^{1}

^{2}

^{1}

^{3}

^{4}

^{*}

The authors have declared that no competing interests exist.

Conceived and designed the experiments: DLC SBH MEH IML. Performed the experiments: DLC SBH MEH IML. Analyzed the data: DLC SBH MEH IML. Contributed reagents/materials/analysis tools: DLC MEH IML. Wrote the paper: DLC SBH MEH IML.

Dengue is a mosquito-borne infectious disease that constitutes a growing global threat with the habitat expansion of its vectors

We developed an individual-level (including both humans and mosquitoes), stochastic simulation model for dengue transmission and control in a semi-rural area in Thailand. We calibrated the model to dengue serotype-specific infection, illness and hospitalization data from Thailand. Our simulations show that a realistic roll-out plan, starting with young children then covering progressively older individuals in following seasons, could reduce local transmission of dengue to low levels. Simulations indicate that this strategy could avert about 7,700 uncomplicated dengue fever cases and 220 dengue hospitalizations per 100,000 people at risk over a ten-year period.

Vaccination will have an important role in controlling dengue. According to our modeling results, children should be prioritized to receive vaccine, but adults will also need to be vaccinated if one wants to reduce community-wide dengue transmission to low levels.

An estimated 40% of the world's population is at risk of infection with dengue, a mosquito-borne disease that can lead to hospitalization or death. Dengue vaccines are currently being tested in clinical trials and at least one product will likely be available within a couple of years. Before widespread deployment, one should plan how best to use limited supplies of vaccine. We developed a mathematical model of dengue transmission in semi-rural Thailand to help evaluate different vaccination strategies. Our modeling results indicate that children should be prioritized to receive vaccine to reduce dengue-related morbidity, but adults will also need to be vaccinated if one wants to eliminate local dengue transmission. Dengue is a challenging disease to study because of its four interacting serotypes, seasonality of its transmission, and pre-existing immunity in a population. Models such as this one are useful coherent framework for synthesizing these complex issues and evaluating potential public health interventions such as mass vaccination.

Dengue is a mosquito-borne disease, caused by a flavivirus with four serotypes, responsible for an estimated 500,000 hospitalizations and 20,000 deaths per year, mostly in the tropics

Several dengue vaccine candidates are currently in development or in clinical trials

Here, we investigate the potential effectiveness of different dengue vaccination strategies using a model of dengue transmission in a Thai population. The individual-level stochastic model was developed to match the epidemiology of dengue in a population in semi-rural Thailand that has experienced hyperendemic dengue transmission for many years. We modeled both single-year campaigns, in which part of the population is vaccinated well before the dengue season, and multi-year roll-outs, in which young children are vaccinated first and progressively older individuals are vaccinated in subsequent years as part of a catch-up campaign.

We developed an agent-based model of dengue transmission. The model is described in detail in

(A) Natural history of dengue model. Susceptible individuals are infected by mosquitoes, and mosquitoes are infected by humans. (B) Population density of the 20 km by 30 km region surrounding Bang Phae, Thailand, at a 1 km^{2} resolution. Red indicates high population density, yellow and white for low density, as indicated in the legend in units of people per km^{2}. Population density data is from GRUMP

Secondary cases may have severe outcomes (i.e., DSS/DHF) at an age-specific proportion (

We describe the synthetic population created for the model in detail in

Within each square kilometer, individual households, schools, and workplaces are assigned random locations. Children of the appropriate age are sent to the elementary school (ages 5 to 10 years), lower secondary school (ages 11 to 14 years), or upper secondary school (15 to 17 years). People of the appropriate age are assigned workplaces according to a gravity model in which people tend to commute to locations that are nearby and have a relatively high population density. Workplaces have an average of 20 workers, who occupy the same location during the workday.

During the morning and evening hours, people are at home, and they may go to school or work during the rest of the day (

To simulate multi-year epidemics, we make two simplifying assumptions: 1) there is no correlation of prior exposure to dengue within households and 2) household structures do not change over time. After simulating a single year of dengue transmission, we “age” the population by setting the immune status (both prior infections and vaccination) of all individuals of age

In the model, individuals are assigned to have immunity from prior exposure to the four serotypes of dengue based on their age. The age-specific immune profile is based on two sources of data on the prevalence of serotypes in Thailand. Thailand's Ministry of Public Health releases an “Annual epidemiological surveillance report” that summarizes dengue serotype surveillance data. Reports from 2000–2009 are available at epid.moph.go.th, which we summarize in

We estimate the age-specific immunity to the four dengue serotypes in our model. We assume that the level of exposure to dengue each year was such that 11% of nave individuals would be infected, based on studies in nearby Vietnam

The data are from two sources: Thailand's Health Ministry from 2000–2009 (available at

For each of the years for which we have serotype prevalence estimates, we randomly selected 11% of the population who was alive in that year (i.e., was 0 years old or older) to be exposed to dengue, and for each individual simulated exposure to a single serotype drawn from that year's prevalence data. Individuals exposed to a serotype are considered to be permanently immune. For years before 1973, we performed the same procedure, except that we assumed that the serotype prevalence was the mean serotype prevalence from 1973–2009. The mean serotype prevalences are 9.8%, 14.6%, 7.5%, and 5.2% for DENV-1, DENV-2, DENV-3, and DENV-4, respectively. In other words, we assumed that there was a constant 11% exposure to dengue (sufficient to infect) for all individuals, regardless of age or immune status, and that exposure to a serotype at any point in an individual's past grants sterilizing immunity to that serotype. In other words, each person who is exposed to dengue each year is exposed to exactly one serotype of dengue, and he or she gains sterilizing immunity to that serotype if he or she was not already immune from prior exposure.

Because the four serotypes have different symptomatic fractions, surveillance data give a skewed representation of the number of individuals infected by each serotype. We re-scaled the number of cases for each of the four serotypes in the historical data as described in

Simulated pre-existing immunity to the dengue serotypes in Bang Phae by age. We assume constant relative serotype prevalence before 1973, which corresponds to the vertical dotted line. Surveillance data is scaled as described in

We simulated a single year of dengue transmission in Ratchaburi, Thailand (

Each plot shows the daily number of newly infected and symptomatic people from a single representative stochastic simulation. (A) A simulation in which no vaccination took place (baseline scenario). (B) A randomly selected 70% of the population aged 2 to 14 years was vaccinated. (C) A randomly selected 70% of the population aged 2 to 46 years was vaccinated. The vaccine confers protection to 70% of vaccinees in the model.

The simulated dengue season produced a 5% infection attack rate with some stochastic variation among runs (

Plotted are the average age-specific (A) infection incidence, (B) dengue fever incidence, and (C) hospitalized DSS/DHF incidence per year from fifty stochastic simulations and aggregated in 5-year age brackets.

pre-vaccination | infected per | cases per | hosp per | vaccinated | hosp averted |

100,000 | 100,000 | 100,000 | per 100,000 | per 100,000 vacs | |

0% | 5,027 |
1,691 |
38.9 |
— | — |

Ages 2–14, 30% | 3,796 |
1,376 |
27.9 |
6,673 | 165 |

50% | 3,191 |
1,198 |
22.6 |
11,121 | 147 |

70% | 2,615 |
996 |
17.9 |
15,570 | 135 |

Ages 2–46, 30% | 2,353 |
816 |
17.6 |
22,303 | 96 |

50% | 1,434 |
493 |
10.5 |
37,172 | 76 |

70% | 904 |
316 |
6.5 |
52,040 | 62 |

We report the total number of uncomplicated and severe (DSS/DHF) cases produced by our model assuming perfect surveillance. Estimates of reporting rates would be needed to compare our modeling results with actual surveillance data. Wichmann et al. estimated that, among children, total dengue cases in Thailand may be underreported by a factor of 8.7 and severe (inpatient) dengue cases by 2.6, with less underreporting in school-aged children than in younger children

During the simulated seasonal peak of dengue transmission, a single person infected an average of 1.9 to 2.3 others, depending on the serotype (

We simulated vaccinating the population to protect them before a single dengue season. Recently, an observer-blind, randomized, controlled, phase 2b vaccine trial was conducted with a tetravalent dengue vaccine

Vaccinating 70% of children 2 to 14 years old would reduce the number of dengue infections by 48%, uncomplicated dengue fever cases by 41%, and severe dengue cases (DSS/DHF) by 54% in a single year (

Those who are

Certain age groups could be prioritized to receive vaccine. Younger people have the least prior exposure, so they would be the most likely to become infected with and transmit dengue. Simulations demonstrated that vaccinating children (2–14 years old) would reduce dengue infections in the total population more than using the same number of doses to cover both children and adults (2–46 years old) (

The larger points represent the median attack rates (y-axis) of ten stochastic simulations run when a percentage of the total population (x-axis) is pre-vaccinated. The results from the individual simulations are plotted as small points to show the stochastic variation. The black Os, connected by lines, represent the effect of pre-vaccinating different fractions of individuals from 2 to 46 years old, from 0 to 80% of this cohort, which translates to 0 to 59.5% of the total population. The other symbols are the results from targeting 70% of individuals in narrower age cohorts (expressed in years in the legends) for pre-vaccination. Pre-vaccinating a particular age group can be considered more efficient than vaccinating an equivalent number of people from ages 2 to 46 if it results in a lower attack rate (i.e., the point falls below the line). The vaccine confers immunity to 70% of those vaccinated in the model. (A) Overall infection attack rate vs. pre-vaccination fraction. (B) Overall symptomatic (uncomplicated dengue fever) attack rate vs. pre-vaccination fraction. (C) Overall DSS/DHF cases (hospitalizations) vs. pre-vaccination fraction.

Due to limited vaccine availability and the logistics of mass vaccination programs, dengue vaccine will probably be deployed in multi-year vaccine roll-out campaigns

(A) No vaccination. (B) Vaccine roll-out that covers only children ages 2 to 14 years. (C) Vaccine roll-out that covers children and adults ages 2 to 46 years. 70% of each age cohort is vaccinated, and the vaccine confers all-or-none protection to 70% of vaccinees. The points indicate the number of newly infected or symptomatic people during a single day in a single representative stochastic simulation.

We also simulated a vaccine roll-out that extended the catch-up to include adults up to age 46. This roll-out targets the same age groups for the first three years as the previously described roll-out, but after this point both 2-year-olds and the youngest four unvaccinated age cohorts are vaccinated, as shown in

Baseline | Roll-out ages 2–14 | Roll-out ages 2–46 | ||||

year | cum cases per | cum hosp per | cum cases averted | cum hosp averted | cum cases averted | cum hosp averted |

100,000 | 100,000 | per 100,000 | per 100,000 | per 100,000 | per 100,000 | |

1 | 1,706 |
40 |
— | — | — | — |

2 | 3,424 |
80 |
225 | 6 | 225 | 6 |

3 | 5,128 |
119 |
689 | 22 | 689 | 22 |

4 | 6,837 |
159 |
1,367 | 43 | 1,367 | 43 |

5 | 8,516 |
198 |
2,061 | 65 | 2,186 | 68 |

6 | 10,179 |
237 |
2,717 | 87 | 3,098 | 94 |

7 | 11,824 |
276 |
3,329 | 109 | 4,106 | 122 |

8 | 13,469 |
315 |
3,992 | 131 | 5,212 | 152 |

9 | 15,143 |
355 |
4,700 | 155 | 6,426 | 185 |

10 | 16,809 |
394 |
5,418 | 179 | 7,699 | 217 |

We used a dengue simulation model to estimate that vaccination of 50% of the population of rural Thailand could be sufficient to reduce local dengue transmission to low levels. Based on our modeling study, we conclude that at least 70% efficacy against infection for all serotypes is desirable if one wants to control dengue in a hyperendemic area, and a higher efficacy vaccine would require less careful targeting of vaccine to reduce community-wide transmission of dengue. We further showed that vaccinating children is the most efficient use of vaccine to reduce cases and hospitalizations, but control of dengue transmission would also require vaccinating adults. In addition, both vaccinated and unvaccinated people would receive protection from mass vaccination because of the considerable indirect effects of dengue vaccination. A vaccine that only protects against only three serotypes could lead to a significant reduction in overall vaccine effectiveness. Further work will be needed in order to understand how to use vaccines that may not protect against all four serotypes. Using a detailed model of dengue transmission allows one to explore strategies that target vaccines most efficiently.

To capture the complex interactions required to evaluate the effectiveness of mass vaccination with tetravalent dengue vaccines, the model includes vector population seasonality

An estimated 40% of the world's population is at risk of dengue infection

Large-scale vaccination campaigns would be both challenging and costly but could be more cost-effective than relying solely on vector control and other non-pharmaceutical interventions

(EPS)

(PDF)

(PDF)

(PDF)

(PDF)

(PDF)

(PDF)

We thank Aubree Gordon for helpful conversations. The authors wish to acknowledge Thailand's National Statistical Office for providing part of the population data that made this research possible.