These results showed that individuals targeted by the larviciding intervention in Dar es Salaam were significantly less likely to have used a bed net the night before the survey. The magnitude of this effect increased with time such that, three years after the initiation of larviciding activities, individuals in intervention areas were 10% less likely to use their bed net as compared to individuals living in non-intervention areas. There was also a decline in household heads’ knowledge of malaria symptoms and this effect was more pronounced for individuals of low SES. No differences between larviciding and non-intervention areas, with respect to knowledge of malaria transmission, were found.
With regard to bed nets, several studies have suggested that their use is a function of night-time temperature, perceived malaria risk and density of nuisance biting insects [30–32]. Thus, the significant reduction in the probability of using a bed net in UMCP intervention areas could result from two factors. First, the UMCP made a programmatic decision to control larval stages of nuisance biting insects such as Culex quinquefasciatus (a mosquito involved in the transmission of lymphatic filariasis, but not malaria), as an effort to gain community support. A significant reduction in nuisance biting rates could deter individuals from using bed nets if personal protection against mosquito bites is not perceived as being necessary anymore. Nevertheless, data from the first phase of the UMCP intervention suggest that routine larviciding was not successful in suppressing nuisance biting, and culicine mosquitoes were still responsible for more than 100 bites per exposed person per night in the intervention wards . The impact of controlling nuisance biting insects will be context specific, however, depending on the relative abundance of different species of mosquitoes. Second, the reduction in the prevalence of malaria infection from 20.8% in 2004 to 1.7% in 2008 following larval control [21, 26] can potentially change the individual perception of malaria risk. In this case, the disease may not be perceived as a threat to health anymore, leading to varied behaviour changes, including reduced adoption of personal protective measures, such as bed net use. The reported results tend to support this hypothesis.
Despite the significant reduction in the probability of using a bed net following the larviciding intervention, the proportion of individuals using a net in non-intervention areas increased throughout the study period. In October 2004, shortly after the initiation of this study, the Tanzania National Voucher Scheme was launched. The aim of this programme was to provide every pregnant woman with a printed voucher valued at TZS2,750 (USD2.75 in 2004) to purchase a discounted-price bed net . In October 2006, a second voucher was introduced targeting mothers and caretakers of infants aged nine months at the time of measles vaccination  and, in January 2007, the value of the voucher was increased to TZS3,250 . The subsequent introduction and improvements of these financial incentives could thus have resulted in higher bed net ownership and usage.
A decline in the knowledge of malaria symptoms, particularly in areas under the UMCP larval control intervention, is also worrisome. Caregivers’ inability to recognize malaria symptoms has been cited as an impeding factor for early treatment of severe malaria in Tanzania . With lower transmission intensities, population-level immunity is expected to decrease and the clinical spectrum of severe malaria may change with cerebral malaria accounting for a higher proportion of cases . Therefore, early and proper recognition of symptoms is crucial to reduce malaria morbidity and mortality . Of particular concern is the finding that SES is modifying the relationship between larviciding and knowledge of malaria symptoms. Given that out-of-pocket expenditure for malaria treatment usually consumes a larger proportion of low SES households’ budget , inappropriate or delayed treatment could potentially be exacerbated in these disadvantaged households by their inability to recognize malaria symptoms.
If knowledge is formed based on experience, one could hypothesize that as malaria transmission goes down, and fewer cases are observed, personal experience with malaria episodes also reduces, and thus the ability of individuals to properly identify disease symptoms may be compromised. That would be maximized if malaria was not perceived as a major threat. While intuitively it is reasonable to assume that these changes would increase over time (assuming that transmission remains fairly low or declines even further), this study’s results do not support that. In addition, the available data do not allow assessing the mechanisms through which knowledge of malaria symptoms is changed.
Regarding knowledge that mosquitoes transmit malaria, there is no evidence of changes following the UMCP larval control. Two factors could explain this result. First, community sensitization and participation are a central component of an integrated vector management strategy as endorsed by the World Health Organization . In Dar es Salaam, each TCU has a leader and the UMCP worked closely with them to foster support for the larviciding activity, and to guarantee unrestricted access to breeding habitats, many located on private properties. Therefore, the population living in the UMCP area was aware of the presence and the purpose of larval control teams. Second, larval control personnel conducted their work wearing a UMCP T-shirt, displaying the name of the project and the life cycle of the mosquito. Thus, the weekly presence of the larval control teams may have acted as a regular reminder of the importance of mosquitoes for malaria transmission. These two factors could potentially overcome the expected decline in knowledge in scenarios of low malaria transmission.
The strengths of this study include its large geographic and temporal extents, availability of reliable baseline information, control of many potential confounders, reporting of effect size estimates on the risk ratio scale, a large sample size, and detailed use of robustness checks and sensitivity analyses. The study has some limitations. First, the order of the rollout of the intervention was not randomly allocated. If ward-level characteristics are correlated with the intervention, the reported effect size estimates could be biased. Nevertheless, including fixed effects at the ward level, which control for ward-level time-invariant confounders, did not affect the reported effect size estimates. Second, information on knowledge of malaria symptoms and transmission was only collected from household heads. Intra-household decisions about health expenditure and treatment-seeking behaviour follow a complex process that involves trade-offs and bargaining among household members. This paper’s inferences are thus based on the assumption that the household head’s level of malaria knowledge is representative of that of other household members involved in this decision making process. The fact that gender was not found to be an effect modifier tends to support this assumption.
This study’s findings need to be discussed in light of the current efforts of intensified malaria control with the goal of eradication. In countries considering elimination, and in areas where transmission has been reduced to very low levels for a few years, acquired immunity is low and thus sustaining gains of malaria control becomes crucial to prevent outbreaks and resurgence of the disease , such as that occurred in Sri Lanka during the late 1960s . If knowledge and behaviour change follows successful interventions that reduce malaria transmission to low levels, then sustainability of control efforts and gains may be at risk. A potential strategy to address these issues, currently largely neglected by national malaria control programmes, is the implementation of a comprehensive behavioural change communication process, which addresses gaps in knowledge and problems in disease risk perception.