A novel model fitted to multiple life stages of malaria for assessing efficacy of transmissionblocking interventions
 Ellie SherrardSmith^{1}Email authorView ORCID ID profile,
 Thomas S. Churcher^{1},
 Leanna M. Upton^{2},
 Katarzyna A. Sala^{2},
 Sara E. Zakutansky^{2},
 Hannah C. Slater^{1},
 Andrew M. Blagborough^{2} and
 Michael Betancourt^{3}
DOI: 10.1186/s1293601717823
© The Author(s) 2017
Received: 23 January 2017
Accepted: 17 March 2017
Published: 4 April 2017
Abstract
Background
Transmissionblocking interventions (TBIs) aim to eliminate malaria by reducing transmission of the parasite between the host and the invertebrate vector. TBIs include transmissionblocking drugs and vaccines that, when given to humans, are taken up by mosquitoes and inhibit parasitic development within the vector. Accurate methodologies are key to assess TBI efficacy to ensure that only the most potent candidates progress to expensive and timeconsuming clinical trials. Measuring intervention efficacy can be problematic because there is substantial variation in the number of parasites in both the host and vector populations, which can impact transmission even in laboratory settings.
Methods
A statistically robust empirical method is introduced for estimating intervention efficacy from standardised population assay experiments. This method will be more reliable than simple summary statistics as it captures changes in parasite density in different lifestages. It also allows efficacy estimates at a finer resolution than previous methods enabling the impact of the intervention over successive generations to be tracked. A major advantage of the new methodology is that it makes no assumptions on the population dynamics of infection. This enables both hosttovector and vectortohost transmission to be densitydependent (or other) processes and generates easytounderstand estimates of intervention efficacy.
Results
This method increases the precision of intervention efficacy estimates and demonstrates that relying on changes in infection prevalence (the proportion of infected hosts) alone may be insufficient to capture the impact of TBIs, which also suppress parasite density in secondarily infected hosts.
Conclusions
The method indicates that potentially useful, partially effective TBIs may require multiple infection cycles before substantial reductions in prevalence are observed, despite more rapidly suppressing parasite density. Accurate models to quantify efficacy will have important implications for understanding how TBI candidates might perform in field situations and how they should be evaluated in clinical trials.
Keywords
Atovaquone Plasmodium berghei Transmissionblocking drugs Transmissionblocking vaccinesBackground
Malaria remains a serious public health concern; in 2015, an estimated 214 million (range 149–303 million) new cases and 438,000 (236,000–635,000) deaths resulted from the infection [1]. Malaria is transmitted between vertebrate hosts by anopheline mosquitoes. Circulating male and female gametocytes are ingested by mosquitoes during blood feeding where they undergo sexual reproduction before developing into oocysts on the wall of the midgut. On rupturing, an average 1250 (interquartile range 313–2400) salivary gland sporozoites are released from a single Plasmodium falciparum oocyst [2]. Sporozoites are then injected onward to the vertebrate host during a subsequent feed. Within the host, sporozoites travel to the liver and invade cells, divide and proliferate to produce tens of thousands of haploid forms, merozoites, per liver cell. Merozoites go on to reproduce asexually, resulting in bloodstage parasites. These can differentiate into gametocytes that complete the life cycle when transmitted back to the insect vector.
There are multiple transmissionblocking interventions (TBIs) currently under development that aim to reduce transmission by inhibiting the development of oocysts in the mosquito midgut or by targeting sexual, sporogonic and/or mosquito antigens (SSMVIMT) [3]. Different assays have been developed to capture the data required to evaluate the efficacy of different TBIs and triage the most promising candidates [4]. The current gold standard is the standard membrane feeding assay (SMFA) [5, 6]. Here, mosquitoes are fed on cultured, infected blood through a membrane and oocysts can develop. Additionally, multiple derivative assays such as the direct membrane feeding assay (DMFA) and the direct (skin) feeding assay (DFA) are used [5]. The efficacy of drugs and vaccines against malaria have been assessed using data from these assays by measuring the relative reduction in the mean number or prevalence of oocysts in the vector, or prevalence for treated host populations in comparison to control groups [7, 8]. The highly aggregated (overdispersed) distribution of oocysts means that analysis of the reduction in the mean count of oocysts is inherently uncertain [9] and large sample sizes are needed to accurately capture infections [5]. One approach to overcome this is the use of the zeroinflated negative binomial distribution [10] of parasitic life stages that has been used to assess vaccines [11], and suggested as an improvement to capture the influence of aggregated infections on efficacy estimates [9].
A second concern for TBI assessment has been that although SMFA (or its derivatives) capture reductions in infections to mosquitoes, they fail to capture the subsequent onward infectivity to the human host or the onward transmission over multiple cycles. To do this in laboratory studies, multigenerational experiments have been developed that simulate the passage of Plasmodium berghei between populations of mice and Anopheles stephensi mosquitoes [12, 13]. Using this approach, the transmissionblocking drug atovaquone (ATV), titrated to give a 32% reduction in oocyst intensity (ATV32%) in a DFA, was found to have an estimated effect size of 20.4% [12]. The effect size measures the ability of a treatment to reduce the basic reproduction number of the parasite (the average number of secondary infections caused by an infected host). These estimates were made using a chainbinomial model [14, 15], but the assumptions inherent in the model affect estimation and need further validation. Firstly, it assumes that all infectious mosquitoes and mice are equally infectious [15]. (In fact, the probability of infection from vertebratestovectors or vice versa increases when hosts have higher infection intensities [16].) Secondly, the number of parasites in successive life stages is highly variable; for example, mosquitoes biting on the same blood source are likely to have widely differing oocyst or sporozoite counts [17]. The importance of parasite intensity on transmission means that it is necessary to account for stochastic variability. Thirdly, the chain binomial model assumes that the probabilities of infection from multiple infectious bites are independent. Finally, although the reproduction number and effect size calculations are widely understood by mathematical modellers, these terms are not generally used by those working in TBI development who tend to measure efficacy as either the transmissionblocking activity ((TBA), the reduction in the number of infected hosts or vectors) or transmissionreduction activity [(TRA), the reduction in mean parasite intensity of the host or vector population]. To overcome these issues, a probabilistic Bayesian model is developed that captures the generative structure of the complex data. This makes it possible to capture the relationship between successive stages of malaria across multiple vertebratetovertebrate transmission cycles and generate TBA and TRA estimates. The zeroinflated negatively binomially distributed data for each parasitic life stage is used to inform the transmission probabilities between vector and host or host and vector. In this way, the transmission probability of a more heavily infected vector or host population (where each individual, on average, harbours a relatively high number of parasitic stages) could be higher than one with lower density infections even where prevalence (the proportion of individuals in the population with any infection) is matched. The Bayesian approach incorporates the variability between transmission events, which is much harder to incorporate into frequentist approaches without relying on approximations. This allows the uncertainty in the system to be captured in the model and enables relatively precise estimates of efficacy (the percentage reduction in parasite prevalence or infection intensity achieved by the TBI) to be broken down by transmission cycle and biting group (the number of potentially infectious mosquitoes biting each mouse during a transmission cycle) to provide a more detailed analysis of data.
Methods
Experimental design
For each treatment regime (ATV32% and control), mousetomouse transmission operated as described previously [12, 13], the experiment is graphically demonstrated in Fig. 1 and described in the figure’s legend. All care and handling of animals strictly followed the Guidelines for Animal Care and Use prepared by Imperial College London, and was performed under the UK Home Office Licences 70/7185 and 70/8788.
Statistical methods
In this experiment, it was not possible to measure the number of sporozoites reaching salivary glands from each oocyst or the number of bloodstage infections in mice resulting from injected sporozoites. These relationships are uncertain and are incorporated as nuisance parameters in the model (Additional file 1). Each of these stages is modelled sequentially, similarly to a hidden Markov model. The number of parasites in each stage is represented by a zeroinflated negative binomial distribution (NBD) to account for both infected and uninfected individuals. A bimodal structure is fitted using a zeroinflation parameter, π, which determines the proportion of mice or mosquitoes that are uninfected and therefore cannot transmit. The shape of the relationships between different parasite life stages is unknown. This is accounted for by including a random effects component that allows the mean number of parasites in each group to vary according to the observed data. A comprehensive description of the model is provided in Additional file 1.
Model fitting
All parameters were fitted jointly using a Bayesian posterior distribution in RStan (version 2.11, [18]). A noncentred parameterization method was employed [19, 20]. The model parameter fitting used a Hamiltonian Monte Carlo method [18], burnin was 500 and the subsequent 500 samples from each chain (n = 4) were used for posterior predictive checks. The model R code is provided in Additional file 2), with the accompanying data (Additional file 2). The parameter estimates are supplied in Additional file 3.
Model output
Two measures of efficacy are presented: (1) the TBA, the percentage difference in the proportion of infected hosts between the control (t = 0) and treatment (t = 1) arms of the experiment; and, (2) the TRA, the percentage difference in parasite density between the control and treatment arms of the experiment. Efficacy estimates (denoted TBA _{ i,m } for prevalence, TRA _{ i,m } for intensity), were generated for each biting rate (m) and transmission cycle (i) from the simulated posterior predictive model outputs (n iterations = 2000) using the equations: \(\begin{aligned} TBA_{i,m} = \frac{{P_{0,i,m}  P_{1,i,m} }}{{P_{0,i,m} }} \times 100 \hfill \\ TRA_{i,m} = \frac{{I_{0,i,m}  I_{1,i,m} }}{{I_{0,i,m} }} \times 100 \hfill \\ \end{aligned}\).
Efficacy estimates
Group  Arithmetic mean (±1.96 × SE)  Median  

TBA  TRA  TBA  TRA  
Overall  N bites (m)  N cycles (i)  28.42 (28.0–28.9)  22.63 (21.7–23.6)  29.1  25.5 
Per bite  1  28.42 (28.0–28.9)  22.63 (21.7–23.6)  29.1  25.5  
2  70.77 (69.8–71.7)  91.18 (90.7–91.6)  75.0  93.9  
3  58.60 (57.5–59.7)  88.58 (88.1–89.1)  62.5  91.2  
4  28.37 (27.2–29.5)  34.53 (32.9–36.1)  30.8  40.9  
5  5.87 (5.0–6.8)  −4.72 (−6.5 to −3.0)  6.3  −4.0  
Per cycle  1  4.80 (3.9–5.7)  −8.78 (−10.6 to −7.0)  6.3  −12.2  
2  21.15 (20.3–22.0)  44.88 (43.8–45.9)  22.7  49.4  
3  30.20 (29.4–31.1)  28.63 (27.1–30.1)  31.3  35.0  
4  32.60 (31.7–33.5)  13.51 (11.7–15.3)  33.3  16.6  
Per bite and cycle  1  1  28.42 (27.4–29.5)  4.16 (2.2–6.1)  28.6  1.3 
1  2  34.5 (32.0–37.0)  67.30 (65.7–68.9)  50.0  78.0  
1  3  75.53 (73.7–77.4)  94.33 (93.7–95.0)  100  100  
1  4  97.60 (97.0–98.2)  99.56 (99.3–99.8)  100  100  
2  1  99.70 (99.5–99.9)  99.9 (99.8–100)  100  100  
2  2  16.14 (13.4–18.9)  60.97 (59.3–62.6)  33.3  71.8  
2  3  55.14 (52.6–57.7)  90.18 (89.3–91.0)  66.7  96.9  
2  4  88.9 (87.6–90.2)  98.07 (97.7–98.5)  100  100  
3  1  97.93 (97.2–98.6)  99.74 (99.6–99.8)  100  100  
3  2  25.95 (23.8–28.1)  41.85 (39.8–43.9)  33.3  53.1  
3  3  31.60 (29.2–34.0)  44.80 (42.5–47.1)  33.3  57.5  
3  4  34.4 (31.7–37.2)  47.7 (45.3–50.2)  50.0  61.6  
4  1  38.60 (35.7–41.5)  49.7 (57.0–52.4)  50.0  66.3  
4  2  10.65 (8.9–12.4)  33.6 (31.6–35.6)  20.0  44.0  
4  3  8.88 (7.0–10.8)  15.18 (12.8–17.5)  0.0  22.1  
4  4  8.29 (6.2–10.4)  7.87 (5.3–10.4)  0.0  14.2  
5  1  3.23 (0.6–9.5)  4.00 (1.2–6.8)  0.0  9.5  
5  2  10.15 (8.6–77.4)  34.9 (33.0–36.9)  20.0  44.1  
5  3  8.65 (6.7–10.6)  14.98 (12.5–17.4)  0.0  21.8  
5  4  4.12 (1.59–6.7)  5.99 (3.3–8.7)  0.0  11.2 
To justify model assumptions, it was important to investigate the difference between the overall distribution of sporozoite scores in control and treatment groups, parasitaemia (%) and gametocytaemia (%). Data exploration was conducted in R, version 3.2.2 [21].
Results
Evaluating efficacy
With a full probabilistic model of the mousetomouse measurements there is potential to define many possible measures of efficacy depending on exactly what elements of the model impact future interventions. Here two are presented: TBA and TRA (Table 1).
The prevalence efficacy TBA
The impact of ATV was greatest at low mosquito biting rates (where each mouse received a single mosquito bite, the TBA was 70.8% (69.8–71.7%); for two mosquito bites per mouse: 58.6% (57.5–59.7%); and, for three bites per mouse: 28.4% (27.2–29.5%)) whilst the impact was much reduced at higher mosquito biting rates [for four bites: 5.9% (5.0–6.8%); for five bites: 4.8% (3.9–5.7%)].
Per bite TBAs were consistent after the second completed transmission cycle (Fig. 3) suggesting that three cycles would be sufficient to capture the impact of ATV at a concentration expected to reduce transmission by 32% from vertebratetovector [12].
The intensity efficacy TRA
The overall TRA was 22.63% (21.7–23.6%). The biggest impact against intensity was achieved after the first transmission cycle (Table 1; Fig. 3).
Like parasite prevalence efficacies, the impact of ATV on parasite density was greatest at low mosquito biting rates; 91.2% (90.7–91.6%), 88.6% (88.1–89.1%) and 34.5% (32.9–36.1%) for one to three mosquito bites per mouse, respectively. There was no positive impact on parasite intensity at higher mosquito biting rates; −4.7% (−6.1–3.0%) and −8.8% (−10.6–7.0%) for four and five mosquito bites per mouse (Fig. 2).
Parasite intensity, transmission and heavy infections
Assessing efficacy estimates by transmission cycle and biting rate allows more detailed examination of the intervention’s impact. For example, in the treatment arm of the experiment, two of five mice in the transmission cycle 3 cohort, which received four mosquito bites each, had exceptionally high parasitaemia (≥12% of 1200 red blood cells were infected). Both mice had, by chance, been fed on by mosquitoes with very high postfeeding sporozoite counts (indicated by high mean sporozoite scores, >2). Once exposed, these mice can be fed upon by any of the 500 naïve mosquitoes in the next cohort. Consequently, higher than expected parasite densities are propagated throughout the system (Fig. 2).
Although the infectiousness of mice depends on the mature gametocytes present in the blood, rather than the asexual parasites (that are used in the model), there is less confidence in microscopic measurements of gametocyte counts. A linear relationship between mean gametocytaemia (the percentage of erythrocytes infected with sexual stage parasites) and mean parasitaemia (the percentage of erythrocytes infected with asexual stage parasites) was consistent with the data (mean gametocytaemia ~0.11 × mean parasitaemia + 0.03, σ = 0.12: F_{40,38} = 197.1, p < 0.001, _{adj}R^{2} = 0.83), justifying the use of parasitaemia as a dependent variable in addition to infection status.
Discussion
Accurate methods to assess potential TBIs are important to ensure that the best candidates go forward to expensive clinical trials. Here, TBI efficacy from mouse population assays is estimated by fitting a probabilistic Bayesian model to multiple parasitic life stages with a structure that reflects the experiment structure used to collect the data. Whilst it may be more computationally expensive to fit a multistage Bayesian model such as the one presented here, the method: (i) is able to identify chance events that can occur at any point in the experiment; (ii) is statistically robust to natural fluctuations in the starting parasite density, which impacts the capacity of the treatment to reduce secondary infections; and (iii) has the flexibility to capture the density dependent (or other) relationships between parasite life stages. The chain binomial model produced a relatively similar estimate for ATV32% (20.4% reduction in secondary cases) when compared to the Bayesian approach (28.4% efficacy against prevalence).
Although the two metrics are not directly comparable, they provide complementary understanding of the intervention impact. Firstly, the chain binomial model presents an overall effect size that measures the reduction in secondary cases in hosts, the R _{0}, that can be achieved by using vaccine. This can be used directly in mathematical models to understand the epidemiology of diseases. Conversely the Bayesian method directly calculates the reduction in the prevalence and density of infection between the different arms of the experiment. This generates more intuitive efficacy estimates which are more comparable to the results of clinical trials and are easier to understand by nonmathematical modellers. Secondly, the chain binomial model assumes that all infected mice and mosquitoes have the same infectivity irrespective of the density of infection. This can bias estimates depending on the quantity of parasite in the system. The new method allows the infectivity to vary according to the density of the parasite at the particular lifestages which provides additional information for the actions of an intervention. Finally, the chain binomial model generates a single effect size estimate whilst the Bayesian model enables intervention efficacy to be broken down by treatment round and biting rate. This finer scale of resolution allows a more detailed understanding of the dynamics of the disease, for example by showing that parasite density efficacy is higher in earlier transmission cycles compared to the efficacy against parasite prevalence. This finer resolution also enables chance fluctuations in parasite intensity to be identified, which could bias overall estimates in populations with small sample sizes and generate misleading results. One of the advantages of the new method is that it makes no assumptions about the likelihood of infection of multiply bitten hosts unlike the chain binomial model which assumes that mice are more likely to have detectable parasites the more infected bites they receive (following a binomial distribution). This makes results from the low biting groups relatively uncertain (due to stochastic fluctuations) so these groups should be interpreted in the context of other results.
Measuring impact at different mosquito biting rates and transmission cycles shows that ATV works first to reduce parasite density and later to reduce secondary infections. Single mousetomouse transmission cycle population experiments have been performed previously [13]. Although this would generate accurate predictions of efficacy for highly effective interventions (which eliminate all onward transmission), using a single cycle of mousetomouse transmission could potentially underestimate the impact of partially effective TBIs. The model also indicates that, for the intervention investigated here, efficacy estimates remain relatively stable after three cycles of transmission. This may allow the length of the experiment to be shortened, lowering costs and reducing the number of mice required.
The mouse population assay outlined in [12] is a standardized experiment and will be used to address future hypotheses including proof of concept questions such as the impact of combining interventions (for example a transmissionblocking vaccines with preerythrocytic vaccines) or changing levels of vaccine coverage. These types of experiments allow the consequences of different epidemiological characteristics or control interventions to be triaged in a controlled environment that would be infeasible to do in a human system or would require multiple, very expensive randomised control trials.
Given the relatively sparse data for each trial this analysis can be sensitive to systematic variation caused by fluctuations in the amount of malaria per individual in any given transmission cycle. The hierarchical structure used, however, does prevent a fluctuation in any one distribution (caused by unusually heavily infected individuals) from strongly biasing the inferences of other lifestages. Moreover, the weaklyinformative priors used in the analysis limit the potential for overfitting and the predictive checks showed no indication of such pathologies. The analysis could be improved by incorporating greater biological realism into the model by increasing the number of intermediate lifestages (for example including a gametocyte lifestage or better capturing host immune responses). However, the benefit of this enhanced realism in the model depends on the applicability of the rodent system to human malaria. There is much debate about the capacity for mouse models to inform on aspects of human malaria [23–25] so extrapolating to P. falciparum must be done with caution.
Conclusions
Here, a standardised statistical methodology is outlined which uses the Hamiltonian Monte Carlo method (RStan, version 2.11 [18]) to fit to population assay experimental data. To the author’s knowledge, this is the first time these methods have been used to examine parasitological data and their flexibility and precision makes them ideally suited for this purpose. In many parasitic infections the population dynamics of transmission and host morbidity and mortality depends on the number of parasites in a host and/or vector and not their presence/absence [26]. The distribution of parasites between hosts is typically highly aggregated [16, 27, 28]. This is particularly the case for many human parasitic infections, which are the target of elimination programmes [29, 30, 31]. As the parasite becomes increasingly rare, meanbased estimates of parasite intensity become increasingly uncertain. The range of statistical distributions available in Stan enables a more precise population estimate of parasite intensity. Many of the processes governing parasite transmission are densitydependent [27, 32]. Using a probabilistic approach enables these nonlinear processes to be captured at the individual level whilst considering parasite distributions without the need for overly simplifying the experimental structure in the statistical model or running computationally expensive individualbased transmission models [33].
Abbreviations
 TBI:

transmissionblocking interventions
 TBV:

transmissionblocking vaccine
 SMFA:

standard membrane feeding assay
 DMFA:

direct membrane feeding assay
 DFA:

direct feeding assay
 ATV:

atovaquone
 TBA:

transmissionblocking activity
 TRA:

transmissionreducing activity
 NBD:

negative binomial distribution
Declarations
Authors’ contributions
TSC and AMB secured the funding for the work. ESS, MB and TSC conceived the method. AMB, LMU, KAS, and SEZ performed the experiments and collated the data. ESS analysed and formatted the data. MB and ESS constructed the model. ESS and HSC prepared the figures. ESS, MB and TSC drafted the manuscript. All authors read and approved the final manuscript.
Acknowledgements
The authors would like to acknowledge the PATH Malaria Vaccine Initiative (MVI funded the authors of this work). TSC would like to thank the UK Medical Research Council (MRC)/UK Department for International Development (DFID) under the MRC/DFID Concordat agreement. AMB thanks the MRC (Grant number MR/N00227X/1) for funding. MB is supported under EPSRC Grant EP/J016934/1. We would also like to thank Mark Tunnicliff for mosquito production.
Competing interests
The authors declare that they have no competing interests.
Availability of data and materials
All data generated and analysed during the current study are included in the published article and supplementary information files.
Ethical approval
All care and handling of animals strictly followed the Guidelines for Animal Care and Use prepared by Imperial College London, and was performed under the UK Home Office Licences 70/7185 and 70/8788.
Funding
PATH Malaria Vaccine Initiative; UK Medical Research Council (MRC)/UK Department for International Development (DFID).
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. 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.
Authors’ Affiliations
References
 WHO. World malaria report 2015. Geneva: World Health Organization, 2015. p. 243.
 Stone WJR, Eldering M, van Gemert GJ, Lanke KHW, Grignard L, van de VegteBolmer MG, et al. The relevance and applicability of oocyst prevalence as a readout for mosquito feeding assays. Sci Rep. 2013;3:3418.View ArticlePubMedPubMed CentralGoogle Scholar
 Delves MJ, Ramakrishnan C, Blagborough AM, Leroy D, Wells TNC, Sinden RE. A highthroughput assay for the identification of malarial transmissionblocking drugs and vaccines. Int J Parasitol. 2012;42:999–1006.View ArticlePubMedGoogle Scholar
 Sinden RE, Blagborough AM, Churcher T, Ramakrishnan C, Biswas S, Delves MJ. The design and interpretation of laboratory assays measuring mosquito transmission of plasmodium. Trends Parasitol. 2012;28:457–65.View ArticlePubMedGoogle Scholar
 Bousema T, Churcher TS, Morlais I, Dinglasan RR. Can fieldbased mosquito feeding assays be used for evaluating transmissionblocking interventions? Trends Parasitol. 2013;29:53–9.View ArticlePubMedGoogle Scholar
 Stone WJR, Bousema T. The standard membrane feeding assay: advances using bioluminescence. In: Vaughan A, editor. Malaria vaccines: methods and protocols. Berlin: Springer; 2015. p. 101–12.View ArticleGoogle Scholar
 Matsuoka H, Kobayashi J, Barker GC, Miura K, Chinzei Y, Miyajima S, et al. Induction of antimalarial transmission blocking immunity with a recombinant ookinete surface antigen of Plasmodium berghei produced in silkworm larvae using the baculovirus expression vector system. Vaccine. 1996;14:120–6.View ArticlePubMedGoogle Scholar
 Blagborough AM, Delves MJ, Ramakrishnan C, Lal K, Butcher G, Sinden RE. Assessing transmissionblocking in Plasmodium spp. In: Menard R, editor. Malaria methods and protocols. Methods in molecular biology. vol. 923. New York: Springer; 2013. p. 577–600.Google Scholar
 Churcher TS, Blagborough AM, Delves M, Ramakrishnan C, Kapulu MC, Williams AR, et al. Measuring the blockade of malaria transmission—an analysis of the standard membrane feeding assay. Int J Parasitol. 2012;42:1037–44.View ArticlePubMedGoogle Scholar
 Bolker BM, Brooks ME, Clark CJ, Geange SW, Poulsen JR, Stevens MH, et al. Generalized linear mixed models: a practical guide for ecology and evolution. Trends Ecol Evol. 2008;24:127–35.View ArticleGoogle Scholar
 RTSS Clinical Trials Partnership. Efficacy and safety of RTS, S/AS01 malaria vaccine with or without a booster dose in infants and children in Africa: final results of a phase 3, individually randomised, controlled trial. Lancet. 2015;6736:31–45.Google Scholar
 Blagborough AM, Churcher TS, Upton LM, Ghani AC, Gething PW, Sinden RE. Transmissionblocking interventions eliminate malaria from laboratory populations. Nat Commun. 2013;4:1812.View ArticlePubMedPubMed CentralGoogle Scholar
 Upton LM, Brock PM, Churcher TS, Ghani AC, Gething PW, Delves MJ, et al. Lead clinical and preclinical antimalarial drugs can significantly reduce sporozoite transmission to vertebrate populations. Antimicrob Agents Chemother. 2015;59:490–7.View ArticlePubMedGoogle Scholar
 Greenwood M. On the statistical {measure} of infectiousness. J Hyg (Lond). 1931;31:836–51.Google Scholar
 Becker N. A general chain binomial model for infectious diseases. Biometrics. 1981;37:251–8.View ArticlePubMedGoogle Scholar
 Sinden RE, Dawes EJ, Alavi Y, Waldock J, Finney O, Mendoza J, et al. Progression of Plasmodium berghei through Anopheles stephensi is densitydependent. PLoS Pathog. 2007;3:e195.View ArticlePubMedPubMed CentralGoogle Scholar
 Pichon G, AwonoAmbene HP, Robert V. High heterogeneity in the number of Plasmodium falciparum gametocytes in the bloodmeal of mosquitoes fed on the same host. Parasitology. 2000;121(Pt 2):115–20.View ArticlePubMedGoogle Scholar
 Stan Development Team. Stan modeling language users guide and reference manual. 2016.
 Papaspiliopoulos O, Roberts GO, Sköld M. A general framework for the parametrization of hierarchical models. Stat Sci. 2007;22:59–73.View ArticleGoogle Scholar
 Betancourt M, Girolami M. Hamiltonian Monte Carlo for hierarchical models. In: Upadhyay SK, Singh U, Dey DK, Logananthan A, editors. Current trends in Bayesian methodology with applications. Boca Raton: Chapman and Hall/CRC Press; 2015. p. 680.Google Scholar
 R Core Team. R: A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing; 2014. http://www.rproject.org/.
 Slater HC, Ross A, Ouédraogo AL, White LJ, Nguon C, Walker PGT, et al. Assessing the impact of nextgeneration rapid diagnostic tests on Plasmodium falciparum malaria elimination strategies. Nature. 2015;528:S94–101.View ArticlePubMedGoogle Scholar
 Wykes MN, Good MF. What have we learnt from mouse models for the study of malaria? Eur J Immunol. 2009;39:2004–7.View ArticlePubMedGoogle Scholar
 Fidock DA, Rosenthal PJ, Croft SL, Brun R, Nwaka S. Antimalarial drug discovery: efficacy models for compound screening. Nat Rev Drug Discov. 2004;3:509–20.View ArticlePubMedGoogle Scholar
 Langhorne J, Buffet P, Galinski M, Good M, Harty J, Leroy D, et al. The relevance of nonhuman primate and rodent malaria models for humans. Malar J. 2011;10:23.View ArticlePubMedPubMed CentralGoogle Scholar
 Anderson RA, May RM. Infectious diseases in humans: dynamics and control. Oxford: Oxford Scientific; 1991.Google Scholar
 Shaw DJ, Dobson AP. Patterns of macroparasite abundance and aggregation in wildlife populations: a quantitative review. Parasitology. 1995;1995(111 Suppl):S111–27.View ArticleGoogle Scholar
 Woolhouse ME, Dye C, Etard JF, Smith T, Charlwood JD, Garnett GP, et al. Heterogeneities in the transmission of infectious agents: implications for the design of control programs. Proc Natl Acad Sci USA. 1997;94:338–42.View ArticlePubMedPubMed CentralGoogle Scholar
 Galvani AP, May RM. Epidemiology: dimensions of superspreading. Nature. 2005;438:293–5.View ArticlePubMedGoogle Scholar
 Linehan M, Hanson C, Weaver A, Baker M, Kabore A, Zoerhoff KL, et al. Integrated implementation of programs targeting neglected tropical diseases through preventive chemotherapy: proving the feasibility at national scale. Am J Trop Med Hyg. 2011;84:5–14.View ArticlePubMedPubMed CentralGoogle Scholar
 Anderson RM, Truscott JE, Pullan RL, Brooker SJ, Hollingsworth TD. How effective is schoolbased deworming for the communitywide control of soiltransmitted helminths? PLoS Negl Trop Dis. 2013;7:e2027.View ArticlePubMedPubMed CentralGoogle Scholar
 Shaw DJ, Grenfell BT, Dobson AP. Patterns of macroparasite aggregation in wildlife host populations. Parasitology. 1998;117(Pt 6):597–610.View ArticlePubMedGoogle Scholar
 Churcher TS, Filipe JAN, Basáñez MG. Density dependence and the control of helminth parasites. J Anim Ecol. 2006;75:1313–20.View ArticlePubMedGoogle Scholar
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.