Ethical approval
The seroprevalence study was conducted as part of the Paraguay Ministry of Health’s public health emergency response and, therefore, did not require ethical approval. The seroprevalence study has not been published (either in full or in part) elsewhere. Following a standard operating process, the Centro Nacional de Servicios de Sangre (CENSSA), with the signed authorization of the responsible director, provided all the donated blood samples used in this study. All blood samples were collected anonymously, with no identifiers that could link the sample to the original patient; there was no enrollment procedure, and the patient consent form was the standard used by the CENSSA. The biochemical analyses were conducted at the Central Public Health Laboratory of the Ministry of Health in Paraguay. The use of epidemiological surveillance data was authorized by the signature of the director of the DGVS (Dirección General de Vigilancia de la Salud). The mathematical modeling work was conducted on aggregate data and, therefore, did not constitute personally identifiable information.
Epidemic case data
We used weekly case data from September 2022 to September 2023. All suspected chikungunya cases were reported to the DGVS, Ministerio de Salud Pública y Bienestar Social de Paraguay. Approximately 60% of suspected cases undergo confirmatory testing through PCR, with an additional 23% tested using IgM ELISA. Suspected infection cases are defined as persons with sudden onset of fever and arthralgia or disabling arthritis unexplained by another medical condition. Probable infections are defined as any suspected case with a positive laboratory result for CHIKV (IgM ELISA) or an epidemiological link to a confirmed case. Confirmed infections are suspected or probable CHIKV cases with positive results in real-time reverse transcription followed by PCR or viral isolation tests12,24.
New seroprevalence study
We conducted a seroprevalence study using blood bank serum samples in four of the five subregions (defined by the Ministry of Health, ‘Ejes’) of Paraguay (Metropolitana, Centro Sur, Centro Este and Centro Norte). We were not able to obtain serum samples from the final subregion (Chaco). In each subregion, we worked with the local blood collection service. Samples were collected from 25 July 2023 to 23 August 2023 from persons aged 18–65 years attending a blood donation service, with a target sample size of 250 persons per axis. We used IgG Euroimmun ELISA kits to test the samples for evidence of IgG antibodies to CHIKV. Testing was conducted in the Laboratorio Central de Salud Pública, Paraguay. We adjusted our seroprevalence estimates to account for the 98.6% sensitivity and 98% specificity of the kits.
Statistical analyses
Age- and sex-specific probability of disease
We assumed that outside the capital subregion (Metropolitana), all individuals were susceptible before the outbreak. In the capital subregion, we assumed that 5% of individuals were seropositive before the outbreak, according to a household-based seroprevalence study in 1,000 individuals aged 5–65 years conducted in 2017 (personal communication, Paraguay Ministry of Public Health; Extended Data Fig. 4). Uncertainty on seroprevalence was accounted for by estimating the 95% CI around point estimates assuming a binomial distribution on samples tested. Seroincidence was calculated as the difference between the sensitivity-adjusted seroprevalence after the outbreak and the seroprevalence at baseline. Uncertainty from the seroprevalence estimates was propagated to the estimates on the underlying number of infections, the IFR and the probability of detection.
Given nonsignificant differences in seroprevalence estimates across sex and age groups, we assumed equal risk of exposure by age and estimated the age- and sex-specific number of infections by subregion using the attack rate and demographic data from the national census. We then estimated the probability of disease by dividing the total number of cases by age and sex strata by the estimated number of infections in the strata. We separately estimated the age- and sex-specific IFR by dividing the number of deaths in each age and sex strata by the estimated number of infections in the same strata (Extended Data Table 1). Uncertainty from the seroprevalence estimates was propagated to the estimates on the underlying number of infections, the IFR and the probability of detection.
Epidemic model
To characterize the chikungunya epidemic trajectory, we developed a compartmental SIR (susceptible–infected–removed) model in which the transmission rate (β) varies over time. We chose the SIR framework as it limits complexity and has been shown to outperform SEIR (susceptible–exposed–infected–removed) and other forms that include mosquito-specific compartments25,26. Weekly transmission rates were assumed to be independent and estimated as free parameters. We assumed that the generation time for CHIKV (defined as the average time interval between consecutive infections) was 2 weeks20.
We modeled the number of incident cases, assuming a negative binomial observation process. The likelihood of observing cobs(t) incident cases on week t given the expected number of infections iexp(t) for that week is given by the density of a negative binomial distribution:
$$P({c}_{\textrm{obs}}(t)|{i}_{\textrm{exp}}(t))={\mathrm{dNegBin}}({c}_{\textrm{obs}}(t),{i}_{\textrm{exp}}(t)\cdot \rho ,{\mathrm{shape}})$$
where ρ is the detection probability and ‘shape’ is the overdispersion parameter of the negative binomial distribution.
Within the same analytical framework using a joint likelihood approach, we incorporated the observed number of seropositive individuals in our seroprevalence study, assuming a binomial observation process. The likelihood of observing npos(t) positive samples out of ntot(t) on week t given the expected proportion of susceptible individuals in the population s(t) for that week is given by the density of a binomial distribution:
$$P({n}_{\textrm{pos}}(t)|{n}_{\textrm{tot}}(t),s(t))={\mathrm{dBin}}({n}_{\textrm{tot}}(t),1-s(t))$$
The number of infectious individuals at the start of the outbreak (first week of November 2022) was fixed at 100. The detection probability is the probability that the surveillance system detected an infected individual and was estimated as a free parameter. Parameters were estimated using a Markov chain Monte Carlo (MCMC) method with a Metropolis–Hastings algorithm. We used four chains of 25,000 iterations, including a burn-in phase of 10,000 iterations and sampling with uniform noninformative priors. Effective sample sizes and R-hat values were computed for each parameter (Extended Data Table 4).
Estimating the impact of the vaccine
Given the absence of a traditional phase 3 trial, we do not have appropriate estimates of IXCHIQ’s efficacy against infection or disease. Therefore, we held a key stakeholder meeting during which we discussed potential parameter values with experts from Gavi, the World Health Organization and academia. Following this meeting, we developed two main scenarios: (1) a conservative scenario in which the vaccine blocked disease only with an efficacy of 75% and (2) a scenario in which the vaccine also blocked infection at the same level. We also included sensitivity analyses with a higher level of protection of 98%, consistent with other live vaccines.
To model the potential effect of a vaccine had it been available during the outbreak, we adjusted our transmission model. We assumed that the vaccine was a ‘leaky’ vaccine, such that vaccinated individuals had a probability of acquiring disease or infection despite vaccination at a predefined vaccine efficacy level. The target population was individuals aged 12 years and older, consistent with the planned initial target age group of the IXCHIQ vaccine. We initially assumed a vaccine efficacy of 0.75, meaning that 75% of vaccinated individuals were protected against disease but not against infection. We assumed a delay of 2 weeks from vaccination to acquisition of immunity27,28. We considered a reactive rollout campaign that started in the first week of October 2022. This date represents the moment when the Ministry of Health officially reported the outbreak. We then assumed that it takes 3 months to reach a coverage of 40% of the target population, assuming a fixed weekly rate of vaccination.
We conducted a sensitivity analysis in which we (1) varied the level of vaccination coverage and (2) allowed for different delays between the start of the outbreak and the start of campaign deployment. We also included a sensitivity analysis wherein we assumed a vaccine efficacy of 98%.
For the second main scenario, we conducted a separate analysis wherein we assumed that the vaccine blocked onward transmission at 75% efficacy. In this scenario, we assumed that vaccinated individuals had a 75% reduction in the probability of infection. Those who nevertheless became infected (that is, breakthrough infections) had the same probability of disease as infected nonvaccinated individuals.
For each scenario, we measured the number of infections, cases and deaths averted, as well as the number of doses used. In this scenario, we assumed that vaccinated individuals had a 75% reduction in the probability of infection. Those who nevertheless became infected (that is, breakthrough infections) had the same probability of disease as infected nonvaccinated individuals.
Software version
R v.4.4.2 (‘Pile of Leaves’) was used to run the analysis presented here.
Ethics and inclusion statement
The project is the result of a close collaboration between the Universidad Nacional de Asunción, CENSSA and University of Cambridge. Researchers from Paraguay were integral in all stages of the research process, including study design, study implementation, data analysis and manuscript development. The first author and nine other coauthors are from Paraguay. The project was part of a knowledge exchange that was established quickly during the epidemic wherein a researcher from Cambridge (G.R.d.S.) initially worked in Paraguay to help integrate modeling efforts into the epidemic response. A researcher from Paraguay (P.E.P.-E.) then spent 2 months in Cambridge to be trained in mathematical modeling. The manuscript was developed taking existing local research into account. As there had been very little chikungunya in Paraguay historically, most of our regional understanding comes from Brazil.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
