Epidemiologic Assessment of the Protective Effects of Smallpox Vaccination

Despite smallpox eradication, the widely discussed possibility of a bioterrorist attack using a variola virus makes it necessary to review the epidemiology of smallpox and the effects of various vaccination schemes. This paper provides a literature review concerning the epidemiologic assessment of the protective effects of smallpox vaccination, with particular emphasis on the statistical and theoretical points of view. Although smallpox vaccination has the longest history of all vaccinations, we lack precise epidemiologic estimates of its effectiveness. Vaccination practice continually evolved and many places experienced vaccinations with various strains over time. Despite the weak statistical evidence, it can be stated with certainty that smallpox vaccines prevented infection for a few decades after primary vaccination and that vaccinated individuals had the benefit of a longer lasting partial protection when they contracted the disease. Confronted with the huge uncertainties and with the necessity to rely on laboratory evidence, appropriate preparedness plans for countermeasures using vaccination must be based on the best available evidence.


INTRODUCTION
Smallpox is the only infectious disease that has been eradicated through concerted efforts [1,2] and thereby provides one of the biggest success stories of immunization and public health efforts [3] . Despite its eradication, the widely discussed possibility of a bioterrorist attack using a variola virus makes it necessary to review the epidemiology of smallpox and the effects of various vaccination schemes [4] . There has never been a population-based evaluation of the effectiveness of smallpox vaccination (e.g., a randomized controlled trial). Being the first disease against which a vaccine became available [2] , some of the earliest epidemiologic vaccination studies were performed on smallpox and both epidemiologic and statistical methods matured in parallel as new clinical and epidemiological observations were made [5,6] . Thus, from a contemporary point of view, some of these pioneering epidemiologic assessments of the protective effects of vaccination should be regarded as technically flawed or exaggerated [7] .
Some of the best insights in the effects of widespread vaccination can be gained from the technical and political discussions and the earliest epidemiologic observations on compulsory vaccination in the late 19th and early 20th century [8,9] . To evaluate the impact of vaccination on the epidemiology of smallpox, statistical evidence, especially on the protective effect of primary vaccination, was extensively accumulated during that period [7] . During the Smallpox Eradication Programme intensified since 1967, considerable efforts were undertaken to evaluate alternative vaccination strategies (e.g., post-exposure and ring vaccinations). Now that we do not see any smallpox cases and are not allowed to perform challenge inoculations, population-based estimates of the protective effects can only be obtained through the investigation of such epidemiologic data. This paper provides a literature review concerning the epidemiologic assessment of the protective effects of smallpox vaccination, with particular emphasis on the statistical and theoretical points of view.

Different vaccines and efficacy:
As several different kinds of vaccine were used [2] , statistical records of smallpox vaccination must be interpreted with caution. Variolae vaccine, for example, can either denote the vaccine first developed by Jenner for cowpox [10] or one of the vaccine strains obtained from the 'retrovaccination' technique which was primarily effected by taking the variola virus from human lesions back to a cow, yielding a vaccine of relatively low potency [11] . We know neither the origins of the vaccinia virus nor the time of its introduction [12] . During the Smallpox Eradication Programme, highly potent vaccines were introduced whose immunogenicity was evaluated using pock counts, but prior vaccines usually did not undergo any strict evaluations [13] . These facts indicate the necessity to assume that the vaccine efficacy may have changed over time.
The efficacy of smallpox vaccination, i.e. the extent to which application of a specific vaccine produced a beneficial result in a population, also differs from the direct effectiveness (i.e. the word 'efficacy' of smallpox vaccine was frequently used to indicate potency or individual-based successful vaccination) [14] . Traditionally, the efficacy of smallpox vaccination was measured by the 'take' rate which was determined by an observation of a pustular lesion at the injection site 6 to 10 days after vaccination [15] . Since the localized infection that expressed itself as the vaccine 'take' lesion was also interpreted as the generation of an immune response, the sign was gradually accepted as a method to determine the vaccine's efficacy [16,17] . In contrast, most recent laboratory-based techniques interpret specific immune responses as measure of immunogenicity and protection [17,18] .

Effectiveness of primary vaccination
Early epidemiologic observations: In the period which followed Jenner's discovery of the variolae vaccine in the late 18th century [10] , the evidence of the protective effect of vaccination mostly depended on 'minute inquiry into individual cases' [7] . With a few noteworthy exceptions, statistical reports from the early 19th century give only approximate numbers of cases by vaccination history, which were often based on somewhat unclear definitions and which in general tended to exaggerate the effectiveness of vaccination [19,20] . By the end of the 19th century, the need arose to accumulate comprehensive evidence on the 'protectiveness of vaccination' in order to convince the public of the benefits of compulsory vaccination [7,21] . Comparisons of the situation before and after the introduction of voluntary or compulsory vaccination or of places with and without vaccination painted a rather crude picture, but sufficed to demonstrate the general benefits of vaccination [22] . One of the oldest assessments of the impact of vaccination shows the smallpox death rate in time series from Sweden during 18th and 19th centuries ( Fig. 1) [22][23][24] .
Protection against disease and death: Despite the impressive decline in incidence after the introduction of vaccination, considerable discussion arose on the evidence of such historical and geographical comparisons, attacking the often highly overstated vaccine effects [21] . To accumulate data that more directly reflected the effect of vaccination, it gradually became customary to perform comparisons within the same outbreak [7] : the case fatality of vaccinated cases was compared with that of unvaccinated ones. This shift in the way epidemiologic observations were performed also changed the viewpoint on the protective effect of vaccines [2,7] . Whereas the population-based observations had crudely measured the protection against smallpox (direct effectiveness) that also included some effect on herd immunity, the observations based on defined case-cohorts allowed to examine the vaccine's partial protection [5,7] . These studies demonstrated that the probability of dying from smallpox was strongly reduced for vaccinated cases (Fig. 2). Table 1 compares crude estimates of the direct effectiveness of vaccination (as proposed by Greenwood and Yule [25] ) and also includes some of the earliest findings available.  [7,19,[91][92][93] Population-based evaluations of the protection against disease unfortunately are relatively rare compared to investigations on the partial protection of cases (Table  1). In the study in Norwich, in 1819, the numbers were based on direct observations of 112 families [26] . Whereas the numbers for Marseilles (1828) are lacking precision [20] , similar studies comparing the 'attack rates' (which now would be called incidence) were gradually  accumulated [21] . Based on a similar point of view, a recent study confirmed that vaccination reduced the risk for death among pregnant smallpox cases [6] .
Direct effectiveness during the eradication programme: The Smallpox Eradication Programme accepted many different vaccination schedules (e.g. revaccination, post-exposure vaccination, ring vaccination etc.), which makes it difficult to estimate the effectiveness of vaccination after World War II ( Table 1). As vaccination efforts always coincided with other countermeasures like isolation or quarantine, crude estimates on the direct vaccination effectiveness may be biased [2] . On the other hand, technical improvements in epidemiologic methods included the consideration of transmission within households [27][28][29] and the adjustment for age [30] . Some of the latest estimates of the direct effectiveness of vaccination, obtained from outbreaks in Brazil [30][31][32] , demonstrated an extremely high effectiveness ( Table 1). As the corresponding case fatalities were extremely low, these epidemics were considered as variola minor outbreaks. Therefore, the corresponding vaccine effectiveness may have been overestimated as mild cases among vaccinated individuals may have been under-diagnosed [33] . Another crude estimate of 81.6% for the direct effectiveness (95% confidence interval (CI): 64.4, 92.2) was obtained in a statistical study of a variola major outbreak in Nigeria, 1967 [34,35] .
Indirect effectiveness during the eradication programme: A major development during the Eradication Programme was the consideration of herd immunity [36] . Whereas Dixon had recommended vaccination and revaccination of the entire population at short intervals to achieve eradication [37] , it has been suggested that an indirect effectiveness of vaccination would predict eradication at a lower threshold [38] . It is noteworthy that, even during the 19th century, there were implicit discussions on such indirect effects [23,24,39] .
The impact of population density on the eradication threshold during the Eradication Programme has been evaluated [40] . Observations on population density and vaccination coverage in African and Asian countries showed that smallpox tended to disappear when the density of unvaccinated individuals fell below 10 persons per km 2 whereas it was more likely to persist in densely populated regions [40] . Although nationwide estimates had indicated the requirement of rather higher vaccination coverage (i.e, more than 95%) to eradicate smallpox by vaccination alone, this reflected the limitation of the crude estimate: Smallpox tended to die out in some rural areas even without vaccinations, whereas urban areas were likely to sustain the chain of transmission [40] . Due to the frequent combination of vaccination with other intervention measures against the disease and because the disease was accompanied by obvious clinical symptoms, the net impact of vaccination on herd immunity could not be evaluated precisely [38] . Although the threshold condition was only clarified after eradication [41,42] , examination of an earlier recommendation by the WHO implies that the basic reproduction number, R 0 , i.e., the average number of secondary cases per index case in a fully susceptible population, was assumed to be around 5, considering 80% of vaccination coverage as a goal [38,43] . This implicit estimate and the discussions on its variability were consistent with recent estimates [35,44,45] .

Loss of vaccine-induced immunity:
The epidemiology of smallpox becomes even more difficult to interpret when we have a closer look at the waning protective effects of vaccination. It was recognized soon after the introduction of vaccination that vaccinated cases tended to be much older than unvaccinated ones [46,47] , which led to the idea that vaccine-induced protection waned over time [37] . Figure 3 allows to compare age-specific frequencies of vaccinated and unvaccinated cases in three outbreaks in the 19th and 20th centuries.    [50] .
Although we obviously have to ignore the underlying population structures, we can clearly see that the majority of individuals who received primary vaccination must have been protected for a few decades [37,48] . Table 2 shows how the direct effectiveness of vaccination decreases over age, clarifying age-stratified estimates of the direct effectiveness. In a recent study, using similar records, the expected median duration of protection was estimated to be 11.7-28.4 years after primary vaccination [49] . More insight into the effect of waning immunity can be obtained by comparing the case fatalities of vaccinated and unvaccinated cases over age [50] . Given a similar data, Greenwood described that 'evidently, the advantage of the vaccinated brought out by such a table cannot be due to a fortuitous concurrence of age and vaccination groupings' [51] . Figure 4 shows the case fatality among vaccinated individuals slowly increasing but never reaching the corresponding value of unvaccinated cases. A statistical analysis, based on an outbreak in Liverpool, 1902-3 [52,53] , led to a half-life of partial protection against death of 49.2 years (95% CI: 42.0, 57.3).
Whereas the vaccine-derived protection against infection seems to have been lost after a few decades, a considerable fraction of vaccinees appears to have been protected throughout life against dying from smallpox [52,53] . However, as outbreaks of smallpox were not uncommon when these observations were made, vaccinated individuals may have experienced infections before the outbreak which may have boostered their immunity. Thus, Nishiura and Eichner confirmed the similar long-lasting partial immunity based on the epidemic records in Australia where booster events were much less frequent than in other countries [54] . An additional line of evidence comes from recent laboratory studies on humoral immune responses that also support the concept of a long-lasting residual (partial) protection [55][56][57] . In the event of a bioterrorist attack, this fact could significantly decrease the individual burden of disease [58] .

Effects on contagiousness and duration of disease:
Partial protection may not only have reduced the vaccinees' susceptibility and their probability to develop severe disease, but it could also have reduced the degree and duration of contagiousness which again would have modified the transmission dynamics (Fig.  5) [59] . Rao favored a reduced contagiousness among vaccinated cases, based on observations in intra-household transmission [27] . He demonstrated that on average 67.4% fewer susceptible intra-household contacts were infected by vaccinated cases than by unvaccinated ones (10/499 vs. 40/650 transmissions). If the contacts had been vaccinated before, the reduction was 74.8% (2/421 vs. 11/583 transmissions). Some evidence for an effect of prior vaccination on the duration of infectiousness was obtained in an outbreak in Dalian, China, in 1920-1 [60] : The average symptomatic period was 15.8 days (n = 179) for vaccinated and 18.3 days (n = 40) for unvaccinated cases, respectively, indicating a reduction of 13.7%. Observations in another outbreak in Dalian, 1933-4 showed mean symptomatic periods of 15.5 and 30.1 days for vaccinated (n = 448) and unvaccinated (n = 39) cases, respectively, demonstrating a reduction of 48.5% [61] . Toyoda suggested that the difference of symptomatic periods may have originated from residual (partial) protection among vaccinated cases [60] .

Vaccination schemes Revaccination:
When it became apparent that the vaccination effects waned over time, repeated vaccinations were performed to renew the vaccinees' immunity and unscheduled revaccination became a common countermeasure whenever new cases occurred in a community [16,37] . The first compulsory vaccinations and revaccinations were introduced in Germany during the mid-19th century, achieved by demanding a vaccination certificate from children who entered primary school [9,62] . Compulsory and scheduled revaccinations widely stimulated scientific discussions on the benefit of vaccination [63,64] , but remained very uncommon in the UK until the beginning of 20th century [9,23] . Scheduled revaccination became widely accepted and reinforced only after the accumulation of sufficient observations that revaccinated cases were infected less often and had much milder manifestations [7,21,37] . Since the intervals from primary vaccination to revaccinations and the number of revaccinations varied strongly within and between countries [37] , analytical evaluations are very difficult and of limited precision, although there is much circumstantial evidence of the benefits of revaccination. Crude estimates of the increased partial protection against case fatality could frequently be confirmed as can be seen from data collected in Madras during the 1960s (Table 3) [16] .

Post-exposure vaccination:
Post-exposure vaccinations have been reported since the late 19th century in the UK [50,52] . Although many records give the number of cases and the delay between post-exposure vaccination and onset of symptoms (which starts with fever), all these reports lacked a denominator, i.e. the total number of people vaccinated after exposure [65,66] and none of them compared the group receiving post-exposure vaccinations with an exposed group of unvaccinated individuals. Thus, the protective effect of post-exposure vaccination still remains unclear [66] . Based on the observations of several outbreaks, Lyons and Dixon stated that successful vaccination during the first seven days after exposure would almost always prevent the disease, that vaccination during the following three days would modify the eruption and that vaccination during the last four days would merely add to the patient's troubles [67] . Part of this suggestion was later confirmed by comparing the frequency of cases, dichotomized at 10 days after exposure [68] . Fig. 3: Comparisons of the frequency of smallpox cases by age and vaccination in three outbreaks during 19th and 20th centuries.a) London (n = 10403) during 1870s [63] , b) Leipzig (n = 688) during 1860s [94] and c) Madras, India (n = 6932) during 1960s [16] Fig. 4: Comparisons of case fatality of smallpox by age and vaccination in Sheffield, 1887-8 (n = 310,756). For stratified number of cases and deaths, see Table 2 [50] Results of a recent statistical model indicated that vaccination up to 3.2 days (95% CI, 2.9, 3.6) after exposure may protect against disease [69] and a recent Delphi analysis suggested that the post-exposure vaccine efficacy could be 80-93% during the first 3 days after the exposure and 2-25% thereafter [70] . With regard to evaluation of partial effects among vaccinated individuals, it was suggested that the residual immunity due to prior vaccinations lasted for a long time (sometimes lifelong) and thus, that partial protection due to vaccination may have influenced the probability of death or severe disease rather than the post-exposure vaccination [66] . Figure 6 shows the probability of severe manifestation in relation to the time interval between post-exposure vaccination and onset of fever, considering only cases who had not received vaccination before exposure [52,66] . Results of a logit model suggest that a growing interval between vaccination and onset of disease significantly reduced the risk of severe disease (Nishiura and Eichner, submitted): cases who were vaccinated more than 5 days before onset had less than half the risk of developing serious manifestations [52] .

Ring vaccination:
A surveillance-containment measure during the Eradication Programme, ring vaccination, denotes the vaccination and monitoring of all susceptible individuals in a prescribed area around one or several index cases [71] . Especially since the 1960s, ring vaccination, in addition to other public health measures, was considered to be more effective than mass vaccination alone [72] . Ring vaccination was introduced and evaluated mainly in West and Central Africa and in Asia [73] where it was frequently combined with case isolation and special measures aimed to prevent transmission in hospitals [74] . Although it has been difficult to exclude confounders and to evaluate the net effectiveness of ring vaccination precisely, technical discussions suggested benefits of field-based practice and compared ring vaccination with mass vaccinations [75,76] . Observations during the 1960-70s in India support the potential effectiveness of ring vaccination: After halting a 10 year series of mass vaccination campaigns which had failed to eliminate transmission, public health workers started to visit each village and to vaccinate the villagers wherever they found at least one case. This resulted in eradication within 1.5 years [77] . Although the impact of previous vaccination cannot be separated, the finding supports the potential effectiveness of ring vaccination in localized outbreaks [77] . Mathematical models have claimed critical importance of contact tracing if we rely on ring vaccination alone [78,79] .
Future challenges: Although (or maybe because) smallpox vaccination has the longest history of all vaccinations, we lack precise epidemiologic estimates of its effectiveness. Vaccination practice continually evolved and many places experienced vaccinations with various strains over time [2] . Several new vaccines that may be more efficacious and less toxigenic have lately been proposed [80,81] , but even simple population-based comparisons have to rely on laboratory studies in the absence of smallpox. Although statistical evidence is weak, it can be stated with certainty that smallpox vaccines prevented infection for a few decades after primary vaccination and that vaccinated individuals had the benefit of a longer lasting partial protection when they contracted the disease. After all, smallpox was eradicated by combining effective vaccines with other control measures [3,4] . There has been a considerable debate on the re-introduction of mass vaccination and containment strategies in the event of a bioterrorist attack, but mass vaccination prior to an attack is currently not recommended [82,83] . One of the biggest problems we are currently faced with may be that residual immunity is gradually lost among vaccinated individuals and that unvaccinated cohorts further add to the accumulating pool of susceptible [84] . Mathematical models, stressing important aspects of the transmission dynamics (i.e., spatial spread [85] , contact tracing [86] and vaccination schemes [78,79,[87][88][89] ), have been used to study responses to a simulated smallpox attack, but none of them explicitly incorporated all the protective effects of vaccination. Incorporating partial protection in these  [52,66] ) models may have a major impact on the results, especially when targeted interventions that strongly rely on typical symptoms are considered. Confronted with the huge uncertainties and with the necessity to rely on laboratory evidence, appropriate preparedness plans for countermeasures using vaccination must be based on the best available evidence.