Cross-serotype interactions and disease outcome prediction of dengue infections in Vietnam

Abstract

Dengue pathogenesis is extremely complex. Dengue infections are thought to induce life-long immunity from homologous challenges as well as a multi-factorial heterologous risk enhancement. Here, we use the data collected from a prospective cohort study of dengue infections in schoolchildren in Vietnam to disentangle how serotype interactions modulate clinical disease risk in the year following serum collection. We use multinomial logistic regression to correlate the yearly neutralizing antibody measurements obtained with each infecting serotype in all dengue clinical cases collected over the course of 6 years (2004-2009). This allowed us to extrapolate a fully discretised matrix of serotype interactions, revealing clear signals of increased risk of clinical illness in individuals primed with a previous dengue infection. The sequences of infections which produced a higher risk of dengue fever upon secondary infection are: DEN1 followed by DEN2; DEN1 followed by DEN4; DEN2 followed by DEN3; and DEN4 followed by DEN3. We also used this longitudinal data to train a machine learning algorithm on antibody titre differences between consecutive years to unveil asymptomatic dengue infections and estimate asymptomatic infection to clinical case ratios over time, allowing for a better characterisation of the population's past exposure to different serotypes.

Figure 1

Odds ratio of dengue illness in the year following neutralising antibody titre measurements. The odds ratio presented here refers to children with a homologous PRNT50 titre under 40, i.e., non-immune to the reference dengue serotype thus reflecting the ratio between the odds of having clinical dengue illness when immune to heterologous serotypes (conditional on having a PRNT50 < 40 to the homologous serotype) and the odds of becoming ill in the absence of immunity to any serotype.

Figure 2

Evaluation of the asymptomatic infection prediction model. The violin plots in the top two rows display the distribution of the mean rise in titres (across all serotypes) in consecutive years for predicted infections and predicted non-infections compared to the observed cases. The bottom row shows the time series of predicted infections by serotype (right) compared to the observed dengue clinical case time series (left).