Modelling seasonal variations in the age and incidence of Kawasaki disease to explore possible infectious aetiologies

Abstract

The average age of infection is expected to vary during seasonal epidemics in a way that is predictable from the epidemiological features, such as the duration of infectiousness and the nature of population mixing. However, it is not known whether such changes can be detected and verified using routinely collected data. We examined the correlation between the weekly number and average age of cases using data on pre-vaccination measles and rotavirus. We show that age-incidence patterns can be observed and predicted for these childhood infections. Incorporating additional information about important features of the transmission dynamics improves the correspondence between model predictions and empirical data. We then explored whether knowledge of the age-incidence pattern can shed light on the epidemiological features of diseases of unknown aetiology, such as Kawasaki disease (KD). Our results indicate KD is unlikely to be triggered by a single acute immunizing infection, but is consistent with an infection of longer duration, a non-immunizing infection or co-infection with an acute agent and one with longer duration. Age-incidence patterns can lend insight into important epidemiological features of infections, providing information on transmission-relevant population mixing for known infections and clues about the aetiology of complex paediatric diseases.

Figure 1.

Relationship between the average age and incidence of measles notifications in Copenhagen, Denmark. (a) The number of weekly measles notifications (blue) among children less than 15 years of age in Copenhagen from 1905 to 1918, and the average age of measles cases (red). We smoothed the data using a five week moving average. (b) Mean number and average age of measles notifications by week of the year. (c) Correlation coefficients between the number of notifications at time t and the average age of cases at time t + l for the longitudinal analysis and aggregate analysis. The dotted lines represent the range within which 95% of the maximum and minimum correlations fell when we randomly permuted the average age time series. (d) Model-predicted relationship between the number and average age of cases assuming an SEIR structure and RAS-like mixing with school-term forcing.

Figure 2.

Relationship between the average age and incidence of rotavirus hospitalizations in the United States. (a) The number of weekly rotavirus hospitalizations (blue) among children less than 5 years of age in 16 US states from 1997 to 2005, and the average age of rotavirus patients (red). We smoothed the data using a five week moving average. (b) Mean number and average age of rotavirus cases by week of the year. (c) Correlation coefficients between the number of hospitalizations at time t and the average age of patients at time t + l for the longitudinal analysis and aggregate analysis. The dotted lines represent the range within which 95% of the maximum and minimum correlations fell when we randomly permuted the average age time series. (d) Model-predicted relationship between the number and average age of rotavirus cases assuming an SIRS-like structure for a best-fit model [17].

Figure 3.

Relationship between the average age and incidence of Kawasaki disease hospitalizations in the United States. (a) The number of weekly KD hospitalizations (blue) among children less than 10 years of age in 10 US states from 1989 to 2003, and the average age of KD patients (red). We smoothed the data using a five week moving average. (b) Mean number and average age of KD cases by week of the year. (c) Correlation coefficients between the number of hospitalizations at time t and the average age of patients at time t + l for the longitudinal analysis and the aggregate analysis. The dotted lines represent the range within which 95% of the maximum and minimum correlations fell when we randomly permuted the average age time series. (d) Age distribution of KD hospitalizations among children less than 10 years of age in 10 US states compared with model-predicted age distributions under different population mixing assumptions. Black bars, KD hospitalizations; blue bars, homogeneous mixing; yellow bars, assortative mixing; pink bars, self-reported mixing; green bars, RAS mixing.