 Research
 Open access
 Published:
Estimating intervention effectiveness in trials of malaria interventions with contamination
Malaria Journal volumeÂ 20, ArticleÂ number:Â 413 (2021)
Abstract
Background
In cluster randomized trials (CRTs) or stepped wedge cluster randomized trials (SWCRTs) of malaria interventions, mosquito movement leads to contamination between trial arms unless buffer zones separate the clusters. Contamination can be accounted for in the analysis, yielding an estimate of the contamination range, the distance over which contamination measurably biases the effectiveness.
Methods
A previously described analysis for CRTs is extended to SWCRTs and estimates of effectiveness are provided as a function of intervention coverage. The methods are applied to two SWCRTs of malaria interventions, the SolarMal trial on the impact of mass trapping of mosquitoes with odorbaited traps and the AvecNet trial on the effect of adding pyriproxyfen to longlasting insecticidal nets.
Results
For the SolarMal trial, the contamination range was estimated to be 146Â m (\(95\%\) credible interval \([0.052,\,0.923]\)Â km), together with a \(31.9\%\) (\(95\%\) credible interval \([15.3,\,45.8]\%\)) reduction of Plasmodium infection, compared to the \(30.0\%\) reduction estimated without accounting for contamination. The estimated effectiveness had an approximately linear relationship with coverage. For the AvecNet trial, estimated contamination effects were minimal, with insufficient data from the cluster boundary regions to estimate the effectiveness as a function of coverage.
Conclusions
The contamination range in these trials of malaria interventions is much less than the distances Anopheles mosquitoes can fly. An appropriate analysis makes buffer zones unnecessary, enabling the design of more costefficient trials. Estimation of the contamination range requires information from the cluster boundary regions and trials should be designed to collect this.
Background
When testing new malaria control interventions, cluster randomized trials (CRTs) are often the study design of choice, because the intervention is either assigned at the household level, or contamination effects are anticipated between the householdsÂ [1]. With malaria, most transmission happens during the night when Anopheles mosquitoes bite and people are in their home. Movement of mosquitoes while searching for human hosts or oviposition sites is therefore the main cause of contamination in trials of mosquito control interventions against malaria. Whereas this is a challenge in field trials, the practical consequence is that intervention has a beneficial community effect on individuals living close by. To prevent this effect from biasing trial estimates of efficacy towards the null, clusters are usually chosen as geographically contiguous areas of households.
Since contamination may still arise at the cluster boundaries and hence bias trial results, malaria trials are often designed by choosing well separated clusters or enforcing separation by defining buffer zones around each clusterÂ [2,3,4,5,6,7]. Ideally, buffer size might be determined using estimates of the range of the contaminationÂ [8], but very broad buffers are often used since other information is rarely available. In such trials, entire clusters receive the intervention, but only data from cluster cores are analyzed. This allows a standard analytical approachÂ [2] but at the cost of enrolling very large populations. Spatial separation may increase heterogeneity between clusters, and the cluster cores may be unrepresentative of the whole population if clusters correspond to natural units such as villages.
Estimating the spatial contamination is of scientific interestÂ [9,10,11,12,13] because protection of people living nearby is an important property of an intervention. Secondary analyses of several CRTs of malaria interventions have estimated contamination effects using linear models with terms measuring the distance between observations from one arm of the study to the nearest observation from the other study armÂ [10, 14,15,16]. These analyses all found evidence of spatial effects and tended to demonstrate the importance of accounting for contamination in estimating unbiased effects of the intervention. Nevertheless, these linear models cannot simultaneously provide closedform estimates of the range over which the contamination is relevant while adjusting the estimate of effectiveness for the contamination. The authors recently demonstrated that this can be achieved with a sigmoid random effects model for the analysis of CRTs of malaria interventions with contamination arising from mosquito movementÂ [9].
Stepped wedge cluster randomized trials (SWCRTs)Â [2, 17, 18] are a modification of CRTs in which the intervention is introduced progressively to all clusters in random order. To gain a better understanding of the effect of contamination in SWCRTs, the proposed model for CRTsÂ [9] is extended to analyze SWCRTs. It is then shown how the measurable contamination between trial arms (a quantity termed contamination range) leads to an estimate of the effective intervention coverage for each household and how this relates to the intervention effectiveness. These methods are applied to two SWCRTs of malaria interventions; the SolarMal trial assessing the effect of mass trapping with solarpowered odorbaited mosquito traps on Rusinga island, KenyaÂ [19,20,21], and the AvecNet trial investigating the effect of adding pyriproxyfen to longlasting insecticidal nets in Burkina FasoÂ [22, 23].
Methods
The SolarMal trial of odorbaited mosquito traps
The SolarMal SWCRT aimed to reduce mosquito population size, reduce biting intensity and eliminate Plasmodium falciparum malaria on Rusinga island, Lake Victoria, KenyaÂ [19,20,21] using Solarpowered Mosquito Trapping SystemsÂ [24] (SMoTS) which included odorbaited traps to lure and kill hostseeking mosquitoes. All households on Rusinga (area 44Â km^{2}, mean population 24,879Â [21]) were eligible to take part in the trial and were assigned to clusters using a travelling salesman algorithm, resulting in 81 geographically contiguous clusters of 50â€“51 households. Between June 2013 and May 2015, SMoTS were installed in one cluster per week, with a randomized order of clusters, until universal coverage of 4358 households was achievedÂ [20].
The primary outcome was clinical malaria in individuals of any age, measured as fever plus a positive rapid diagnostic test (RDT) result and monitored through repeated household visits, secondary outcomes were malaria prevalence, measured by RDT, and mosquito densities. Data on malaria prevalence were collected at fourmonth intervals resulting in five survey rounds during rollout, at \(22\%\), \(46\%\), \(63\%\), \(76\%\) and \(95\%\) intervention coverage. In each survey, malaria prevalence was recorded in a \(10\%\) random sample of households and clusters were excluded from analysis in the week during which the SMoTS were installed. Further details are given in the study protocolÂ [19].
The clinical incidence of malaria episodes was unexpectedly low, hence the focus of the original analysisÂ [21] shifted to the secondary outcome. Malaria prevalence was reported to be \(31.4\%\) [\(95\%\) confidence intervals (CI) \([27.5,\,35.1]\%\)] lower in intervention clusters (prevalence \(23.7\%\), 1552/6550 people) than in control clusters (prevalence \(34.5\%\), 2002/5795 people). Including random effects for clustering and survey round with generalized linear mixed models, the effectiveness of SMoTS on malaria prevalence was estimated to be \(30.0\%\) (\(95\%\) CI \([20.9,\,38.0]\%\)).
The AvecNet trial of longlasting insecticidal nets
The AvecNet trial assessed the effect of adding pyriproxyfen, an insect growth regulator, to longlasting insecticidal nets (LLINs) in rural Burkina Faso, an area with intense malaria transmission and highly pyrethroidresistant vectors. A baseline demographic survey enumerated 63,903 individuals living in 93 villages in an area of 1250Â km\(^2\)Â [22]. Over a twoyear period, during high malaria transmission seasons, LLINs treated with permethrin were incrementally replaced by LLINs treated with permethrin and pyriproxyfen in a SWCRT with 40 clustersÂ [22, 23], with an overall \(95\%\) coverage of nets. Clusters were based on administrative units and an average of 50 children (aged 6â€“60 months) were selected in each cluster and followed up by passive case detection for clinical malaria at health centers. Each month from June to September in 2014 and 2015, five clusters switched from control to intervention arm.
The primary outcome was clinical malaria, measured as fever plus a positive RDT result for Plasmodium falciparum. The childyears at risk and the incidence rate ratio (IRR), were calculated for each month in each group. Data were collected between June to December 2014 and May to December 2015, resulting in nine months with data from both the intervention and control arms. In these 9Â months, the mean intervention coverage was 17, 33, 44, 50, 51, 51, 56, 64 and \(81\%\). Further details of trial design can be found in the study protocolÂ [22].
The original analysisÂ [23] estimated clinical malaria incidence of 2.0 episodes per childyear in the control group and 1.5 episodes per childyear in the intervention group (IRR 0.88, \(95\%\) CI \([0.77,\,0.99]\), estimated from a Poisson model with offset for logtransformed exposure years, a random effect for cluster and fixed effects for month and health facility).
Analysis of SWCRTs allowing for contamination
Contamination between clusters because of mosquito movement between households might be expected to bias the intervention effects in both the SolarMal and AvecNet trials. This contamination is expected to follow a symmetrical smooth gradient in the boundary regions between intervention and control clusters and can hence be modeled by a sigmoid function of the distance of households to the nearest household in the discordant trial armÂ [9] (Fig.Â 1). Analyses of simulated datasetsÂ [9] found that this approach can provide unbiased and precise estimates of the contamination range and of the effectiveness, given that at least \(50\%\) of the households are at distances greater than the estimated contamination range from the nearest discordant household, hereafter called households in core.
With SWCRT designs, while the cluster size is constant, the assignment to arms changes, leading to variation in the distance to the nearest discordant household throughout the study. By accounting for this, and including a random effect for time, it is possible to extend the proposed sigmoid analysis to SWCRTs. Data with all clusters assigned to the intervention or control arm cannot be included, since the distance to the nearest discordant household is then not defined. Let \(\Delta _{ijk}\) denote the distance of the jth household in the ith cluster at the kth time step to the nearest discordant household, endowed with a negative sign for the households in the control arm. For malaria prevalence, the outcome \(Y_{ijk}\) of the jth household in the ith cluster at the kth time step, with \(i = 1,\dots ,c\), \(j = 1,\dots ,h\) and \(k = 1,\dots ,s\) (hereafter abbreviated with household ijk), can then be described by a Bayesian hierarchical model as followsÂ [9]:
For malaria incidence, a log link function together with an offset for the time at risk should be used. In this model formulation, \(\beta _{1,ik}\) denotes a random effect parameter assigned to each cluster in each survey roundÂ [2], centered on the expected prevalence in the control arm, \(\beta _1\). The other parameters \(\beta _2\), \(\beta _3\), \(\mu\) and \(\tau\) are assigned noninformative priors. The parameter \(\beta _2\) denotes the intervention effect and \(\beta _3\) can be transformed into the contamination range in km as \(\hat{\theta } = \log (0.95/0.05)\beta _3^{1}\).
This estimate \(\hat{\theta }\) can be used to define the area around household ijk that influences the density of infectious mosquitoes, the effective intervention coverage \(\mathcal {R}_{ijk}\). This is defined as the common density of the intervened households relative to the general density of households. The closer any other household m is to ijk, the greater is the contribution of mâ€™s intervention status to the effective intervention coverage (Fig.Â 2). This leads to a simple relationship between \(\mathcal {R}_{ijk}\) and the distance to the nearest discordant household \(\Delta _{ijk}\). Households whose distance to the nearest discordant household is large are only surrounded by households with the same intervention status and hence \(\mathcal {R}_{ijk}\) is either almost zero or one. By approximating this relationship with a sigmoid function whose growth rate depends on \(\hat{\theta }\), and plugging it into the Bayesian hierarchical modelÂ 1, it holds that:
With this formulation and after a back transformation, it is possible to describe the trial outcome as a linear function of the effective coverage, \(\mathcal {R}_{ijk}\). This procedure is formally described in Appendix.
Analysis of the SolarMal and the AvecNet trials
Both SWCRTs were analyzed with the sigmoid random effects model from Eq.Â 1 and fitted using rjagsÂ [25]. Uninformative priors were chosen for all parameters, and \(\beta _3\) was constrained for the contamination range to be positive. All calculations were performed at sciCORE scientific computing core facility at the University of Basel under R version 4.0.0Â [26].
For the SolarMal trial, the random effects parameter \(\beta _{1,ik}\) was varied by survey round (as in the original publicationÂ [21]) and by survey round and households. The results are reported in terms of the reduction in odds ratio (OR), as well as the reduction in relative risk (RR). For the AvecNet trial, the random effects parameter was varied by survey round and health facility and the results are reported in terms of the IRR. Instead of the distance to the nearest discordant household, the distance to the household of the nearest discordant child enrolled in the trial was calculated, because only this data was available. The intervention effectiveness is described in terms of the effective intervention coverage for both trials.
Results
SolarMal trial
For the five survey rounds, the mean distance to the nearest discordant household was 2.3, 0.9, 1.1, 1.6 and 2.7Â km. For an assumed contamination range of 100Â m, this results in 95, 91, 85, 93, and \(98\%\) of households in core, justifying a sigmoid random effects analysisÂ [9]. When including a random effect for survey round only, SMoTS were associated with a \(31.9\%\) reduction (\(95\%\) credible interval (CrI): \([15.3,\,45.8]\%\)) in odds ratio in the two arms, translating to a \(25.2\%\) reduction (\(95\%\)CrI: \([10.9,\,39.0]\%\)) in relative risk (TableÂ 1). The credible intervals were wider than the original confidence intervals. The contamination range was estimated to be 146Â m, also with a wide credible interval (\([0.052,\,0.923]\)Â km). With another random effect included for the household effects, SMoTS were associated with a \(42.1\%\) reduction (\(95\%\)CrI: \([32.2,\,51.3]\%\)) in odds ratio, and a \(34.1\%\) reduction (\(95\%\)CrI: \([24.4,\,44.1]\%\)) in relative risk. The contamination range was estimated to be 133Â m (\([0.052,\,0.943]\)Â km). The effectiveness is almost linear in effective coverage (Fig.Â 3) rising from zero effectiveness at zero coverage to the maximal effectiveness of \(34.1\%\) when intervention households are only surrounded by other intervention households. The credible intervals increase with coverage.
AvecNet trial
The mean distance to the household of the nearest discordant child enrolled was high for all nine survey rounds, with a mean of 4.5Â km. For an assumed contamination range of 100Â m, this resulted in a mean of \(98\%\) of households in core. This indicates that the data to estimate the contamination range are sparse, but a sigmoid random effects analysis can be carried out.
Adding pyriproxyfen to LLINs was associated with a reduction in incidence of clinical malaria in children of \(17\%\) (IRR 0.83, \(95\%\)CrI: \([0.70,\,1.00]\)), with credible intervals comparable to the confidence intervals from the original analysis (TableÂ 2). The contamination range was estimated to be 101Â m, with a wide credible interval (\(95\%\)CrI: \([0.051,\,0.745]\)Â km). The incidence rate ratio decreases as the effective coverage increases, in an almost linear fashion because of the inverse logarithmic transform (Fig.Â 4). The credible intervals become wider with higher the coverage.
Discussion
In CRTs or SWCRTs of malaria interventions, contamination between the trial arms arises because of mosquito movement. In a conventional analysis this may bias effectiveness estimates, but this can be corrected with an appropriate analysis, such as a Bayesian hierarchical model with a sigmoid function for effectiveness as a function of distance to the nearest discordant household, that was recently proposed for CRTsÂ [9]. This model yields a closedform contamination range that quantifies the contamination arising from mosquito movement between trial arms, and adjusts the main estimate of effectiveness for contamination, eliminating the need for buffer zones.
The proposed analytical approach is tailormade for malaria interventions where transmission can be geolocated to the hostâ€™s primary residence, and the main source of contamination between clusters arises from dispersal of adult female Anopheles mosquitoes, for which the proposed model, corresponding to mosquito dispersion by diffusionÂ [27, 28] is a reasonable approximation. In nature dispersal will vary between sites, within sites and by season, and depends on the extent and spatial distribution of aquatic habitats, households and alternative blood sourcesÂ [29], as well as wind strength and direction and obstacles in the environment. For both interventions considered here, with effects mainly depending on mosquito densities, contamination was considered to be symmetrical. The intervention may protect nearby nonusers, while users with many nearby nonusers have reduced intervention effects. With this model, a difference between homogeneously distributed and clustered interventions on the overall intervention effect is not distinguishableÂ [29]. The same modelling approach might be applied where the intervention itself is designed to be dispersed by mosquitoes (for instance sterile insect techniques) and even with humanside interventions such as mass drug administration or mass vaccination, though in the latter cases contamination is less important relative to the overall efficacy since more of the impact is due to the direct effect of individual protection. Contamination also arises in CRTs of many other health interventions, but where transmission is not by nightbiting mosquitoes the geometry is likely to be more complicated. For instance, where the intervention is behavioural and the primary source of contamination is social (and hence nonspatial), or with directly transmitted infections or those transmitted by less mobile and day biting Aedes mosquitoes (where infections often acquired at workplaces or schools, making geographically congruent clusters less desirable), different models of contamination are needed. In any given trial, the appropriateness and fit of the chosen contamination model should be carefully evaluated.
In this work, the sigmoid model is applied to two SWCRTs, the SolarMal and AvecNet trials. SWCRTs can be inferior in terms of power or bias compared to parallel designs and might be vulnerable to imprecision caused by temporal trends in underlying disease ratesÂ [3, 30] but may be required because of logistical, practical or financial constraintsÂ [1, 31] (for example, in the SolarMal trial an objective was to assess whether interruption of transmission would occur at complete coverageÂ [20]). Because of the changing boundaries, the analysis of contamination effects in SWCRTs is more complicated, but in principle SWCRT data could be used to analyze changing patterns of contamination in time and place. At the same time, it is unclear how the imbalance between arms affects the precision and bias of the resulting estimates.
A reanalysis of the SolarMal trial yielded a slightly higher estimate of effectiveness than was reported in the original trial analysisÂ [21], but with less precision. Adding a random effect for the households increased the estimate of effectiveness with reasonably wide credible intervals. Also for the AvecNet trial, the reduction in incidence of clinical malaria in children was higher than in the original analysisÂ [23], with only slightly less precision. The contamination range was consistently around 140Â m in the SolarMal trial and around 100Â m in the AvecNet trial, which is much less than the maximal distance Anopheles mosquitoes can flyÂ [8, 32].
The SolarMal trial was conducted in a small, densely populated area and had many small clusters. The AvecNet trial, in contrast, was conducted in a much larger area, with a population density 10 times lower than that in the SolarMal trial (around 50 people per km\(^2\) compared to more than 500 people per km\(^2\)). The settlement patterns where these trials were conducted are also different: in the region where the SolarMal trial took place around Lake Victoria, households are scattered, while the area where the AvecNet trial was conducted has villages with tight aggregations of houses, typical of the West African Sahel. These factors affect the percentage of households in core, the percentage of households unaffected by the contamination across cluster boundaries, where a balance is needed for the proposed analysis to yield unbiased and precise estimates. In the AvecNet trial, a subset of children was chosen from each village, to allow for clusters to be chosen as administrative units. This resulted in a high percentage of households in core, though this number is not comparable to the SolarMal trial, because only the distance to the household of the nearest discordant child was calculated. Informed by a previous simulation studyÂ [9], it is assumed that with so little information from the boundary regions, the contamination range cannot be estimated reliably and the proposed model is not working.
Like AvecNet, many trials define clusters based on administrative units with cluster boundaries passing through uninhabited areas. However, for contamination effects to be estimable, the trial must be designed to collect information from the boundary zones where contamination is likely. If cluster boundaries can pass through inhabited areas, as in the SolarMal trial, equalpopulation clusters can be assigned giving a more balanced design with optimal power, therefore requiring fewer participants. When there is contamination there is also empirical information about every level of local coverage from within either a CRT or SWCRT, even without universal overall coverage. This enables extension of the analysis using kernel density estimation to infer from the contamination range how effectiveness depends on intervention coverage. These estimates could be used to support allocation decisions when interventions are deployed, but where resource constraints mean universal coverage is not achievable.
Conclusions
It was shown how trials with anticipated contamination effects arising from mosquito movement can be analyzed to give unbiased and precise estimates of effectiveness. Guidance is now needed on how to plan trials with adequate power and precision to allow for contamination. Without the need for buffer zones, or for clusters to correspond to villages, cluster size can be reduced to a minimum determined by operational factors or contamination effects, reducing the required numbers of participants in field trials of malaria interventions. This should lead to more costefficient trials and a better understanding of the indirect effects of interventions in protecting nearby nonusers.
Availability of data and materials
The data underlying this article will be shared on reasonable request to the corresponding author.
Abbreviations
 CRTs:

Cluster randomized trials
 SWCRTs:

Stepped wedge cluster randomized trials
 SMoTS:

Solarpowered mosquito trapping systems
 CI:

Confidence intervals
 LLIN:

Longlasting insecticidal net
 RDT:

Rapid diagnostic test
 IRR:

Incidence rate ratio
 OR:

Odds ratio
 RR:

Relative risk
References
Wilson A, Boelaert M, Kleinschmidt I, Pinder M, Scott T, Tusting L, et al. Evidencebased vector control? Improving the quality of vector control trials. Trends Parasitol. 2015;31:380â€“90.
Hayes R, Moulton L. Cluster randomised trials. Biostatistics series. Boca Raton: Chapman & Hall/CRC, Taylor & Francis Group; 2009.
Wolbers M, Kleinschmidt I, Simmons C, Donnelly C. Considerations in the design of clinical trials to test novel entomological approaches to dengue control. PLoS Negl Trop Dis. 2012;6:e1937.
Protopopoff N, Wright A, West P, Tigererwa R, Mosha F, Kisinza W, et al. Combination of insecticide treated nets and indoor residual spraying in northern Tanzania provides additional reduction in vector population density and malaria transmission rates compared to insecticide treated nets alone: a randomised control trial. PLoS ONE. 2015;10:e0146629.
Protopopoff N, Mosha J, Lukole E, Charlwood J, Wright A, Mwalimu C, et al. Effectiveness of a longlasting piperonyl butoxidetreated insecticidal net and indoor residual spray interventions, separately and together, against malaria transmitted by pyrethroidresistant mosquitoes: a cluster, randomised controlled, twobytwo factorial design trial. Lancet. 2018;391:1577â€“88.
Eisele T, Bennett A, Silumbe K, Finn T, Chalwe V, Kamuliwo M, et al. Shortterm impact of mass drug administration with dihydroartemisinin plus piperaquine on malaria in southern province Zambia: a clusterrandomized controlled trial. J Infect Dis. 2016;214:1831â€“9.
Delrieu I, Leboulleux D, Ivinson K, Gessner B, Chandramohan D, Churcher T, et al. Design of a phase III cluster randomized trial to assess the efficacy and safety of a malaria transmission blocking vaccine. Vaccine. 2015;33:1518â€“26.
Guerra C, Reiner R, Perkins T, Lindsay S, Midega J, Brady O, et al. A global assembly of adult female mosquito markreleaserecapture data to inform the control of mosquitoborne pathogens. Parasit Vectors. 2014;7:276.
Multerer L, Glass T, Vanobberghen F, Smith T. Analysis of contamination in cluster randomized trials of malaria interventions. Trials. 2021;22:613.
Jarvis C, Di Tanna G, Lewis D, Alexander N, Edmunds W. Spatial analysis of cluster randomised trials: a systematic review of analysis methods. Emerg Themes Epidemiol. 2017;14:12.
Halloran M, Auranen K, Baird S, Basta N, Bellan S, Brookmeyer R, et al. Simulations for designing and interpreting intervention trials in infectious diseases. BMC Med. 2017;15:223.
BenjaminChung J, Abedin J, Berger D, Clark A, Jimenez V, Konagaya E, et al. Spillover effects on health outcomes in low and middleincome countries: a systematic review. Int J Epidemol. 2017;46:1251â€“76.
AnayaIzquierdo K, Alexander N. Spatial regression and spillover effects in cluster randomized trials with count outcomes. Biometrics. 2020;77:490â€“505.
Binka F, Indome F, Smith T. Impact of spatial distribution of Permethrinimpregnated bed nets on child mortality in rural northern Ghana. Am J Trop Med Hyg. 1998;59:80â€“5.
Howard S, Omumbo J, Nevill C, Some E, Donnelly C, Snow R. Evidence for a mass community effect of insecticidetreated bednets on the incidence of malaria on the Kenyan coast. Trans R Soc Trop Med Hyg. 2000;94:357â€“60.
Hawley W, PhillipsHoward P, ter Kuile F, Terlouw D, Vulule J, Ombok M, et al. Communitywide effects of Permethrintreated bed nets on child mortality and malaria morbidiy in western Kenya. Am J Trop Med Hyg. 2003;68:121â€“7.
Hussey M, Hughes J. Design and analysis of stepped wedge cluster randomized trials. Contemp Clin Trials. 2007;28:182â€“91.
Mdege N, Man M, Taylor (nee Brown) C, Torgerson D. Systematic review of stepped wedge cluster randomized trials shows that design is particularly used to evaluate interventions during routine implementation. J Clin Epidemiol. 2011;64:936â€“48.
Hiscox A, Homan T, Mweresa C, Maire N, Di Pasquale A, Masiga D, et al. Mass mosquito trapping for malaria control in western Kenya: study protocol for a stepped wedge clusterrandomised trial. Trials. 2016;17:356.
Silkey M, Homan T, Maire N, Hiscox A, Mukabana R, Takken W, et al. Design of trials for interrupting the transmission of endemic pathogens. Trials. 2016;17:278.
Homan T, Hiscox A, Mweresa C, Masiga D, Mukabana W, Oria P, et al. The effect of mass mosquito trapping on malaria transmission and disease burden (SolarMal): a steppedwedge clusterrandomised trial. Lancet. 2016;388:1193â€“201.
Tiono A, Pinder M, Nâ€™Fale S, Faragher B, Smith T, Silkey M, et al. The AvecNet trial to assess whether addition of Pyriproxyfen, an insect juvenile hormone mimic, to longlasting insecticidal mosquito nets provides additional protection against clinical malaria over current best practice in an area with pyrethroidresistant vectors in rural Burkina Faso: study protocol for a randomised controlled trial. Trials. 2015;16:113.
Tiono A, OuÃ©draogo A, Ouattara D, Bougouma E, Coulibaly S, Diarra A, et al. Efficacy of Olyset Duo, a bednet containing pyriproxyfen and permethrin, versus a permethrinonly net against clinical malaria in an area with highly pyrethroidresistant vectors in rural Burkina Faso: a clusterrandomised controlled trial. Lancet. 2018;392:569â€“80.
Hiscox A, Otieno B, Kibet A, Mweresa C, Omusula P, Geier M, et al. Development and optimization of the Suna trap as a tool for mosquito monitoring and control. Malar J. 2014;13:257.
Plummer M. rjags: Bayesian Graphical Models using MCMC. R Foundation for Statistical Computing. 2019;R package version 4â€“10.
R Core Team. R: a language and environment for statistical computing. Vienna: R Foundation for Statistical Computing; 2018.
Malinga J, Maia M, Moore S, Ross A. Can trials of spatial repellents be used to estimate mosquito movement? Parasit Vectors. 2019;12:421.
Malinga J, Mogeni P, Omedo I, Rockett K, Hubbart C, Jeffreys A, et al. Investigating the drivers of the spatiotemporal patterns of genetic differences between Plasmodium falciparum malaria infections in Kilifi County, Kenya. Sci Rep. 2019;9:19018.
Lutambi A, Penny M, Smith T, Chitnis N. Mathematical modelling of mosquito dispersal in a heterogeneous environment. Math Biosci. 2013;241:198â€“216.
Kotz D, Spigt M, Arts I, Crutzen R, Viechtbauer W. The stepped wedge design does not inherently have more power than a cluster randomized controlled trial. J Clin Epidemiol. 2013;66:1059â€“60.
Mdege N, Man M, Taylor (nee Brown) C, Torgerson D. There are some circumstances where the steppedwedge cluster randomized trial is preferable to the alternative: no randomized trial at all. Response to the commentary by Kotz and colleagues. J Clin Epidemiol. 2012;65:1253â€“4.
Service M. Mosquito (Diptera: Culicidae) dispersalthe long and short of it. J Med Entomol. 1997;34:579â€“88.
Hazelton M. Kernel smoothing methods. In: Lawson A, Banerjee S, Haining R, Ugarte M, editors. Handbook of spatial epidemiology. Boca Raton: Chapman & Hall/CRC; 2016. p. 195â€“207.
Kelsall J, Diggle P. Kernel estimation of relative risk. Bernoulli. 1995;1:3â€“16.
Waller L. Point process models and methods in spatial epidemiology. In: Gelfand A, Diggle P, Fuentes M, Guttorp P, editors. Handbook of spatial statistics. Boca Raton: Chapman & Hall/CRC; 2010. p. 403â€“23.
Acknowledgements
We thank CK Mweresa, WR Mukabana and T Homan for logistical assistance during the SolarMal trial. Furthermore, we thank the field teams who performed data collection in the AvecNet trial.
Funding
This work was supported by the Swiss National Science Foundation (under grant number 310030_162837 to LM and FV); the COmON Foundation (through the University Fund Wageningen, to SolarMal); and the EU Seventh Framework Programme (FP7/20072013; number 265660, to AvecNet).
Author information
Authors and Affiliations
Contributions
LM carried out the analysis, LM and TS wrote the manuscript. All authors provided critical input on the manuscript. All authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Ethics approval and consent to participate
Not applicable.
Consent for publication
Not applicable.
Competing interests
The authors declare that they have no competing interests.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
Estimation of effectiveness as a function of intervention coverage
Effective intervention coverage
Assume that the N households in the study are ordered such that the first \(N_1\) receive the intervention. This means that \((x_m,y_m)\), \(m = 1, \dots , N_1\), denote all the coordinates of the \(N_1\) intervened households and \((x_m,y_m)\), \(m = N_1 + 1, \dots , N\), the coordinates of the control households. For any household with coordinates (x,Â y) the effective intervention coverage \(\mathcal {R}(x,y)\) is then defined as:
where K(x,Â y) is an appropriate function.
If K(x,Â y) is chosen as 1 whenever \(x^2 + y^2 \leqslant \hat{\theta }\), the expression \(\mathcal {R}(x,y)\) compares the number of intervened households to the total number of households within the contamination range \(\hat{\theta }\) and hence describes the percentage of households within an estimated contamination range \(\hat{\theta }\) that receive the intervention, as it has been previously definedÂ [16, 20]. This discrete measurement is imprecise, since it does not consider the closeness between households within the contamination range. Instead, let K(x,Â y) be a radially symmetric probability density function, for instance a twodimensional Gaussian kernel. The expression \(\mathcal {R}(x,y)\) then represents the common density of the intervened households relative to the general density of households and can be interpreted as a spatial relative risk function, as used in kernel density estimationÂ [33,34,35].
Choice of bandwidth
The bandwidth \(\varepsilon\) of the twodimensional Gaussian kernel
is chosen such that \(95\%\) of its distribution lies within a radius of the contamination range \(\hat{\theta }\) around each household. Taking
results in a bivariate normal distribution with \(95\%\) of the distribution laying within the contamination range \(\hat{\theta }\). This is because the contamination range \(\hat{\theta }\) was calculated from \(\beta _3\), the growth rate of a sigmoid function. This sigmoid function is the cumulative density function of the onedimensional logistic distribution with variance \(\pi ^2/(3\beta _3^2)\). On the other side, a covariance matrix of a twodimensional Gaussian kernel gives rise to a variance in one direction quantified by \(2\varepsilon ^2\)Â [9]. The effective intervention coverage of the ith household in the jth cluster in the kth survey round with coordinates \((x_{ijk},y_{ijk})\) can then be calculated as \(\mathcal {R}_{ijk} = \mathcal {R}(x_{ijk},y_{ijk})\) and is defined analogously for CRTs, with less indexing.
Approximation of the effective intervention coverage
The effective intervention coverage \(\mathcal {R}_{ijk}\) can be approximated by
with the same \(\beta _3\) as was fitted in the Bayesian hierarchical model
This relationship between the nearest discordant household \(\Delta _{ijk}\) and \(\mathcal {R}_{ijk}\) is illustrated in Fig.Â 5 for the SolarMal trial. The percentage error
between the effective intervention coverage \(\mathcal {R}_{ijk}\) and its approximation \(\hat{\mathcal {R}}_{ijk}\) was estimated to be \(6\%\) for the SolarMal trial and \(3\%\) for the AvecNet trial. These small errors justify this approximation.
Estimation of the intervention effectiveness at the effective coverage
Plugging in the inverse relationship between \(\Delta _{ijk}\) and \(\mathcal {R}_{ijk} \approx \hat{\mathcal {R}}_{ijk}\) in Eq.Â 2 it holds that
and the trial outcome can be described as a function of the effective intervention coverage.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
About this article
Cite this article
Multerer, L., Vanobberghen, F., Glass, T.R. et al. Estimating intervention effectiveness in trials of malaria interventions with contamination. Malar J 20, 413 (2021). https://doi.org/10.1186/s12936021039247
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s12936021039247