malaria including coma or prostration or multiple seizures, or both;

(c) severe life‐threatening anaemia (WHO 2015b).

Malaria‐related deaths.

Mean haemoglobin levels (g/dL).

Anaemia prevalence defined using WHO cut‐offs (WHO 2011).

Hospital admissions for malaria.

Adverse events

Any indicators of adverse events of the intervention, including the following.

Non‐target effects such as the larvicide killing other animals in the water body.

Reports of poisoning in humans due to exposure to larviciding chemicals.

Environmental impacts such as changes to the biodiversity and ecosystem due to the use of larvicides.

Search methods for identification of studies

We attempted to identify all relevant trials regardless of language or publication status (published, unpublished, in press, and in progress).

Electronic searches

We searched the following databases up to 6 June 2019, using the search terms and strategy described in Appendix 1:

Cochrane Infectious Diseases Group Specialized Register;

Cochrane Central Register of Controlled Trials (CENTRAL) published in the Cochrane Library (Issue 6, 2019)

MEDLINE (Pubmed, from 1966);

Embase (OVID, from 1974);

CAB Abstracts, from 1973 (Web of Science);

Latin American Caribbean Health Sciences Literature (LILACS) (BIREME, from 1982).

We also searched ClinicalTrials.gov, the WHO International Clinical Trials Registry Platform (ICTRP; www.who.int/trialsearch), and the International Standard Randomized Controlled Trials Number (ISRCTN) registry (www.isrctn.com/) for trials in progress, using "malaria", "mosquito", and "larvicid*" as search terms.

Searching other resources

Tusting 2013 handsearched the US Armed Forces Pest Management Board Defense Pest Management Literature Retrieval System and the Tropical Diseases Bulletin using the terms: malaria or mosquito and larvicides up to the end of 2010 and incorporated the results into the Cochrane Review ‘Mosquito larval source management for controlling malaria'. We had planned to update this search but decided it was unlikely any new studies that would fit the review's inclusion criteria would be found.

We contacted researchers in the field to identify unpublished data, and checked the reference lists of studies identified using electronic searches.

Data collection and analysis

Selection of studies

Two review authors (LC and AW) independently assessed the titles and abstracts of trials identified by the literature searches. We obtained the full‐text articles of any potentially relevant articles. The same two review authors assessed the full‐text articles of potentially relevant studies for inclusion using an eligibility form based on predetermined inclusion criteria. We resolved any disagreements by discussion and consensus, with arbitration by a third review author (SM), when necessary. We ensured that multiple publications of the same trial were included only once. We listed studies excluded after full‐text assessment, together with their reasons for exclusion, in the Characteristics of excluded studies table. We illustrated the study selection process in a PRISMA flow chart (Moher 2009).

Data extraction and management

Two review authors (LC and AW) independently extracted information from the trials using pre‐piloted electronic data extraction forms. SM was a primary investigator and author of one included study. He was not involved in the screening, data extraction or risk of bias assessment, and analysis for this particular study. When differences in extracted data arose, the two review authors discussed these differences to reach consensus and involved a third review author (SM), where necessary. For missing data, we contacted the original study author(s) for clarification.

We extracted the following data.

Trial design: type of trial; method of participant selection; adjustment for clustering (for cRCTs); sample size; method of blinding of participants and personnel.

Participants: trial settings and population characteristics; recruitment rates; withdrawal and loss to follow‐up.

Intervention: description of intervention (active ingredient, dose, formulation, method, frequency and timing of application, buffer zone between clusters); quality control of the larvicide (e.g. WHO Pesticide Evaluation Scheme (WHOPES) approved); quality assurance of implementation of larviciding; co‐interventions; description of control; duration of follow‐up; passive or active case detection; coverage of larvicide (as reported by the study authors) and co‐interventions (e.g. vector control, vaccines, chemoprophylaxis, diagnosis, and treatment); duration of the activity of the larvicide; compliance (with application of larvicide and co‐interventions).

Outcomes: definition of outcome; diagnostic method or surveillance method; number of events; number of participants or unit time; time point at which outcome was assessed in relation to larviciding implementation, statistical power; unit of analysis; incomplete outcomes or missing data.

Other:

primary and secondary vector(s) species; vector(s) behaviour (nature, stability, and extent (number and size) of aquatic habitats, proximity of aquatic habitats to human habitation, adult habitat, peak biting times, exophilic/endophilic, exophagic/endophagic, anthropophilic/zoophilic); method of mosquito collection(s); phenotypic insecticide resistance (based on WHO definitions if WHO cylinder assays, Centers for Disease Control and Prevention (CDC) bottle bioassays, intensity assays or synergist assays were performed while the trial was running); genotypic insecticide resistance profile (either performed during the trial or if the trial references data from previous studies done on the same local vector population within the previous five years); insecticide and larvicide resistance detected in the larvae (as reported by study authors);

malaria endemicity; eco‐epidemiological setting; population proximity and density; Plasmodium species.

For dichotomous outcomes, we extracted the number of participants experiencing each outcome and the number of participants in each treatment group. For count data outcomes, we extracted the number of outcomes in the treatment and control groups, the total person time at risk in each group or the rate ratio, and a measure of variance (e.g. standard error). For continuous outcomes, we extracted the mean and a measure of variance (e.g. standard deviation).

For cRCTs we recorded the number of clusters randomized; number of clusters analyzed; measure of effect (such as risk ratio (RR), odds ratio (OR), rate ratio, or mean difference (MD)) with confidence intervals (CI) or standard deviations; number of participants; and the intracluster correlation coefficient (ICC) value.

For non‐randomized studies (NRS), we extracted adjusted measures of intervention effects that attempt to control for confounding.

Assessment of risk of bias in included studies

Two review authors (LC and AW) independently assessed the risk of bias for each cRCT using the Cochrane ‘Risk of bias' tool and the five additional criteria listed in Section 16.3.2 of the Cochrane Handbook for Systematic Reviews of Interventions that relate specifically to cRCTs (Higgins 2011a; Higgins 2011b). For assessing the risk of bias for randomized cross‐over trials, we used the Cochrane ‘Risk of bias' tool also and the additional criteria listed in Section 16.4.3 of the Cochrane Handbook for Systematic Reviews of Interventions that relate specifically to randomized and non‐randomized cross‐over trials (Higgins 2011a). We planned to assess non‐randomized controlled studies and ITS for risk of bias using the Cochrane Effective Practice and Organisation of Care (EPOC) ‘Risk of bias' tool. We resolved any discrepancies through discussion or by consulting a third review author (SM). We judged studies at low, high, or unclear risk of bias, and used summary graphs (‘Risk of bias' summary and ‘Risk of bias' graph) to display results.

Measures of treatment effect

We compared intervention and control data using RRs if the outcome was dichotomous. Where effect sizes from studies were presented as an OR, we converted these to RRs following the methodology stated in Section 12.5.4.4 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a). We presented rate data as rate ratios. We calculated the MD for continuous measures. We used adjusted measures of effect to summarize treatment effect from NRS. We presented all results with their associated 95% CIs.

We aimed to report any accounts of possible adverse effects. We appreciated that the specified inclusion criteria were not designed to detect effects on animals in the water, people exposed to the larvicides, and the ecosystem overall, and we intended to note this in the discussion, but there were no adverse events.

Unit of analysis issues

For cRCTs, we planned to extract adjusted measures of effect where possible. If the study authors did not perform any adjustment for clustering, we planned to adjust the raw data using an ICC value. If the study did not report an ICC value, we contacted the study authors, obtained this from similar studies, or estimated the ICC. When we estimated the ICC, we performed sensitivity analyses to investigate the robustness of our analyses (Richardson 2016).

For cross‐over trials, we applied the principles stated in Sections 16.4.4 and 16.4.5 of the Cochrane Handbook for Systematic Reviews of Interventions that relate specifically to randomized and non‐randomized cross‐over trials (Higgins 2011a).

Dealing with missing data

In case of missing data, we intended to apply available‐case analysis and to only include data on the known results. The denominator would have been the total number of participants who had data recorded for the specific outcome. For outcomes with no missing data, we planned to perform analyses on an intention‐to‐treat basis. We intended to include all participants randomized to each group in the analyses and analyse participants in the group to which they were randomized.

Assessment of heterogeneity

We inspected forest plots for overlapping CIs and assessed statistical heterogeneity in each meta‐analysis using the I² statistic and Chi² test. We regarded heterogeneity as moderate if I² statistic values were between 30% and 60%; substantial if they were between 50% and 90%; and considerable if they were between 75% and 100%. We regarded a Chi² test statistic with a P value ≤ 0.10 as indicative of statistically significant heterogeneity. We explored clinical and methodological heterogeneity through consideration of the trial populations, methods, and interventions, and by visualization of trial results.

Assessment of reporting biases

If there were 10 or more trials included in each meta‐analysis, we intended to investigate reporting biases (such as publication bias) using funnel plots. We planned to assess funnel plot asymmetry both visually and using formal tests (Harbord 2006), and explore possible reasons for asymmetry.

Data synthesis

We analyzed data using Review Manager 5 (Review Manager 2014). We used a fixed‐effect meta‐analysis to combine data if heterogeneity was absent. If there was considerable heterogeneity, we combined data using a random‐effects meta‐analysis and reported a mean treatment effect (RRs and ORs for dichotomous outcomes and rate ratio for count data). We decided whether to use a fixed‐ or random‐effects model based on the consideration of clinical and methodological heterogeneity between trials.

We combined data across follow‐up time points for each included study.

Certainty of the evidence

We assessed the certainty of evidence using the GRADE approach (Guyatt 2011). We rated each primary epidemiological outcome (malaria incidence and prevalence) as described by Balshem 2011.

High: we are very confident that the true effect lies close to that of the estimate of the effect.

Moderate: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect.

Low: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect.

Very low: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

RCTs started as high‐certainty evidence but we downgraded the certainty of the evidence if there were valid reasons within the following five categories: risk of bias, imprecision, inconsistency, indirectness, and publication bias. We upgraded the certainty of the evidence for studies where there was a large effect, a dose–response effect, and if all plausible residual confounding would reduce a demonstrated effect or would suggest a spurious effect if the was no effect (Balshem 2011). We presented the GRADE assessments in a ‘Summary of findings' tables.

Subgroup analysis and investigation of heterogeneity

We initially analyzed all types of larvicide (e.g. surface films, synthetic organic chemicals, insect growth regulators, and microbial larvicides) together. If there was a sufficient number of studies then we grouped these and analyzed them separately.

We explored reasons for substantial heterogeneity using subgroup analysis. We intended to perform the following subgroup analyses.

Seasonality of malaria:

perennial, defined as year‐round transmission;

seasonal as reported by study authors in the manuscript or defined as 75% or more of all malaria episodes occurring in six or fewer months of the year (Roca‐Feltrer 2009);

epidemic, defined as a sharp rise in malaria incidence, higher than typical levels.

Extent of aquatic habitat:

container habitat;

habitats smaller than 1 km² (excluding containers);

habitats larger than 1 km².

Continent:

Africa;

non‐Africa.

We only performed a subgroup analysis based on the extent of aquatic habitat as there were insufficient studies to perform the other subgroup analyses.

Sensitivity analysis

We performed sensitivity analysis on the primary outcome to determine the effect of exclusion of trials at high risk of bias (for allocation concealment and incomplete outcome data) on the overall results. If the ICC value was estimated, we undertook sensitivity analyses to investigate the impact of varying the ICC value on meta‐analysis results.

Results

Description of studies

Results of the search

We identified 2510 reports using electronic searches. We removed one duplicate and screened all remaining 2509 abstracts against the review's inclusion criteria. Abstract screening resulted in 98 unique reports for full‐text screening (Figure 2).

2

Study flow diagram.

Included studies

Design

Five reports detailing four separate studies met the inclusion criteria and these are described in the Characteristics of included studies table. One study was a cRCT (Yapabandara 2001), one was a non‐randomized cross‐over trial (Majambere 2010), and two were CBAs (Fillinger 2009; Maheu‐Giroux 2013a).

Transmission

Three studies were each conducted in sub‐Saharan Africa with one in Kenya where P falciparum was present and malaria transmission was moderate (Fillinger 2009), one in Gambia where P falciparum was present and malaria transmission was seasonal (Majambere 2010), and one in Tanzania where P falciparum accounted for more than 90% of cases and malaria transmission was perennial with peaks in incidence after the two rainy seasons a year (Maheu‐Giroux 2013a). The remaining study was conducted in Sri Lanka where both P falciparum and P vivax were present (Yapabandara 2001). Yapabandara 2001 did not report on the level of malaria transmission.

Habitat and vectors

The four studies targeted a range of aquatic habitats and vector species and all applied larvicides by hand. In the Elahera gem‐mining area situated in Matale District, Sri Lanka, Yapabandara 2001 targeted shallow pits dug by gem miners that harbour An culicifacies and An subpictus. In The Gambia, investigators larvicided large flooded areas of the floodplain of the lower reaches of the Gambia River which are ideal larval habitats for An gambiae s.s., An melas, and An arabiensis (Majambere 2010). Fillinger 2009 targeted aquatic habitats in the valley bottoms in the Kenyan highlands. These habitats harbour predominantly An gambiae s.s. and An funestus and are becoming more important as papyrus swamps are deforested to create agricultural land. Maheu‐Giroux 2013a evaluated the effect of the Urban Malaria Control Programme (UMCP) in Dar es Salaam, Tanzania. Here there were numerous aquatic habitats including natural habitats (swamps, river beds, springs), agricultural habitats (rice paddies, ridge and furrow agriculture) and artificial non‐agricultural habitats (drains, construction pits, etc.). Habitats harboured predominantly An gambiae s.l. but An funestus and An coustani were also present, along with large numbers of culicines.

Interventions

A summary of the interventions implemented is given in Table 4. There are two main microbial larvicides: Bs and Bti, and these exist in two formulations – water dispersible and corn cob granule. Two studies used both water‐dispersible and corn cob granule formulations of Bti (Majambere 2010; Maheu‐Giroux 2013a). Fillinger 2009 used a water dispersible formulation of Bs for six months but switched to corn cob Bti after six months due to a lack of residual effect. Maheu‐Giroux 2013a also used a corn cob Bs formulation to treat closed aquatic habitats that predominantly harboured culicines. All studies that used Bti or Bs used commercial formulations manufactured by Valent BioSciences LLC. Yapabandara 2001 used the insect growth regulator pyriproxyfen as a larvicide in Sri Lanka.

1

Characteristics of larviciding

Study	Active ingredient, formulation, dose, and manufacturer	Frequency of application	Targeted aquatic habitats	Who carried out the larviciding	Vector species
Fillinger 2009	Commercial strains of Bs (water‐dispersible, Valent BioSciences LLC)	Weekly intervals for first 6 months of the study	All water bodies	Project staff	An gambiae s.l. and An funestus s.l.
	Commercial strains of Bti (water‐dispersible, Valent BioSciences LLC)	Weekly intervals for remainder of the study
Maheu‐Giroux 2013a	Commercial strains of Bs (0.04 g/m², water‐dispersible, Valent BioSciences LLC)	Weekly intervals	All open light‐exposed water bodies	Community‐owned resource person	An gambiae s.s., An funestus s.l., and An coustani
	Commercial strains of Bti (0.1 g/m², corn cob Valent BioSciences LLC)	Once every 3 months	All closed, covered, often highly polluted water bodies
Majambere 2010	Commercial strains of Bti (0.2 kg/hectare, water‐dispersible, Valent BioSciences LLC)	Weekly intervals	Areas of low vegetation across the Gambia river	Project staff using knapsack compression sprayers	An gambiae s.s., An melas, and An arabiensis
	Commercial strains of Bti (0.5 kg/hectare, corn cob, Valent BioSciences LLC)		Areas of high vegetation across the Gambia river	Project staff by hand
Yapabandara 2001	Pyriproxyfen, S31183 (Adeal 0.5% G) applied at a rate of 0.01 mg active ingredient/L (2 g of granules/m³)	3 applications: December 1994, June–July 1995, end of November 1995	Gem mining pits	Project staff	An culicifacies, An subpictus, and An aruna

Abbreviations: An: Anopheles; Bs:Bacillus sphaericus;Bti:Bacillus thuringiensis israeliensis.

The frequency of application varied across the studies that used a microbial larvicide. Two studies applied the larvicide at weekly intervals (Fillinger 2009; Majambere 2010). In the Tanzanian study, open habitats were treated every week with Bti whereas closed habitats were treated every three months with Bs (Maheu‐Giroux 2013a). Yapabandara 2001 applied pyriproxyfen on three occasions during the study, one in December 1994, the second between June and July 1995 and the last application at the end of November 1995. Only one study reported on the duration of the activity of the larvicide (Fillinger 2009).

In Sri Lanka, local volunteers helped field staff to access villages and locate gem pits and assisted with administering the intervention (Yapabandara 2001). In The Gambia, field applicators were recruited from communities within each zone to make use of their local knowledge of the environment (Majambere 2010). They were supervised by one field supervisor in each zone and trained for one month before larviciding. In the Kenyan study, larviciding was implemented by project staff (Fillinger 2009). The UMCP in Dar es Salaam utilized community‐owned resource people, each assigned to a particular neighbourhood, to deliver the larvicides (Maheu‐Giroux 2013a).

In the Sri Lankan study, the control arm received no interventions (Yapabandara 2001). Both CBA studies had two arms: the control arm received standard practice vector control (insecticide‐treated nets (ITNs) in Fillinger 2009 and predominantly untreated bednets in Maheu‐Giroux 2013a), while the intervention arm received larviciding plus standard practice vector control. The cross‐over trial had two intervention arms; standard practice vector control (ITNs), and standard practice vector control (ITNs) plus larviciding (Majambere 2010). There were two units per arm, called zones. Zones 1 and 3 had the larviciding in the first year and served as control in the second year. Zones 2 and 4 received the interventions in the reverse order.

Fillinger 2009 reported an increase in ITN use from 4.8% (95% CI 3.0% to 6.6%; range in control valleys 1% to 9% and in intervention valleys 2% to 6%) at baseline to 40.8% (95% CI 36.7% to 45.0%; range in control valleys 24% to 51% and in intervention valleys 25% to 51%) during the intervention year. Majambere 2010 also reported an increase in net use during the study period, from between 6.1% to 38.3% in 2006 to between 37.2% to 81.4% in 2007.

Outcomes

Two studies measured clinical outcomes in children aged six months to 10 years only (Fillinger 2009; Majambere 2010). The other two studies measured outcomes in participants of all ages. Two studies measured the incidence of clinical malaria (Yapabandara 2001; Majambere 2010). Four studies measured the prevalence of Plasmodium infection (Yapabandara 2001; Fillinger 2009; Majambere 2010; Maheu‐Giroux 2013a). Fillinger and colleagues also reported the incidence of new Plasmodium infections (Fillinger 2009). They used children with no parasites at the first cross‐sectional survey of the season who had become infected two months later to calculate the incidence rate of new parasite infections over the two‐month follow‐up. One study measured mean haemoglobin concentration (Majambere 2010).

Three studies reported EIR (Fillinger 2009; Majambere 2010; Maheu‐Giroux 2013a). Three studies presented adult mosquito density as a biting rate (Yapabandara 2001; Fillinger 2009; Maheu‐Giroux 2013a). One study presented this outcome as totals caught in traps (Majambere 2010). One study measured sporozoite rate (Majambere 2010).

Excluded studies

We excluded 90 full‐text articles for the following reasons (see Characteristics of excluded studies table).

No relevant outcomes (40 articles).

Study design did not match inclusion criteria (29 articles).

Intervention did not match inclusion criteria (eight articles).

Full text not available (four articles).

Duplicate but under a different journal (three articles).

Conference abstract (two articles).

Protocol (three articles).

Protocol of a study not performed (one article).

We found three studies awaiting classification (see Characteristics of studies awaiting classification table). We contacted the authors of Fuseini 2017, Javadian 1974, and Zhou 2013 for additional data to determine whether the studies would meet the review's inclusion criteria but we did not receive the necessary information.

Risk of bias in included studies

Judgement of the risk of bias in the included studies is summarized in Figure 3. We listed individual risk of bias assessments in the Characteristics of included studies table.

3

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Allocation

Three studies were not randomized trials, and, therefore. we judged them at high risk of selection bias (Fillinger 2009; Majambere 2010; Maheu‐Giroux 2013a). We judged the cRCT to have an unclear risk of selection bias as the trial authors did not explicitly state whether they conducted random sequence generation or allocation concealment (Yapabandara 2001).

Blinding

We judged studies to have a low (Yapabandara 2001; Fillinger 2009; Maheu‐Giroux 2013a), or unclear (Majambere 2010) risk of bias for performance bias and low (Fillinger 2009; Majambere 2010; Maheu‐Giroux 2013a) or unclear (Yapabandara 2001) detection bias.

Incomplete outcome data

Two studies were at high risk of attrition bias. Fillinger 2009 reported absences from cross‐sectional surveys and the magnitude of these absences differed by study arm. There was also over 10% loss to follow‐up in study groups in Majambere 2010. Maheu‐Giroux 2013a had low risk of attrition bias and Yapabandara 2001 was at unclear risk.

Selective reporting

All studies had unclear risk of reporting bias.

Other potential sources of bias

We considered the risk of additional types of bias in the studies (baseline characteristics, contamination, incorrect analysis, and baseline outcome). We judged Majambere 2010 at high risk of bias due to significant differences in baseline characteristics between the zones. For example, the prevalence of P falciparum infections was much higher in zone 1 (38.4%) compared to the others (range 9.5% to 16.8%).

Although not a form of bias, Yapabandara 2001 used an inappropriate analysis technique which did not adjust for the clustered nature of the data in their analysis. Unadjusted estimates from cRCTs contribute disproportionately to the pooled result in meta‐analysis since they receive too much weight.

Effects of interventions

See: Table 1; Table 2

Summary of findings for the main comparison

Summary of findings table 1

Larviciding versus no larviciding where mosquito aquatic habitats are < 1 km²
Patient or population: people at risk of malariaSetting: areas where mosquito aquatic habitats are < 1 km² (one RCT carried out in Sri Lanka, and two CBA studies carried out in Kenya and Tanzania (Yapabandara 2001; Fillinger 2009; Maheu‐Giroux 2013a)).Intervention: larvicidingComparison: no larviciding
Outcomes	Anticipated absolute effects* (95% CI)		Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Rate or risk with no larviciding	Rate or risk with larviciding
Malaria incidence	23 episodes per 100 person‐years	5 episodes per 100 person‐years	Rate ratio 0.24	4649 person‐years(1 cluster‐RCT)	⊕⊕⊝⊝ Lowa,bdue to imprecision	Larviciding may decrease malaria incidence
Parasite prevalence	4 per 100	1 per 100	RR 0.26	5868(1 cluster‐RCT)	⊕⊕⊝⊝ Lowcdue to imprecision	Larviciding may decrease parasite prevalence
	12 per 100	9 per 100 (9 to 11)	RR 0.79 (0.71 to 0.89)	70,902(2 controlled before‐and‐after studies)	⊕⊕⊝⊝ Lowddue to non‐randomized design
*The risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).Abbreviations: CI: confidence interval; RCT: randomized controlled trial; ICC: intracluster correlation coefficient; RR: risk ratio.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

aDowngraded two levels for imprecision: the rate ratio and CIs reported in the study were not adjusted for clustering. Sensitivity analysis with a mean cluster population of 675 showed the most conservative estimate of an ICC of 0.1 gave a rate ratio of 0.24 (95% CI 0.05 to 1.08) whereas the least conservative estimate of an ICC of 0.01 gave a rate ratio of 0.24 (95% CI 0.14 to 0.4). This created uncertainty around the point estimate.
 bAn additional study measured incidence of new infections. As this study was not a RCT, it was not combinable. However, the study showed a large effect consistent with the results of the RCT (RR 0.62, 95% CI 0.38 to 1.01) (Fillinger 2009). On GRADE assessment, the point estimate of 0.62 was very low‐certainty evidence. This was due to the study being a non‐randomized controlled trial, therefore baseline GRADE assessment started at ‘low'. Further downgraded one level for imprecision due to wide CIs.
 cDowngraded two levels for imprecision: the odds ratio and CIs reported in the study were not adjusted for clustering. Sensitivity analysis with a mean cluster population of 675 showed the most conservative estimate of an intracluster coefficient of 0.1 gave a RR of 0.26 (95% CI 0.03 to 2.42) whereas the least conservative estimate of an ICC of 0.01 gave an odds ratio of 0.08 (95% CI 0.02 to 0.46). The wide range in CIs generated very serious uncertainty around the point estimate.
 dNon‐RCTs, so baseline GRADE assessment started at ‘low', therefore no further downgrading required for risk of bias.

Summary of findings 2

Summary of findings table 2

Larviciding versus no larviciding where mosquito aquatic habitats are > 1 km²
Patient or population: people at risk of malariaSetting: areas where the extent of mosquito aquatic habitats are > 1 km² (one non‐randomized cross‐over study in The Gambia (Majambere 2010)).Intervention: larvicidingComparison: no larviciding
Outcomes	Anticipated absolute effects* (95% CI)		Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Rate or risk with no larviciding	Rate or risk with larviciding
Malaria incidence	23 episodes per 100 child‐years	36 episodes per 100 child‐years (22 to 61)	RR 1.58 (0.94 to 2.65)	1793 child‐years (1 non‐randomized cross‐over trial)	⊕⊝⊝⊝ Very lowa,bdue to inconsistency and imprecision	We are uncertain of the effects on malaria incidence.
Parasite prevalence	14 per 100	16 per 100 (6 to 45)	RR 1.15 (0.41 to 3.20)	3574 (1 non‐randomized cross‐over trial)	⊕⊝⊝⊝ Very lowa,bdue to inconsistency and imprecision	We are uncertain of the effects on parasite prevalence.
*The risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). The assumed risk of the comparison group is calculated from the total number of events/total number of participants in the control arms of the trials contributing to the meta‐analysis.Abbreviations: CI: confidence interval; RR: risk ratio.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

aDowngraded one level for inconsistency: both comparisons indicated an effect favouring no larviciding, but there was considerable quantitative heterogeneity (I² = 77%).
 bDowngraded two levels for imprecision: very wide CIs.

Primary epidemiological outcomes

Clinical malaria incidence

Cluster‐randomized controlled trial

Yapabandara 2001 found a reduction of 76% in the incidence of clinical malaria when communities in four villages received larviciding compared to those in four villages that did not receive larviciding. The study authors did not adjust for clustering. Using an ICC value of 0.1 (a conservative estimate), gave wide CIs ranging from a 95% reduction to an 8% increase (Analysis 1.1). Using an ICC value of 0.01 resulted in a smaller range (rate ratio 0.24, 95% CI 0.14 to 0.40).

1.1

Analysis

Comparison 1 Larviciding versus no larviciding, Outcome 1 Malaria incidence (cluster‐randomized controlled trial (cRCT)).

Non‐randomized studies

Two NRS investigated the impact of larviciding on malaria incidence. Majambere 2010 measured clinical malaria incidence and Fillinger 2009 measuring malaria parasitaemia incidence. They were subgrouped by the extent of aquatic habitats due to considerable heterogeneity (Analysis 1.2; I² statistic = 89%). Majambere 2010 reported two time points. The first time point in 2006 showed an increase in the risk of clinical malaria in the study group receiving larviciding by 101% compared to the control arm which did not receive larviciding (95% CI 51% to 168% increase). The second time point showed no effect of larviciding on the incidence of clinical malaria (RR 1.18, 95% CI 0.79 to 1.78).

1.2

Analysis

Comparison 1 Larviciding versus no larviciding, Outcome 2 Malaria incidence with subgrouping by extent of aquatic habitat (non‐randomized study (NRS)).

Malaria parasitaemia incidence

Fillinger 2009 found a reduction in the risk of new infections in the study group receiving larviciding of 38% compared to the control arm which did not receive larviciding (95% CI 62% decrease to 1% increase).

Parasite prevalence

Yapabandara 2001 reported two separate time periods postintervention for parasite prevalence. Similar to the malaria incidence outcome measured in this study, sensitivity analysis to take account of clustering showed some uncertainty around the precision. The most conservative ICC gave a very imprecise result, a RR of 0.25 (95% CI 0.03 to 2.42), while the least conservative gave a RR of 0.08 (95% CI 0.02 to 0.46), somewhat more precise but CIs were still wide.

All three NRSs reported the effect of larviciding on parasite prevalence (Analysis 1.5). Fillinger 2009 and Maheu‐Giroux 2013a reported an adjusted OR. When converted to a RR and pooled, the effect size showed a reduction of 21% in parasite prevalence in areas receiving larviciding compared to areas not receiving larviciding (95% CI 11% to 29% reduction).

1.5

Analysis

Comparison 1 Larviciding versus no larviciding, Outcome 5 Parasite prevalence with subgrouping by extent of aquatic habitat.

Majambere 2010 reported an unadjusted OR. We took the two northern zones as one comparison and the two southern zones as another. We also took each follow‐up year as separate comparisons. This led to four separate comparisons for Majambere 2010. The pooled analysis across the three NRS taking the least conservative estimate for the ICC in Majambere 2010 gave a RR of 0.88 with CIs crossing 1 (95% CI 0.66 to 1.16). However, when we took the most conservative estimate for the ICC for Majambere 2010, the pooled RR showed a reduction in parasite prevalence associated with larviciding (RR 0.80, 95% CI 0.71 to 0.89). This was because less weighting was given in the meta‐analysis for the Majambere 2010 comparisons as the ICC increased.

Due to concerns of a high risk of bias for baseline imbalance, we also conducted a sensitivity analysis excluding two comparisons from Majambere 2010 (Analysis 1.4). Excluding the northern zones of Majambere 2010 and taking an estimated ICC of 0.01, the pooled result suggested a lower 21% in parasite prevalence in the area that received larviciding compared to the area that did not receive larviciding (95% CI 11% to 29% lower).

1.4

Analysis

Comparison 1 Larviciding versus no larviciding, Outcome 4 Parasite prevalence (NRS).

Subgroup analyses

There was moderate heterogeneity in Analysis 1.4 when the comparisons from Majambere 2010 with an estimated ICC of 0.01 were pooled with the other two NRS (Fillinger 2009; Maheu‐Giroux 2013a) (I² = 59%; P = 0.003). This was explained when we subgrouped the studies by extent of the aquatic habitat in Analysis 1.5. The subgroup analysis showed there was a reduction in parasite prevalence when ground application of larvicides was conducted in areas where the extent of aquatic habitats were smaller than 1 km² (RR 0.79, 95% CI 0.71 to 0.89). The analysis did not show an effect of larviciding on parasite prevalence when larvicides were administered by ground application in areas where the extent of aquatic habitats exceeded 1 km² (RR 1.15, 95% CI 0.41 to 3.20). We did not conduct the other subgroup analyses planned due to the small number of studies identified.

Secondary outcomes

Entomological

Several studies reported EIR in the intervention and control arms (summarized in Table 5). Entomological outcomes could not be pooled due to issues with the way the figures were reported in the manuscripts. All studies reported a mean number with 95% CIs except for Majambere 2010. Both Fillinger 2009 and Maheu‐Giroux 2013a analyzed the data by using a model to adjust for confounders, whereas Majambere 2010 reported the raw data with no analysis. Fillinger 2009 and Maheu‐Giroux 2013a reported a statistically significant lower EIR in areas receiving larviciding compared to control areas. For the first year in Majambere 2010, there was no difference in EIR in the northern zones (each EIR 0) and an increase in EIR in the southern intervention zone (EIR 5.82) compared to the southern control zone (EIR 3.13). For the second year of the study, there was barely any difference in EIR between the northern intervention zone (EIR 2.32) and the northern control zone (EIR 2.24). The southern zones for 2007 showed a large difference between arms with an EIR of 17 in the southern control zone compared to 3.91 in the southern intervention zone.

2

Entomological inoculation rate from included studies

Study	Study arm	Mean numbers (95% CI)
Fillinger 2009a	No larviciding	1.68 (1.16 to 2.43)
	Larviciding	0.39 (0.16 to 0.79)
Maheu‐Giroux 2013aa	No larviciding	1.28
	Larviciding	0.683 (0.491 to 0.952)
Majambere 2010 (2006 data)	No larviciding North Zone	0
	Larviciding North Zone	0
	No larviciding South Zone	3.13
	Larviciding South Zone	5.82
Majambere 2010 (2007 data)	No larviciding North Zone	2.24
	Larviciding North Zone	2.32
	No larviciding South Zone	17.00
	Larviciding South Zone	3.91

Abbreviations: CI: confidence interval.
 aThere is a statistically significant difference between the study arms (P < 0.05).

Epidemiological

The mean haemoglobin level reported by study arm by Majambere 2010 was converted into an MD. The pooled result from the four comparisons showed no statistically significant difference in the mean haemoglobin level of children living in areas that received larviciding compared to areas that received no larviciding (MD –0.13, 95% CI –0.40 to 0.13). We adjusted using an ICC of 0.01 for this analysis as the CIs were already very wide.

Discussion

Summary of main results

See Table 1 and Table 2. We included four studies: one cRCT in Sri Lanka, and three NRSs in sub‐Saharan Africa.

Primary outcomes

Malaria incidence

The cRCT reported a protective efficacy against malaria of 76% and adjusting using a conservative ICC of 0.1 gave the same effect estimate but wide CIs (95% CI 0.05 to 1.08) (Yapabandara 2001).

The pooled estimate of the two comparisons from the cross‐over trial showed a statistically non‐significant 58% increase (95% CI 6% decrease to 165% increase) in the incidence of malaria in children in the group that received larviciding compared to those who did not receive larviciding (Majambere 2010). However, the trial authors found that the year of study was a potential effect modifier which generated uncertainty around this pooled analysis.

Fillinger 2009 reported a protective efficacy of 38% against incidence of new parasite infections in the study group receiving larviciding compared to the control arm (95% CI 62% reduction to 1% increase).

Yapabandara 2001 reported on the effect of larviciding on parasite prevalence. Utilizing the most conservative ICC value gave a statistically non‐significant protective efficacy of 74% (95% CI 97% reduction to 142% increase). Pooled estimates from two NRS showed a significant protective efficacy of larviciding against parasite prevalence of 21% (95% CI 11% to 29% reduction). The extent of aquatic habitat explained the moderate heterogeneity present when all NRS studies were pooled for this outcome.

Majambere 2010 was the only study to measure mean haemoglobin level. The pooled analysis showed there was no statistically significant difference between the two study arms (MD –0.13, 95% CI –0.40 to 0.13).

For entomological outcomes, three studies reported on the impact of larviciding on EIR. Due to missing data and differences in how EIR was measured in each study, it was not possible to pool across studies. Two studies reported a statistically significant reduction in EIR when an area received larviciding compared to an area that did not receive larviciding (Fillinger 2009; Maheu‐Giroux 2013a). The entomological results from Majambere 2010 were mixed.

Certainty of the evidence

We appraised the certainty of evidence using the GRADE approach. The GRADE assessments are presented in Table 1 and Table 2.

Three studies (one cRCT, two NRS) evaluated the efficacy of ground application of larvicides where aquatic habitats were less than 1 km². One cRCT provided low‐certainty evidence that ground application of larvicides could have had a large impact on malaria incidence. The analysis of this study was not adjusted for clustering and so the CIs may have been misleadingly narrow, and thus we downgraded by two levels due to imprecision. The same trial also reported a large beneficial effect of ground larviciding on parasite prevalence but again this was not adjusted for clustering. We downgraded by two levels to low‐certainty evidence. We judged the certainty of the evidence for the pooled estimate for the prevalence of malaria from NRS to be low.

One study evaluated the efficacy of ground application of larvicides where the extent of aquatic habitats was more than 1 km². We judged the certainty of the evidence for the pooled OR for the incidence of malaria as very low, as we downgraded by one level due to unexplained heterogeneity and by two levels due to imprecision. The level of certainty for the pooled estimate for parasite prevalence was also very low, as we downgraded by two levels for imprecision and by one level for inconsistency.

Overall completeness and applicability of evidence

Larviciding is a context‐dependent intervention which requires knowledge of malaria transmission dynamics, vector ecology, behaviour, and the extent of water bodies and population density in proposed target areas. It is currently listed as a supplementary intervention for malaria control (WHO 2012b), compared to LLINs and IRS which are potentially more broadly applicable to different settings.

All studies included in this review looked at the efficacy of ground application of larvicides. Our review provided low‐certainty evidence of the efficacy of larviciding where habitats could be feasibly treated by hand, such as drainage channels, irrigation channels, ponds, and pools. The certainty of evidence was downgraded due to imprecision; however, point estimates of effect from the three studies consistently showed a high reduction in malaria incidence and prevalence.

There was very low‐certainty evidence on the efficacy against malaria of ground larvicide application where aquatic habitats were extensive, such as flood plains. The larval habitats treated in the Majambere study were huge as marshland areas stretched for several kilometres along the River Gambia and some larval habitats were up to 2 km wide (Bogh 2003; Majambere 2008). Not all aquatic habitats were treated in this study due to deep water which prevented access by ground staff, especially during high tides. Another reason for the lack of effect in this study could have been that mosquitoes can fly long distances in this part of The Gambia (Bøgh 2007), potentially leading to spillover of mosquitoes from non‐intervention areas into study zones treated with larvicide. Because there was only one study with very low‐certainty evidence conducted in such a setting using ground application of larvicides, it was difficult to draw any conclusions.

LSM, primarily larviciding, is the basis of contemporary mosquito control in large parts of the USA (Mosquito Abatement Districts) and Europe (e.g. Rhine valley, Germany) (Becker 1997; Floore 2006). Programmes in the USA and Europe routinely use aerial application technologies such as planes and helicopters which enable higher coverage of extensive larval habitats compared to ground larviciding, and also target cryptic aquatic habitats. There is also a vast body of historical literature on the programmatic use of LSM including larviciding. For example, LSM was the primary intervention responsible for the eradication of An gambiae from Brazil (Soper 1943) and in Wadi Haifa, Egypt (Najera 2001). Unfortunately, this evidence could not be included in this review due to the study designs employed and extensive literature searches only identified four studies. Many large‐scale field trials were excluded from this review as they only measured entomological outcomes, while epidemiological outcomes are typically required to demonstrate the public health benefit of an intervention (Wilson 2015). Larviciding, as with other wide area vector control interventions like environmental management, does not lend itself to cRCTs since the cost of studies with sufficient numbers of large clusters would be prohibitive.

Potential biases in the review process

We identified no potential biases in the review process. SM is a trial author of one of the included studies but was not involved in the screening or data extraction of this study.

Agreements and disagreements with other studies or reviews

There is an existing Cochrane Review that included all LSM interventions (Tusting 2013). Aside from that review, we are not aware of any other systematic reviews on larviciding for malaria control.

Authors' conclusions

The WHO currently recommends larviciding and other larval source management (LSM) interventions as a supplementary malaria control intervention. Unlike insecticide‐treated nets (ITNs) and indoor residual spraying (IRS) which target indoor vectors, LSM could potentially target outdoor as well as indoor transmission. As a result, many programmes in the elimination phase are now considering LSM including larviciding to tackle the remaining foci of malaria transmission. This review supports the use of ground larviciding for non‐extensive larval habitats. We do not know if larviciding by hand in extensive habitats, largely inaccessible on foot or where water is tidal has any effect on malaria based on the results of one study of very low‐certainty evidence. Operational research could strengthen the evidence base in these particular settings, with an aim of identifying effective methods for distributing larvicides over large areas

Most countries do not have the capacity or capability to conduct larviciding. If malaria control programmes are to implement larviciding, then support will be required to assess feasibility, and implement, and monitor and evaluate the intervention.

The findings of this review indicate low‐certainty evidence of benefit from controlled studies; however, the reality is that few, if any, studies will be conducted in the coming years. Further evidence on the effects of larviciding should be generated through monitoring and evaluation of programmatic implementation using concurrent control areas, perhaps in pragmatic stepped wedge designs.

Although not evaluated or discussed in this review, evaluation of new technologies for identifying aquatic habitats (such as high‐resolution imaging) and aerial application of larvicides in malaria‐endemic areas may well be relevant to further refine larviciding strategies.

What's new

Zhou 2013

Methods	Study design: cRCT assessing a combination of vector control interventions. The study spanned 2 years (2010 to 2011), but for this review, only the data from 2011 were included as this was when larviciding was used.Unit of allocation: clustersNumber of units: 3 study sites which each had 3 paired clusters, making a total of 9 paired clusters. Each of these pair would then be randomly assigned the control (ITNs) or the intervention (ITNs + larviciding). Furthermore, within each cluster, an area was targeted with IRS.Outcome assessment/surveillance type: active case detection in cohort of 350 participants per cluster in 2010 and 450 per cluster in 2011.Cross‐sectional surveys were done during February and March 2010 before the IRS application and 2011 before the IRS and larvicide application. Another survey was conducted post intervention in May 2010, May 2011, and July 2011. Blood smears were taken from randomly selected participants of different ages within each cluster: approximately 150 in 2010 and 250 in 2011.Only 12/18 clusters were monitored in 2011.Length of follow‐up: around 2 years: February 2010 to July 2011Adjustment for clustering: yes
Participants	Number of participants: in 2010, numbers ranged from 2884 to 2906 for the cross‐sectional survey. In 2011, numbers ranged from 4323 to 5139. For the cohort, there were 6248 participants in 2010 and 5574 in 2011.Population characteristics: not reported but no particular age group was targeted.Withdrawal and loss to follow‐up: not reported.
Interventions	Larvicide:Active ingredient and dosage: commercial strains of Bti (VectoMax)Formulation: corn granulesManufacturer: Valent BioSciences LLC, USAQuality control of the larvicide: aquatic habitats were searched thoroughly by a team of technicians accompanied by field assistants from local villages.Duration of the activity of the larvicide: not reported.Method of application: not reported.Frequency of application: first round of application was completed in February/March 2011 and the second in March/April 2011, 4 weeks after the first round.Buffer size between clusters: 250 m buffer zone between control and intervention clusters; however, there was no buffer between the IRS targeted zones and the non‐targeted zones.Cointervention: ITNsActive ingredient and dosage: not reported.Method of distribution: Global Fund supported mass distribution in June to September 2006 (targeting < 5 s and pregnant women) in all study areas. A second round was accomplished during May to July 2011 (targeting everyone at risk).Coverage: increased from 40.7% (range 34.3 to 47.8%) in 2010 to 93.0% (range 81.6 to 100%) in 2011.Compliance: not reported
Outcomes	Malaria incidence defined as fever/history and plasmodium parasites detected by smearParasite prevalenceIndoor resting density of mosquitoes
Notes

Abbreviations: An: Anopheles; Bti: Bacillus thuringiensis israeliensis; CBA: controlled before‐and‐after; cRCT: cluster‐randomized controlled trial; DDT: dichlorodiphenyltrichloroethane; IRS: indoor residual spraying.

Comparison 1

Larviciding versus no larviciding

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Malaria incidence (cluster‐randomized controlled trial (cRCT))	1		Rate Ratio (Fixed, 95% CI)	Subtotals only
1.1 Sensitivity analysis with estimated mean cluster population 675 and estimated intracluster correlation coefficient (ICC) 0.01	1		Rate Ratio (Fixed, 95% CI)	0.24 [0.14, 0.40]
1.2 Sensitivity analysis with estimated mean cluster population 675 and estimated ICC 0.05	1		Rate Ratio (Fixed, 95% CI)	0.24 [0.08, 0.70]
1.3 Sensitivity analysis with estimated mean cluster population 675 and estimated ICC 0.1	1		Rate Ratio (Fixed, 95% CI)	0.24 [0.05, 1.08]
2 Malaria incidence with subgrouping by extent of aquatic habitat (non‐randomized study (NRS))	2		Risk Ratio (Random, 95% CI)	1.17 [0.60, 2.26]
2.1 Habitats < 1 km²	1		Risk Ratio (Random, 95% CI)	0.62 [0.38, 1.01]
2.2 Habitats > 1 km²	1		Risk Ratio (Random, 95% CI)	1.58 [0.94, 2.65]
3 Parasite prevalence (cRCTs)	1		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only
3.1 Sensitivity analysis with estimated mean cluster population 675 and estimated ICC 0.01	1	763	Risk Ratio (M‐H, Fixed, 95% CI)	0.08 [0.02, 0.46]
3.2 Sensitivity analysis with estimated mean cluster population 675 and estimated ICC 0.05	1	169	Risk Ratio (M‐H, Fixed, 95% CI)	0.19 [0.02, 1.68]
3.3 Sensitivity analysis with estimated mean cluster population 675 and estimated ICC 0.1	1	87	Risk Ratio (M‐H, Fixed, 95% CI)	0.26 [0.03, 2.42]
4 Parasite prevalence (NRS)	3		Risk Ratio (Random, 95% CI)	Subtotals only
4.1 Adjusted data	2		Risk Ratio (Random, 95% CI)	0.79 [0.71, 0.89]
4.2 Sensitivity analysis including Majambere 2010 with estimated ICC 0.01	3		Risk Ratio (Random, 95% CI)	0.88 [0.66, 1.16]
4.3 Sensitivity analysis including Majambere 2010 with estimated ICC 0.05	3		Risk Ratio (Random, 95% CI)	0.80 [0.71, 0.89]
4.4 Sensitivity analysis including Majambere 2010 with estimated ICC 0.1	3		Risk Ratio (Random, 95% CI)	0.80 [0.71, 0.89]
4.5 Sensitivity analysis excluding Majambere 2010 northern zones due to large baseline imbalance; estimated ICC 0.01	3		Risk Ratio (Random, 95% CI)	0.79 [0.71, 0.89]
5 Parasite prevalence with subgrouping by extent of aquatic habitat	3		Risk Ratio (Random, 95% CI)	0.88 [0.66, 1.16]
5.1 Habitats < 1 km²	2		Risk Ratio (Random, 95% CI)	0.79 [0.71, 0.89]
5.2 Habitats > 1 km²	1		Risk Ratio (Random, 95% CI)	1.15 [0.41, 3.20]
6 Mean haemoglobin level	1	3586	Mean Difference (IV, Fixed, 95% CI)	‐0.13 [‐0.40, 0.13]
6.1 Sensitivity analysis with estimated ICC 0.01	1	3586	Mean Difference (IV, Fixed, 95% CI)	‐0.13 [‐0.40, 0.13]