Household immunity and individual risk of infection with dengue virus in a prospective, longitudinal cohort study Marco Hamins-Puértolas1 , Darunee Buddhari2, Henrik Salje3,4, Derek A.T. Cummings4,5, Stefan Fernandez2, Aaron Farmer2, Surachai Kaewhiran6, Direk Khampaen6, Sopon Iamsirithaworn6, Anon Srikiatkhachorn7,8, Adam Waickman9, Stephen J. Thomas9,11, Alan L. Rothman7, Timothy Endy9,10, Isabel Rodriguez-Barraquer1*, Kathryn B. Anderson9,11* * These authors jointly supervised this work 1. Department of Medicine, University of California, San Francisco, San Francisco, USA 2. Department of Virology, Armed Forces Research Institute of Medical Sciences, Thailand 3. Department of Genetics, University of Cambridge, UK 4. Department of Biology, University of Florida, USA 5. Emerging Pathogens Institute, University of Florida, USA 6. Ministry of Public Health, Tiwanond, Nonthaburi, Thailand. 7. Institute for Immunology and Informatics, Department of Cell and Molecular Biology, University of Rhode Island, Providence, RI 02903 8. Faculty of Medicine, King Mongkut's Institute of Technology Ladkrabang, Bangkok, Thailand. 9. Department of Microbiology and Immunology, SUNY Upstate Medical University, Syracuse, NY, USA. 10. Coalition for Epidemic Preparedness Innovations (CEPI), Washington DC, USA 11. Institute for Global Health and Translational Sciences, SUNY Upstate Medical University, Syracuse, NY, USA. Correspondence to: AndeKath@upstate.edu mailto:AndeKath@upstate.edu ABSTRACT Although it is known that household infections drive transmission of dengue virus (DENV) it is unclear how household composition and the immune status of inhabitants affects the individual risk of infection. Most population-based studies to date have focused on pediatric cohorts because more severe forms of dengue mainly occur in children, and the role of adults in dengue transmission is understudied. We analysed data from a multigenerational cohort study of 470 households, comprising 2860 individuals, in Kamphaeng Phet, Thailand, to evaluate risk factors for DENV infection. Using a gradient boosted regression model trained on annual haemagglutination inhibition (HAI) antibody titre inputs, we identified 1049 infections, 90% of which were subclinical. . By analysing imputed infections, found that individual antibody titres, household composition, and antibody titres of other members in the same householdaffect an individual’s risk of DENV infection. Those individuals living in households with high average antibody titres, or households with more adults, had a reduced risk of infection. We propose that herd immunity to dengue acts at the household level and may provide insight into the drivers of the recent change in the shifting age distribution of dengue cases in Thailand. Introduction The number of individuals infected with DENV ranges from 100-400 million per year 1–3, primarily in tropical and sub-tropical regions of the world. A substantial proportion of DENV transmission occurs in and around the home, with infections having a high likelihood of being spatiotemporally correlated 4–9. However, individuals living in neighbouring but separate households can experience persistent differences in risk of infection 9–11. The drivers of heterogeneity in risk of DENV infection among households and villages are unknown, potentially limiting the capacity for targeted interventions. There is evidence supporting focal transmission at either the school 12,13 or household level 9–11,14,15. Immunity, or susceptibility, of household members may impact the individual risk of infection. Analysing the role of immunity in household transmission is complicated. This is because most infections are subclinical, and are therefore missed by surveillance systems 16,17. Studies characterizing risk factors for DENV infection are therefore biased towards symptomatic infections rather than the entire population of infected individuals. In addition, DENV has historically been concentrated in children so most studies have focused on understanding infection dynamics in this subpopulation 2,18. This has resulted in large gaps in knowledge about risk factors for DENV infection in either adults or entire households. Recent shifts in the average age of dengue cases towards adults in several countries in South Asia.19–23. For example, the mean age of dengue hemorrhagic fever (DHF) has risen in Thailand from approximately eight to 24 years of age between 1981 and 2017 19, making understanding risk factors for DENV infection in adults now more pressing than ever. Identifying subclinical DENV infections in individuals is difficult, because it either requires data from longitudinal serological testing of large cohorts 24–27 or from follow-up of index cases and their close contacts at the household level 28–32. Estimates of the proportion of subclinical cases vary substantially 33in published observational studies, due to differences in how susceptible the population is to the major circulating DENV serotype, definitions of symptomatic and subclinical infections, and differences among follow-up monitoring protocols . Most studies that have analysed longitudinal serological data to identify subclinical infections have defined infections as ‘a four-fold increase in antibody levels between two samples for both HAI 24,26,34 and ELISA assays’ 25,27. However, while there is good support for using cut-off points in the context of acute/convalescent samples obtained weeks apart, their accuracy in identifying infections from samples obtained months or years apart in this way, is unclear. Due to antibody decay in the months following an acute infection35 the sensitivity of the ‘four-fold’ approach to identifying infection is likely to diminish over time, resulting in underestimates of the true number of infections. Additionally, it is unclear if the ‘four-fold’ method underperforms in individuals with high initial antibody titres. Other approaches have reconstructed subclinical infections by fitting full probabilistic models that simultaneously characterize antibody kinetics and infection histories, but this method is data intensive, requiring large numbers of longitudinal serum samples collected frequently and virologically confirmed infections to estimate antibody kinetics 36. Such detailed and prospective datasets are not commonly available, and therefore alternative approaches are needed to study the transmission of dengue, understand its drivers, and quantify the impact of interventions including vaccines. Here, we analyse data from an ongoing longitudinal study in Kamphaeng Phet, Thailand in order to characterize risk factors of DENV infection and disease. A key feature of our https://paperpile.com/c/Jvvymh/WGkIs+fDhH6+NCNC1 https://paperpile.com/c/Jvvymh/GNvVC+BbW81+cbuWT+1iGj8+Ir5ey+m7NE6 https://paperpile.com/c/Jvvymh/cLv8f+ChcHm+m7NE6 https://paperpile.com/c/Jvvymh/OS2EN+ca6Qj https://paperpile.com/c/Jvvymh/cLv8f+ChcHm+m3d2I+L8aVP+m7NE6 https://paperpile.com/c/Jvvymh/5k5Mi+QdSI8 https://paperpile.com/c/Jvvymh/fDhH6+UT3HB https://paperpile.com/c/Jvvymh/XzXY+SNT7j+ZbZ1Z+H0CEC+JYDIF https://paperpile.com/c/Jvvymh/XzXY https://paperpile.com/c/Jvvymh/mbyPM+uy2sY+m8M3u+mBU72 https://paperpile.com/c/Jvvymh/NNrg6+FWA3e+0RWec+CIoS3+HRrhb https://paperpile.com/c/Jvvymh/Xf16 https://paperpile.com/c/Jvvymh/m8M3u+rdE9N+mbyPM https://paperpile.com/c/Jvvymh/uy2sY+mBU72 https://paperpile.com/c/Jvvymh/qZHm https://paperpile.com/c/Jvvymh/rVDi7 longitudinal study is that it enrolled multigenerational households, which enabled us to study the risk profiles of children, adults and full households in parallel 37. Instead of relying on fixed cut- points, like the four-fold approach, we applied a flexible classification algorithm that takes yearly paired antibody titers to determine whether an individual was infected between sampling events. Using confirmed DENV infections to train this algorithm we characterized the dynamics of DENV infections in this cohort, including the association between infection and individual and household factors, and report our findings here. MAIN Cohort description This study used data from an ongoing cohort study in Kamphaeng Phet, Thailand, that has enrolled 3514 individuals living in 515 households (Supplemental Information Figure 2). The study started in September 2015 with the aim of defining immunological correlates of protection from DENV and illness as well as factors shaping DENV transmission in multi-generational households. A second stage of this cohort is planned to continue through 2028. This study included yearly follow up of participants where serum samples were obtained as well as active illness investigations and household investigations triggered whenever a participant reported a fever (defined as an index case). Yearly serum samples were tested using HAI while illness and household investigations included multiple assays (RT-PCR, immunoglobulin M, immunoglobulin G, and HAI). Our analysis included 2868 individuals within 470 households who had been followed up at least once after enrollment and before March 2022. The analysed dataset contains data on 11131 “yearly” intervals, with an average of 3.90 intervals per enrolled individual (95% CI 1-6). Characteristics of the analysed intervals are reported in Table 1 and the age pyramid is shown in Extended Data Figure 1a. The intervals were an average of 407.8 days long (95% CI 229 - 642.75) and took place over six sampling periods (Figure 1a). Not all individuals in a household consented to being sampled at every visit, such that approximately 80% of potential individuals were sampled . Over the study period there were 469 index cases, which resulted in laboratory confirmation of 90 infections between paired yearly samples. These 90 infections consisted of 61 PCR positive and with the remaining 29 cases being identified using serological evidence and constitute the gold standard data used in model training. Model performance We fit gradient boosted regression models to infer subclinical infections in individuals from antibody titres measured during yearly visits after training on gold standard infections. Our best fitting model was able to classify our training data with 93.3% sensitivity and 98.0% specificity (Extended Data Figure 2a). The longitudinal design of the cohort study allows visualization of HAI trajectories across time for enrolled individuals. Figure 2a illustrates imputed infections for three individuals enrolled in the cohort. The average and maximum ratios of pre to post interval HAI titres are the features of greatest importance for accurate classification defined by the information gain metric (Figure 2b). https://paperpile.com/c/Jvvymh/vxURl Characterizing subclinical infections Using our best fitting prediction models on the evaluation dataset (n=9885), we imputed 959 subclinical infections. When incorporating the 90 laboratory confirmed infections a total of 1049 infections are identified in 11131 intervals of observation, or 9.4% of intervals. This translates to 12.44 infections per 100 susceptible people per year (95% CI 11.01 - 13.88). Application of the four-fold increase in antibody levels to infer infections, as done previously to interpret paired serological data, identified a total of 956 infections, suggesting that our method identifies ~10% more infections. This is similar to an estimated annual proportion of seronegative individuals being infected per year of 10.8% derived using a serocatalytic model from age-stratified seroprevalence data (95% CI, 9.9-11.8% using seropositive cutoff as HAI >=20) (Figure 1b). We note that the model had high certainty in the assigning of infections for the majority of infections, with 673 of the 1049 intervals with infection being given a probability of greater than 90%. Similarly, the model had high certainty for the absence of infections in the remaining intervals with 8458 of the 11131 intervals being assigned a probability of less than 10% (Extended Data Figure 2b). Figure 2c shows where these imputed infections fall when comparing the average HAI across all four DENV serotypes pre and post interval while a breakdown by serotype can be found in Extended Data Figure 3. We found that the incidence of infections varied by year, with 2018 having higher incidence (Figure 3a). Hospitalizations peaked in Kamphaeng Phet in 2018 during the analysed study period (Figure 1a). Incidence of infection rates peaked amongst school aged children (Figure 3b). As expected, the proportion of primary infections (infections occurring in individuals without detectable antibodies to any serotype in any prior visit) was directly related to age, with almost all infections being post-primary (occurring in individuals with HAI antibody titers against at least one serotype greater than 20) after age 25. The ratio of subclinical to symptomatic infections was 13.8:1 (95% CI: 10.0-17.8:1) in the cohort. There was some variability across years and age, with the highest risk of symptomatic disease occurring between the ages of 15- 25 (Figure 3d-e). We note that there were only 77 symptomatic infections out of 1049 total infections, leading to wide confidence intervals for these ratios, particularly for years of age- groups with few cases. It is possible that additional mildly symptomatic infections were missed by the surveillance platform during yearly follow up and in turn these estimates likely represent lower bounds on the true number of symptomatic infections. Out of these 1049 infections, 139 individuals had multiple infection events throughout the study (Figure S3a), with the average time between infections found to be 733 (95%CI 677-791). The probability of having a second or third infection given that the individual had a previous infection peaks between the ages of 10- 15, similar to the age range of highest incidence (Figure S3b). Risk factors for DENV infection Using imputed infections from our classification algorithm we investigated which individual and household risk factors were associated with infection risk. We found that individuals between 5 and 18, and those aged between 18 and 30 were at higher risk of infection with an aOR of 1.44 (95% CI 1.16-1.77) and 1.41 (95% CI 1.06-1.89) respectively compared with children aged 1-5y. In an unadjusted analysis there was no significant difference in odds of infection by sex (OR 1.11, 95% CI 0.98 - 1.27). However, our data is consistent with an observed interaction between age and sex in infection risk of women between the ages of 18-40, who had an increased risk of infection compared with their male counterparts (Extended Data Figure 1c). We also found no significant association between occupation and risk of infection in an adjusted analysis (Table S2). We studied how household level factors affect an individual’s risk of infection. No covariates describing the surrounding built environment had a significant impact on dengue risk. However, we found strong associations between household composition and risk of infection. While the number of individuals living in the household was not associated with risk of infection (aOR 1.00, 95% CI, 0.97 - 1.04), we found that each additional adult in the household reduced the likelihood of infection in the other household members with an aOR of 0.95 (95% CI, 0.90 - 0.99). The presence of each additional newborn and individual between 5 and 18 increased the odds of infection for the other household members with an aOR of 2.13 (95% CI, 1.65 2.75) and 1.09 (95% CI, 1.01 - 1.19) respectively. Although not significant, the presence of each additional individual between one and five increased the odds of infection for household members with an aOR of 1.13 (95% CI, 1.00 - 1.28) (Figure 4a). Analyses stratified by sex revealed a more complex association between household composition and risk. For either sex, each additional newborn increased infection risk for the other individuals living in that household. For older age groups however the associations varied by sex. Each additional male between the ages of 1 and 5 and between 5 and 18 increased risk with an aOR of 1.25 (95% CI, 1.08 - 1.44) and 1.18 (95% CI, 1.06 - 1.31) respectively while additional adult male had no impact on risk. Additional females provided no changes in risk except for adults, where each additional female adult reduced risk with an aOR of 0.88 (95% CI, 0.81 - 0.95) (Figure 4b). Beyond characterizing the association between household characteristics and composition on dengue risk, we sought to understand the impact of individual and household immunity. Consistent with previous findings, the most important predictor of infection risk during an interval was an individual’s HAI titers at the beginning of the interval (Figure 5). In our analysis, the magnitude of average HAI log2 titres was inversely associated with risks of both subclinical and symptomatic infections. On average, each log2 increase in titres was associated with a 26.4% (95% CI: 23.5 - 29.2%) decrease in risk of infection and a 38.7% (95% CI: 27.9 - 47.9%) decrease in having a symptomatic infection. Interestingly, we also found that household immunity impacted an individual's risk of infection even when accounting for that individual’s antibody titre. The distribution of these variables is found in Figure S4. Individuals living in households with high immunity (average HAI titres greater than 66) had decreased risk with an aOR of 0.78 (95% CI, 0.63 - 0.96) when compared to those with an average below 40 (Figure 4d). Since household titres are likely to be associated with recent household infection history, we also investigated how household attack rates during a preceding interval (the proportion of individuals within a household that had an imputed DENV infection in the preceding interval) impact future risk. Individuals living in households that had moderate to high attack rates (greater than 20% of household members experiencing an infection) during the previous year were at decreased risk of infection with an aOR of 0.61 (95% aOR CI, 0.49 - 0.77) in comparison to individuals coming from a household with no infections the previous year (Figure 4c). We also found that higher proportions of immune individuals at the household level decreased the risk of infection for household members (Figure S5). Sensitivity analyses were generally consistent with original findings. We want to highlight that if infections are imputed using a four-fold increase in any DENV serotype HAI titre, instead of the classification model developed here, the protective association between individual and household titres and infection risk remains (Extended Data Figure 4). Specifically, for each log2 increase in an individual’s titres, there was an associated 28.5% (95% CI: 25.5 - 31.4%) decrease in risk of infection and a 40.1% (95% CI: 2.1 - 50.7%) decrease in having a symptomatic infection. Sensitivity analyses in which we restrict the data to households with more than 80% of individuals sampled and just seronaive individuals were also consistent with the main findings (Extended Data Figures 5-6). Discussion We developed a classification algorithm using longitudinal data from a multigenerational cohort in Kamphaeng-Phet, Thailand to reconstruct subclinical DENV infections. Inferring subclinical infections with more precision enabled us to analyse individual and household level factors that affect risk of DENV infection. We report a protective effect of higher HAI titres at both the individual and household level. Although previous work has demonstrated that higher antibody titresprotects individuals against infection 36,49,50, we report an independent indirect effect of household immunity and composition on infection risk. We studied how several household factors including composition, immunity, and infection history each independently affect risk of infection with DENV. We found that all three factors determine an individual’s risk. When analyzing household composition we found that each additional adult reduced the likelihood of an infection, while each additional young child (1- 5 years old) increased the likelihood of infection. These findings might be explained by the fact that children are more susceptible to infection than their adult counterparts who have already experienced infection and developed immunity in the past. We also found that higher levels of household immunity, and higher attack rates in the previous year, have protective effects against infection. These associations were evident even though there are other potential locations in an individual’s daily routine outside the home that impact risk of DENV infection, limiting the indirect protection in the home. Taken together, households with more adults or more recent infections will have more immunity to DENV and in turn reduce subsequent infection risk for household members. At the individual level, our results are consistent with prior studies showing that individual antibody titres are the most important predictor of future DENV infection risks 36,49,50. How this relationship varies across adults is less understood. Here we find that the risk of infection for adults over the age of 30 remains high, at approximately half that of younger individuals. These infections occur in individuals who have been infected two or more times and are in turn multi- typically protected. This is particularly relevant to the open question of how long boosting post- infection confers immunity and protection from clinical manifestations. For these same individuals we find a higher subclinical to symptomatic ratio, suggesting that these adults are likely exposed to DENV while simultaneously not experiencing symptoms. We hypothesized previously that the aging population of Thailand resulted in a decrease in the force of infection , potentially driven by longer living adults that have multitypic immunity who reduce the risk that younger individuals living in the same household experience an infection 19,51. Our results lead us to propose that a combination of immunity and recent infection history in a household can confer a form of ‘herd immunity’ for an individual, regardless of their own immune status. Children are more likely to be seronaive than adults, and may https://paperpile.com/c/Jvvymh/HOkU9+rVDi7+4p6fv https://paperpile.com/c/Jvvymh/HOkU9+rVDi7+4p6fv https://paperpile.com/c/Jvvymh/XzXY+vxkNM present a means by which DENV can be introduced into the household. Introduction of DENV would subsequently increase the risk that the virus will be transmitted (by mosquitoes) to others in the household, a mechanism that would explain some of the spatial correlations found in another study of the same population 52. It is intriguing that household composition, immunity and infection history have a significant impact on infection risk, whereas covariates measuring the surrounding built environment were not. Our work provides a framework upon which machine learning classification models could be used to predict infection events from yearly serological data. Although application of a four- fold rise in titres as a barometer for infection can be useful when analyzing acute and convalescent titres , our approach is a more robust and sensitive way to characterize subclinical infections. Previously, Bayesian-based approaches have been successful at reconstructing dengue infection events 36,53, but they require substantial temporal information to inform the underlying mechanistic model of antibody kinetics. Our method provides a flexible framework that removes some of the bias of potential model misspecification and instead takes a fully data driven approach to reconstruct infection events. This methodology could more broadly be applied to other infectious diseases where longitudinal serological data is collected. Our results highlight the importance of multigenerational household studies in order to fully understand the population dynamics of infectious diseases. The protective effects of household immunity had been hidden in previous analyses, some in the same setting, that have focused on children. However, our work has some limitations. Our model training is limited by the fact that there are only 90 data points used to inform the classification algorithm. If these illness investigations are biased , this would propagate to our predictions. In particular since primary and subclinical DENV infections are underrepresented then we may be less capable of identifying these DENV infections. In addition, development of the training data required that we define individuals who had no infection event during an interval, a difficult task that could further limit our approach. We were unable to differentiate between homologous and heterologous infections due to HAI data being cross-reactive across DENV serotypes (Figure S6). Instead we are only able to determine whether an individual had an exposure or not during a given interval. If plaque reduction neutralization test (PRNT) data were utilized instead, it is possible some additional serotype specific information could be elucidated, however cross-reactivity is also an issue for PRNTs in post-primary infections. Another limitation of the study was the fact that serum samples were taken at yearly intervals. This made it impossible to fully disentangle the timing of infections which would provide important information on how infections propagate in a household. Incorporating additional active sampling events throughout the year in household studies like this one could provide important temporal information to understand this further. Finally, due to study design, most female participants of reproductive age give birth upon enrollment. We are therefore not in a position to examine whether the sex differences found between the ages of 18-40 are due to age or to other biological or behavioral factors (Extended Data Figure 1) related to pregnancy and giving birth. Further work must be done to fully understand this relationship. There is a critical need to better understand how immunity impacts the spread of infectious diseases like DENV. With DENV infections being highly spatiotemporally correlated in endemic settings, the success of future intervention efforts hinges on the ability to accurately quantify infection risk. Disentangling risk into the component contributions from individual, https://paperpile.com/c/Jvvymh/Y4vlx https://paperpile.com/c/Jvvymh/EIedm+rVDi7 household, and community level factors could help direct these efforts. Individuals with higher immunity are protected from infection and disease, while entire populations can also experience similar protective effects from population level immunity. Here we show evidence of protective effects of immunity at the household level, something that has not previously been studied in DENV nor in other vector borne diseases, to our knowledge. If household immunity is a major driver of spatiotemporal clustering, interventions may be effectively targeted towards households with lower immunity.Considering immunity at multiple scales when mapping dengue risk and making public health decisions is important. Acknowledgments We are thankful for all efforts from the data collection team as well as the children and adults involved in the study. The authors were supported in this work by the following: The National Institutes of Health (NIH) Grant 5P01AI034533-22: entire team; Military Infectious Disease Research Program (MIDRP): DB, SF, AF, KBA; NIH 1R01AI175941-01: entire team; European Research Council 804744: HS; and NIH 1R35GM138361-01: MHP and IRB. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. Author Contributions The study was conceived and designed by DB, HS, DATC, SF, AF, ALR, TPE, IRB, and KBA. The data was collected by DB, SF, SK, DK, SI, and AS. The analysis and interpretation of results was performed by MHP, HS, DATC, AF, SJT, AW, ALR, TPE, IRB, and KBA. The draft manuscript was prepared by MHP, IRB, and KBA. All authors reviewed the results and approved the final version of the manuscript. Competing interests The authors declare no competing interests. ONLINE METHODS Kamphaeng-Phet family-based cohort study This study used data from an ongoing family-based longitudinal cohort study in Kamphaeng- Phet, Thailand. Details of the design have been previously described 37. Briefly, we enrolled pregnant mothers and their multi-generational households. Per the inclusion criteria, a household was eligible for enrollment if a minimum of three members in addition to the newborn consented/assented to participation; the pregnant woman, another child, and an older adult aged at least 50 years. Active surveillance began with the birth of the newborn, with enrollment specimens for the remainder of the household collected prior to the birth of the newborn. In order to ascertain subclinical infections, serum samples were obtained from all participants roughly annually after enrollment. Acute febrile illness events were detected through a combination of active and passive surveillance strategies. Individuals were instructed to notify study staff if an acute febrile illness event occurred. In addition, participants were contacted by the study team on a weekly basis to determine if any member of the household had been febrile https://paperpile.com/c/Jvvymh/vxURl since the last contact. Upon discovery of a febrile episode, an illness investigation was triggered, where acute and convalescent blood samples were obtained from the febrile case. If the illness investigation identified a PCR-confirmed case a household investigation was triggered where acute and convalescent samples were taken for the remaining household members. The convalescent samples were taken at 14 and 28 days after the acute sample collection. This study was approved by Thailand Ministry of Public Health Ethical Research Committee, Siriraj Ethics Committee on Research Involving Human Subjects, Institutional Review Board for the Protection of Human Subjects State University of New York Upstate Medical University, and Walter Reed Army Institute of Research Institutional Review Board (protocol #2119). Laboratory methods All samples obtained during routine visits were tested using hemagglutination inhibition assays (HAIs) to quantify antibody titres against all four DENV serotypes and Japanese encephalitis virus (JEV) 38. Additionally, all acute and convalescent samples were tested by HAI for all four DENV serotypes and JEV as well as immunoglobulin M (IgM) and immunoglobulin G (IgG) capture ELISAs for DENV and JEV 39. All acute samples also underwent DENV reverse- transcriptase polymerase chain reaction (RT-PCR) 40. For the purpose of this analysis, we defined a confirmed DENV infection as any case that is RT-PCR positive for any DENV serotype or where both HAI and ELISA results using the acute and convalescent samples were diagnostic of an infection 37. Further details on the specific laboratory methods utilized have been described in previous work 41–43 and are summarized in the supplemental information. Statistical analysis The purpose of this analysis was to investigate individual and household risk factors for DENV infection in this multigenerational cohort. To do this, we first fit a classification algorithm to the yearly HAI data in order to identify subclinical infections. We then used these imputed infections to investigate individual and household-level drivers of infection. Training data We define a positive and negative person period as follows. A total of 90 confirmed DENV infections were identified through the case investigations. Data from the yearly HAIs surrounding these confirmed DENV infections were defined as the positive person periods. For negative person periods we took the remaining full dataset and removed any interval where an individual had a confirmed DENV infection via the illness investigation, or where individuals had a larger than fourfold increase in any one of their yearly DENV serotype HAIs. We then removed any individuals living in the same household during these aforementioned intervals since DENV transmission is known to be clustered within households. This left 3466 intervals that could potentially be used as negative controls from the available 11131 observed intervals (Figure S1). We randomly sampled a third of these to be added to the training data creating a total of 1246 intervals in our training set. The first interval of sampling for newborns was excluded in this https://paperpile.com/c/Jvvymh/9gPM8 https://paperpile.com/c/Jvvymh/JHUVe https://paperpile.com/c/Jvvymh/DdI5G https://paperpile.com/c/Jvvymh/vxURl https://paperpile.com/c/Jvvymh/ecUwq+ZNBQb+mn1ny analysis due to limited representation in the serologically supported infections that could provide information on maternal antibody kinetics. Prediction model Using training data described above, we ran a gradient boosted regression using the R package xgboost 44,45. Unlike in random forest models where multiple independently trained decision trees are combined to determine the overall likelihood of a model, in gradient boosted regression each decision tree is fit on what the previous trained ensemble of trees have misclassified, allowing for refinement on difficult classification problems. The candidate predictors we used to train this model are listed in Table S1. Variables used to summarize the ratio and difference between pre and post-interval DENV titres across serotypes (maximum, minimum, geometric mean, and variance) were calculated at the individual serotype level and then the summary statistic of interest was quantified across all four serotypes. Model fit For hyperparameter tuning we utilized a random search approach within a nested cross validation approach where we initially split the training data into four cross-validation sets and subsequently performed hyperparameter tuning on each subset using five-fold cross validation. Model performance was quantified using the hold out set. Prior to each random search run we randomly downsampled the dataset to balance the number of positive and negative person periods. We performed this random search approach a total of 5000 times and saved the top 100 performing models evaluated on the held out cross validation set with the lowest log-loss value. The average predicted classification score (bounded between zero and one) for these 100 runs was taken to be the probability the individual was infected in that yearly interval. Intervals assigned a value greater than or equal to 0.5 were considered to be DENV infections. Predicting subclinical DENV infections We subsequently fit the models with the lowest log-loss values on the entire training dataset and predicted the presence or absence of infections in the remaining intervals that make up the evaluation data set. We used the training labels as ground truth and subsequently analysed risk factors for the entire dataset. Characterizing risk factors of DENV infection and disease We fit a series of univariate and adjusted logistic regressions to characterize how DENV infection risk is a function of temporal, individual, and household factors. These models were run using the glmmTMB function found within the glmmTMB package in R 46,47. We fit all models using a binomial GLM with a logit link function. All generalized linear models were optimized using the nlminb method found in the stats package. Only household random effects were incorporated into the model as the inclusion of both household and individual random effects led to singular fits. Individual and household level risk We first tested whether demographic factors were associated with risk including age, sex, and employment. We binned individuals into five age bins (1-5, 5-18, 18-30, 30-50, and 50+). https://paperpile.com/c/Jvvymh/BjfCp+G1SCg https://paperpile.com/c/Jvvymh/1DlDk+4FI23 Individuals under 1 year were excluded as they will usually have maternal antibodies, which would complicate this analysis due to different kinetics. Sex was defined upon enrollment into the cohort. Further information on individual and household related covariates can be found in the supplemental information. We subsequently performed analyses to quantify how household composition and infection history impacted risk for an individual. Data on household composition consisted of the number of newborns, individuals between 1 and 5, individuals between 5 and 18, and adults all of which were broken down by sex. We fit models to estimate how the number of individuals in each of these bins impacted infection risk. For the analysis on infection history we fit models to assess how the attack rate, the proportion of the household members who were inferred to have an infection, in the previous interval (categorized into three sets defined as containing strictly 0, (0.2), and [0.2,1]) impacted the infection risk. The distribution of these values was zero inflated and skewed right due to many households having no infections in the previous interval. Note that we removed the individual of interest in determining both the household composition and attack rate of the household in order to isolate how the household is impacting risk. For both of these covariates we fit three logistic regression models, a univariate model, a univariate model with random effects, and a multivariate model with random effects. Since the goal of these models was to characterize the independent effect of the household-level covariates, each of these multivariate models also accounts for the individual’s average pre-interval HAI titer as well as the month and year of sampling as these have been shown to be important predictors of risk. This ensured the individual’s age, titers, and infection history did not impact subsequent calculations. Confounding effects of household related factors were accounted for in adjusted analyses where household random effects were incorporated. Note that as this analysis required at least two consecutive intervals, around 25% of the intervals were not included leaving 6913 intervals. We then performed logistic regressions to understand how individual and household immunity impact DENV infection risk. We defined individual immunity to be the geometric mean of HAI titers transformed into log2 space. We defined household immunity for a particular individual to be the geometric mean of HAI titers of the household transformed into log2 space with the individual of interest removed from the calculation. HAI cutoffs of 40 and 66 were chosen for the household immunity covariate as these constituted the 33rd and 66th percentiles. Similar to the previous analysis, we fit three logistic regressions for each, a univariate model, a univariate model with random effects, and a multivariate model with random effects. In addition to these random effects and the covariate of interest, each multivariate model also accounts for the month and year of sampling. The household immunity adjusted model also accounted for the individual’s average pre-interval HAI titer. Sensitivity Analysis To assess whether our main findings were robust to methodological assumptions or potential biases in the data, we performed three sensitivity analyses. The first sensitivity analysis we performed was based on the fact that at times not all individuals in a household were sampled. Those that went unsampled in a household were more likely to be adult males potentially leading to confounding effects of households with more missing data. We in turn reran the analyses with the four-fold increase in titers rule often used as the standard in longitudinal serological studies. This sensitivity analysis allows for the direct comparison between our prediction algorithm and the most commonly implemented approach in the literature (Extended Data Figure 4). We also conducted analyses on all intervals taken from households with more than 80% of their members sampled, limiting the analysis to 6453 intervals (Extended Data Figure 5). Lastly, we reran the analyses in individuals who were seronaive at the beginning of the interval in order to investigate whether the identified associations were also observable in the fully susceptible subpopulation, limiting the analysis to 2066 intervals (Extended Data Figure 6). Further descriptions of these results can be found in the Sensitivity Analysis section of the Supplemental Information. Data Exclusion Note that we excluded data from newborns in the analysis in order to avoid the potential biases from maternal antibodies 54. FIGURE LEGENDS Figure 1. Cohort data summary (n=11131). (a) Hospitalization counts for Kamphaeng Phet from 2015-2021 (blue solid line). Bars represent the timing of the confirmed DENV infections used to train the model (n=90). Serotype information was ascertained via RT-PCR. Shaded time periods represent active sampling periods during the cohort study when yearly blood draws were taken. (b) Age-stratified seropositive at enrollment for subjects enrolled before 2017. Points are mean seroprevalence found at each tenth percentile age bin while the line is the resulting fit using the serocatalytic model for non-newborns under 30 years old (n=6197), details found in the Supplemental Information. (c) Average DENV HAI titers at enrollment age binned into each tenth percentile. Confidence bounds (95%) for B and C are found using a basic nonparametric bootstrap while a generalized linear model is fit in black. Mean and 95% confidence interval are presented as the line and shaded region. Figure 2. Model performance and fit. (a) Three examples of HAI titer trajectories for all four DENV serotypes and JEV in three subjects. Alternating white and gray time periods represent distinct intervals, separated by the sampled HAIs. The imputed probability of an infection having occurred within an interval is presented by point size at the post-interval sample date while red arrows represent an imputed infection. Black arrows represent JEV vaccination events. (b) Feature importance for model fits (n=100). Gain represents the relative contribution of each feature. Bars represent the mean and 95% credible intervals. (c) Pre and post interval HAI titers for the DENV serotype with the largest ratio grouped by age at post- interval sampling event and colored by whether the model predicted a seroconversion. Yellow and blue dots represent points that were or were not identified as infections respectively by the model while black and red points represent a similar dichotomy but in laboratory confirmed seroconversions. A four-fold increase in titers between samples is represented by the black https://paperpile.com/c/Jvvymh/UHKi diagonal line. Figure 3. Incidence, proportion of primary infections, ratio of subclinical to symptomatic cases. (a-b) Incidence (infections per person-year) for both symptomatic (red, n=77) and subclinical (yellow, n=972) infections across interval year (a) and age (b). (c) Proportion of primary infections as a function of age. (d-e) ratio of subclinical to symptomatic DENV incidence in the cohort as a function of interval year (d) and age (e). Mean and 95% CIs for the ratio of subclinical to symptomatic DENV incidence are represented by the dotted line and gray regions respectively. Mean and 95% CIs for polynomial fits to time and age are represented by the solid blue line and regions. Figure 4. Household composition and risk of infection across n=11131 intervals. Household composition (a-b), infection history (c) and immunity (d) and their impact on the risk of infection are shown. (a) Odds ratio for the number of total individuals in various age bins (newborn [NB], from 1-5 years old [LT5], from 5 to 18, and those older than 18 [GE18]) defined at the time of the post-interval sample. (b) Odds ratio for the number of males and females of various age bins (newborn [NB], from 1-5 years old [LT5], from 5 to 18, and those older than 18 [GE18]) defined at the time of the post-interval sample. (c) Previous interval’s attack rate and subsequent odds ratio of infection risk relative to having no infections in the previous interval. (d) Geometric mean of DENV HAI titers for the rest of the household members and subsequent odds ratio of infection risk relative to having an average household HAI titer under 40. All models are adjusted for household random effects, individual pre- interval titers, as well as the year and month of post-interval sample and have both means and 95% CIs presented. Figure 5. Impact of pre-interval DENV titres and probability of infection and symptoms across n=11131 intervals. We first calculated the annualized probability of infection and then fit splines of order three to the data using a generalized logistic regression with 95% confidence intervals presented as shaded regions. All panels also contain means and 95% confidence intervals derived from basic nonparametric bootstraps. (a) The annual probability of infection is a function of their pre-interval DENV titers. (b) The annual probability an individual is symptomatic is a function of their pre-interval DENV titers. (c) The annual probability that an individual is symptomatic given that they were infected is a function of their pre-interval DENV titers. Table 1. Covariates and infection prediction. Predicted infections are further subdivided into those with symptomatic and subclinical infections. Covariate No infection (N=10,082) Symptomatic infection (N=77) Subclinical infection (N=972) Overall (N=11,131) Sex Male 4192 (41.6%) 39 (50.6%) 425 (43.7%) 4656 (41.8%) Female 5890 (58.4%) 38 (49.4%) 547 (56.3%) 6475 (58.2%) Age (years) Mean (SD) 29.6 (22.2) 14.5 (11.1) 22.5 (20.8) 28.9 (22.2) Median [Min, Max] 26.2 [1.00, 100] 12.4 [1.18, 57.3] 14.8 [1.02, 88.3] 25.0 [1.00,100] [1,5) 1696 (16.8%) 16 (20.8%) 229 (23.6%) 1941 (17.4%) [6,18) 2166 (21.5%) 38 (49.4%) 310 (31.9%) 2514 (22.6%) [18,30) 1732 (17.2%) 17 (22.1%) 153 (15.7%) 1902 (17.1%) [30,50) 2121 (21.0%) 5 (6.5%) 135 (13.9%) 2261 (20.3%) 50+ 2367 (23.5%) 1 (1.3%) 145 (14.9%) 2513 (22.6%) JE vaccination in interval Yes 453 (4.5%) 1 (1.3%) 68 (7.0%) 522 (4.7%) No 9629 (95.5%) 76 (98.7%) 904 (93.0%) 10,609 (95.3%) Pre interval titer (HAI) Mean (SD) 117 (200) 31.1 (37.0) 55.8 (83.6) 111 (193) Median [IQR] 67.3 [14.1,134.5] 14.1 [10, 33.6] 28.3 [10, 67.3] 56.6 [14.1,134.5] Data availability The dataset analysed in this study is available at https://github.com/marcohamins/role-of-HH- immunity. Code availability All code associated with the work is available at https://github.com/marcohamins/role-of-HH- immunity. References 1. Cattarino, L., Rodriguez-Barraquer, I., Imai, N., Cummings, D. A. T. & Ferguson, N. M. Mapping global variation in dengue transmission intensity. Sci. Transl. Med. 12, (2020). 2. Wilder-Smith, A., Ooi, E.-E., Horstick, O. & Wills, B. Dengue. Lancet 393, 350–363 (2019). 3. Bhatt, S. et al. The global distribution and burden of dengue. Nature 496, 504–507 (2013). 4. Santos, G. R. dos et al. Individual, Household, and Community Drivers of Dengue Virus Infection Risk in Kamphaeng Phet Province, Thailand. The Journal of Infectious Diseases Preprint at https://doi.org/10.1093/infdis/jiac177 (2022). 5. Salje, H. et al. Dengue diversity across spatial and temporal scales: Local structure and the effect of host population size. Science 355, 1302–1306 (2017). 6. Yoon, I.-K. et al. Fine scale spatiotemporal clustering of dengue virus transmission in children and Aedes aegypti in rural Thai villages. PLoS Negl. Trop. Dis. 6, e1730 (2012). 7. Anders, K. L. et al. Households as foci for dengue transmission in highly urban Vietnam. PLoS Negl. Trop. Dis. 9, e0003528 (2015). 8. Cuong, H. Q. et al. Spatiotemporal dynamics of dengue epidemics, southern Vietnam. https://github.com/marcohamins/role-of-HH-immunity https://github.com/marcohamins/role-of-HH-immunity https://github.com/marcohamins/role-of-HH-immunity https://github.com/marcohamins/role-of-HH-immunity http://paperpile.com/b/Jvvymh/WGkIs http://paperpile.com/b/Jvvymh/WGkIs http://paperpile.com/b/Jvvymh/WGkIs http://paperpile.com/b/Jvvymh/WGkIs http://paperpile.com/b/Jvvymh/WGkIs http://paperpile.com/b/Jvvymh/WGkIs http://paperpile.com/b/Jvvymh/fDhH6 http://paperpile.com/b/Jvvymh/fDhH6 http://paperpile.com/b/Jvvymh/fDhH6 http://paperpile.com/b/Jvvymh/fDhH6 http://paperpile.com/b/Jvvymh/fDhH6 http://paperpile.com/b/Jvvymh/NCNC1 http://paperpile.com/b/Jvvymh/NCNC1 http://paperpile.com/b/Jvvymh/NCNC1 http://paperpile.com/b/Jvvymh/NCNC1 http://paperpile.com/b/Jvvymh/NCNC1 http://paperpile.com/b/Jvvymh/NCNC1 http://paperpile.com/b/Jvvymh/NCNC1 http://paperpile.com/b/Jvvymh/GNvVC http://paperpile.com/b/Jvvymh/GNvVC http://paperpile.com/b/Jvvymh/GNvVC http://paperpile.com/b/Jvvymh/GNvVC http://paperpile.com/b/Jvvymh/GNvVC http://paperpile.com/b/Jvvymh/GNvVC http://paperpile.com/b/Jvvymh/GNvVC http://dx.doi.org/10.1093/infdis/jiac177 http://paperpile.com/b/Jvvymh/GNvVC http://paperpile.com/b/Jvvymh/BbW81 http://paperpile.com/b/Jvvymh/BbW81 http://paperpile.com/b/Jvvymh/BbW81 http://paperpile.com/b/Jvvymh/BbW81 http://paperpile.com/b/Jvvymh/BbW81 http://paperpile.com/b/Jvvymh/BbW81 http://paperpile.com/b/Jvvymh/BbW81 http://paperpile.com/b/Jvvymh/BbW81 http://paperpile.com/b/Jvvymh/cbuWT http://paperpile.com/b/Jvvymh/cbuWT http://paperpile.com/b/Jvvymh/cbuWT http://paperpile.com/b/Jvvymh/cbuWT http://paperpile.com/b/Jvvymh/cbuWT http://paperpile.com/b/Jvvymh/cbuWT http://paperpile.com/b/Jvvymh/cbuWT http://paperpile.com/b/Jvvymh/cbuWT http://paperpile.com/b/Jvvymh/1iGj8 http://paperpile.com/b/Jvvymh/1iGj8 http://paperpile.com/b/Jvvymh/1iGj8 http://paperpile.com/b/Jvvymh/1iGj8 http://paperpile.com/b/Jvvymh/1iGj8 http://paperpile.com/b/Jvvymh/1iGj8 http://paperpile.com/b/Jvvymh/1iGj8 http://paperpile.com/b/Jvvymh/1iGj8 http://paperpile.com/b/Jvvymh/Ir5ey http://paperpile.com/b/Jvvymh/Ir5ey http://paperpile.com/b/Jvvymh/Ir5ey Emerg. Infect. Dis. 19, 945–953 (2013). 9. Salje, H. et al. Revealing the microscale spatial signature of dengue transmission and immunity in an urban population. Proc. Natl. Acad. Sci. U. S. A. 109, 9535–9538 (2012). 10. Santos, G. R. dos et al. Individual, Household, and Community Drivers of Dengue Virus Infection Risk in Kamphaeng Phet Province, Thailand. The Journal of Infectious Diseases Preprint at https://doi.org/10.1093/infdis/jiac177 (2022). 11. Salje, H. et al. Dengue diversity across spatial and temporal scales: Local structure and the effect of host population size. Science 355, 1302–1306 (2017). 12. Ratanawong, P. et al. Spatial Variations in Dengue Transmission in Schools in Thailand. PLoS One 11, e0161895 (2016). 13. Chen, Y. et al. Measuring the effects of COVID-19-related disruption on dengue transmission in southeast Asia and Latin America: a statistical modelling study. Lancet Infect. Dis. 22, 657–667 (2022). 14. Yoon, I.-K. et al. Fine scale spatiotemporal clustering of dengue virus transmission in children and Aedes aegypti in rural Thai villages. PLoS Negl. Trop. Dis. 6, e1730 (2012). 15. Anders, K. L. et al. Households as foci for dengue transmission in highly urban Vietnam. PLoS Negl. Trop. Dis. 9, e0003528 (2015). 16. Undurraga, E. A., Halasa, Y. A. & Shepard, D. S. Use of expansion factors to estimate the burden of dengue in Southeast Asia: a systematic analysis. PLoS Negl. Trop. Dis. 7, e2056 (2013). 17. Clapham, H. E., Cummings, D. A. T. & Johansson, M. A. Immune status alters the probability of apparent illness due to dengue virus infection: Evidence from a pooled analysis across multiple cohort and cluster studies. PLoS Negl. Trop. Dis. 11, e0005926 (2017). 18. Stanaway, J. D. et al. The global burden of dengue: an analysis from the Global Burden of Disease Study 2013. Lancet Infect. Dis. 16, 712–723 (2016). http://paperpile.com/b/Jvvymh/Ir5ey http://paperpile.com/b/Jvvymh/Ir5ey http://paperpile.com/b/Jvvymh/Ir5ey http://paperpile.com/b/Jvvymh/Ir5ey http://paperpile.com/b/Jvvymh/m7NE6 http://paperpile.com/b/Jvvymh/m7NE6 http://paperpile.com/b/Jvvymh/m7NE6 http://paperpile.com/b/Jvvymh/m7NE6 http://paperpile.com/b/Jvvymh/m7NE6 http://paperpile.com/b/Jvvymh/m7NE6 http://paperpile.com/b/Jvvymh/m7NE6 http://paperpile.com/b/Jvvymh/m7NE6 http://paperpile.com/b/Jvvymh/cLv8f http://paperpile.com/b/Jvvymh/cLv8f http://paperpile.com/b/Jvvymh/cLv8f http://paperpile.com/b/Jvvymh/cLv8f http://paperpile.com/b/Jvvymh/cLv8f http://paperpile.com/b/Jvvymh/cLv8f http://paperpile.com/b/Jvvymh/cLv8f http://dx.doi.org/10.1093/infdis/jiac177 http://paperpile.com/b/Jvvymh/cLv8f http://paperpile.com/b/Jvvymh/ChcHm http://paperpile.com/b/Jvvymh/ChcHm http://paperpile.com/b/Jvvymh/ChcHm http://paperpile.com/b/Jvvymh/ChcHm http://paperpile.com/b/Jvvymh/ChcHm http://paperpile.com/b/Jvvymh/ChcHm http://paperpile.com/b/Jvvymh/ChcHm http://paperpile.com/b/Jvvymh/ChcHm http://paperpile.com/b/Jvvymh/OS2EN http://paperpile.com/b/Jvvymh/OS2EN http://paperpile.com/b/Jvvymh/OS2EN http://paperpile.com/b/Jvvymh/OS2EN http://paperpile.com/b/Jvvymh/OS2EN http://paperpile.com/b/Jvvymh/OS2EN http://paperpile.com/b/Jvvymh/OS2EN http://paperpile.com/b/Jvvymh/OS2EN http://paperpile.com/b/Jvvymh/ca6Qj http://paperpile.com/b/Jvvymh/ca6Qj http://paperpile.com/b/Jvvymh/ca6Qj http://paperpile.com/b/Jvvymh/ca6Qj http://paperpile.com/b/Jvvymh/ca6Qj http://paperpile.com/b/Jvvymh/ca6Qj http://paperpile.com/b/Jvvymh/ca6Qj http://paperpile.com/b/Jvvymh/ca6Qj http://paperpile.com/b/Jvvymh/ca6Qj http://paperpile.com/b/Jvvymh/m3d2I http://paperpile.com/b/Jvvymh/m3d2I http://paperpile.com/b/Jvvymh/m3d2I http://paperpile.com/b/Jvvymh/m3d2I http://paperpile.com/b/Jvvymh/m3d2I http://paperpile.com/b/Jvvymh/m3d2I http://paperpile.com/b/Jvvymh/m3d2I http://paperpile.com/b/Jvvymh/m3d2I http://paperpile.com/b/Jvvymh/L8aVP http://paperpile.com/b/Jvvymh/L8aVP http://paperpile.com/b/Jvvymh/L8aVP http://paperpile.com/b/Jvvymh/L8aVP http://paperpile.com/b/Jvvymh/L8aVP http://paperpile.com/b/Jvvymh/L8aVP http://paperpile.com/b/Jvvymh/L8aVP http://paperpile.com/b/Jvvymh/L8aVP http://paperpile.com/b/Jvvymh/5k5Mi http://paperpile.com/b/Jvvymh/5k5Mi http://paperpile.com/b/Jvvymh/5k5Mi http://paperpile.com/b/Jvvymh/5k5Mi http://paperpile.com/b/Jvvymh/5k5Mi http://paperpile.com/b/Jvvymh/5k5Mi http://paperpile.com/b/Jvvymh/5k5Mi http://paperpile.com/b/Jvvymh/QdSI8 http://paperpile.com/b/Jvvymh/QdSI8 http://paperpile.com/b/Jvvymh/QdSI8 http://paperpile.com/b/Jvvymh/QdSI8 http://paperpile.com/b/Jvvymh/QdSI8 http://paperpile.com/b/Jvvymh/QdSI8 http://paperpile.com/b/Jvvymh/QdSI8 http://paperpile.com/b/Jvvymh/QdSI8 http://paperpile.com/b/Jvvymh/UT3HB http://paperpile.com/b/Jvvymh/UT3HB http://paperpile.com/b/Jvvymh/UT3HB http://paperpile.com/b/Jvvymh/UT3HB http://paperpile.com/b/Jvvymh/UT3HB http://paperpile.com/b/Jvvymh/UT3HB http://paperpile.com/b/Jvvymh/UT3HB http://paperpile.com/b/Jvvymh/UT3HB 19. Huang, A. T. et al. Assessing the role of multiple mechanisms increasing the age of dengue cases in Thailand. Proc. Natl. Acad. Sci. U. S. A. 119, e2115790119 (2022). 20. Limkittikul, K., Brett, J. & L’Azou, M. Epidemiological trends of dengue disease in Thailand (2000-2011): a systematic literature review. PLoS Negl. Trop. Dis. 8, e3241 (2014). 21. Chareonsook, O., Foy, H. M., Teeraratkul, A. & Silarug, N. Changing epidemiology of dengue hemorrhagic fever in Thailand. Epidemiol. Infect. 122, 161–166 (1999). 22. Rodríguez-Barraquer, I. et al. Revisiting Rayong: shifting seroprofiles of dengue in Thailand and their implications for transmission and control. Am. J. Epidemiol. 179, 353–360 (2014). 23. Yang, X., Quam, M. B. M., Zhang, T. & Sang, S. Global burden for dengue and the evolving pattern in the past 30 years. J. Travel Med. 28, (2021). 24. Endy, T. P. et al. Epidemiology of inapparent and symptomatic acute dengue virus infection: a prospective study of primary school children in Kamphaeng Phet, Thailand. Am. J. Epidemiol. 156, 40–51 (2002). 25. Kuan, G. et al. The Nicaraguan pediatric dengue cohort study: study design, methods, use of information technology, and extension to other infectious diseases. Am. J. Epidemiol. 170, 120–129 (2009). 26. Endy, T. P. et al. Determinants of inapparent and symptomatic dengue infection in a prospective study of primary school children in Kamphaeng Phet, Thailand. PLoS Negl. Trop. Dis. 5, e975 (2011). 27. Gordon, A. et al. The Nicaraguan pediatric dengue cohort study: incidence of inapparent and symptomatic dengue virus infections, 2004-2010. PLoS Negl. Trop. Dis. 7, e2462 (2013). 28. Ly, S. et al. Asymptomatic Dengue Virus Infections, Cambodia, 2012-2013. Emerg. Infect. Dis. 25, 1354–1362 (2019). 29. Reyes, M. et al. Index cluster study of dengue virus infection in Nicaragua. Am. J. Trop. Med. Hyg. 83, 683–689 (2010). http://paperpile.com/b/Jvvymh/XzXY http://paperpile.com/b/Jvvymh/XzXY http://paperpile.com/b/Jvvymh/XzXY http://paperpile.com/b/Jvvymh/XzXY http://paperpile.com/b/Jvvymh/XzXY http://paperpile.com/b/Jvvymh/XzXY http://paperpile.com/b/Jvvymh/XzXY http://paperpile.com/b/Jvvymh/XzXY http://paperpile.com/b/Jvvymh/SNT7j http://paperpile.com/b/Jvvymh/SNT7j http://paperpile.com/b/Jvvymh/SNT7j http://paperpile.com/b/Jvvymh/SNT7j http://paperpile.com/b/Jvvymh/SNT7j http://paperpile.com/b/Jvvymh/SNT7j http://paperpile.com/b/Jvvymh/ZbZ1Z http://paperpile.com/b/Jvvymh/ZbZ1Z http://paperpile.com/b/Jvvymh/ZbZ1Z http://paperpile.com/b/Jvvymh/ZbZ1Z http://paperpile.com/b/Jvvymh/ZbZ1Z http://paperpile.com/b/Jvvymh/ZbZ1Z http://paperpile.com/b/Jvvymh/H0CEC http://paperpile.com/b/Jvvymh/H0CEC http://paperpile.com/b/Jvvymh/H0CEC http://paperpile.com/b/Jvvymh/H0CEC http://paperpile.com/b/Jvvymh/H0CEC http://paperpile.com/b/Jvvymh/H0CEC http://paperpile.com/b/Jvvymh/H0CEC http://paperpile.com/b/Jvvymh/H0CEC http://paperpile.com/b/Jvvymh/JYDIF http://paperpile.com/b/Jvvymh/JYDIF http://paperpile.com/b/Jvvymh/JYDIF http://paperpile.com/b/Jvvymh/JYDIF http://paperpile.com/b/Jvvymh/JYDIF http://paperpile.com/b/Jvvymh/JYDIF http://paperpile.com/b/Jvvymh/mbyPM http://paperpile.com/b/Jvvymh/mbyPM http://paperpile.com/b/Jvvymh/mbyPM http://paperpile.com/b/Jvvymh/mbyPM http://paperpile.com/b/Jvvymh/mbyPM http://paperpile.com/b/Jvvymh/mbyPM http://paperpile.com/b/Jvvymh/mbyPM http://paperpile.com/b/Jvvymh/mbyPM http://paperpile.com/b/Jvvymh/mbyPM http://paperpile.com/b/Jvvymh/uy2sY http://paperpile.com/b/Jvvymh/uy2sY http://paperpile.com/b/Jvvymh/uy2sY http://paperpile.com/b/Jvvymh/uy2sY http://paperpile.com/b/Jvvymh/uy2sY http://paperpile.com/b/Jvvymh/uy2sY http://paperpile.com/b/Jvvymh/uy2sY http://paperpile.com/b/Jvvymh/uy2sY http://paperpile.com/b/Jvvymh/uy2sY http://paperpile.com/b/Jvvymh/m8M3u http://paperpile.com/b/Jvvymh/m8M3u http://paperpile.com/b/Jvvymh/m8M3u http://paperpile.com/b/Jvvymh/m8M3u http://paperpile.com/b/Jvvymh/m8M3u http://paperpile.com/b/Jvvymh/m8M3u http://paperpile.com/b/Jvvymh/m8M3u http://paperpile.com/b/Jvvymh/m8M3u http://paperpile.com/b/Jvvymh/m8M3u http://paperpile.com/b/Jvvymh/mBU72 http://paperpile.com/b/Jvvymh/mBU72 http://paperpile.com/b/Jvvymh/mBU72 http://paperpile.com/b/Jvvymh/mBU72 http://paperpile.com/b/Jvvymh/mBU72 http://paperpile.com/b/Jvvymh/mBU72 http://paperpile.com/b/Jvvymh/mBU72 http://paperpile.com/b/Jvvymh/mBU72 http://paperpile.com/b/Jvvymh/mBU72 http://paperpile.com/b/Jvvymh/NNrg6 http://paperpile.com/b/Jvvymh/NNrg6 http://paperpile.com/b/Jvvymh/NNrg6 http://paperpile.com/b/Jvvymh/NNrg6 http://paperpile.com/b/Jvvymh/NNrg6 http://paperpile.com/b/Jvvymh/NNrg6 http://paperpile.com/b/Jvvymh/NNrg6 http://paperpile.com/b/Jvvymh/NNrg6 http://paperpile.com/b/Jvvymh/FWA3e http://paperpile.com/b/Jvvymh/FWA3e http://paperpile.com/b/Jvvymh/FWA3e http://paperpile.com/b/Jvvymh/FWA3e http://paperpile.com/b/Jvvymh/FWA3e http://paperpile.com/b/Jvvymh/FWA3e http://paperpile.com/b/Jvvymh/FWA3e http://paperpile.com/b/Jvvymh/FWA3e 30. Beckett, C. G. et al. Early detection of dengue infections using cluster sampling around index cases. Am. J. Trop. Med. Hyg. 72, 777–782 (2005). 31. Mammen, M. P. et al. Spatial and temporal clustering of dengue virus transmission in Thai villages. PLoS Med. 5, e205 (2008). 32. Yoon, I.-K. et al. Underrecognized mildly symptomatic viremic dengue virus infections in rural Thai schools and villages. J. Infect. Dis. 206, 389–398 (2012). 33. Asish, P. R., Dasgupta, S., Rachel, G., Bagepally, B. S. & Girish Kumar, C. P. Global prevalence of asymptomatic dengue infections - a systematic review and meta-analysis. Int. J. Infect. Dis. 134, 292–298 (2023). 34. Endy, T. P. et al. Spatial and temporal circulation of dengue virus serotypes: a prospective study of primary school children in Kamphaeng Phet, Thailand. Am. J. Epidemiol. 156, 52– 59 (2002). 35. Sabin, A. B. Research on dengue during World War II. Am. J. Trop. Med. Hyg. 1, 30–50 (1952). 36. Salje, H. et al. Reconstruction of antibody dynamics and infection histories to evaluate dengue risk. Nature 557, 719–723 (2018). 37. Anderson, K. B. et al. An Innovative, Prospective, Hybrid Cohort-Cluster Study Design to Characterize Dengue Virus Transmission in Multigenerational Households in Kamphaeng Phet, Thailand. Am. J. Epidemiol. 189, 648–659 (2020). 38. Clarke, D. H. & Casals, J. Techniques for Hemagglutination and Hemagglutination- Inhibition with Arthropod-Borne Viruses. The American Journal of Tropical Medicine and Hygiene vol. 7 561–573 Preprint at https://doi.org/10.4269/ajtmh.1958.7.561 (1958). 39. Innis, B. L. et al. An enzyme-linked immunosorbent assay to characterize dengue infections where dengue and Japanese encephalitis co-circulate. Am. J. Trop. Med. Hyg. 40, 418–427 (1989). 40. Lanciotti, R. S., Calisher, C. H., Gubler, D. J., Chang, G. J. & Vorndam, A. V. Rapid http://paperpile.com/b/Jvvymh/0RWec http://paperpile.com/b/Jvvymh/0RWec http://paperpile.com/b/Jvvymh/0RWec http://paperpile.com/b/Jvvymh/0RWec http://paperpile.com/b/Jvvymh/0RWec http://paperpile.com/b/Jvvymh/0RWec http://paperpile.com/b/Jvvymh/0RWec http://paperpile.com/b/Jvvymh/0RWec http://paperpile.com/b/Jvvymh/CIoS3 http://paperpile.com/b/Jvvymh/CIoS3 http://paperpile.com/b/Jvvymh/CIoS3 http://paperpile.com/b/Jvvymh/CIoS3 http://paperpile.com/b/Jvvymh/CIoS3 http://paperpile.com/b/Jvvymh/CIoS3 http://paperpile.com/b/Jvvymh/CIoS3 http://paperpile.com/b/Jvvymh/CIoS3 http://paperpile.com/b/Jvvymh/HRrhb http://paperpile.com/b/Jvvymh/HRrhb http://paperpile.com/b/Jvvymh/HRrhb http://paperpile.com/b/Jvvymh/HRrhb http://paperpile.com/b/Jvvymh/HRrhb http://paperpile.com/b/Jvvymh/HRrhb http://paperpile.com/b/Jvvymh/HRrhb http://paperpile.com/b/Jvvymh/HRrhb http://paperpile.com/b/Jvvymh/Xf16 http://paperpile.com/b/Jvvymh/Xf16 http://paperpile.com/b/Jvvymh/Xf16 http://paperpile.com/b/Jvvymh/Xf16 http://paperpile.com/b/Jvvymh/Xf16 http://paperpile.com/b/Jvvymh/Xf16 http://paperpile.com/b/Jvvymh/Xf16 http://paperpile.com/b/Jvvymh/rdE9N http://paperpile.com/b/Jvvymh/rdE9N http://paperpile.com/b/Jvvymh/rdE9N http://paperpile.com/b/Jvvymh/rdE9N http://paperpile.com/b/Jvvymh/rdE9N http://paperpile.com/b/Jvvymh/rdE9N http://paperpile.com/b/Jvvymh/rdE9N http://paperpile.com/b/Jvvymh/rdE9N http://paperpile.com/b/Jvvymh/rdE9N http://paperpile.com/b/Jvvymh/qZHm http://paperpile.com/b/Jvvymh/qZHm http://paperpile.com/b/Jvvymh/qZHm http://paperpile.com/b/Jvvymh/qZHm http://paperpile.com/b/Jvvymh/qZHm http://paperpile.com/b/Jvvymh/qZHm http://paperpile.com/b/Jvvymh/rVDi7 http://paperpile.com/b/Jvvymh/rVDi7 http://paperpile.com/b/Jvvymh/rVDi7 http://paperpile.com/b/Jvvymh/rVDi7 http://paperpile.com/b/Jvvymh/rVDi7 http://paperpile.com/b/Jvvymh/rVDi7 http://paperpile.com/b/Jvvymh/rVDi7 http://paperpile.com/b/Jvvymh/rVDi7 http://paperpile.com/b/Jvvymh/vxURl http://paperpile.com/b/Jvvymh/vxURl http://paperpile.com/b/Jvvymh/vxURl http://paperpile.com/b/Jvvymh/vxURl http://paperpile.com/b/Jvvymh/vxURl http://paperpile.com/b/Jvvymh/vxURl http://paperpile.com/b/Jvvymh/vxURl http://paperpile.com/b/Jvvymh/vxURl http://paperpile.com/b/Jvvymh/vxURl http://paperpile.com/b/Jvvymh/9gPM8 http://paperpile.com/b/Jvvymh/9gPM8 http://paperpile.com/b/Jvvymh/9gPM8 http://paperpile.com/b/Jvvymh/9gPM8 http://paperpile.com/b/Jvvymh/9gPM8 http://dx.doi.org/10.4269/ajtmh.1958.7.561 http://paperpile.com/b/Jvvymh/9gPM8 http://paperpile.com/b/Jvvymh/JHUVe http://paperpile.com/b/Jvvymh/JHUVe http://paperpile.com/b/Jvvymh/JHUVe http://paperpile.com/b/Jvvymh/JHUVe http://paperpile.com/b/Jvvymh/JHUVe http://paperpile.com/b/Jvvymh/JHUVe http://paperpile.com/b/Jvvymh/JHUVe http://paperpile.com/b/Jvvymh/JHUVe http://paperpile.com/b/Jvvymh/JHUVe http://paperpile.com/b/Jvvymh/DdI5G detection and typing of dengue viruses from clinical samples by using reverse transcriptase-polymerase chain reaction. Journal of Clinical Microbiology vol. 30 545–551 Preprint at https://doi.org/10.1128/jcm.30.3.545-551.1992 (1992). 41. Nisalak, A. LABORATORY DIAGNOSIS OF DENGUE VIRUS INFECTIONS. Southeast Asian J. Trop. Med. Public Health 46 Suppl 1, 55–76 (2015). 42. Sirikajornpan, K. et al. COMPARISON OF ANTI-DENV/JEV IG-A ENZYME-LINKED IMMUNOSORBENT ASSAY AND HEMAGGLUTINATION INHIBITION ASSAY. https://www.tm.mahidol.ac.th/seameo/2018-49-4/10-75055-629.pdf (2018). 43. Sirikajornpan, K. et al. Standardization and Evaluation of an Anti-ZIKV IgM ELISA Assay for the Serological Diagnosis of Zika Virus Infection. Am. J. Trop. Med. Hyg. 105, 936–941 (2021). 44. Chen, T. et al. xgboost: Extreme Gradient Boosting. Preprint at https://CRAN.R- project.org/package=xgboost (2022). 45. Chen, T. & Guestrin, C. XGBoost. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining Preprint at https://doi.org/10.1145/2939672.2939785 (2016). 46. R Core Team. R: A Language and Environment for Statistical Computing. https://www.R- project.org/ (2022). 47. Brooks, M. et al. glmmTMB Balances Speed and Flexibility Among Packages for Zero- inflated Generalized Linear Mixed Modeling. The R Journal vol. 9 378 Preprint at https://doi.org/10.32614/rj-2017-066 (2017). 48. Chan, K. R. et al. Serological cross-reactivity among common flaviviruses. Front. Cell. Infect. Microbiol. 12, 975398 (2022). 49. Moodie, Z. et al. Neutralizing Antibody Correlates Analysis of Tetravalent Dengue Vaccine Efficacy Trials in Asia and Latin America. J. Infect. Dis. 217, 742–753 (2018). 50. Katzelnick, L. C., Montoya, M., Gresh, L., Balmaseda, A. & Harris, E. Neutralizing antibody http://paperpile.com/b/Jvvymh/DdI5G http://paperpile.com/b/Jvvymh/DdI5G http://paperpile.com/b/Jvvymh/DdI5G http://paperpile.com/b/Jvvymh/DdI5G http://paperpile.com/b/Jvvymh/DdI5G http://dx.doi.org/10.1128/jcm.30.3.545-551.1992 http://paperpile.com/b/Jvvymh/DdI5G http://paperpile.com/b/Jvvymh/ecUwq http://paperpile.com/b/Jvvymh/ecUwq http://paperpile.com/b/Jvvymh/ecUwq http://paperpile.com/b/Jvvymh/ecUwq http://paperpile.com/b/Jvvymh/ecUwq http://paperpile.com/b/Jvvymh/ecUwq http://paperpile.com/b/Jvvymh/ZNBQb http://paperpile.com/b/Jvvymh/ZNBQb http://paperpile.com/b/Jvvymh/ZNBQb http://paperpile.com/b/Jvvymh/ZNBQb https://www.tm.mahidol.ac.th/seameo/2018-49-4/10-75055-629.pdf http://paperpile.com/b/Jvvymh/ZNBQb http://paperpile.com/b/Jvvymh/mn1ny http://paperpile.com/b/Jvvymh/mn1ny http://paperpile.com/b/Jvvymh/mn1ny http://paperpile.com/b/Jvvymh/mn1ny http://paperpile.com/b/Jvvymh/mn1ny http://paperpile.com/b/Jvvymh/mn1ny http://paperpile.com/b/Jvvymh/mn1ny http://paperpile.com/b/Jvvymh/mn1ny http://paperpile.com/b/Jvvymh/mn1ny http://paperpile.com/b/Jvvymh/BjfCp http://paperpile.com/b/Jvvymh/BjfCp http://paperpile.com/b/Jvvymh/BjfCp https://cran.r-project.org/package=xgboost https://cran.r-project.org/package=xgboost http://paperpile.com/b/Jvvymh/BjfCp http://paperpile.com/b/Jvvymh/G1SCg http://paperpile.com/b/Jvvymh/G1SCg http://paperpile.com/b/Jvvymh/G1SCg http://paperpile.com/b/Jvvymh/G1SCg http://paperpile.com/b/Jvvymh/G1SCg http://dx.doi.org/10.1145/2939672.2939785 http://paperpile.com/b/Jvvymh/G1SCg http://paperpile.com/b/Jvvymh/1DlDk http://paperpile.com/b/Jvvymh/1DlDk http://paperpile.com/b/Jvvymh/1DlDk https://www.r-project.org/ https://www.r-project.org/ http://paperpile.com/b/Jvvymh/1DlDk http://paperpile.com/b/Jvvymh/4FI23 http://paperpile.com/b/Jvvymh/4FI23 http://paperpile.com/b/Jvvymh/4FI23 http://paperpile.com/b/Jvvymh/4FI23 http://paperpile.com/b/Jvvymh/4FI23 http://paperpile.com/b/Jvvymh/4FI23 http://paperpile.com/b/Jvvymh/4FI23 http://dx.doi.org/10.32614/rj-2017-066 http://paperpile.com/b/Jvvymh/4FI23 http://paperpile.com/b/Jvvymh/LUM7 http://paperpile.com/b/Jvvymh/LUM7 http://paperpile.com/b/Jvvymh/LUM7 http://paperpile.com/b/Jvvymh/LUM7 http://paperpile.com/b/Jvvymh/LUM7 http://paperpile.com/b/Jvvymh/LUM7 http://paperpile.com/b/Jvvymh/LUM7 http://paperpile.com/b/Jvvymh/LUM7 http://paperpile.com/b/Jvvymh/HOkU9 http://paperpile.com/b/Jvvymh/HOkU9 http://paperpile.com/b/Jvvymh/HOkU9 http://paperpile.com/b/Jvvymh/HOkU9 http://paperpile.com/b/Jvvymh/HOkU9 http://paperpile.com/b/Jvvymh/HOkU9 http://paperpile.com/b/Jvvymh/HOkU9 http://paperpile.com/b/Jvvymh/HOkU9 http://paperpile.com/b/Jvvymh/4p6fv titers against dengue virus correlate with protection from symptomatic infection in a longitudinal cohort. Proc. Natl. Acad. Sci. U. S. A. 113, 728–733 (2016). 51. Cummings, D. A. T. et al. The Impact of the Demographic Transition on Dengue in Thailand: Insights from a Statistical Analysis and Mathematical Modeling. PLoS Medicine vol. 6 e1000139 Preprint at https://doi.org/10.1371/journal.pmed.1000139 (2009). 52. Ribeiro Dos Santos, G. et al. Estimating the effect of the wMel release programme on the incidence of dengue and chikungunya in Rio de Janeiro, Brazil: a spatiotemporal modelling study. Lancet Infect. Dis. 22, 1587–1595 (2022). 53. Salje, H. et al. Evaluation of the extended efficacy of the Dengvaxia vaccine against symptomatic and subclinical dengue infection. Nat. Med. (2021) doi:10.1038/s41591-021- 01392-9. 54. O’Driscoll, M. et al. Maternally derived antibody titer dynamics and risk of hospitalized infant dengue disease. Proc. Natl. Acad. Sci. U. S. A. 120, e2308221120 (2023). http://paperpile.com/b/Jvvymh/4p6fv http://paperpile.com/b/Jvvymh/4p6fv http://paperpile.com/b/Jvvymh/4p6fv http://paperpile.com/b/Jvvymh/4p6fv http://paperpile.com/b/Jvvymh/4p6fv http://paperpile.com/b/Jvvymh/4p6fv http://paperpile.com/b/Jvvymh/vxkNM http://paperpile.com/b/Jvvymh/vxkNM http://paperpile.com/b/Jvvymh/vxkNM http://paperpile.com/b/Jvvymh/vxkNM http://paperpile.com/b/Jvvymh/vxkNM http://paperpile.com/b/Jvvymh/vxkNM http://paperpile.com/b/Jvvymh/vxkNM http://dx.doi.org/10.1371/journal.pmed.1000139 http://paperpile.com/b/Jvvymh/vxkNM http://paperpile.com/b/Jvvymh/Y4vlx http://paperpile.com/b/Jvvymh/Y4vlx http://paperpile.com/b/Jvvymh/Y4vlx http://paperpile.com/b/Jvvymh/Y4vlx http://paperpile.com/b/Jvvymh/Y4vlx http://paperpile.com/b/Jvvymh/Y4vlx http://paperpile.com/b/Jvvymh/Y4vlx http://paperpile.com/b/Jvvymh/Y4vlx http://paperpile.com/b/Jvvymh/Y4vlx http://paperpile.com/b/Jvvymh/EIedm http://paperpile.com/b/Jvvymh/EIedm http://paperpile.com/b/Jvvymh/EIedm http://paperpile.com/b/Jvvymh/EIedm http://paperpile.com/b/Jvvymh/EIedm http://paperpile.com/b/Jvvymh/EIedm http://dx.doi.org/10.1038/s41591-021-01392-9 http://dx.doi.org/10.1038/s41591-021-01392-9 http://paperpile.com/b/Jvvymh/EIedm http://paperpile.com/b/Jvvymh/UHKi http://paperpile.com/b/Jvvymh/UHKi http://paperpile.com/b/Jvvymh/UHKi http://paperpile.com/b/Jvvymh/UHKi http://paperpile.com/b/Jvvymh/UHKi http://paperpile.com/b/Jvvymh/UHKi http://paperpile.com/b/Jvvymh/UHKi http://paperpile.com/b/Jvvymh/UHKi