Assessing the role of multiple mechanisms increasing the age of dengue cases in Thailand

Abstract

The mean age of dengue hemorrhagic fever (DHF) cases increased considerably in Thailand from 8.1 to 24.3 y between 1981 and 2017 (mean annual increase of 0.45 y). Alternative proposed explanations for this trend, such as changes in surveillance practices, reduced mosquito–human contact, and shifts in population demographics, have different implications for global dengue epidemiology. To evaluate the contribution of each of these hypothesized mechanisms to the observed data, we developed 20 nested epidemiological models of dengue virus infection, allowing for variation over time in population demographics, infection hazards, and reporting rates. We also quantified the effect of removing or retaining each source of variation in simulations of the age trajectory. Shifts in the age structure of susceptibility explained 58% of the observed change in age. Adding heterogeneous reporting by age and reductions in per-serotype infection hazard to models with shifts in susceptibility explained an additional 42%. Reductions in infection hazards were mostly driven by changes in the number of infectious individuals at any time (another consequence of shifting age demographics) rather than changes in the transmissibility of individual infections. We conclude that the demographic transition drives the overwhelming majority of the observed change as it changes both the age structure of susceptibility and the number of infectious individuals. With the projected Thai population age structure, our results suggest a continuing increase in age of DHF cases, shifting the burden toward individuals with more comorbidity. These insights into dengue epidemiology may be relevant to many regions of the globe currently undergoing comparable changes in population demographics.

Keywords: Thailand; aging demography; dengue epidemiology; force of infection; infectious disease.

Fig. 1.

The increase in the mean ages of (A) the population (black) and of reported cases with DHF (red) from 1981 to 2017 in the 72 provinces of Thailand calculated using midpoints of the age strata. (B) Country-level age distribution at 9-y intervals; the underlying population is in gray, and DHF cases are in red. Bin widths reflect the age strata reported for the cases at those times. SI Appendix, Fig. S2 shows the age distribution of all years in the dataset.

Fig. 2.

Hypotheses of factors driving the age increase of DHF cases encoded into (A) model parameters. (B) Diagrammatic representation of the model. Model parameters are grouped according the hypothesized drivers of changes in age: changes in reporting rate (brown), clinical detectability (purple), and transmission intensity (black and green).

Fig. 3.

Parameter estimates of the country-wide best-fitting model. The model includes (A) time-varying per-serotype infection hazards, $\overline{\tau }(t)$; (B) age-specific multiplicative modifiers for infection hazards, $\kappa (a)$; (C) time-varying reporting rates, $\varphi (t)$; (D) age-specific reporting rates, $\varphi (a)$; and (F) time-varying clinical detectability of infections Q(i, t) shown in colors matching the model diagram in Fig. 2B. Shades represent provincial 95% credible intervals, points are medians of provincial medians, and whiskers are 95% IQRs of the medians. The piecewise constant time-varying reporting rates are shown in repeats across the ranges within their bins. The estimates are compared against (E) the time series of reported country-level DHF case counts per 1,000 population.

Fig. 4.

Provincial mean age of simulated cases when (A) parameters were kept as fitted and when (B) all variations, whether in age or time, were removed. (C) The mean age difference between when variations in each of the component estimates were removed (by replacing with its mean) relative to when all variations were retained as fitted. (D) The mean age difference between when only a single component retains variations relative to when variations in all components were removed.

Fig. 5.

Biweekly infection histories in the Thai population. (A) Biweekly time series of population susceptibility reconstructed from fitted provincial infection hazards. (B) Proportions of total dengue infections at each point in time that were first, second, third, and fourth infections and the infectious fraction in the population. (C) Biweekly FoI of DENV calculated by averaging age-specific FoI over all individuals in the population accounting for their susceptibility status. (D) Biweekly transmission efficiencies calculated from the FoI and infectious fractions. (E) Relationship between transmission efficiencies and FoI (log scaled).