The prospective cohort study that generated these data was co-administered by the University of Massachusetts and the Armed Forces Research Institute of Medical Sciences (AFRIMS) for the years 1998–2002. TE, AR, DL, and AN were involved in the design and conduct of the cohort study, for which all subjects provided written informed consent. ST and RG are among the lead scientists presently managing the Department of Virology at AFRIMS, where specimens and data from this study are maintained. The lead author, KA, used only preexisting demographic and laboratory data for the purposes of this retrospective analysis.

The cohort study was approved by the Human Use Review and Regulatory Agency of the Office of the Army Surgeon General, the Institutional Review Board of the University of the Massachusetts Medical School, and the Ethical Review Board of the Ministry of Public Health, Thailand. Secondary data analysis for the purpose of this publication was approved by the Institutional Review Board at Emory University.

Data were collected during a five-year, school-based, prospective cohort study for DENV infections in children in Northern Thailand. The study design and methods have been described previously [27], [28]. Briefly, the study was conducted in Kamphaeng Phet Province from 1998–2002. In January 1998, 2,214 children were recruited from grades 1 through 5 at twelve primary schools. New participants were enrolled from the 1^st grade class in January of each year and eligible participants were re-enrolled. The numbers of children enrolled at the start of the active surveillance period each year were 2,044 in 1998; 1,915 in 1999; 2,203 in 2000; 2,011 in 2001; and 1,759 in 2002. A total of 3,687 children were enrolled for at least one year during the study period, with an average of 2.9 years of follow-up per child. At enrollment, the children or their parents were asked about prior JEV vaccination and age of vaccination.

Serum samples were collected for dengue serology four times each year (January, June, August, and November). All serological and virological testing was performed at AFRIMS in Bangkok, Thailand. Active case surveillance of the participants was conducted from June 1 to November 1, with potential illnesses identified based on absence from school, visit to a school nurse or public health clinic, or admission to the hospital. Absent students were visited by village health workers and evaluated with a symptom questionnaire and an oral temperature. Acute blood samples were obtained for students with a history of fever within 7 days of fever onset, as were 14-day convalescent samples.

Acute and convalescent blood specimens from incident febrile illnesses were tested using immunoglobulin M (IgM) and G (IgG) enzyme immunoassays for DENV and JEV. Acute DENV infections were defined serologically as a DENV-specific IgM level ≥ 40 units and with DENV-IgM > JEV-IgM. The infecting DENV serotype was identified from acute blood specimens using serotype-specific reverse-transcriptase polymerase chain reaction (RT-PCR) or virus isolation. Symptomatic infections were defined as a documented history of febrile illness with virologic or serologic evidence of acute DENV infection. Charts of hospitalized children were independently reviewed and classified as DF or DHF and assigned a severity grade following WHO criteria [29]. If a child experienced a febrile DENV illness but did not meet the criteria for DHF, they were characterized as having DF.

The duration of illness was derived from home visit data and hospitalization records. For non-hospitalized illnesses, the duration of illness was the length of time from the date of school absence or clinic visit to the date of the last home visit at which the child exhibited symptoms consistent with DENV infection. For hospitalized illnesses, the duration of illness was the pre-hospitalization time plus the time in the hospital.

Routine specimens were tested for the presence of hemagglutination inhibiting (HI) antibodies against DENV-1 – DENV-4 and JEV using standard methods [30]. The reference virus strains were DENV-1 (Hawaii), DENV-2 (New Guinea ‘C’), DENV-3 (H87), DENV-4 (H241), and JEV (Nakayama). Asymptomatic DENV infections were defined as a four-fold or greater rise in hemagglutination inhibition (HI) titers for any of the four DENV serotypes between two consecutive routine serum samples and without a concurrent four-fold rise in JEV HI titers, or with a concurrent four-fold rise in JEV HI titers but with higher HI titers for any DENV serotype than for JEV.

Plaque reduction neutralization titers (PRNTs) were obtained for pre-infection samples using standard methods [31]. Briefly, LLC-MK2 cell monolayers were infected with DENV1 – DENV4 and JEV in the presence of serial dilutions of heat-inactivated patient plasma. The reference virus strains were: DENV-1 (16007), DENV-2 (16681), DENV-3 (16562), DENV-4 (1036), and JEV (Nakayama), for which the sources and passage histories have been previously described [32]. The lowest dilution of serum tested was 1∶10 (corresponding to a final dilution of 1∶20 when combined with an equal volume of DENV); the dilutions and titers reported herein refer to the initial serum dilution. The concentration of patient plasma that resulted in a 50% reduction in plaque formation was calculated using log probit regression. The reciprocal titer of this dilution was defined as the PRNT₅₀. A PRNT₅₀ <10 was defined as undetectable or ‘negative’ and a titer ≥10 as ‘positive.’ PRNT assays were performed for asymptomatic infections only if the child had not missed school during the observation period. This was done in an attempt to exclude missed symptomatic DENV illnesses that could have been misclassified as asymptomatic seroconversions.

Children were grouped into three categories of DENV immunity based upon their pre-infection DENV antibody profiles: DENV-naïve (pre-infection PRNT₅₀<10 for all DENV serotypes), DENV-monotypic (pre-infection PRNT₅₀ ≥10 for a single DENV serotype), and DENV-multitypic (pre-infection PRNT₅₀ ≥10 for two or more DENV serotypes). To attempt to discern cross-reactive immunity from serotype-specific immunity, we further stratified DENV-multitypics according to mean age of infection, with the logic that older children would have had more time to experience multiple DENV infections and generate multiple serotype-specific responses, while younger children with multitypic profiles would, on average, reflect more cross-reactivity. Preexisting JEV immunity was dichotomized as JEV-positive (pre-infection PRNT₅₀≥10) and JEV-negative (pre-infection PRNT₅₀<10).

By definition, acute phase blood samples were not available from children with asymptomatic DENV infections for direct detection of the infecting virus. Approximately one-quarter of symptomatically infected individuals were RT-PCR negative, likely because the acute specimen was drawn when the child was no longer viremic. Based upon evidence that DENV transmission in this community is highly clustered, we assumed that the serotypes detected among RT-PCR positive cases attending a given school during a given epidemic year were representative of the serotypes causing asymptomatic and RT-PCR negative symptomatic infections at that school [33], [34]. To minimize the possible biases inherent to this assumption, imputation of an individual's unknown infecting serotype using community-level data was performed only for children residing in communities that had a single serotype detected during that epidemic year. If more than one serotype was detected at a given school during a given year, the infecting serotype was left as missing.

The JEV vaccine used in Thailand and throughout Southeast Asia, JEVAX, is a formalin-inactivated, mouse brain-derived JEV vaccine. The first dose of JEVAX in Thailand is administered concurrently with the diphtheria/tetanus/pertussis vaccine at 18 months of age. A second dose is administered 1 to 4 weeks later and a third (booster) dose is administered at 30 to 36 months of age. The seroconversion rates with JEVAX in Thai children have been estimated to be 50% following primary vaccination, 64–80% following secondary vaccination, and 100% following receipt of the third dose [35]. Based upon these observations of suboptimal seroconversion rates with the two dose regimen, the third dose was added to the Expanded Program for Immunization (EPI) regimen in Thailand in 2000.

A Phase III trial of JEVAX was conducted in 1984 in Kamphaeng Phet, Thailand, which demonstrated favorable efficacy [8]. JEVAX began to be slowly incorporated into the EPI in Thailand in 1988 and became part of Kamphaeng Phet's EPI in 1992. Vaccination histories for the cohort study were obtained by self-report and subsequent verification with clinic or family documents was not possible. However, given the ages of the children enrolled (five to fourteen years of age) and the timing of the study (1998 to 2002), it is likely that the majority of children enrolled in the cohort study would have received two doses of JEV vaccine. Younger children would likely have been vaccinated as infants, as part of the EPI program, and older children would likely have been vaccinated during the large “catch-up" vaccination programs taking place in the province in the 1990s. The JEVAX vaccine strain in use during the 1990s in Thailand was the Nakayama strain, the country has since switched to the Beijing strain.

Bivariate analyses were performed using chi-square testing for categorical variables and ANOVA or nonparametric testing for continuous variables. Logistic regression models for symptomatic versus asymptomatic infection were constructed using SAS' GENMOD procedure, accounting for the clustering of observations by school. The best model was then chosen by backward regression. Analyses were performed using SAS software, version 8 (SAS Institute, Cary, NC), SPSS for Windows version 10.0 (SPSS Inc., Chicago, IL), and R version 2.10.1 (R Foundation for Statistical Computing, Vienna, Austria).

Children with JEV NAbs were more likely to experience symptomatic infection than children without JEV NAbs (57% versus 46%, p = 0.021 by χ² testing, Table 1). Of the 569 first detected DENV infections, 479 had pre-infection DENV and JEV neutralizing antibody titer information available for analysis (84.2%). 90% of the missing NAb data were associated with asymptomatic infections, for reasons described above.

There were no differences by age, gender, or school in the proportions of infections that were symptomatic. The proportion of infections that were symptomatic varied by year, from 26% in 2000 to 51% in 2001 (p = 0.032, by χ² testing across all strata of epidemic year). There were no differences in the proportion symptomatic by pre-infection DENV antibody status. There was no difference in the proportion symptomatic by reported JEV vaccination history and, among those reporting a history of vaccination, no difference in the time between vaccination and the first detected infection (mean±SD 5.01±2.22 years for asymptomatics, 4.93±2.11 for symptomatics).

Children developing DF were most likely to be JEV NAb positive, those experiencing asymptomatic infection the least (50% versus 39%, p = 0.017, by χ² testing across all strata of infection severities) (Table 2). There were no differences in the proportions of children that were JEV NAb positive by age or gender. The proportion JEV NAb positive varied by school and year. DENV-naives were most likely to be JEV NAb positive (54%), then DENV-multitypics (47%), then DENV-monotypics (26%) (p = 0.001 by χ² testing). Those reporting a history of JEV vaccination were more likely to be JEV NAb positive (46% versus 31% p = 0.012 by χ² testing). Among those reporting a history of vaccination, there was no difference in time since vaccination (mean±SD 5.01±2.27 years for JEV NAb positives, 4.96±2.19 for JEV NAb negatives).

JEV NAb positivity was associated with an increase in the odds of symptomatic infection in unadjusted analysis (OR = 1.55, 95% CI: 1.08 to 2.23) (Table 3). There were no significant differences in the occurrence of hospitalized illness or the occurrence of DHF.

Individuals with JEV NAbs were more likely to experience symptomatic DENV infection for all strata of preexisting DENV immunity, though this association was strongest and significant only for DENV-naives (OR = 2.75, 95% CI: 1.12 to 6.72) (Figure 1a). The association was non-significant and weakest for children with DENV-monotypic immunity prior to infection. Younger and older children with DENV-multitypic immunity had approximately the same increased odds of symptomatic infection with JEV NAbs; neither association was significant.

The directions of the associations between JEV NAbs and hospitalized illness (Figure 1b) and JEV NAbs and DHF (Figure 1c) were not consistent across strata of preexisting DENV immunity and the associations were largely non-significant. The association between JEV NAbs and DHF was significant only for younger DENV-multitypics, who had increased odds of DHF with JEV NAbs (OR = 4.05, 95% CI: 1.18 to 13.87).

The presence of JEV NAbs was associated with an increased duration of DENV illness in DENV-naives (5.70 versus 2.69 days, p = 0.045) (Table 4). For DENV-monotypics and multitypics, no difference in the duration of illness was observed.

Among those with JEV NAbs prior to infection, there was no difference in the geometric mean titer between asymptomatically and symptomatically infected individuals (p = 0.45 by Mann-Whitney U test, Figure 2). The distributions of the titers were similar.

The greatest number of RT-PCR-positive illnesses in the cohort was associated with DENV-2; the highest proportion of hospitalized infections was observed with DENV-3 (Figure 3a). The infecting DENV serotype was unknown for all asymptomatic infections and for 25.2% of symptomatic infections. 40.0% of children missing serotype data resided in communities with a single serotype in circulation the year of their infection and were therefore eligible for imputation. The probability of symptomatic infection was increased with JEV NAbs for all DENV serotypes, but the association was not significant for any serotype in subgroup analysis (Figure 3b). Limiting the comparisons to symptomatic, RT-PCR positive infections (with no imputation), the direction of the association between JEV NAbs and hospitalized illness was highly variable (Figure 3c). DENV-3 infection in the setting of preexisting JEV NAbs was associated with decreased hospitalized illness (15.0% hospitalized with JEV NAbs versus 46.7% hospitalized without, p = 0.004 by χ2 test).

The positive, significant association between the presence of JEV NAbs and symptomatic infection remained after controlling for age, pre-infection DENV immunity, and epidemic year and accounting for clustering of observations by school (data not shown). The adjusted odds ratio was 1.70 (95% CI 1.15 to 2.51).