Deciphering the impacts of vaccination and immunity on pertussis epidemiology in Thailand

Abstract

Pertussis is a highly infectious respiratory disease that is currently responsible for nearly 300,000 annual deaths worldwide, primarily in infants in developing countries. Despite sustained high vaccine uptake, a resurgence in pertussis incidence has been reported in a number of countries. This resurgence has led to critical questions regarding the transmission impacts of vaccination and pertussis immunology. We analyzed pertussis incidence in Thailand--both age-stratified and longitudinal aggregate reports--over the past 30 y. To dissect the contributions of waning pertussis immunity and repeat infections to pertussis epidemiology in Thailand following a pronounced increase in vaccine uptake, we used likelihood-based statistical inference methods to evaluate the support for multiple competing transmission models. We found that, in contrast to other settings, there is no evidence for pertussis resurgence in Thailand, with each model examined pointing to a substantial rise in herd immunity over the past 30 y. Using a variety of empirical metrics, we verified our findings by documenting signatures of changing herd immunity over the study period. Importantly, this work leads to the conclusion that repeat infections have played little role in shaping pertussis epidemiology in Thailand. Our results are surprisingly emphatic in support of measurable impact of herd immunity given the uncertainty associated with pertussis epidemiology.

Fig. 1.

Time series of monthly pertussis incidence per 100,000 individuals (black), annual vaccine uptake (red), and the annual per capita birth rate (green) in Thailand. (Inset) Incidence data from 1995–2010 at a finer resolution. The background shading represents three distinct vaccine eras: low vaccine uptake followed by a steep transition to high uptake, which subsequently remains at high levels.

Fig. 2.

Schematic diagram of competing model structures. The baseline SIR model is denoted by light blue lines; the SIRS model additionally includes dark blue; the model includes green; and the model includes red lines and text. Here, λ is the force of infection, p is the fraction of infants vaccinated, σ is the rate of immunity loss, ε is the probability that a susceptible individual who was previously vaccinated or infected has their immunity boosted upon exposure, and κ is the boosting coefficient.

Fig. 3.

(A) Plot of the log-likelihood value corresponding to the MLE for each value of waning immunity over the entire duration of the time series for the SIRS (blue, model II), (green, model III), and (red, model IV) models. Likelihood values are rescaled so that zero corresponds to the best likelihood for each model, and the vertical dashed line indicates the upper bound for the 95% confidence interval ( years is the lower bound of the confidence interval for models II and III). The duration of immunity is not identifiable for model IV. (B) Simulations of the reported cases for each model using the associated MLE parameters. In each, the dark line is the average over 1,000 realizations of the model, and light shading represents the upper and lower quartiles. (Upper) Simulations of models II–IV. (Lower) Black line is the data, and the light blue represents simulations from the best-fitting model (model I, SIR). The corresponding values (computed as 1 − SSE/SST where SSE is the error sum of squares and SST is the total sum of squares) are 0.93, −1.0, 0.22, and 0.92 for models I–IV, respectively. (Inset) Cumulative conditional log-likelihood values for each model. All models perform well during the first vaccine era, but the performance of models II and III significantly declines during the final vaccine era because these models are unable to capture the transition to very low incidence, even with long durations of immunity.

Fig. 4.

Mean number of fadeouts for each of Thailand’s provinces in all vaccine eras. Exponential curves [i.e., where N is the population size and A and b are the estimated parameters] are fit to each vaccine era using standard regression techniques to determine an estimate of the CCS; similar protocols have been used elsewhere (23). Bangkok is excluded from the fit as it is an outlier due to its high population size relative to other provinces (∼5 × 10⁶). The horizontal dashed gray line represents a mean of one fadeout per year, and the intersection of the vertical dashed red and black lines with the gray line represents the estimated CCS in the first and second vaccine eras, respectively. Below the CCS, frequent disease extinctions are expected to occur. Though some of the shift may be attributed to an increase in the age of infection, leading to more asymptomatic cases, the magnitude of the shifts strongly points to reduced transmission following vaccination (SI Materials and Methods).

Fig. 5.

Log transformation of the average annual incidence (Upper) and in infants only (Lower) for each vaccine era for all of Thailand’s provinces.

Fig. 6.

Mean age structure of case notifications in Thailand with vaccine uptake of <46% (1981–1983, red) and high vaccine uptake of >95% (1996–2000, blue). Shading indicates the SEM, noting that the high variation results from province-level differences in the magnitude of incidence.