Types of studies

• Randomized controlled trials (RCTs) with: (a) the unit of randomization being a cluster and (b) at least two clusters per arm. As the two interventions are distributed at a community level, we do not expect to find trials with individual randomization.

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

• Interrupted time series designs (ITS) with: (a) a clearly defined point in time when the intervention occurred and (b) at least three data points before and three after the intervention.

Types of participants

All people living in a rural or urban malarious area, all levels of endemicity including both stable and unstable transmission.

Types of interventions

Types of outcome measures

Electronic searches

We will search the following databases using the search terms and strategy described in Appendix 1: Cochrane Infectious Diseases Group Specialized Register; Central Register of Controlled Trials (CENTRAL), published in the Cochrane Library; MEDLINE (PubMed); Embase (OVID); and LILACS, using the search terms detailed in Appendix 1. We will also check the WHO International Clinical Trials Registry Platform (WHO ICTRP; http://www.who.int/ictrp/en/) and ClinicalTrials.gov (https://clinicaltrials.gov/ct2/home) for ongoing trials using the terms: indoor residual spraying; IRS; insecticide‐treated nets; bednets; ITNs; LLIN.

Searching other resources

We will contact researchers working in the field for unpublished data. We will also check the citations of all trials identified by the above methods.

Selection of studies

Two review authors (LC and JP) will independently assess the titles and abstracts of trials identified by the searches. The same two review authors will assess full‐text copies of potentially relevant trials for inclusion using an eligibility form based on inclusion criteria. We will compare the results of our assessments and will resolve any disagreements by discussion and consensus, with arbitration by a third review author (PG) if necessary. We will ensure that multiple publications of the same trial are included once. We will list excluded studies, together with their reasons for exclusion, in a ‘Characteristics of excluded studies' table. We will illustrate the study selection process in a PRISMA diagram.

Data extraction and management

Two review authors (LC and JP) will independently extract information from the trials using pre‐piloted, electronic data extraction forms. In case of differences in extracted data, the two review authors will discuss these differences to reach consensus. If unresolved, we will consult a third review author (PG). In case of missing data, we will contact the original study author(s) for clarification.

We will extract data on the following.

• Trial design: type of trial; method of participant selection; adjustment for clustering (for cluster‐RCTs (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 and control (active ingredient, dose, formulation, method, frequency and timing of application, buffer zone between clusters); co‐interventions; description of control; coverage of intervention, control, and co‐interventions; compliance of intervention, control, and any co‐interventions.

• Outcomes: definition of outcome; diagnostic method or surveillance method; passive or active case detection; duration of follow‐up; time points at which outcomes were assessed; number of events; number of participants or unit time; statistical power; unit of analysis; incomplete outcomes/missing data.

• Other:

  • primary and secondary vector(s) species; vector(s) behaviour (nature, stability, adult habitat, peak biting times, exo/endophilic, exo/endophagic, anthro/zoophilic); method of mosquito collection(s); phenotypic insecticide resistance (based on WHO definitions if supplementary WHO cylinder assays or CDC bottle bioassays, or both, were performed whilst 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);

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

For dichotomous outcomes, we will extract the number of patients experiencing each outcome and the number of patients in each treatment group. For count/rate data outcomes, we will extract the number of outcomes in the treatment and control groups, and the total person time at risk in each group or the rate ratio, and a measure of variance (for example, standard error). For continuous outcomes, we will extract the mean and a measure of variance (standard deviation).

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

For non‐randomized studies, we will extract adjusted measures of intervention effects that attempt to control for confounding.

Assessment of risk of bias in included studies

Two review authors (LC and JP) will independently assess the risk of bias for each included 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 cluster‐randomized trials (Higgins 2011a; Higgins 2011b). We will assess non‐randomized controlled trials and ITS trials for risk of bias using Cochrane EPOC’s ‘Risk of bias’ tool. We will resolve any discrepancies through discussion or by consulting a third review author (PG). We will classify judgements of risk of bias as either at low, high, or unclear risk of bias, and we will use summary graphs (‘Risk of bias’ summary and ‘Risk of bias’ graph) to display results.

Measures of treatment effect

We will compare intervention and control data using risk ratios and for count/rate data, rate ratios. We will use adjusted measures of effect to summarize treatment effect from non‐randomized studies. We will present all results with their associated 95% CIs.

Unit of analysis issues

For cRCTs, or cluster non‐randomized trials, we will extract adjusted measures of effect where possible. If the study authors did not perform any adjustment for clustering, we will adjust the raw data ourselves using an ICC value. If an ICC is not reported in the paper, we will obtain this from similar studies, or estimate the ICC value. We will not present results from cluster‐randomized trials that are not adjusted for clustering. If we estimate the ICC, we will perform sensitivity analyses to investigate the robustness of our analyses.

If we identify studies for inclusion that have multiple intervention arms, we will include data from these studies by either combining treatment arms, or by splitting the control group so that we only include these participants in the meta‐analysis once.

Dealing with missing data

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

Assessment of heterogeneity

We will inspect forest plots for overlapping CIs and will assess statistical heterogeneity in each meta‐analysis using the I² statistic values and Chi² statistics. We will regard heterogeneity as moderate if I² statistic values are between 30% to 60%; substantial if they are between 59% to 90%; and considerable if they are between 75% to 100%. We will regard a Chi² test statistic with a P value ≤ 0.10 indicative of statistically significant heterogeneity. We will explore 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 are 10 or more trials included in each meta‐analysis, we will investigate reporting biases (such as publication bias) using funnel plots. We will assess funnel plot asymmetry visually, and use formal tests for funnel plot asymmetry (Harbord 2006). If we detect asymmetry in any of these tests or by a visual assessment, we will explore the reasons for asymmetry.

Data synthesis

We will analyse data using Review Manager 5 (RevMan 5) (RevMan 2014). We will use fixed‐effect meta‐analysis to combine data if heterogeneity is absent. If considerable heterogeneity is present, we will combine data using random‐effects meta‐analysis and report an average treatment effect. We will decide whether to use fixed‐effect or random‐effects models based on the consideration of clinical and methodological heterogeneity between trials, as described previously.

Subgroup analysis and investigation of heterogeneity

We will explore reasons for substantial heterogeneity using subgroup analysis. We plan to perform the following subgroup analyses.

• Use of LLINs/ITNs defined by individual use from the previous night:

  • high (80% to 100%);

  • moderate (50% to 79%);

  • low (less than 50%).

• Coverage of IRS:

  • high (80% to 100%);

  • moderate (50% to 79%);

  • low (less than 50%).

• Seasonality of malaria:

  • perennial;

  • seasonal;

  • epidemic.

• Mode of action of insecticides used for IRS:

  • voltage‐gated sodium ion channels;

  • acetylcholinesterase.

We will assess differences between subgroups using the Chi² test, with a P value less than 0.05 indicating statistically significant differences between subgroups.

Sensitivity analysis

We will perform sensitivity analysis on the primary outcome to see 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 is estimated, we will undertake sensitivity analyses to investigate the impact of varying the ICC value on meta‐analysis results.