Reconciling heterogeneous dengue virus infection risk estimates from different study designs

Abstract

Uncovering rates at which susceptible individuals become infected with a pathogen, i.e., the force of infection (FOI), is essential for assessing transmission risk and reconstructing distribution of immunity in a population. For dengue, reconstructing exposure and susceptibility statuses from the measured FOI is of particular significance as prior exposure is a strong risk factor for severe disease. FOI can be measured via many study designs. Longitudinal serology is considered gold standard measurements, as they directly track the transition of seronegative individuals to seropositive due to incident infections (seroincidence). Cross-sectional serology can provide estimates of FOI by contrasting seroprevalence across ages. Age of reported cases can also be used to infer FOI. Agreement of these measurements, however, has not been assessed. Using 26 y of data from cohort studies and hospital-attended cases from Kamphaeng Phet province, Thailand, we found FOI estimates from the three sources to be highly inconsistent. Annual FOI estimates from seroincidence were 1.75 to 4.05 times higher than case-derived FOI. Seroprevalence-derived was moderately correlated with case-derived FOI (correlation coefficient = 0.47) with slightly lower estimates. Through extensive simulations and theoretical analysis, we show that incongruences between methods can result from failing to account for dengue antibody kinetics, assay noise, and heterogeneity in FOI across ages. Extending standard inference models to include these processes reconciled the FOI and susceptibility estimates. Our results highlight the importance of comparing inferences across multiple data types to uncover additional insights not attainable through a single data type/analysis.

Keywords: catalytic model; force of infection; seroincidence; seroprevalence; susceptibility.

Fig. 1.

Data to inform infection risk estimates.

Fig. 2.

Study data. (A) Map of Kamphaeng Phet province showing spatial coverage of cohort studies (colored) and location of Kamphaeng Phet Provincial Hospital (KPPH, blue point). (B) Number of bleeds by year (Top) and percentages of with GMT ≥ 10 by age and year of collection (Bottom). (C) Number of dengue cases reported at KPPH per thousand population by year (Top), and by year and age (Bottom).

Fig. 3.

Estimates from standard force of infection (FOI) inference models. (A) Annual FOI estimated from each of the data sources: seroincidence data (red) and seroprevalence data (yellow) using seropositivity threshold of GMT >= 10, and case data (blue). Annual FOI estimated from longitudinal samples of KPS1 [black, (8)] are included for comparison. (B) Relationship between serology-derived (y-axis, respective colors) and case-derived susceptibility reconstructions (x-axis). Each point in the reconstruction represents the proportion in each age-year that has not been infected with DENV (naive), has been infected by one serotype (monotypic) or more than one serotype (multitypic).

Fig. 4.

Biases in serology-derived force of infection (FOI) using simulated data with known true parameters. (A) Illustration of anti-DENV antibody kinetics as an individual acquires a cross-reactive (CXR) virus infection or vaccination (i.e., not DENV), one DENV infection, and >1 DENV infections. Measured titers distribute around the true underlying titers with variability depending on the assay characteristics. (B) Schematic of biases in serology-derived FOI and their correction efficiencies at low and high seropositivity thresholds. (C–F) Antibody kinetics, assay characteristics (rows), and distribution of infection risk in age among susceptible individuals (columns) used to generate observed titer measurements and FOI inferred from those respective simulations using standard models for seroincidence (red) and seroprevalence (yellow).

Fig. 5.

Infection risks in Kamphaeng Phet. (A) Distribution of infection risk by age, (B) annual per-serotype FOI inferred from the joint serology model (brown) and the extended case-based model (blue), and (C) relationships between the susceptibility reconstructions. Each point in the reconstruction represents the proportion in each age-year that has not been infected with DENV (naive), has been infected by one serotype (monotypic) or more than one serotype (multitypic). (D) Effects of presumed test positive probabilities in DENV-naives (x-axis) and long-term test positive probabilities in monotypically infected individuals (y-axis) on the correlation and ratio between temporal FOIs inferred from the extended case-based model and temporal FOIs inferred from a single data source (either seroincidence or seroprevalence at seropositivity threshold of 10) imposed with age-specific risk inferred from the extended case-based model. Test positive probabilities estimated from the joint serology model are annotated as crosses for comparison.