Dar es Salaam is situated at the shores of the Indian Ocean coast with a hot and humid climate which is ideal for mosquito proliferation and malaria transmission, satisfying the climatic requirement for stable transmission of temperatures between 22°C and 32°C and a rainfall of around 80 mm per month for at least five months per year 83. There are typically two rainy seasons: a main rainy season from March to June and a shorter, more erratic rainy season from October to December. Dar es Salaam has around 2.5 million inhabitants and covers a total area of 1,400 km² 84. The city is divided into three municipalities; Temeke, Ilala and Kinondoni which collectively comprise 73 wards. Each ward is further subdivided into neighbourhoods known as mitaa (singular mtaa) which typically comprise between 20 and 100 mashina (singular shina) or Ten Cell Units (TCU). The TCU is the smallest subunit of local government in Tanzania which, in principle, comprises a cluster of 10 houses with an elected representative known as a mjumbe although in practice most TCUs include 20–30 houses and some may even exceed 100. This study was based within the project area of the ongoing Urban Malaria Control Programme (UMCP) implemented by the Dar es Salaam City Council 5281. The main project area includes five wards from each municipality with a total of 67 mitaa. Overall, this study area covers an area of 55 km² with a total population of 609,514 people 84. The houses surveyed here were located in five wards, eight mitaa (Figure 1).

For comparison, the results obtained in Dar es Salaam are contrasted with those obtained with similar methodology in the Kilombero Valley, a rural setting with intense perennial malaria transmission in southern Tanzania 85.

For the purposes of routine monitoring and programme management, the UMCP surveys mosquito biting densities at 268 locations (four in each mtaa), distributed across the study area every four weeks. Initial trials proved that existing trapping technologies were not sufficiently sensitive to monitor the low densities of An. gambiae which occur across the study area. Therefore, outdoor human landing catch (HLC) 86 has been implemented as the standard sampling tool for adult mosquitoes as an interim measure until a suitable alternative is proven practical, effective and affordable. Once every four weeks at each location, HLC is conducted from 6 pm to 6 am for 45 minutes of each hour, allowing 15 minute breaks for rest, hot drinks and snacks. All collected mosquitoes are identified morphologically to genus and, in the case of Anopheles to species complex level 8788. Members of the Anopheles gambiae species complex are further resolved to sibling species level by polymerase chain reaction (PCR) 89. The sporozoite infection status of each mosquito was determined by enzyme-linked immunoabsorbent assay as previously described 90.

HLCs between April and December 2005 were used to identify the primary vectors of malaria in Dar es Salaam and to test for variation by location in the distribution of An. gambiae biting activity across the night. Members of the An. gambiae species complex were identified as the major malaria vectors in Dar es Salaam (See Results) so only these species were considered in the following analysis and study design. The influence of location as a determinant of An. gambiae biting habits was tested by treating TCU unique ID for each sampled site as a fixed factor in a logistic model with the proportion of mosquitoes caught during typical sleeping hours of city residents (10 pm to 6 am; see results) as the outcome variable. This data set was also used to identify sites with the highest densities of An. gambiae s.l. for the detailed and intensive mosquito behavioural surveys described below.

The 12 TCU in Temeke municipality and 2 TCU in Ilala municipality, which had the highest An. gambiae s.l. densities in the UMCP surveillance system, were selected for further, more detailed, surveys of the behavioural patterns of mosquitoes and humans. Informed consent was obtained from 216 houses in order to conduct HLC both indoors and outdoors. In each house, HLC was conducted for one night from 6 pm to 7 am as described above except that catchers switched between indoor and outdoor stations every hour in order to preclude biases resulting from variations in individual attractiveness 919293. These human landing catch surveys took place during 10 weeks of the main rainy season between April and June 2006. In order to estimate the biting rate for a full hour, total catches per hour were divided by 0.75.

A brief interview was conducted with all household members present at the time of the interview. They were asked where they usually eat dinner, where they stay after dinner before going to bed, what time they go to bed and what time they typically get out of bed in the morning. Furthermore, they were asked which preventive measures, such as bednets or insecticides, they use to avoid mosquito bites. The quality of their houses, i.e. the quality of screening and availability of ceiling boards was examined in each household. In order to verify the sleeping and resting behaviours reported by residents during interviews, also surveys were conducted based on direct observation by walking through these TCUs once every hour of the night and counting the number of people seen outdoors. Direct observation surveys were conducted for three nights in each TCU. Once validated by direct observation (see results), the questionnaire reports were used to estimate proportion of the inhabitants in each of the three behavioural compartments (outdoor, indoor awake, indoor asleep) at each hour of the night.

Data from the human and mosquito behavioural surveys described above were integrated to evaluate the interaction between them using an extension of a recently developed mathematical model 85. EIR is the product of the biting rate experienced by humans exposed to a vector population and the sporozoite infection prevalence of that mosquito population 94. The latter is only reduced by community-level impacts of malaria interventions 9596 so here personal protection purely in terms of biting rates and the impact that protective measures such as ITNs have upon them were estimated. First B_u,t, the mean biting rate experienced by an unprotected individual at each time of the night (t), based on the proportion of time spent outdoors multiplied by the outdoor biting rate at that time (B_o,t) plus the proportion of that hour spent indoors multiplied by the indoor biting rate at that time (B_i,t) was calculated. The main difference between this model and the one of Killeen et al. is that, because of the available information from the questionnaires, there was the possibility to divide the indoor compartment into being indoor but not asleep (and therefore not under a bednet) and being indoor and asleep (and, therefore, protected if using a bednet). The proportion of people sleeping or trying to sleep in bed and indoors (S_t) is not the same as the proportion of people staying indoors asleep or not asleep (I_t). If people are unprotected because they do not have a bednet, it only matters if they are indoors or outdoors and thus they experience the following biting rate:

B_u,t = B_o,t (1-I_t) + B_i,t I_t

The number of bites experienced per night, or nightly biting rate, for an unprotected non-user (B_u) can thus be calculated by summing the relevant biting rates for each hour:

B u = ∑ t = 1 24 B u , t MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqqGcbGqdaWgaaWcbaGaeeyDauhabeaakiabg2da9maaqahabaGaeeOqai0aaSbaaSqaaiabbwha1jabcYcaSiabbsha0bqabaaabaGaeeiDaqNaeyypa0JaeGymaedabaGaeGOmaiJaeGinaqdaniabggHiLdaaaa@3CDF@

Note that an unprotected individual is defined as someone lacking any net whereas a protected individual is defined as someone regularly using an effectively insecticidal net. The nightly biting rate of a protected individual (B_p) based on the combined nightly profiles of mosquito biting rate (B_u,t) over time (t), the protective efficacy of ITNs (P), which is assumed to be constant, and the behaviour of humans which results in fluctuating adherence of ITN users over the course of the night was modelled. As here a more detailed behavioural survey was taken into account, the nightly biting rate of a protected individual is calculated by multiplying the proportion of time spend outdoors at a certain time of the night by the outdoor biting rate at that time (B_o,t) plus the proportion of that hour being indoors but not asleep (I_t - S_t) multiplied by the indoor biting rate during that hour (B_i,t) plus the proportion of that time spent indoors being asleep under an ITN multiplied by the indoor biting rate at that hour (B_i,t) times the proportion of bites which can not be prevented by an ITN (1-P), as measured in experimental hut trials 449798. The effective adherence to ITN use at a given time of the night was assumed to be equivalent to the proportion of people sleeping at that time (S_t). This assumption allows us to express the overall effect of this interaction as follows:

B p = ∑ t = 1 24 B p , t = ∑ t = 1 24 [B o,t  (1 − I t ) + B i,t  (I t − S t  ) + B i,t S t  (1 − P)] MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@7163@

Based on existing evidence from experimental hut trials 499798, a conservative minimum protective efficacy level of 80% for ITNs (P = 0.8), equivalent to a relative exposure to bites of 20% when, and only when, actually sleeping under the net, was assumed. In this study, it was possible to take into account the proportion of people staying indoors or outdoors during waking hours and experiencing the corresponding biting rate. Furthermore, there was the possibility even to do the same for people living in different house quality who spent different amount of time in different compartments. During sleeping hours, people staying indoors were presumed sleeping under an ITN if available, whereas people sleeping outdoors were presumed not using a net and being fully exposed to the outdoor biting rate.

Taking the data for nightly human and mosquito behaviour profiles, the relative biting rate for ITN users which is equivalent to relative availability of protected individuals (λ_p) as previously defined (See equations 8 and 14 in reference 95), could be estimated. λ_p was calculated by comparing the total biting rate that protected individuals are exposed to (B_p) with that of non-users (B_u) who are unprotected:

λ_p = B_p/B_u

The true protective efficacy of an ITN (P*) against transmission exposure is then calculated as the overall nightly reduction of biting rate:

P* = 1 - λ_p

This estimate of protective efficacy differs from that previously reported from experimental hut trials as well as previous applications of this approach 85, because it allows for typical shortcomings in adherence resulting from the time people typically spend outside of their ITN indoor, as well as outdoors and even considering people staying or sleeping the whole night outdoors. Note, however, that this estimate is merely a comparison between the biting rates experienced by those who use an ITN and those who do not. It does not include the community-level protection of both groups when ITNs reach sufficient levels of coverage to reduce vector biting densities and sporozoite prevalence over large areas 95.

Distinct and useful indicators with which to interpret the results of the above equations are the proportion of exposure which occur indoors and the proportion that occurs during sleeping hours. The proportion of bites that occur during the observed peak sleeping hours (π_s) for an unprotected individual can thus be calculated as the nightly biting rate experienced during these hours divided by the total nightly biting rate:

π s = ∑ t = 10 pm 6 am B u , t / ∑ t = 1 24 B u , t MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqaHapaCdaWgaaWcbaGaee4Camhabeaakiabg2da9maaqahabaGaeeOqai0aaSbaaSqaaiabbwha1jabcYcaSiabbsha0bqabaGccqGGVaWldaaeWbqaaiabbkeacnaaBaaaleaacqqG1bqDcqGGSaalcqqG0baDaeqaaaqaaiabbsha0jabg2da9iabigdaXaqaaiabikdaYiabisda0aqdcqGHris5aaWcbaGaeeiDaqNaeyypa0JaeGymaeJaeGimaaJaeeiCaaNaeeyBa0gabaGaeGOnayJaeeyyaeMaeeyBa0ganiabggHiLdaaaa@5075@

Note that π_s describes the proportion of human exposure during which an ITN is in use and is used as a key parameter for modelling the community- and individual-level effects of ITNs upon malaria transmission 95. Overall, π_s was usually calculated using median reported values of 10 pm to 6 am for the whole study area but was evaluated separately for individual houses or houses with different quality of screening and ceiling boards for some analysis.

The proportion of bites occurring indoors but while awake and, therefore, not protected by a bednet (π_a) can be calculated as the estimated number of bites estimated to occur indoors while awake, divided by the total number of bites estimated to occur both indoors and outdoors:

π a = ∑ t = 1 24 [ B i , t ( I t − S t ) ] / ∑ t = 1 24 [ B o , t ( 1 − I t ) + B i , t I t ] MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@66BF@

The proportion of bites occurring indoors (π_i) for an unprotected individual can be calculated as the total number of bites estimated to occur indoors, divided by the total number of bites estimated to occur both indoors and outdoors. It should be noted that this equivalent to summing π_a and π_s:

π i = π a + π s = ∑ t = 1 24 [ B i , t I t ] / ∑ t = 1 24 [ B o , t ( 1 − I t ) + B i , t I t ] MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@6A02@

All activities of the UMCP, including these field surveys are approved by the Medical Research Coordination Committee of the National Institute for Medical Research, Ministry of Health, Government of Tanzania (Reference numbers NIMR/HQ/R.8a/Vol. IX/279 and 324). No persons in high risk groups, namely people under 18 years or women of reproductive age, were recruited to conduct human landing catches. Furthermore, the human landing catchers were screened every week for malaria microscopic examination of thick smear peripheral blood samples and treated with artemisinin-based combination therapy when diagnosis was positive.

In the areas in Dar es Salaam which were covered by the urban malaria control programme (UMCP) during the first three rounds of the household surveys, bed net usage was quite high and mosquito-proofed houses were common with many being made of concrete or bricks with a corrugated iron roof (Table 1). Around half of the houses had a complete ceiling board and/or good screening although a small proportion of residents didn't use any protection measures at all. The same was true in the TCUs which were selected for the more detailed study (Table 2). When compared to historical reports from Dar es Salaam, bednet usage had increased whereas the use of other protective measures had decreased 34. In contrast, in the Kilombero Valley in southern Tanzania, where ITNs have been promoted since 1997, bednet use is currently approximately at the same level, but both treatment of these nets and the use of other protective measures (coil, spray or repellent) are higher in Dar es Salaam (Killeen et al, Unpublished). Bed net usage in two contemporary Kenyan cities in 2001 was slightly lower and it should be noted that while screening of houses was less common than in Dar es Salaam, use of personal protection measures was more common 99.

A total of 1,388 An. gambiae s.l. (meaning members of the species complex as a whole in the absence of further identification to species by cytological or molecular methods) were caught in 1,650 catcher-nights, through routine monitoring activities of the UMCP during the preliminary survey of the entire study area (Figure 2). The majority of these proved to be An. gambiae (often referred to as An. gambiae sensu stricto): 75.6%, 21.3% and 3.1% of 1099 successfully amplified specimens proved to be An. gambiae s.s., An. arabiensis and Anopheles merus, respectively. During the same preliminary surveys, only 55 An. funestus were caught, indicating that although it is usually a very efficient vector 87, its contribution to transmission in urban Dar es Salaam is minor. Nevertheless, sporozoite infection and local transmission within urban Dar es Salaam was confirmed for An. gambiae s.s. (0.24%; 2/831) and An. funestus (2.32%, 1/43), but not An. arabiensis (0.0%, 0/234) and An. merus (0.0%, 0/34). Estimates of actual transmission intensity and its spatio-temporal heterogeneity over longer, more representative time periods will be reported in detail elsewhere. The only other Anopheles species caught was Anopheles coustani (370), of which none were found to be sporozoite-infected, so it is thought to contribute little or no vectorial capacity as described elsewhere 87.

Anopheles gambiae s.s. was by far the most important vector in the study area so all subsequent analysis focus upon this species and, to a lesser extent, An. arabiensis. Based on preliminary surveys of the total study area, location had no influence upon the proportion of An. gambiae s.l. bites which occurred between 10 pm and 6 am when residents of Dar es Salaam typically slept (An. gambiae s.s.: P = 0.519 by logistic regression, N = 72 locations, n = 714 mosquitoes, An. arabiensis: P = 0.398 by logistic regression, N = 32 locations, n = 133 mosquitoes). The great majority of the combined bites of these species occurred during sleeping hours (π_s = 83.16%; equation 6). Subsequent detailed surveys of mosquito and human behaviours therefore focussed upon the 14 TCUs with the highest An. gambiae densities observed during the preliminary site-wide surveys (Figure 1).

A total of 2,153 people were living in these 216 houses at the time of survey, of whom approximately half were under the age of 22 (Table 3). All the TCU were either near a swamp or close to a depression with poorly functioning drains and most of these areas were partially flooded during the rains. Although these were mostly poorer, unplanned areas, half of the houses had intact screening or screening with small holes. Almost three quarters of these houses did not have a ceiling board and it was typically observed that the eaves of most houses in Dar es Salaam were accessible to mosquitoes. Although more than three quarters of residents slept under a net, only a third of these nets had ever been treated with insecticide. Very few residents reported using alternative protective measures such as repellents, mosquito coils or insecticidal sprays (Table 3).

The reported and observed behaviours of humans were largely consistent (Figure 3A). The minor discrepancies can be explained as follows. Less people were observed than reported outdoors in the evenings and mornings, because it was not possible for us to enter all courtyards and some individuals may be elsewhere during these hours. More people were observed than reported to be outdoors towards midnight but, based on direct experience, this was attributed to the transition of people through the TCU who do not live there. The residents reported that shortly after 10 pm, 50% of the people had gone to bed and at around 6 am 50% of the people were still asleep. A small, but noteworthy, proportion of residents slept outdoors all night (Table 3), often citing heat and poor ventilation inside the house as their primary motivation.

During the intensive entomological study in the selected sites with high An. gambiae densities, 432 catcher-nights yielded 2,484 An. gambiae s.l., 63 An. funestus, 370 An. coustani, 41,290 Culex, 70 Aedes and 97 Mansonia. Of the 2,027 An. gambiae s.l. which were successfully amplified, 83.9%, 15.9% and 0.2% were identified as An. gambiae s.s., An. arabiensis and An. merus, respectively. Only 0.41% (7/1700) of An. gambiae s.s. and 0.31% (1/322) and An. arabiensis were found to be infected with sporozoites. An. gambiae s.s., An arabiensis, An. funestus, An. coustani and Mansonia were all exophagic, meaning that they mainly bite outdoors 100 as evidenced by the proportion of mosquitoes caught outside being significantly greater than half (Figures 3 and 4). Anopheles gambiae s.l. is generally endophagic in rural Tanzania 3687101 and the proportion of An. gambiae s.l. caught outdoors was higher in Dar es Salaam than in Kilombero valley (Figure 4; 63 versus 34%, respectively; Χ² = 597.1, P < 0.001), considering only catches up to 6 am because the studies in Kilombero valley stopped at this time. In Dar es Salaam, the proportion of An. arabiensis caught outdoors was significantly higher than the proportion of An. gambiae s.s. caught outdoors (Χ² = 23.4, P-value < 0.001). Culex sp. and Aedes sp. exhibited neither exo- nor endophagic tendencies in Dar es Salaam.

Hourly biting pattern almost exactly followed classically reported patterns of An. gambiae s.l. 87 with an increase of Anopheles gambiae s.s. densities towards midnight, and a second peak around 4 – 5 am, followed by a decline towards dawn (Figure 3B). In fact, the proportions of An. gambiae s.s mosquitoes caught during peak sleeping hours was greater in the city than in the rural area (Χ² = 112.9, P < 0.001) with peak sleeping hours in Kilombero valley from 9 pm to 5 am and in Dar es Salaam from 10 pm to 6 am. As summarized in Figure 4, biting activity was more intense outdoors than indoors throughout the night and was highest during sleeping hours (Figure 3B). An. gambiae s.s. constituted 84% of An. gambiae s.l. and therefore dominates the shape of the curve for the pooled sibling species (Figure 3D). Nevertheless, it is noteworthy that An. arabiensis had its peak biting time at 10 pm, when more than three quarters of the residents were still awake, and then slowly declined towards the morning (Figure 3C).

Combining the human and mosquito behavioural surveys, and using the model described in the methods section, allowed estimation of the biting rates experienced by residents at each hour of the night (Figure 5). This approach also allowed dissection of these mosquito-human interactions into distinct domestic compartments (Figure 6) where specific interventions may or may not reduce exposure. For example, ITNs are expected only to provide personal protection while sleeping so their protective efficacy is limited to those times of the night when users sleep and cannot exceed the proportion of exposure which would otherwise occur while asleep (π_s; equation 6). In contrast, interventions which prevent house entry, such as mosquito proofing 6061 or spatial repellents such as DDT 100, could prevent any indoor exposure regardless of whether occupants are awake or in bed (π_i; equation 8). It should be noted that the simpler form of this approach applied previously 85 did not allow estimation of exposure indoors while awake so it is not possible to compare Dar es Salaam with this rural precedent in terms of the relative contributions of exposure indoors and outdoors while awake. Nevertheless, it is possible to compare the proportion of exposure which an ITN might be expected to prevent (π_s; equation 6).