The authors have declared that no competing interests exist.
Current address: Centre for Tropical Medicine and Global Health, Nuffield Department of Medicine, University of Oxford, Oxford, United Kingdom
Current address: Oxford University Clinical Research Unit, Ho Chi Minh City, Vietnam
Current address: Department of Biology, University of Florida, Gainesville, FL, United States of America
Current address: Emerging Pathogens Institute, University of Florida, Gainesville, FL, United States of America
Dengue is an important vector-borne pathogen found across much of the world. Many factors complicate our understanding of the relationship between infection with one of the four dengue virus serotypes, and the observed incidence of disease. One of the factors is a large proportion of infections appear to result in no or few symptoms, while others result in severe infections. Estimates of the proportion of infections that result in no symptoms (inapparent) vary widely from 8% to 100%, depending on study and setting. To investigate the sources of variation of these estimates, we used a flexible framework to combine data from multiple cohort studies and cluster studies (follow-up around index cases). Building on previous observations that the immune status of individuals affects their probability of apparent disease, we estimated the probability of apparent disease among individuals with different exposure histories. In cohort studies mostly assessing infection in children, we estimated the proportion of infections that are apparent as 0.18 (95% Credible Interval, CI: 0.16, 0.20) for primary infections, 0.13 (95% CI: 0.05, 0.17) for individuals infected in the year following a first infection (cross-immune period), and 0.41 (95% CI: 0.36, 0.45) for those experiencing secondary infections after this first year. Estimates of the proportion of infections that are apparent from cluster studies were slightly higher than those from cohort studies for both primary and secondary infections, 0.22 (95% CI: 0.15, 0.29) and 0.57 (95% CI: 0.49, 0.68) respectively. We attempted to estimate the apparent proportion by serotype, but current published data were too limited to distinguish the presence or absence of serotype-specific differences. These estimates are critical for understanding dengue epidemiology. Most dengue data come from passive surveillance systems which not only miss most infections because they are asymptomatic and often underreported, but will also vary in sensitivity over time due to the interaction between previous incidence and the symptomatic proportion, as shown here. Nonetheless the underlying incidence of infection is critical to understanding susceptibility of the population and estimating the true burden of disease, key factors for effectively targeting interventions. The estimates shown here help clarify the link between past infection, observed disease, and current transmission intensity.
Dengue disease severity is known to vary widely from the very severe to asymptomatic. There is a wide range of estimates of how many infections result in each of these outcomes. It is known that after a first infection the outcome of a second infection with a different serotype varies over time, but this has not been taken into account in these previous estimates. In this paper, we use modelling methods, combined with information from published dengue research in which individuals are followed over time, to estimate the proportion of infections that result in symptoms at different times after infection. We estimated the proportion of infections that are symptomatic for first infections as 0.18 (95% Credible Interval, CI: 0.16, 0.20), 0.13 (95% CI: 0.05, 0.17) for individuals infected in the year following a first infection and 0.41 (95% CI: 0.36, 0.45) for those experiencing secondary infections after this first year. The estimates here will help understand how cases relate to underlying transmission, which is vital for understanding how much of the population are susceptible to infection and for effectively targeting interventions.
Dengue is an important vector-borne disease found across much of the world [
Previous estimates of the proportion of dengue infections that are apparent have come from cohort studies and cluster studies. Cohort studies follow the same individuals over time, usually recording antibody titres at consistent intervals (months to years), as well as recording whether individuals experienced a symptomatic dengue infection in these intervals. Cluster studies focus on testing individuals living in close proximity to known dengue cases and recording whether those individuals have experienced disease. An asymptomatic or inapparent infection is usually defined as a substantial rise in antibody titres between two measurements in a participant not experiencing symptoms. Symptomatic or apparent infections are infections concurrent with compatible symptoms, with the infection usually virologically confirmed. A recent review found that inapparent proportion estimates varied from 8–100% across relevant studies [
We performed a search in PubMed with the terms dengue cohort and dengue cluster study. We also searched the references of a recent review of dengue inapparent infections [
HI: Haemagglutination inhibition, PRNT: Plaque reduction neutralisation titre, ELISA: Enzyme linked immunosorbent assay.
Study | Study type | Age group (yrs) | Inapparent infection identification | Apparent case identification | Serotype data in paper | Analysis |
---|---|---|---|---|---|---|
Philippines[ |
Cohort | 0.5–85 | HI: 4-fold increase | Fever AND RT-PCR: IgM positive or 4-fold IgG increase | Symptomatic infections only | A |
Brazil, Colombia, Puerto Rico (A) and Mexico [ |
Cohort | 9–16 | ELISA: IgG seroconversion (primary only) | 2 days fever AND RT-PCR: positive OR ELISA: IgM positive or 4-fold IgG increase | None | A |
Nicaragua [ |
Cohort | 2–9 | HI: 4-fold increase | Fever AND RT-PCR: positive OR ELISA: IgM positive or 4-fold inhibition increase | Apparent infections only | A, B, D |
Sri Lanka [ |
Cohort | 0–12 | ELISA: IgG seroconversion (primary only), PRNT: 2-fold increase (secondary) | Fever AND RT-PCR: positive OR ELISA: IgM positive or 4-fold IgG increase | Apparent infections only | A, D |
Peru [ |
Cohort | 0–75 | PRNT: seroconversion | Fever and one other dengue symptom AND RT-PCR: positive OR ELISA: 4-fold IgM increase | All infections | A, D |
Vietnam [ |
Cohort | 2–15 | ELISA: IgG seroconversion (primary only) | Fever and suspected dengue or viral disease AND RT-PCR: positive OR ELISA: IgM positive or 4-fold IgG increase | Apparent infections only | A, B, D |
Thailand (A) (Bangkok) [ |
Cohort | 4–16 | HI or PRNT: seroconversion (primary) or 4-fold increase (secondary) | 2 day school absence for fever AND ELISA: 4-fold IgM increase OR HI or PRNT: seroconversion (primary) or 4-fold increase (secondary) | Apparent infections only | A |
Thailand (B) (Kamphaeng Phet) [ |
Cohort | 4–15 | HI: 4-fold increase AND PRNT: 4-fold increase | 2 day school absence OR fever AND ELISA: IgM positive or 4-fold IgG increase | Apparent infections only | A, B, D |
Puerto Rico (B) [ |
Cohort | 10–18 | PRNT: 4-fold increase | Fever AND RT-PCR: positive OR ELISA: IgM positive | Apparent infections only | A |
Thailand (B) [ |
Cluster | 1–15 | ELISA: IgM positive OR 4-fold IgG increase | Any symptoms AND ELISA: IgM positive or 4-fold IgG increase | Index case only | C |
Vietnam [ |
Cluster | 5–55 | ELISA: seroconversion OR RT-PCR: positive OR NS1: positive | Fever AND ELISA: seroconversion OR RT-PCR: positive OR NS1: positive | Index case only | C |
Nicaragua [ |
Cluster | 2–60+ | ELISA: seroconversion HI: 4-fold increase | WHO definition of DF or undifferentiated fever AND ELISA: IgM seroconversion OR HI: 4-fold increase | All infections | C |
Indonesia [ |
Cluster | 9–55 | RT-PCR: positive HI: 4-fold increase | Fever AND RT-PCR: positive OR ELISA: seroconversion | All infections | C |
We made several key assumptions about the infection risk and the risk of symptomatic disease in order to formulate our model. First, we assumed that for a given study (j) and time period (i), the infection risk (ρi, j) was equal for immunologically naïve individuals and for individuals with a previous infection. We then assumed that the proportion of infections resulting in symptomatic disease (γ) for each infection group (primary: γ’, or secondary: γ”) was equivalent across all time periods and studies (though we also made study-specific estimates). We then used study data to identify for each study and year: (1) the number of subjects with no previous dengue exposure (Nnaïve, i, j) who were susceptible to primary infection (indicated as IgG negative), (2) the number of subjects with previous dengue exposure (Nprev, i, j) (IgG positive), (3) the number of inapparent infections (seroconversion), O’inapp, i, j and O”inapp, i, j, for primary and secondary infections, respectively, and (4) the number of symptomatic infections (acute seroconversion or detection of viral RNA), O’app, i, j and O”app, i, j for primary and secondary infections, respectively. For each class of observation (O), we assumed that number of observed infections came from a binomial distribution with the respective population of each group from that the study, N, and a year and location-specific probability of infection (ρi, j) and the group-specific probability of having apparent disease (γ’ or γ”):
We extended the model to include a period of possible altered immunity (cross-immunity) in the year following infection. This was modelled by including a third susceptibility group Nrec,i,j, which was individuals who had experienced an infection in the preceding cohort year, and allowing the model to fit a different probability of apparent infection during this period (γrec,i,j).
Each model was fit to the data in a Bayesian framework using rStan [
The PubMed search for “dengue cohort” returned 357 papers. The references for Grange et al. added an additional 3 papers and 2 more were discovered through personal communication. Of these, 9 papers on 12 different cohorts contained sufficient published information on both apparent and inapparent infections in the same cohort. The search for “dengue cluster study” returned 180 papers, with 4 papers on 4 different cluster studies containing enough information on the number of non-index cases in the cluster during the follow up period.
We present the results of four analyses in this manuscript. For Analysis A, we estimated the apparent proportion without controlling for serotype or temporary cross-protection for each of 12 studies individually and for all 12 studies together. For Analysis B, we used multi-year cohort studies (3) to estimate apparent proportions for secondary infection in two different groups: (1) individuals with recent primary infection (the previous year), who may experience cross-protection; and (2) individuals with more distant primary exposure (more than one year). For Analysis C, we estimated the apparent proportion using only cluster studies (4 studies). Finally using the cohort studies, for Analysis D, we estimated serotype specific apparent proportions for primary and secondary infections (5 studies). The final column of
For each individual cohort study, the estimated apparent proportion for secondary infection was close to or slightly higher than the estimate for primary infection (
Probability densities of estimates for the apparent proportion in primary (i) and secondary (ii) infection for each study (Analysis A for each study separately).
In the analysis including all 12 cohort studies with shared parameters for the proportion of infections experiencing disease and different local infection risks, we estimated an overall apparent proportion that was significantly higher for secondary infections (0.24, 95% Credible Interval, CI: 0.22, 0.26) than for primary infections (0.18, 95% CI: 0.16, 0.19) (
(i) Probability densities of estimates of the apparent proportion in primary and secondary infections from cohort studies (Analysis A). (ii) Probability densities of estimates including a period of cross-immunity (Analysis D). (iii) Probability densities of estimates from cluster studies (Analysis C). For (i) and (iii) estimates for primary infection shown in green, secondary infection in orange and for (ii) estimates for primary infection shown in green, secondary infections in the year after infection in brown and secondary infections in the subsequent years in orange.
The modelling framework incorporated local data on primary and secondary cases with apparent and inapparent infection as well as global parameters for the probability of apparent disease in primary and secondary infection to estimate transmission intensity (
The yearly probability of infection in the cohort studies for each study year (from Analysis A all studies together).
Three studies had data collected across multiple years and therefore sufficient information to estimate the impact of short-term cross-protective immunity. The estimated apparent proportion for primary infections among these studies was similar to Analysis A, 0.18 (95% CI: 0.16, 0.20) (
We then estimated the apparent proportion using data exclusively from the four cluster studies that included sufficient data. The primary infection estimate of 0.22 (95% CI: 0.15, 0.29) from these data was similar to that from the cohort studies (Analyses A and B). However, the secondary infection estimate of 0.57 (95% CI: 0.49, 0.68) was significantly higher than the general estimate for all cohorts (Analysis A) and closer to the estimates for secondary infections more than one year after primary infection (Analysis B) (
The probability of infection in the time of follow up for those in the cluster around an index case (Analysis C).
Finally, we estimated serotype-specific apparent proportions for the five cohort studies that had sufficient data available (
Probability densities of estimates for the apparent proportion in primary (i) and secondary (ii) infection across serotypes (Analysis B).
We developed a statistical framework to assess the proportion of dengue virus infections that result in apparent disease and how that proportion depends on immune status. This framework allowed us to assess this proportion across different geographical areas, study types, and transmission intensities. The most comprehensive data, from multi-year cohort studies predominantly of children, showed that approximately 18% of primary infections experienced apparent disease. This proportion remained low, approximately 13%, for infections in the year following, but then increased substantially for secondary infections beyond the first year to approximately 45%. The other data analysed substantiated these differences: in individual cohort studies, secondary infections tended to have higher apparent proportions; across 12 cohort studies, the average estimated apparent proportion was significantly higher for secondary infections; and in cluster studies, the difference was even more pronounced.
This finding substantiates evidence from individual cohort studies that have shown that the apparent proportion varies by year and that this variation is related to the incidence of infection in the previous year [
Grange et al. [
Using data from cluster studies, we estimated higher apparent proportions for both primary and secondary infections. It is possible that cluster studies are better at capturing milder infections than cohort studies as there may be increased effort to identify illness and follow up tends to happen over a shorter time period when recall may be better. Indeed, Grange et al. [
The analytical approach used here enabled simultaneous estimation of multiple unknowns (infection risk and multiple apparent proportions) in a context of limited outcome information. We drew on data from different locations where infection history and risk may drive differences in apparent incidence. As suggested here, the apparent proportion may actually be quite similar across locations, despite possibly appearing different due to different histories of recent exposure. Combining data from multiple locations has great benefit in increasing overall sample size and generalizability and allowed us to disentangle the effects of each covariate. An additional benefit of this method is that we were able to estimate the probability of infection over the study period, as has been estimated before [
This analysis represents an important first step towards aggregating knowledge of dengue transmission and disease dynamics globally. This work could be further extended to estimate the contribution of other important factors such as age; available cohort data largely focuses on children. Additional data would also allow assessment of possible differences between serotypes. The current serotype-specific estimates were mainly derived from information from one study per serotype, limiting our ability assess the possibility of serotype-specific effects. Also, variations in the way infections were detected and confirmed (e.g. case definitions, follow-up methods, assay or change in titre) across studies could be better controlled for with the individual level data. These differences may contribute to the variability in the apparent proportion estimates for each study. We have also not explicitly addressed third and fourth infections, which likely also exhibit different patterns. Indeed, some of the infections considered as secondary infections here may have been third or fourth infections, as noted particularly in Peru [
Only by drawing on detailed data across multiple years and populations experiencing different infection histories were we able to make general inferences about the relationship between the proportion of infections that are apparent and immune status. These estimates show a clear difference in the apparent proportion between primary and secondary infections, and are helpful for understanding the relationship between the incidence of dengue virus infection and the incidence of disease. For example, even in the presence of an effective intervention, the incidence of disease may increase as there may be an increased delay between primary and secondary infection [
(PDF)