Dynamics of Pertussis Transmission in the United States

F. M. G. Magpantay; P. Rohani


Am J Epidemiol. 2015;181(12):921-931. 

In This Article


The resurgence of pertussis in some countries with high vaccine coverage has identified gaps in our understanding of the pathogen.[5] In the United States, pertussis has rebounded in almost every state, yet the timing of its resurgence has been highly variable, spread out over 3 decades[36]—an observation that has yet to be explained. Part of the answer may lie in the differential historical epidemiology of the bacterium across states. Studies of a variety of infectious diseases have indicated that past patterns of transmission can have long-term consequences for disease dynamics due to their effects on the susceptibility profile of the population.[18,37–41] For this reason, we scrutinized records on pertussis incidence during the transition to routine infant immunization in the United States.

We began by examining the interepidemic periods of pertussis in the continental United States from 1938 to 1955. Using wavelet decomposition, we identified 4 distinct patterns of periodicity (Figure 4), according to which dynamics in each state were classified. We found that most commonly (20 group 3 states), pertussis epidemics during this era occurred 2–5 years apart. In a further 17 states (group 2), initially annual outbreaks gave way, by the late 1940s, to multiennial cycles, with a period ranging from 2 years to 6 years. Finally, 5 group 1 states exhibited annual epidemics throughout the time period.

We sought to explain these contrasting patterns among states assuming that they arose from differences in demographic and epidemiologic factors (Figure 5). For instance, epidemiologic theory predicts that annual epidemics can result from low transmission seasonality, high rates of baseline transmission, and rapid recruitment of susceptibles. Thus, we correctly predicted that school attendance and the proportion of families with no children would be lower in group 1 states than in group 3 states and that crowding, per capita birth rates, and the proportion of families with more than 1 child would be higher in group 1 than in group 3, with group 2 having intermediate values. Additionally, because per capita health spending correlates with estimates of pertussis vaccine uptake in the contemporary setting (see Web Appendix 7 and Web Figure 3 http://aje.oxfordjournals.org/content/181/12/921/suppl/DC1), we predicted that group 3 states would be associated with higher historical estimates of health spending than states in group 1. While indeed the mean level of health spending in group 3 states was higher, the difference was not statistically significant (in Web Appendix 8 http://aje.oxfordjournals.org/content/181/12/921/suppl/DC1, we use cross-validation to demonstrate the robustness of this conclusion to the removal of outliers). The remaining demographic covariates we examined were not significantly different across groups.

Our prediction on the association between periodicity and school attendance assumed the latter to be a surrogate for transmission seasonality.[42,43] However, in previous studies of seasonality in pertussis incidence (e.g., 1996–2006 in the Netherlands[44] and 1977–1982 in England and Wales[45]), investigators concluded that seasonality of pertussis incidence is not driven by the school calendar. Although these findings were based on data from other countries collected in different time periods, they indicate that alternative mechanisms may underpin our observed association between school attendance and periodicity. Specifically, because school attendance was inversely correlated with household crowding (regression slope 95% confidence interval (CI): −0.58, −0.29; adjusted R 2 = 0.047), our detection of a correlation may indeed be mostly due to the association with household crowding.

Somewhat surprisingly, the proportion of the population living in an urban area—which intuitively may be assumed to be a driver of transmission—was not significantly different across groups in either test. Subsequent analyses revealed a negative correlation between urban population and crowding within households (regression slope 95% CI: −1.57, −0.86; adjusted R 2 = 0.53). Hence, although a larger urban concentration could have provided greater opportunities for contacts and transmission between households, we found its main transmission impact to be neutral, perhaps as a result of improved living conditions in historical urban settings.

As an additional test of our conclusions regarding the influence of demographic factors in shaping pertussis epidemiology in the 1938–1955 data, we fitted a multinomial logistic model. As detailed in Web Appendix 9 (Web Table 5 and Web Figure 4) http://aje.oxfordjournals.org/content/181/12/921/suppl/DC1, we classified states into group 1, 2, or 3 based on those demographic parameters that determine the susceptible recruitment rate, including the per capita birth rate and the fraction of families with no children. Reassuringly, the signs of the model coefficients were consistent with the relationships described in Figure 5.

The next set of state characteristics we explored was that associated with the geography of states and the historical epidemiology of pertussis. Of these, latitude, historical weekly incidence, and the seasonal peak were significantly different across groups. We emphasize that historical pertussis incidence was higher in group 1 states than in group 3 states. Assuming comparable reporting fidelity across groups, this finding suggests higher rates of transmission in group 1, as also indicated by our analyses of demographic factors.

The association with latitude is intriguing. However, because of the negative correlations between latitude and household crowding (regression slope 95% CI: −36.6, −17.2; adjusted R2 = 0.42) and per capita birth rate (regression slope 95% CI: −13.2, −1.3; adjusted R2 = 0.11), we suspect that the difference in mean latitude between groups may be due to these demographic differences among northern and southern states. Thus, we submit that differences in periodicity with respect to latitude may be more parsimoniously attributed to latitudinal demographic variation[46] than to climatic drivers, such as day length or average temperature.[47] Further analyses of geographical variation in US pertussis dynamics can be found in Web Appendices 10 and 11 (Web Figures 5–7) http://aje.oxfordjournals.org/content/181/12/921/suppl/DC1.

We also attempted to relate the historical epidemiology of pertussis to recent pertussis dynamics. We compared summary measures of resurgence across groups by performing segmented linear regression on the NNDSS data from 1951 to 2000. This pinpointed the start date (the breakpoint) and the speed (the subsequent slope) of the resurgence. We also compared the 2001–2010 mean monthly incidence rates across groups. Interestingly, the slope of the resurgence and recent incidence were lowest for group 2 states. To better understand this finding, we explored the association between contemporary vaccine coverage (1995–2012) and historical grouping. As shown in Web Table 4 http://aje.oxfordjournals.org/content/181/12/921/suppl/DC1, recent vaccine uptake is highest in group 2 states and lowest in group 1. If this hierarchy in coverage across groups was also in effect during the early vaccine era, and perhaps more pronounced, then it would likely explain the relationship between the historical grouping of states and recent resurgence and incidence. Lastly, there was no significant difference in breakpoints across groups.

To our knowledge, this is the first study of pertussis epidemiology in the United States during the early vaccine era. Despite the absence of information on vaccine uptake, we were able to paint a comprehensive picture of the epidemiology of pertussis during this period. Our analyses revealed how the dynamics of pertussis incidence are associated with state-specific demographic characteristics. In particular, in contrast to those states with multiennial epidemics, states in which pertussis incidence was rigidly annual tended to have higher per capita birth rates, greater household crowding, more children per family, and lower school attendance. These factors would lead to rapid recruitment of susceptibles, high transmission rates within households, and possibly a lower impact of school terms on transmission seasonality. Epidemiologic theory has shown these conditions to favor annual epidemics.[33,34,48]

These findings add to the back-story of contemporary pertussis in the United States, providing clues as to the historical processes that have shaped its modern epidemiology. Finally, our analyses revealed substantial geographical variation in the patterns of pertussis incidence, highlighting the potential pitfalls of drawing conclusions based solely on analyses of national-level incidence data.