The present study emphasizes the important role that host reservoirs with low-level gametocyte and asexual parasite density can play in sustaining malaria transmission despite LLIN-based malaria control. For any given control scenario, the model which includes a large pool of hosts with sub-microscopic gametocyte and asexual parasite densities predicts considerably higher transmission and therefore a need for greater coverage and adherence to LLIN usage. When LLIN coverage is set to levels at or around values currently achieved with the up-scaled efforts enabled by The Global Fund to Fight AIDS, Tuberculosis and Malaria (GFATM) 
, only M1 in a low transmission setting predicts malaria eradication. M2 does not predict interrupted transmission even in the low transmission setting (). For high transmission environments, the models presented here predict the need for much higher LLIN coverage and usage. Furthermore our predictions emphasize the importance of implementing sensitive diagnostic approaches that are used for monitoring and evaluating malaria control strategies and the progress of malaria control campaigns.
Although previous models acknowledge submicroscopic gametocytemia 
, our current approach is, to the best of our knowledge, the first to evaluate the impact of these infections on malaria transmission and control. In our models, LLIN is the only malaria control strategy. In real-life field situations, efficient malaria control utilizes LLIN as part of a range of co-ordinated strategies. The next phase of worldwide intensified malaria control aims to build on the success of LLIN distribution in 2008 by deploying rapid diagnostic tests and effective antimalarial therapies as widely as possible in order to reduce preventable deaths from malaria to near zero by 2015 
. It should be noted that efficient treatment of symptomatic malaria cases, while undoubtedly an urgent priority, will only partially reduce malaria transmission because asymptomatic carriers will continue to transmit the disease'.
Adult populations in holo- or hyperendemic regions have high levels of acquired immunity to the severe manifestations of the infection. Although parasitemic, these people do not usually feel ill and thus see no reason to visit a health center or seek treatment, but they are likely to harbor infectious gametocytes that will sustain malaria transmission regardless of a successful local reduction in symptomatic malaria. Asymptomatic malaria infection is normally characterized by very low asexual parasite densities and the absence of gametocytes, especially in adults 
. A recent review by Okell et al. 
revealed that standard LM failed to detect asexual malaria parasites in 49.2% of all cases and 91.3% of gametocytemic cases. Furthermore, Bonnet et al. estimate the contribution of hosts with undetectable gametocytemia to malaria transmission at 23.7% 
while Schneider et al. determined an even higher value at 54.8% 
. Sustained active case detection using rapid diagnostic tests and LM will, therefore, be inadequate because the sensitivity of these methods is too low. It is important to maximize detection of cases infectious to mosquitoes for a prolonged period, presumably decades, to achieve true malaria eradication 
If a fraction of malaria cases go undetected and the intensified efforts of malaria control are reduced prematurely, malaria will increase such as it did in India 
, Sri Lanka 
, Zanzibar 
and Madagascar 
after the halt of the eradication programs in the last century. Resurgent malaria can have devastating effects on populations with levels of immunity that have waned following partially effective control. For example, the incidence of malaria in the highlands of Madagascar fell dramatically during the control program in the 20th
century but more than 40,000 fatalities were attributed to malaria during the 5 years after the control program was stopped 
It will be very important to complement current methods of surveillance with enhanced diagnostic techniques to more accurately plan, monitor progress and guide malaria control which is aimed at eliminating malaria. Recently developed methods to detect gametocytes at low densities such as RT-PCR, QT-NASBA and HFGMF can reduce the threshold of gametocyte detection to cover a wider epidemiologically relevant range. Application of these methods and their potential refinement to maximize time and cost effectiveness appears a vital component of successful malaria elimination.
It is particularly important to detect gametocytes because they are more resilient to treatment. While there is evidence of a gametocytocidal effect of artemisinin drugs, gametocyte clearance times are still very long in ACT-treated patients with a mean of 49 h for asexual stages and 220 h (>9 days) for gametocytes in a study of 559 patients conducted in Thailand in from 1998 to 2006 
. Another recent study detected submicroscopic gametocytes up to 48 days after ACT treatment 
. Primaquine, the most potent gametocytocidal drug, can provoke hemolysis in glucose-6-phosphate dehydrogenase (G-6-PD) deficient patients and is thus not used routinely in the many tropical countries in which G-6-PD deficiency has been observed 
. This means that some patients can be infective for weeks following clearance of asexual parasites even after ACT treatment. ACT or other treatments were not included in the current analyses, in part because of factors such as drug-specific post-treatment gametocyte viability 
and a variable if brief period of carriage of gametocytes without asexual forms. We believe that the problem of submicroscopic gametocytemia sustaining malaria transmission is mostly relevant in asymptomatic malaria carriers. Since they will rarely seek treatment, ACT/other therapy will not have a great impact on this part of the population. The exception may be mass drug administration which is not likely to play a big role in malaria control efforts, especially with the fear of emerging artemisinin resistance 
The aim of the present study was to highlight the importance of sensitive gametocyte and asexual parasite detection. For this we have used data from two very different transmission settings to calibrate the models. Other calibration procedures that constrain the range of admissible parameters could also be used. The models presented here have several limitations. Ross-Macdonald methodology is a very simplistic approach, since the real-world dynamics of malaria infection depend on many more variables (such as the age distribution of the population, seasonality of transmission and protective host factors including red cell polymorphisms) that cannot be encompassed by such models. Nevertheless, our adaptation of the most commonly used form of mathematical models for malaria can be viewed as a necessary refinement in their further development and an important step in identifying practical strategies for parasite detection necessary to enhance control and elimination efforts.