key: cord-0431033-bo7ppwvv authors: Tran, Q. M.; Soda, J.; Siraj, A.; Moore, S. M.; Clapham, H.; Perkins, A. title: Expected endpoints from future chikungunya vaccine trial sites informed by serological data and modeling date: 2022-05-25 journal: nan DOI: 10.1101/2022.05.23.22275466 sha: 49ceaf7a7ea8377c3a8b4899422cf730a9afd5ce doc_id: 431033 cord_uid: bo7ppwvv In recent decades, there has been an increased interest in developing a vaccine for chikungunya. However, due to its unpredictable transmission, planning for a chikungunya vaccine trial is challenging. To inform decision making on the selection of sites for a vaccine efficacy trial, we developed a new framework for projecting the expected number of endpoint events at a given site. In this framework, we first accounted for population immunity using serological data collated from a systematic review and used it to estimate parameters related to the timing and size of past outbreaks, as predicted by an SIR transmission model. Then, we used that model to project the infection attack rate of a hypothetical future outbreak, in the event that one were to occur at the time of a future trial. This informed projections of how many endpoint events could be expected if a trial were to take place at that site. Our results suggest that some sites may have sufficient transmission potential and susceptibility to support future vaccine trials, in the event that an outbreak were to occur at those sites. In general, we conclude that sites that have experienced outbreaks within the past 10 years may be poorer targets for chikungunya vaccine efficacy trials in the near future. Our framework also generates projections of the numbers of endpoint events by age, which could inform study participant recruitment efforts. Vaccine development for CHIKV faces a serious obstacle when it comes to demonstrating efficacy [15] . This owes to the fact that it is notoriously difficult to predict when and where chikungunya outbreaks will occur due to the sporadic epidemiology of this disease. This means that vaccine trials most likely need to be conducted during one or more outbreaks [20] . The considerable logistical difficulties associated with conducting a trial during an outbreak place even more importance on being prepared to carry out trials in locations where outbreaks may be likely. Continuing outbreaks of chikungunya globally [21] offer encouragement about the prospect of efficacy trials being possible, although current vaccine candidates have so far not been able to leverage these outbreaks as opportunities for phase-III efficacy trials required for licensure. Completion of a phase-III vaccine efficacy trial depends not just on a chikungunya outbreak occurring somewhere, but in a particular population that can be enrolled and followed during a trial. Advancements in the ability to appropriately select sites with a high probability of a future outbreak are needed to speed up the evaluation of these vaccine candidates. Mathematical modeling is playing an increasingly important role in vaccine trial planning in general [22] , including in the problem of site selection [23] [24] [25] . Site selection is a major challenge for chikungunya vaccine efficacy trial planning due to its sporadic epidemiology. This problem is traditionally solved with the help of epidemiological data to estimate recent incidence of the proposed efficacy endpoint, such as reported cases of symptomatic disease. For chikungunya, incidence is dynamic and not fixed over time. Moreover, it is likely that past chikungunya outbreaks may have gone unreported or misdiagnosed [26] . Analysis of agestratified serological data offers one way to reconstruct recent patterns of CHIKV transmission relevant to vaccine efficacy trial planning [27, 28] , but such analyses have only been performed in a limited number of settings and have not been leveraged to directly inform vaccine efficacy trial planning. In this study, we use age-stratified serological data and mathematical modeling to project vaccine trial endpoint incidence in the event of a future outbreak. Our approach consists of three steps: 1) infer epidemic parameters from age-stratified serological data; 2) determine the population immunity; and 3) project outbreak size and endpoint events if an outbreak were to occur at the time of a vaccine trial. We first demonstrate our framework in a simulation study, with the purpose of understanding the accuracy of our inferences when the true values of the underlying parameters are known. We then apply this approach to 35 sites for which agestratified serological data were available, although we also propose this as a method that could be applied to age-stratified serological data collected as part of pre-trial epidemiological studies. Inform vaccine efficacy trial endpoints from serological data For a given location, we 1) estimate the basic reproduction number R0, the timing of past outbreaks y, and average initial susceptible proportion before the first outbreak happened S0(0) from serological data, and 2) determine population immunity at the time of data collection. Assuming a hypothetical outbreak happens during a vaccine efficacy trial in 2022, we then 3) project the IAR of the outbreak and the number of endpoints expected among a given number of subjects. This framework is summarized visually in Fig. 1 . All of the code for this analysis was written in the R language [29] and is available, along with all data required to reproduce this study, in a GitHub repository [30]. Our framework consists of three stages: 1) using serological data available before the vaccine trial, we can estimate the basic reproduction number R0, the average initial susceptible proportion before the first outbreak S0(0) (used to inform the susceptible proportion before the first outbreak S1(0) (Supplementary Text)), and the timing of past outbreaks y (panel A). 2) From the estimated parameters in A, we can derive the population immunity during the serological study at the site (panel B). 3) The population immunity at the time of a future vaccine trial can then be projected during the vaccine trial (panel C). In the event that an outbreak happens during that future trial, we project the infection attack rate (IAR) of the outbreak using the estimated R0 and the projected population immunity. The number of expected endpoint events can be derived subsequently. Note: S1(0) and S2(0) are the susceptible proportion before the first and second outbreaks respectively; y1 and y2 are the timing of the first and second CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 25, 2022. We assumed that the number positive pi in each age group i follows a binomial distribution, such that the log likelihood informed by serological data is where ni is the number of samples and Pi is the model's prediction of the proportion positive. We can derive Pi from the age-stratified susceptibility in the population Sa,Y in the year Y and age a at which the data were collected, which equals where al and au are the lower bound and upper bound of age group i and popa,Y is the proportion of people in age group a at year Y. This age-specific proportion was derived from annual population by age data from United Nations World Population Prospects [31] , assuming the site has the same age demographic as in country level. Consistent with the sporadic nature of chikungunya outbreaks, we assume that the agestratified susceptibility Sa,Y has a step-like shape, where each "step" of the seroprevalence curve corresponds to an outbreak in the past [32] . The particular shape can be modeled in our framework with an SIR model in which the years of past outbreaks are specified, along with the basic reproduction number R0 and the population susceptibility prior to the first outbreak within the lifetime of the oldest study participants S0(0). The time between outbreaks corresponds to the width of each step, while the susceptible-infected-recovered (SIR) model informs the proportion immune following each outbreak, which corresponds to the height of each step (Fig. 1A ). Details on how to derive Sa,Y for given values of R0, the initial susceptible proportion, and the timing of past outbreaks y are described in the Supplementary Text. For each seroprevalence data set, we fitted five models, which differed in terms of the number of historical outbreaks (1 to 5 outbreaks). We estimated the parameters R0, S0(0), and y in a Bayesian framework, using the BayesianTools package [33] . We used the DREAMzs algorithm, with one million iterations and a burn-in period of 500,000 iterations. The assumed prior distributions for the parameters were with the latter two being non-informative and the range for the R0 prior being defined broadly within the range of values generally observed for chikungunya [34] . Convergence was assessed by calculating the potential scale reduction statistic Ȓ using the coda package in R [35] . We applied a post-processing reversible-jump Markov chain Monte Carlo algorithm from the rjmcmc package [36] to derive ensemble weights of the five models for each data set. These weights were then used to generate ensemble predictions. From the estimates of R0, y, and S0(0), we can reconstruct the step-like shape of population immunity in every age group at the time of the study. This reconstruction was done based on the relationship between the basic reproduction number (R0), the proportion susceptible in a population before an outbreak S(0), and the IAR over the course of an outbreak from a SIR model as stated in step 1 and in the Supplementary Text. For the next step, we projected the age-stratified positive proportion in the population to year 2022 to derive the positive proportion in the population before a hypothetical outbreak in 2022. The age-stratified positive proportion was estimated individually from each of the five models with different numbers of outbreaks. Consistent with the endpoint for chikungunya vaccine trials proposed by WHO [37] , we chose symptomatic cases as the endpoint for the hypothetical vaccine trials in 2022 that we focused on in our analysis. The number of symptomatic cases among trial participants is generated by multiplying the predicted IAR by the proportion of infections that result in symptoms, which is sampled from a beta distribution with shape and scale parameters of 21.80 and 3.18, respectively. We derived this distribution from empirical estimates of the asymptomatic proportion of CHIKV infections ranging from 3% to 28% [38] [39] [40] , such that 95% of samples would fall within this range. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) To assess the behavior of our model, we calculated the IAR during a hypothetical outbreak in 2022 across a range of parameter values. The range of R0 values was chosen to be from 1 to 3, which is the typical R0 values estimated from previous outbreaks [41] . We assume the study was conducted in 2000 and that past outbreaks may have happened between 1900 to 2000 in increments of ten years. We also explore how IAR will change if there is only one or multiple past outbreaks. We expect that S0(0) < 1 would have the same effect as an additional outbreak before the first one, which is already covered by exploring how multiple outbreaks affect IAR as just mentioned. Therefore, for simplicity, S0(0) was set to 1 so there was no immunity in the population before the first outbreak. To assess the performance of our model at estimating parameters, we applied our inference method to simulated data from hypothetical serological studies with known parameters. These hypothetical studies involved 2,000 simulated study participants, stratified into 20 age groups of five years each. The outbreak times y were set to be 5, 45, and 85 years before the study time. For these simulations, each outbreak had R0 = 1.5 and S0(0) = 1. These simulated datasets were used to estimate back the parameters R0, y, and S0(0), which were then used to project the IAR during a hypothetical outbreak in 2022. To assess the performance of our approach when data is more limited, we also conducted a sensitivity analysis by changing the sample size (only ten participants for each of 20 age groups) or the age bins (only five age groups of 20 participants each), but keeping everything else the same. The age distribution of Kenya [31] was used in all of these simulations. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 25, 2022. ; https://doi.org/10.1101/2022.05.23.22275466 doi: medRxiv preprint Applying to published serosurveys After validating our method on simulated data, we applied it to published data, which were obtained from a systematic review of CHIKV serological data [42] . Criteria for inclusion of studies in this review were that a study should report data from immunoglobulin G (IgG) enzyme-linked immunoassay (ELISA) test, plaque reduction neutralization tests (PRNT), or hemaglutination inhibition (HI) on at least three age groups collected as part of a cross-sectional study. We used estimates of population by age appropriate to the country and year of the study as estimated by the United Nations World Population Prospects [31] . To understand the relationships between the model's parameters and the predicted IAR, we derived the possible values of IAR for a hypothetical outbreak in 2022, given a range of possible values of R0 and the timing of past outbreaks. In the case of a site with only one outbreak, we observed that a longer time since the last outbreak and larger R0 would be associated with a higher IAR in 2022 (Fig. 2 ). When the time since the last outbreak was less than 10 years, the IAR was less than 5%, regardless of the R0 value. generated is colored as shown in the legend. We observed that in the model with one outbreak, higher R0 and the further the time from the last outbreak will result in higher IAR. When there have been multiple outbreaks at a site, the relationship between R0 and IAR is more complicated (Fig. S1, Fig. S2 ). We observed that as the time since the last outbreak and the value of R0 increase, the average values of IAR tend to increase (Fig. S1 ). This relationship is similar to what we observed when we simulated data on a single outbreak. However, as the time between the past outbreaks increases, the average values of IAR tend to decrease ( To evaluate the performance of our model for estimating parameters and predicting IAR, we applied it to simulated data on age-stratified seroprevalence with predefined parameter values of R0 = 1.5, S0(0) = 1, and outbreak times of 5, 45, and 85 years before the serological study. Our reconstruction of age-stratified seroprevalence, as expected, exhibited three step-like increases corresponding to three past outbreaks (Fig. S4) . The model probabilities for the five models placed the most weight on the model with three outbreaks (94.8%, Fig. S3 ). Overall, the uncertainty intervals of our estimates covered the true values well (Fig. S3) . Furthermore, the ensemble posterior prediction from the five models produced a good prediction to the agestratified serological data (Fig. S4) . We did assess our model performance on simulated data with more outbreaks (not presented here) but found out that with R0 = 1.5, more than three outbreaks within 100 years depleted enough susceptibles that the final outbreak sizes became minimal and unable to be inferred. To assess how well the inference framework performs on less ideal data, we carried out a sensitivity analysis in which we reduced the sample size or increased the size of the age bins is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 25, 2022. ; https://doi.org/10.1101/2022.05.23.22275466 doi: medRxiv preprint outbreaks may not be strictly necessary if population susceptibility can still be ascertained. To explore this possibility, we compared IAR values projected using true parameter values versus parameter values inferred from different simulated data sets. When we used data sets with smaller sample sizes and larger age bins, we found that the projected IAR values were comparable to the true IAR values in terms of their central tendency, but more uncertain in comparison to the more detailed simulated data set (Fig. S9 ). Applying the model to published serosurveys We applied our framework to age-stratified seroprevalence data identified through a literature review [42] . In total, we used data from 18 studies consisting of 35 site-level datasets, given that some studies reported data from multiple sites. A detailed description of each study is available in the Supplementary Text and the colated data can be found in [dataset] [30]. The study sites were labeled alphabetically from A to R for convenience. All of the model runs converged according to the potential scale reduction statistic Ȓ. Estimates of R0 for all studies were within the range of 1 to 3 (Fig. 3) , which is typical for chikungunya [41] . Most sites had model weights for the one-outbreak model of >90%, showing strong support for a one-outbreak model. However, several sites showed more support for a model with two outbreaks (Fig. 3; sites A, H1 , J, I2, I4, I5, I7, I8, L2), and one site in the Philippines (M) appears to have had five outbreaks (model weight of five-outbreaks model is 99.5%). Estimates of S0(0) for most sites were highly uncertain, which is to be expected given that most information in the data about susceptibility pertained to the period in which study participants were alive, rather than before then. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 25, 2022. ; https://doi.org/10.1101/2022.05.23.22275466 doi: medRxiv preprint Utilizing the estimated parameters, we computed the population immunity of the population in 2022 before and after a hypothetical outbreak (Fig. S11) , and subsequently projected the expected number of endpoint events for every site. The numbers of endpoint events were projected to be highly variable across sites. Only two sites (L5 and E) were projected to have the 2.5 quantile of expected endpoint events higher than ten events per thousand study participants (Fig. 4) . These two sites either had higher R0 (site E) or higher susceptibility in the population before the hypothetical outbreak (site L5) (Fig. 4) . When susceptibility was high, increasing R0 increased the expected number of events. However, when susceptibility was low, a much higher R0 value was required to result in an equivalent number of events, as observed for site E (Fig. 4) . A positive interaction between R0 and susceptibility was also observed when we inspected the relationship between the time since the last estimated outbreak and the number of events (Fig. 5) . With high enough R0, site E is expected to still obtain a higher number of events despite more recently experiencing an outbreak. represents studies that have the 95% PPI cross or below the value of 10 per 1000 people. We annotated study sites that can confidently produce more than 10 endpoints per thousand people. Because historical outbreaks can affect the population immunity of older and younger generations differently, we also stratified our results into two age groups-children (<18 years old) and adults (≥18 years old)-and compared results between them. Adults generally have experienced more outbreaks in the past than children; hence, susceptible proportions before hypothetical outbreaks in 2022 are expected to be higher in children. This results in a higher expected number of events in children than adults (Fig. 6 ). (Fig. 7A) , the lower seropositive proportion predicted in 1973, and most importantly the lower estimate of R0, led to a lower projection of the seropositive proportion in older age groups in 2012 and 2013 (Fig. 7B, 7C) . The model runs that included data from all years resulted in a good fit for the serological data in 2012 and 2013 (Fig 7E and 7F ). However, the higher R0 estimate from this model run resulted in a poorer fit to the 1973 serological data (Fig. 7D) . Consistent with these results, the IAR projections right after the five previous outbreaks in the site using only the 1973 dataset were also lower (Fig. 8) . This reflects that there may be variability in the R0 over time that our model does not capture. We also compared our estimates of the timing of past outbreaks to the timing of reported outbreaks (Table S1 ) not used to inform our estimates (Fig. 9) . Overall, most of our estimates match the timing of reported outbreaks. However, the model performed less well for studies A (Singapore) and K (Kerala, India). This is due to the fact that these studies only sampled patients older than 18 years for study A and older than 14 years for study K. Missing information from younger age groups may result in the model picking up the signal from unreported earlier outbreaks instead. Interestingly, data sets from study R (Bolivia) and D (Benin) also apply to the same age bins, yet our model still captured the timing of those reported outbreaks within its credible intervals, though with a high uncertainty. This discrepancy is because these sites may have had only one recent outbreak happen at any time between when the oldest of the missing reported person was born and the time the study was conducted. Our model cannot pinpoint exactly the timing of these outbreaks from these missing ranges, thus explaining the high uncertainty in the estimates. The studies from I1 to I8 were conducted in different places in Kenya; hence, there may be some variation in the estimates of the timing of past outbreaks. In this study, we constructed a framework to make projections of expected endpoints for future vaccine trials based on serological data for a candidate trial site. The framework employs the historical transmission information imprinted in serological data to infer the susceptibility of the population and a time-averaged R0 from previous outbreaks. These quantities can then be used to make projections of IAR for future outbreaks. We first applied this framework to is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 25, 2022. ; https://doi.org/10.1101/2022.05.23.22275466 doi: medRxiv preprint simulated data to assess its performance and then applied it to published serosurveys to make inferences about hypothetical future trials. We propose that this framework is suitable for planning a vaccine trial conducted during an outbreak and is independent of any particular trial design, as all trials need endpoint events to be successful. Our results suggest that, in most cases, outbreaks may be minimal if the time since the last outbreak is 10 years or less, even at relatively high R0 values. This is because it takes time for the susceptible population to build up through births to a sufficient degree for an outbreak to happen. This fits with previous reports of CHIKV outbreaks in some places, where it may take 10 or more years to re-emerge in a given population [43] [44] [45] . Similar expectations have been described for Zika, another arbovirus with similar epidemiology [46] . When applying this framework to published serological data, our estimates of the timing of past outbreaks generally agree with the timing of reported outbreaks at the same site. Although estimates at some sites were not as good due to missing samples in younger age groups, this can be noted and improved by collecting data for all ages in preparation for future vaccine trials. Most of the sites we analyzed here appear to have experienced only a single outbreak within the lifetimes of study participants, which is consistent with the idea that the occurrence of chikungunya outbreaks is sporadic. We inferred that the site in Cebu City, Philippines appeared to have experienced multiple outbreaks, a conclusion reached by a previous analysis using the same data set but using a slightly different method [28] . Based on a WHO consultation for future phase-III chikungunya vaccine trials [37] , future clinical trials to assess vaccine efficacy need to be conducted across multiple sites and multiple countries due to the unpredictable nature of CHIKV outbreaks [37] . These sites also need to be chosen prospectively in a way that they accrue enough endpoint events that the trial has enough power to detect a desired efficacy. Our framework can be used to support this objective by offering a basis for determining which sites are expected to generate a larger number of 23 385 386 387 endpoint events, in the event that an outbreak occurs. Based on the aforementioned WHO consultation, 62 endpoint events are needed to have 90% power to reject the null hypothesis that the vaccine efficacy is less than 30% when the true efficacy is 70% [37] . Our analysis identified two sites that alone could potentially accrue the required 62 endpoint events with fewer than 6,200 study participants if an outbreak were to happen. One additional suggestion from our framework is to potentially target enrollment on individuals who were born after past outbreaks, given that they would be more likely to be susceptible. Our approach addresses only one criterion for site selection: the expected incidence of endpoint events. Conducting a vaccine trial during an outbreak poses other challenges that the site selection process needs to take into account. A suitable site should also have trained personnel, regulatory oversight, and testing facilities. There are also questions on how many sites are needed and allocation of effort across sites. To answer this, a balance between the expected incidence of endpoint events and limited resources is necessary. Many sites across multiple countries also require considerable logistical effort to communicate among sites. These challenges have been noted in a previous Ebola vaccine trial [47] . Ultimately, selecting suitable sites for chikungunya vaccine efficacy trials is a multifaceted problem and the modeling framework we proposed here can act as a screening tool so that the sites it recommends can be further scrutinized according to other criteria, as well. Our study has some limitations. For an arbovirus like CHIKV, heterogeneity plays an important role in transmission, which could affect the relationship between R0 and the IAR [48, 49] . However, estimation of R0 based on observed IAR values is not the end goal of our method. Rather, the goal here is to having a model that is capable of using past values of IAR and an estimate of R0 to predict future IAR values. Thus, the omission of heterogeneity may lead to an inaccurate estimate of R0 but does not necessarily compromise our main goal to project future IAR [48, 49] . Because we assume that R0 for every outbreak is the same, our framework will not perform as well in situations in which there is variation in transmission between outbreaks [50] . This could partly explain why the R0 estimates in the validation for the Philippines site are different and lead to different projections of IAR. Analyzing seroprevalence data from too far back could be a potential problem since we may miss recent outbreaks. This could be remedied by conducting pilot seroprevalence studies, which is usually done before the vaccine trial anyway. Another limitation of our analysis is that the symptomatic proportion is sampled from a distribution derived from a widely cited asymptomatic proportion ranging 3 -28% [38] [39] [40] . However, there is growing evidence that the asymptomatic proportion of CHIKV may be lineage dependent [51] . This could easily be modified in our framework when the circulating lineage in the selected site for the trial is known, or at least suspected. It is also worth noting that our framework does not attempt to suggest when and where future outbreaks will occur, but instead to assess how feasible a vaccine trial would be if an outbreak occurs. Predicting the occurrence and timing of future chikungunya outbreaks remains a challenge and is beyond the scope of this study. In conclusion, we have established a comprehensive framework that can estimate the number of expected endpoint events to inform future chikungunya vaccine efficacy trials using age-stratified serological data. Our analysis shows that this framework is robust when applied to simulated data but can also provide insights when applied to published data. In addition to chikungunya, this framework could also be applied to other emerging viruses with similar epidemiological dynamics. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 25, 2022. ; https://doi.org/10.1101/2022.05.23.22275466 doi: medRxiv preprint Global distribution and environmental suitability for chikungunya virus An epidemic of virus disease in Southern Province Haemorrhagic fever in Calcutta area Chikungunya outbreak in the Caribbean region Epidemiology of Chikungunya in the Americas Post-Epidemic Chikungunya Disease on Reunion Island: Course of Rheumatic Manifestations and Associated Factors over a 15-Month Period Chikungunya virus infection. A retrospective study of 107 cases Rheumatoid arthritic syndrome after chikungunya fever Clinical progression of chikungunya fever during acute and chronic arthritic stages and the changes in joint morphology as revealed by imaging Impact of Chikungunya virus infection on health status and quality of life: a retrospective cohort study Is Dengue Vector Control Deficient in Effectiveness or Evidence?: Systematic Review and Meta-analysis Reflections from an old Queenslander: can rear and release strategies be the next great era of vector control? Optimizing the deployment of ultra-low volume and targeted indoor residual spraying for dengue outbreak response Chikungunya as a paradigm for emerging viral diseases: Evaluating disease impact and hurdles to vaccine development Valneva Successfully Completes Pivotal Phase 3 Trial of Single-Shot Chikungunya Vaccine Candidate -Valneva n A Phase 3 Safety, Immunogenicity, and Lot-Consistency Trial of the VLP-Based Chikungunya Vaccine PXVX0317 in Healthy Adults and Adolescents. clinicaltrials.gov World Health Organization. Emergency Use Listing Procedure World Health Organization. WHO Operational Tool for efficient and effective lot release of SARS-CoV-2 (Covid-19) vaccines Considerations for the design of vaccine efficacy trials during public health emergencies The global epidemiology of chikungunya from 1999 to 2020: A systematic literature review to inform the development and introduction of vaccines Simulations for designing and interpreting intervention trials in infectious diseases Preliminary results of models to predict areas in the Americas with increased likelihood of Zika virus transmission in 2017 Using simulated infectious disease outbreaks to guide the design of individually randomized vaccine trials Using simulation to aid trial design: Ring-vaccination trials Co-circulation and misdiagnosis led to underestimation of the 2015-2017 Zika epidemic in the Americas Evidence of previous but not current transmission of chikungunya virus in southern and central Vietnam: Results from a systematic review and a seroprevalence study in four locations Reconstruction of 60 Years of Chikungunya Epidemiology in the Philippines Demonstrates Episodic and Focal Transmission R: A Language and Environment for Statistical Computing World Population Prospects: Data Booklet -2015 Revision How Modelling Can Enhance the Analysis of Imperfect Epidemic Data BayesianTools: General-Purpose MCMC and SMC Samplers and Tools for Bayesian Statistics Estimating Drivers of Autochthonous Transmission of Chikungunya Virus in its Invasion of the Americas CODA: Convergence Diagnosis and Output Analysis for MCMC R package rjmcmc : reversible jump MCMC using post-processing World Health Organization. WHO consultation on Chikungunya vaccine evaluation Yellow Book | Travelers' Health | CDC n Chikungunya virus: epidemiology, replication, disease mechanisms, and prospective intervention strategies Chikungunya Fever: An Epidemiological Review of a Re-Emerging Infectious Disease Reviewing estimates of the basic reproduction number for dengue, Zika and chikungunya across global climate zones Current challenges and implications for dengue, chikungunya and Zika seroprevalence studies worldwide: A scoping review Chikungunya resurgence in the Maldives and risk for importation via tourists to Europe in 2019-2020: A GeoSentinel case series Human and entomologic investigations of chikungunya outbreak in Mandera, Northeastern Kenya Tracking the reemergence of epidemic chikungunya virus in Indonesia Impact of age-specific immunity on the timing and burden of the next Zika virus outbreak Operationalizing International Regulatory Standards in a Limited-Resource Setting During an Epidemic: The Sierra Leone Trial to Introduce a Vaccine Against Ebola (STRIVE) A Note on the Derivation of Epidemic Final Sizes Edge-based compartmental modelling for infectious disease spread Inter-annual variation in seasonal dengue epidemics driven by multiple interacting factors in Guangzhou, China Conflict of interest TAP receives consulting fees and research support from Emergent Biosolutions for work on chikungunya vaccine trial planning. The study described in this manuscript was completed prior to TAP's relationship with Emergent.