In this comparative study, we have documented pertussis periodicity patterns in 64 countries over more than 30 years, covering regions with little prior information on the epidemiology of the disease, such as Africa and Asia. We applied the same methodological approach to all countries and provided the first global picture of pertussis multi-annual epidemic cycles. Our results are in agreement with previous studies from a limited number of countries, mostly in Europe and North America, suggesting a mean periodicity of 3–4.5 years in the vaccine era (Hethcote 1998
; Gomes et al. 1999
; Rohani et al. 1999
; Skowronski et al. 2002
; Broutin et al
Further, we showed that the periodicity of outbreaks increased by 1.27 years (95% CI: 1.13–1.41 years) following vaccination (p
< 0.001), based on eight countries with sufficiently long epidemiological records. This observation is in clear contrast with the classic study of Fine & Clarkson (1982)
, who examined cyclicity of pertussis in national England and Wales case notification data. They interpreted the failure of immunization to increase pertussis inter-epidemic periods in these data to mean that vaccines are more effective in reducing disease rather than transmission. As echoed by Cherry (1996)
, the implications of such a conclusion for pertussis control are substantial. Our finding that immunization programmes consistently increase the inter-epidemic period point to reduced disease transmission, in line with strong herd immunity effects (Rohani et al. 2000
A key finding of this global study is the demonstration that pertussis periodicity is associated with both birth rate and vaccine coverage, as predicted by theory for strongly immunizing diseases (Anderson & May 1991
; Keeling & Rohani 2007
). This relationship is clearly indicative of what may be termed ‘supply-side’ epidemiology: namely, that outbreak frequency is determined by susceptible births. It is clearly in direct contradiction of the assertion that waning of immunity is the main driver of SR dynamics and thus determines transmission. However, an increasing number of cases in adolescents and older people, including those who have been vaccinated in infancy, has been described during the last 10 years in different countries with long vaccination history (Guris et al. 1999
; Strebel et al. 2001
; Skowronski et al. 2002
; Hellenbrand et al. 2009
). One potential explanation for that discrepancy between our results and case studies is that by estimating the mean periodicity over several decades, our results probably reflect a long-term trend without capturing the possible recent changes. Detailed investigation of these pronounced and age-related increases in incidence has become urgent, though is currently hampered by limited age-stratified longitudinal records.
Interpretation of our results requires a number of caveats. A first limitation relates to heterogeneities in epidemiological data, due to differences in diagnostics tools (i.e. clinical, culture, PCR, serology), case definitions and survey efforts between countries. Under-reporting can be expected and may vary geographically and over time. In particular, detection of disease in adolescents and adults is not necessarily systematic since pertussis was considered a childhood disease until recently. Moreover, diagnostic sensitivity is poor in older age-groups (as well as in vaccinated children) because of non-specific and mild clinical symptoms. As a result, numbers of reported cases are probably much lower than the true number of clinical cases. However, under-reporting should not affect estimates of the mean periodicity, since years considered as epidemic by our method probably reflect a true increase in disease incidence compared with surrounding years, even if the magnitude of reported epidemics may be uncertain. Moreover, most time series used in this study are annual and do not allow us to quantify the seasonal patterns of the disease, potentially leading to overestimation of periodicity in countries where annual outbreaks persist. However, our modelling approach suggests that annual outbreaks are not expected in the 64 countries studied, given the range of SR rates. In addition, the majority of our time series comprise more than 20 years of data, producing robust periodicity estimates for multi-annual cycles.
Finally, for the sake of simplicity, we did not take into account heterogeneity in vaccine types, schedules (timing of vaccination and number of doses) and efficacy across countries or within the same country over time, or between administrative units of the same country. For the period of interest here, ending in 2005 or earlier, whole-cell vaccines were most commonly used, with acellular vaccines phased in at the very end of our study period (with Japan being the only exception; Watanabe & Nagai 2005
). Globally, almost all our time series span the era with no vaccination or when only three doses were recommended. In fewer than 10 countries, a fourth dose is recommended (with USA and Canada being the two exceptions that have recommended five doses) for decades but we do not have access to specific vaccine coverage data for this booster. The impact of boosters on pertussis periodicity remains an interesting and open question. Finally, we used an average estimate of vaccine coverage during the vaccination period, and do not capture short-term time trends in vaccine coverage. However, coverage did not change much over the study period, except in the UK (Rohani et al. 2000
), Japan and Sweden. We note that further research should focus on heterogeneity in disease dynamics and vaccine coverage between administrative units of large and economically diverse developing countries such as Brazil or India, which we could not investigate here. Taken together, both data heterogeneity and methodological issues are expected to obscure the true relationship between vaccination and disease periodicity, but would not lead to spurious and statistically significant associations.
In addition to quantifying the global periodicity patterns of pertussis, we used a stochastic model to qualitatively explore the relationship between periodicity and SR rate, for different scenarios regarding loss of immunity and transmissibility of repeat infections. Simulations showed that two assumptions were equally consistent with periodicity patterns detected in the global disease dataset: (i) reasonably long duration of natural immunity and (ii) very slight transmissibility of repeat infections. Epidemiological studies documenting duration of immunity are scarce and provide imprecise estimates, typically 4–20 years for natural infection and 4–12 years for vaccination (Wendelboe et al. 2005
). Our modelling approach suggests that natural immunity could last longer than previously thought, consistent with strong herd immunity effects, if repeat infections do contribute to the overall transmission of pertussis. While our modelling exercise is intended to provide a general comparison, our overall conclusions are in strong agreement with the work of Wearing & Rohani (2009)
, who used the method of moments (specifically the inter-epidemic period and the extinction profile) to explore parsimony between model predictions and the epidemiology of pertussis in England and Wales. Future work needs to attempt to use statistical inference methods to estimate these parameters of key interest, though this represents a formidable technical challenge.
Our study reinforces the need for more quantitative information on the transmission consequences of repeat infections. Along these lines, a prospective study suggested that adults are responsible for 76 to 83 per cent of transmission of B. pertussis
to young infants in the household in Germany, France, US and Canada (Wendelboe et al. 2007
), implying that teenagers and adults may play a role in pertussis transmission. These values are likely to represent an overestimate, however, since these studies are focused on households with infant cases but do not consider ‘control’ group by including households with no infant case.
Long-term vaccination efforts will inevitably increase the mean age of infection, leading to proportionately more cases in adolescents and adults. In fact, older pertussis cases have become more commonly described in Europe and North America, although they are still probably under-reported (Deville et al. 1995
; Schmitt-Grohe et al. 1995
; Wirsing von König et al. 1995
; Cherry 1999
; Guris et al. 1999
; Senzilet et al. 2001
; Gilberg et al. 2002
; Park et al. 2005
; Rothstein & Edwards 2005
). It is plausible, therefore, that our estimates of inter-epidemic periods from the data are based predominantly on typical primary pertussis cases. It is worth highlighting that primary and repeat infections have highly correlated incidence in our model dynamics, suggesting that periodicity of pertussis outbreaks in adults and children is similar, and that our periodicity estimates are robust. Further, mixing patterns may affect disease periodicity; for instance, adults could be substantially infectious, but mixing may occur preferentially among children. Our model did not take into account age-specific mixing patterns, which are likely to vary geographically; instead, we focused on a generic model that could be applied to all settings.
In this context, future research on the global epidemiology of pertussis should focus on (i) estimation of disease burden in adults, controlling for diagnostic issues; (ii) quantification of the role of adults in pertussis transmission, in particular to infants; and (iii) studies of geographical variation in disease dynamics in large countries where vaccine coverage and demographics are heterogeneous. These issues represent crucial questions to address in order to understand changes in pertussis epidemiology in the vaccination era and move towards a global control of pertussis. More broadly, our study underlines the power of large-scale multi-national epidemiological analyses, bolstered by mathematical modelling, in clarifying infection dynamics and the impact of vaccination.