Dar es Salaam is a hot, humid coastal city and experiences two rainy seasons: the short rains from mid-October to early-December followed by the long, more intense rains from March to June. Dar es Salaam is Tanzania’s biggest and most economically important city with an estimated population of 3.3 million in 2010, living within an administrative region of 1,400 km2[67, 68]. The city is divided into three municipalities, namely Kinondoni, Temeke and Ilala, and these municipalities are further divided into a total of 72 wards. The study site encompasses 31 administrative wards at the heart of the city, comprised of one set of 15 wards previously described as the UMCP study area
 and another 16 neighbouring wards, totalling approximately 2.65 million residents living in an area of 160 km2. Before the initiation of larviciding, the area experienced modest malaria transmission rate with an entomological inoculation rate (EIR) of approximately one infectious bite per person per year
[39, 51]. The main malaria vectors are members of the An. gambiae complex, which prefer to feed outdoors and may therefore be only moderately vulnerable to control with indoor-targeted insecticidal means such as ITNs
The Dar es Salaam UMCP
All UMCP activities are coordinated by the City Medical Office of Health, and fully integrated into the decentralized administrative system of Dar es Salaam
[32, 39]. The UMCP operates on all six administrative levels of the city: the city council, the three municipal councils it oversees, the 15 wards chosen from those municipalities, containing 67 neighbourhoods referred to as mitaa in Kiswahili (singular mtaa, meaning literally street), and more than 3000 housing clusters known as ten-cell-units (TCUs), each of which is subdivided into a set of plots corresponding largely to housing compounds
[39, 51, 56]. The main tasks of the three upper levels within UMCP are programme management and supervision, whereas actual mosquito larval surveillance and control is organized at ward level and implemented at the level of TCUs and their constituent plots. In principle, a TCU is a cluster of ten houses with an elected representative known as an mjumbe, but typically comprises between 20–100 houses in practice
. As a prerequisite for effective management of a larviciding programme, the UMCP implemented routine larval habitat surveillance between 2004 and 2008
[39, 53, 54]. From March 2006 to date, the UMCP implemented regular larviciding of all mosquito breeding habitats as a means to kill aquatic mosquito stages, prevent adult emergence and reduce malaria incidence and prevalence through a community-based but vertically managed delivery system
[32, 39, 52–54]. UMCP began systematic larviciding in three wards (one from each municipality) in April 2006
[51–54], following complete participatory mapping of the area
[55, 56] and CB baseline surveys of the breeding habitats. The programme subsequently scaled-up larvicide application to nine wards in May 2007. In March 2008 the programme was extended to all the 15 wards of the original study area. In this particular study, community-based adult mosquito surveys were set up across the original 15 UMCP wards plus an additional 16 adjacent wards from outside the study area to include non-UMCP wards chosen from the same three municipalities where there was no larviciding taking place. Overall, this 160 km2 area contained 31 wards, 85 mitaa, approximately 8,000 TCUs and approximately 2.65 million residents (Figure
Routine programmatic adult mosquito surveillance by community-based personnel
Based on a pilot-scale evaluation in 12 wards that used the B-design ITT
, a CB scheme for trapping adult mosquitoes using the C-design ITT
 was developed and implemented as a replacement for the previous system that relied on HLC
. ITT-C differs from the earlier ITT-B prototype, in that the netting panel lying between the entry funnels and the bait host is bisected into two compartments within the trap. This enables a person in the process of collecting mosquitoes to stand up within the trap while protected from mosquito bites. In addition, there are two long sealable cotton sleeves hanging from each trap chamber to enable operators to safely remove mosquitoes by using mouth aspirators while protected from bites. In contrast, the B design required the opening of the long zipper across the netting panel and aspirating from within the open trap chamber, thereby exposing the operator to mosquito bites
The entomological survey was initially set up across the previous 15 UMCP intervention wards, each of which comprised of a cluster of 20 sampling sites, making a total of 300 sentinel sites distributed across the UMCP study area that were routinely surveyed on monthly basis. This was primarily meant to serve as a tool for routine monitoring of progress of the larviciding programme activities by identifying areas with residual vector populations and, presumably, malaria transmission. Adult mosquito surveillance was therefore decentralized to ward level to coincide with management practice for concurrent community-based larval surveillance and larvicide application. The system adopted a decentralized sampling protocol
, that enabled unskilled community members, rather than trained entomologists sent from a centralized team, to capture, record and submit mosquito samples, without any night time supervision by the research team, and with only occasional contact with programme staff. This system was modified from that of the original pilot
 so that only one volunteer per ward was recruited, compared to one per neighbourhood or mtaa (3–7 per ward) in the pilot system, to conduct monthly surveys of 20 locations per ward rather than weekly surveys of four locations per neighbourhood (12–28 per ward).
Overall, thirty-one, volunteers including fifteen from the 15 original UMCP wards were recruited and remunerated at a rate of 3500 Tanzanian shillings (2010 US$ 2.70) per night of trapping. Each volunteer took responsibility for trapping mosquitoes for one night per month at each of the 20 locations within his or her assigned ward. They were allowed to choose, at their own discretion, which nights of the week (Monday to Friday) they would sleep in the traps, the sequence they would visit each of their 20 assigned locations, and what time they entered and left the traps, under the condition that they recorded these dates and times in standardized forms. This was considered necessary for promoting a sense of ownership and responsibility for the project, and making working conditions relaxed, conducive and flexible so that the modest remuneration remained sufficiently attractive to retain CORPs and minimize any incentive to fabricate data. Furthermore, there were no consequences to the CORPs for not trapping on a particular night so long as all the 20 sites were sampled at any week day of that particular month. The 20 sampling sites in each ward were deliberately chosen by the local leaders and the CORP, with the intention that they were well-distributed across the ward, close to obvious Anopheles larval habitats, and preferably within walled compounds so that safety of the sleeping volunteer was assured.
The volunteers were supplied with all the necessary materials including paper cups, air-tight containers, aspirators, petroleum ether and bicycles for transport. This allowed them to continuously trap, collect and store mosquitoes for a period of one week, recording their observations and trapping sequence daily on a form they were provided with. Samples were submitted each week to the central laboratory for further processing using the bicycles that each CORP was provided with to assist them in moving the trap between the sites within the ward. Each night the trap was erected outside of the designated house and the volunteer slept in it over night to act as a bait to attract human-feeding mosquitoes. Note that the user is completely protected by the fine netting trap chambers where the mosquitoes are trapped
. Mosquitoes were removed from the trap chambers using aspirators, transferred into paper cups, and then anesthetized with a small ball of cotton wool soaked in petroleum ether. Dead mosquitoes were then transferred into an air-tight (1.5 ml Eppendorf tubes, Nantong Shenhua Laboratory Apparatus Co., Ltd) container half-filled with silicagel for storage and preservation before submission to the central mosquito laboratory each week. To control for and minimize data fabrication by CORPs, standardized forms were supplied ( Additional file
1: Table S1) and they were obliged to record the approximate number of each relevant mosquito taxon caught, early each morning immediately after they finished collecting, and to document confirmation of his visit with the signature of the house owner where the trapping took place that particular night. At the laboratory, the samples were received by a technician who verified their content before formally recording their acceptance in good condition in a registry book.
This protocol for routine CB sampling with ITT-C across the original 15 UMCP wards, where larviciding had already been established as a routine activity, began in February 2009 whereas the 16 non-intervention wards outside this area started in October 2009. These additional wards were included as a preparatory step for scaling up city-wide vector surveillance and larviciding, as well as to enable subsequent evaluation of the protocol as applied at large scale across the full range of vector densities found in the city. Overall, this CB system for routine surveillance of mosquito biting intensities spanned over 620 designated sentinel sites (clusters of twenty in each of the 31 wards) of which 615 were actually sampled on a monthly basis in practice (Figure
Randomized quality assurance entomological surveys
To assess the quality of data collected by the decentralized, routine adult mosquito surveys described above, two quality assurance (QA) adult mosquito surveillance teams were recruited, each comprising five catchers earning slightly more than their counterparts in the routine CB system. The first team, earning 4000 TShs (2010: US$ 3.50 per person per night) was responsible for repeating adult mosquito collection using ITT at five locations scheduled one day after the routine CB mosquito surveillance team had applied the same trapping method in these same locations. The sampling framework for the sites involved randomly selecting five sites from the list of locations where the CB collectors had set their traps the previous night. Therefore, this team was responsible for repeating adult mosquito sampling at randomly chosen locations, over four days of the week (Tuesday to Friday), totalling 20 locations sampled for resurvey by the QA team each week. The second team, earning 8,000 Tanzanian Shillings (2010: US$6.15) per day, was responsible for repeating adult mosquito collections using HLC at the same randomly-selected locations used the previous nights for QA-ITT and the night before that for routine CB collections with ITT. This second team worked three days per week (Wednesday to Friday) at the same five randomly chosen locations as the first QA team, totalling 15 locations sampled per week. Outdoor HLC was conducted at each of these houses from 6 pm to 7 am for a period of 45 minutes every hour, allowing for 15 minutes break each hour, as previously described
[51, 62]. These two QA teams were vertically and regularly supervised, including random night time spot checks by the research team for quality control. The locations selected for QA follow up was not disclosed to either the QA teams nor to the supervising research staff until the day after the routine survey was set up, in the late evening of the day for the first QA surveys using ITT. This was necessary to avoid any possibility of collusion between CORPs in the routine and QA teams and thereby minimize risk of data fabrication. CORPs from the two QA teams were dropped by vehicle at their scheduled stations, accompanied by the field supervisor. The mosquitoes collected by the ITT-C and HLC QA teams were collected by vehicle and taken to the central laboratory the following morning when the catchers had finished their collections.
Laboratory processing and data reporting
In the laboratory, all mosquitoes were identified morphologically using taxonomic keys
 as males or females, and as An. gambiae s.l., Anopheles funestus, Anopheles ziemanni, Culex species, or Aedes species. Abdominal status was scored as gravid/semi-gravid, fed or unfed for all the Anopheles and for Culicines. All Anopheles caught were subsequently desiccated over silica gel and kept at room temperature until they were further processed. These classification and count data were first recorded on standardized paper forms ( Additional file
1: Table S1) and then reported using mobile phones with specifically designed menus and made available to stakeholders and project staff at the following
 This web site was also loaded with automatically generated (pre-coded R script) weekly synthesis report for the UMCP management staff and other stakeholders to review at will. A wing or a leg of every An. gambiae s.l. mosquito caught was analyzed by PCR to identify its exact species within the An. gambiae complex
. An enzyme-linked immunosorbent assay (ELISA) using a monoclonal antibody that recognizes a repetitive epitope on the circumsporozoite-protein of P. falciparum was used to establish malaria sporozoite infection status in each individual An. gambiae s.l. specimen
Cross sectional epidemiological survey
All the 620 sites used for the routine entomological surveillance were mapped to the TCU level
[55, 56] and the households within each were carefully listed. Three teams of four people, comprised of a supervisor, community-based health nurse and two interviewers conducted the cross-sectional household surveys (March to August 2010) in all households of the house or housing compound (median = 4 households) which routine CB mosquito surveillance was conducted. All people occupying the household were included in the survey, excluding children who were three months old or less. Systematic screening of all the inhabitants of each selected household who were present at the time of the survey, and consented to participate, was carried out to determine their malaria infection status. Parasitological examination was carried out by the community-based health nurses by finger prick with a sterile lancet. A small amount (5 μl) of blood was drawn from consenting residents using micro pipettes and placed on MAL-Pf® (ICT Diagnostics, Cape Town, Southa Africa) malaria rapid diagnostic test kits (RDTs) using histidine rich protein-2 as the test antigen (HRP-2). Such HRP-2 RDTs, including this specific kit, have increasingly been proven sensitive, reliable and accurate for routine malaria diagnosis in the field
[74–77]. While this specific test kit is prone to a phenomenon called prozone that results in weak responses to very high density parasitemias, no false negatives were documented in a recent evaluation of this and other comparable HRP-2 based products
. Questionnaire responses and RDT results were recorded electronically in the field using Socket SoMo 650 Series (Socket Mobile, Inc) portable digital assistants programmed in Visual CE.
All the data were entered in coded numeric form, cleaned, restructured and analyzed using SPSS® 18.0 except where described otherwise.
The mean relative sensitivity of the three surveillance methods was estimated by fitting a generalized linear model (GLM) with a negative binomial distribution to the mosquito catch for each recorded trap night, treating surveillance method as a categorical independent variable with location as the subject and date as a within-subject source of variation with first order autocorrelation. Correlation between the mean catch (transformed as logarithm (y + 1)) at each location obtained with the three alternative vector surveillance methods were tested pair-wise using Pearson’s linear correlation test. Associations between the relative sensitivities of CB trapping with ITT and mosquito densities measured by the two QA survey methods were tested for using binary logistic regression
. Specifically, GLMs were fitted to the proportion of all mosquitoes caught by the CB-ITT in a given location and week where all methods were applied.
The catches of female An. gambiae or An. funestus and Culex spp were aggregated by survey method, yielding mean catches for each method per trap night per location. On several occasions, all the three survey methods recoded zero values even after aggregation so an artificial incremental scatter was added to generate the none-zeros and allow separation and visualization of otherwise identical data points. Since divisions by zero gives infinite values, data for each location thus included the sum of several observations of the catches for the specific survey method. In order to establish the density dependence of sampling sensitivity of ITT through either CB or QA methods, the mean catches of the collections by alternative survey methods (CB-ITT and QA-ITT) was divided by the sum of the QA (QA-ITT + QA-HLC) collections, and this denominator was treated as the continuous independent variable in a generalized linear model.
To allow direct comparison of the three surveys in terms of cost-effectiveness only the direct and non-direct expenditures incurred by each system, during the period when all three systems were operating in parallel are considered. These included monthly personnel costs (salaries and volunteer allowances) for each team, supplies and transport costs. Transport costs comprised of the upfront costs for buying a bicycle or a vehicle (for both the CB and QA-surveys, respectively) plus the three years or ten years-depreciated costs (for the bicycles and vehicle, respectively) and their respective monthly-recurrent (service and maintenance) costs. All cost estimates are presented in Tanzanian shillings as recorded at the time they were incurred and then converted into 2010 US$ at a rate of 1408.02 shillings per dollar.
To qualitatively examine differences in age-prevalence profiles associated with malaria transmission hot spots, infection prevalence data from household surveys were initially stratified based on either the presence or absence of any detectable primary vectors (any An. gambiae s.l. or An. funestus caught) by a given survey method. Subsequently, this approach was refined to stratify on the basis of being amongst the 5% highest mean catches of primary vectors. In all cases, differences between the two strata for each vector surveillance method, in terms of the distribution of infection probability among the following age classes, was tested by χ
2 analysis using Microsoft Excel®: less than 5 years, 5 to 19 years and 20 years or more.
Explanatory logistic regression models (GLMM) of malaria infection prevalence were fitted and selected in a forward stepwise manner using R version 2.12.2. The association of malaria prevalence with the following independent variables was assessed: mean catch at a given location with each individual entomological survey type, LLIN use, presence of eaves, presence of ceiling, presence of window screening (good indicators of socioeconomic status), larviciding activity, use of insecticide consumer products, travel in the previous month or residence elsewhere, sex and living with both parents. To adjust for spatial and temporal heterogeneities TCU location identity and date were incorporated into all models as random effects. Only variables exhibiting evidence of association with malaria infection risk (P ≤ 0.05) when tested as a single categorical independent variable was retained in the model
[80, 81]. The variables with the lowest P-value obtained in the exploratory analysis were included first. Based on qualitative examination of age-prevalence relationships in this dataset (see results), this logistic regression analysis was applied only to children and teenagers (<19 years) because the relationship between their exposure and infection prevalence appeared to be higher and to increase with age in areas with higher vector density.
Ethical consideration and informed consent
The study received ethical clearance from the Medical Research Coordination Committee of the Tanzanian National Institute of Medical Research (Reference numbers NIMR/HQ/R.8a/Vol.IX/279 and 324). Informed consent was obtained from all the participants, including the mosquito catchers and the house owners where the sampling took place, as well as the participants in the household surveys. All the volunteers recruited for conducting HLC were provided with prophylactic treatment with atovaquone-proguanil (Malarone®) free-of-charge, which they were obliged to take once a day to prevent malaria infection. In order to deal with the possibility of poor compliance or drug failure, participants in mosquitoes-trapping surveys who developed any symptoms such as fever, chills, headache or nausea, were tested for malaria parasites and would have been offered free treatment if found to be infected but this eventuality never occurred during the study. All participants in either the household surveys found to be infected with malaria were offered supervised treatment with artemether-lumefantrine (Coartem®; Novartis Pharma AG, Basel, Switzerland) prescribed by a clinical officer and provided by the community health nurse, following national treatment policies and guidelines, as soon as the RDT test was complete. However, if the participant refused this offer of treatment, they were referred to a nearby health facility and given all required transport and other logistical assistance to attend. Women of child-bearing age found to be infected with malaria were offered treatment with artemether-lumefantrine unless they were known or suspected to be pregnant and in their first trimester, in which case were instead treated with oral quinine as per national guidelines.