Main

Before eradication, smallpox caused significant epidemics and mortality1. Following the WHO (World Health Organization) campaign, smallpox was eradicated by 1979 and routine vaccination gradually cased worldwide1. Official stocks of virus now remain in only two laboratories, one at CDC in Atlanta, Georgia, and the other at VECTOR, Novosibirsk, Russia1. There is concern, however, that the virus might be held clandestinely and less securely elsewhere3. Release of smallpox back into the population could constitute a significant public health problem on account of the transmissibility of the virus and waning immunity. The greater the transmissibility of the virus the greater its potential to spread, and the more intense are the intervention strategies needed to bring it under control. Significant discrepancies exist in the recent literature4,5,6,7,8, so it is essential to provide a rational assessment of the transmissibility of smallpox by reference to adequate and relevant data.

Reports of outbreaks of smallpox caused by variola major were sought that contained sufficient numerical detail to derive estimates of transmissibility, R0 (see Methods). Such reports are scarce, and those in Table 1 represent all that were found from the extensive published data. Data for four of the outbreaks were from the pre-vaccination era. Analysis of the cumulative smallpox deaths for Boston, USA10 in 1721 and Burford, England11 in 1758 (Fig. 1a, b and Table 1) gave R0 values of 4.3 and 3.4, respectively, which produced extremely good fits to the data. As was often the case for small isolated communities, nearly the whole population of Burford was susceptible. Similarly, outbreaks in Chester10 (1774) and Warrington10 (1773) (Fig. 1c, d, and Table 1) gave estimates for R0 of 5.8 and 4.7 (range for Warrington 4–5.3), respectively. In both of these outbreaks, only children less than 10 years old were afflicted, so the numbers of susceptible individuals were estimated from the relevant birth and age-dependent mortality rates (Table 1). The predicted epidemic curve fitted the data for Chester very well. For Warrington, however, a range of R0 values was derived that depended on the estimated population size, which was calculated to be between 6,000 and 8,000, respectively. The fitted epidemic curve that is shown for Warrington, which is based on an R0 of 4.7 and a population estimate of 7,000, followed the initial course of the observed epidemic well but over-predicted the cumulative number of deaths by the end of the epidemic. It is possible that the case fatality rate of 25% given by Creighton10 was an overestimate as the other contemporaneous epidemics modelled here had case fatality rates of about 15%. Adjusting the case fatality rate to 15% greatly improved the fit to the data, closely fitting all 12 data points and producing an estimated R0 value of 4.8 (see Fig. 1d and Table 1).

Table 1 Population parameters and analysis of smallpox epidemics for estimation of R0
Full size table
Figure 1: Analysis of smallpox epidemics.
figure 1

a–e, Epidemic curves (solid lines) fitted to data (points) from selected smallpox outbreaks. a, Boston monthly deaths; b, Chester monthly deaths; c, Warrington monthly deaths, predicted using published mortality rate of 25% (black line), or 15% (grey line; see text); d, Burford daily deaths; and e, Kosovo daily onset of cases. Disease and population parameters given in the text and Tables 1 and 2, respectively. f, Endemic smallpox in London with epidemics occurring approximately every 4 years12, giving a lower bound for the estimate of R0 of about 5; see text and Methods.

Full size image

Smallpox was endemic in London from the seventeenth to the twentieth century. Weekly mortality statistics were recorded, first in the London Bills of Mortality and later in the Registrar General's reports12. Superimposed on an endemic level of disease, epidemics generated a cyclical pattern of smallpox deaths15 (Fig. 1f). Such patterns are characteristic of the dynamics of endemic diseases and the models that are often used to describe them9. Analysis of these individual epidemics using the methods described above was not possible because independent estimates of the numbers susceptible in the population could not be made. The poorly documented levels of variolation1 and vaccination that were being practised compounded the problem. If the problems due to vaccination are neglected, the inter-epidemic interval15 between 1836 and 1890 suggests that the transmission rate, R (see Methods) was about 5. However, vaccination was being practised, so this provides a lower bound for the estimate of R0. Interepidemic intervals as low as 2–3 years15 and presumably higher values of R0 had also applied previously in London, and these have been variously correlated with greater susceptibility to disease, greater density of population or lack of vaccination. Dietz and Heesterbeek16 also report on an analysis of smallpox in Paris by Bernoulli in 1766. His estimated force of infection, λ, of 0.125, and life expectancy, L, of 32 years, give an estimated R0 of between 4 and 5 for smallpox in this other major European city (given that9 R0 = λ L or R0 = 1 + λ L).

Although no longer endemic in Europe by the early 1950s, smallpox still caused isolated epidemics after the importation of the disease from endemic areas despite the considerable level of herd immunity in European populations1,13,17. One of the best documented epidemics was in Yugoslavia in 1972 (ref. 1). Starting in Kosovo, the outbreak went unrecognized until the second generation of infection, when vaccination and quarantine of contacts was initiated, followed by vaccination of about 95% of the population. We fitted equation (1) to the data, parameterized to take into account these interventions (Table 2), and hence estimated R (see Methods) to be about 5.4; this value fell rapidly to below 1 with the introduction of quarantine and vaccination (Fig. 1e, Table 1). However, about 50% of the population were probably protected by vaccinations that pre-dated the outbreak13,14, such that the initial R0 would have been nearer 10.8. The epidemic curve from equation (1) provided a reasonable overall fit to the data from Kosovo but underestimated the number of cases in the earlier part of the outbreak, probably on account of the small size of the outbreak and the fact that there were only two generations of infection before interventions were applied.

Table 2 Interventions imposed for smallpox outbreak in Kosovo, 1972
Full size table

Thirty European smallpox importations between 1958 and 1973 were also briefly summarized in ref. 1. Stratified by the time delay between importation and recognition of the outbreak, and its notification to the WHO (0–7, 8–14, 15–21 and 22–42 days), the outbreaks that took longer to be notified resulted in a greater average number of cases before the epidemics were halted (5.4, 8.8, 15.8, 67.8 cases, respectively). With the median delay adjusted by the addition of a latent period for the index case and expressed in units of intergeneration time (19 days), a clear relationship was derived between the delay and the total number of cases (Fig. 2a). Each generation's delay increased the average number of cases by e1.71, giving a mean estimate for R of 5.5 (μ ± s.e.m. = 4.9 to 6.3). As before, these estimates are affected by previous vaccination levels; 50% vaccination, as has been suggested13,14, would give an estimate for R0 of about 11. Interestingly, about 50% of cases in most of these European outbreaks, including in Kosovo, were acquired in hospital by staff and patients1 before the outbreak was recognized and hospital infection-control measures had been put in place (Fig. 2b). Without hospital-associated cases, the R0 for smallpox would have been about 5.5 for community-acquired disease.

As expected from data on other infectious agents9, the estimates of R0 derived for smallpox varied with time and place. Over the 1700s to 1900s, they tended to be in the range 3.5–6, but under particular circumstances rose to nearer 10–12 (that is, with crowding and poorer socio-economic conditions in eighteenth century London15 or when exacerbated by hospital-associated infections in more recent European outbreaks1). This analysis is in accordance with inferences from the vaccination level required for smallpox eradication1,9 in industrialized countries. The value of 70–80% that was found to be effective would suggest an R0 between 3.3 and 5 (ref. 9). However, vaccination levels of 80% were not effective in eradicating smallpox in India because of much higher population densities and poorer socio-economic conditions18, which would imply R0 values higher than 5. For contemporary outbreaks in industrialized countries, which now have low levels of herd immunity (18%, see Methods), an R0 value of 4–6 would probably apply for community-acquired infections, but might transiently be higher (for example, 10–12) before the disease was correctly recognized and appropriate hospital infection controls implemented. These values differ from recently applied values. The higher values4,5,6 from 10 to 20 have been estimated from atypical outbreaks with unusually high transmission rates1,19, whereas the lower estimates of around 2 appear to have neglected concurrent herd immunity8. Although our estimate for smallpox represents a relatively modest transmission rate by comparison with some other infectious diseases, such as measles or chickenpox9, significant epidemics could result, particularly if there were delays in detecting the first cases or in setting up effective public health interventions.

Methods

Epidemic modelling

For epidemic modelling, transmissibility is often expressed as the basic reproductive number, R0, the mean number of secondary cases caused by each primary case in a population composed entirely of susceptible individuals9. The reproductive number, R, is an equivalent measure, but without reference to the proportion of susceptible hosts9. Smallpox infection results first in a non-infectious incubation period followed by a short prodrome with non-specific symptoms (assumed to be non-infectious20), which together constitute the latency period. Hosts then become overtly infectious and symptomatic, with a pustular rash and, thereafter, slowly recover and are immune, or die1,17. Whilst infectious, hosts transmit disease to susceptible individuals at a rate dependent on R0. Public health interventions interrupt transmission by removing infectious individuals into quarantine, or by reducing the susceptibility of contacts by vaccination9. These processes were modelled using an appropriately parameterized set of differential equations (1), which were solved using Euler's method with a time-step of 0.1 days.

S and I represent the proportion of the population susceptible and infectious, respectively. U represents the proportion dead and recovered, with the number dead dependent on the specific case fatality rates given in the data sets. The contacts of cases are divided into the following classes: En the number of untraced latent individuals in the population, Ei the number of traced latent contacts, and Ci the number of traced uninfected contacts. The final class of contacts are those untraced and uninfected and so effectively remaining in S. Q represents the proportion in quarantine and V the proportion protected by vaccination. The average rate at which latent individuals become infectious13,21 is α = (latency period)-1 = 0.0685 days-1 and the rate at which infectious individuals in the community recover or die22 is γ = (infectious period)-1 = 0.116 days-1. Two states of quarantine are defined: the first for the traced contacts successfully vaccinated and released into the community at a rate χ1, and the second for the infectious cases, which enter U at a rate χ2. Different vaccine efficacies are assumed for those uninfected, ε1, and infected, ε2. The proportion of contacts found through contact tracing is ρ and the daily rate at which infectious individuals enter quarantine from the community is θ. The proportion of contacts infected is defined as ϕ. The rate at which potentially infected contacts occur is defined as β, as in equation (2), and N is the size of the population in which the epidemic occurs.

Additional assumptions are that no transmission occurs from those quarantined, dead or recovered and the background mortality rate was assumed to be negligible over the time periods examined.

For the Boston, Burford, Warrington and Chester data sets, ρ = θ = 0, which effectively reduces equations (1) above to a simple SEIR model9. Intervention parameters were only required when equations (1) was fitted to the data from Kosovo. Here, interventions were implemented 31 days after the onset of symptoms in the index case1 with the associated parameters shown in Table 2. The number of potentially infected contacts per case was determined as 50 (ref. 1). Values of R0 were derived for each outbreak by minimizing the mean square error between the mortality data and the predictions of mortality from the model, while applying the outbreak-specific case fatality rates to U and adjusting R0 and time of onset of symptoms in the index case. In the case of Kosovo, equations (1) were fitted more simply to the reported number of cases rather than deaths. All the other parameters required for equations (1) were obtained independently from the published source(s) given in Table 1. For epidemics in London, R0 was roughly calculated from the interepidemic interval, T = 2π[L(D + D′)/(R0 - 1)]1/2, where L is life expectancy between 1840 and 1870 adjusted for excess births over deaths, equal to 25 years, and D + D′ is latent + infectious period, equal to 0.063 years9.

Estimation of current vaccination coverage

Given that smallpox vaccination ceased in industrialized countries in the mid to late 1970s (ref. 13), a crude estimate of the immunity of the contemporary UK population was calculated, on the basis of 50% having been vaccinated as infants up to 1972, and estimating that about 60% of these would be alive today from current population statistics. Of these only about 60% would still be protected by the vaccinations done on average 50 years previously, calculated by extrapolating from data on secondary attack rates, which increased from 4 to 12% over 10 years following vaccination23. This suggests that the level of herd immunity may be about 18%, which will continue to decrease with time.

Figure 2: Analysis of 32 European smallpox outbreaks post-1950.
figure 2

a, Regression analysis demonstrating the increase in the mean number of cases per outbreak that occurred (note logarithmic scale of ordinate and abscissa) with increased median delay in notification of the outbreak (delay expressed in units of generation time of the disease (19 days)). Regression: y = 0.95e1.71x, r2 = 0.99. b, Regression analysis showing the relationship between the total number of cases per outbreak and the number of cases that were hospital-acquired. Regression: y = 0.4777x, r2 = 0.94).

Full size image