Discussion
In our study, since we obtained both the cases and the controls in a surveillance setting, it implies that both have a population base and are, therefore, representative of the same.
The results allowed us to evaluate the ‘real-world’ effectiveness of a single pre-exposure dose of a third-generation smallpox vaccine in the prevention of mpox. A statistically significant time–response effect was observed between the time elapsed after vaccination and the protection obtained by vaccination, which made it possible to evaluate VE on a weekly basis. VE values above 90% were obtained from the fourth to the eighth weeks, with 95% confidence that VE is at least more than 80%. Similar figures were obtained in the sensitivity analysis using matched case‒control groups and for the male population.
The time–response relationship was consistent with data from the phase III immunogenicity study of MVA-BN, which showed an increase in response (antibody titre) over time.21 22 The overall adjusted pre-exposure VE of 86.4% (95% CI 62.2% to 95.1%) was also estimated considering the first 2 weeks after vaccine administration as the induction period and disregarding the time thereafter. This finding coincides with the results of the cohort study in men by Wolff Sagy et al,23 who obtained an adjusted HR for infection in the vaccinated population (one dose) compared with the non-vaccinated population of 0.14 (95% CI 0.05 to 0.41); therefore, a VE of 86% (95% CI 59% to 95%).
In our study, we considered the time elapsed in weeks from vaccination to onset of suspected mpox symptoms to be the most relevant variable, not only because of its statistical advantages but also because it reveals the evolution of the response to vaccination since its administration. Additionally, this variable prevents a possible classification bias by considering unvaccinated individuals who had not passed the induction period; therefore, it captures the ‘partial’ effect of the period and the possible involvement of the superinfection exclusion mechanism.24–27
The majority of published studies of pre-exposure VE with a single dose obtain a point estimate in the range of 68.1%–93% with different designs including cohort and case–control studies.28–32 However, Deputy et al33 obtained a low VE, of 35.8% (95% CI 22.1% to 47.1%), such a discrepancy would be conditioned by their selection of the controls.
The underestimation of VE by Deputy et al33 would be the result of three decisions when choosing their secondary-based controls. First, they include only incident HIV infections in their controls, but prevalent HIV infections predominate among their cases—the population that lives with HIV has a higher risk of getting sick from mpox.4 34 Second, when choosing HIV infections and PrEP-HIV users as controls, HIV infections or PrEP-HIV users with risk factors other than men who have sex with men (MSM), who are not the target population for vaccination, have been included (MSM have predominated as cases during the mpox outbreak). Finally, the exclusion of participants ‘who had no in-person medical encounters during the 3 years before their index event or who had only a telehealth visit to serve as the index event during the study period’, which removes from the study the healthiest population or those with the lowest use of healthcare (and therefore not vaccinated) has acted differentially between cases and controls: the difference in the exclusion percentage is 29% (95% CI 27.2% to 30.9%) between cases and HIV-infected controls and 6.5% (95% CI 4.9% to 8.1%) between cases and HIV-PrEP-using controls.
The use of TND to estimate the VE is well established, and its appropriateness has been demonstrated in various infectious diseases.35–42 This design, also called an indirect cohort, can also be thought of as a variant of a cohort design,43 and therefore, the OR can be interpreted as a risk OR. The TND has also been used to determine risk factors associated with susceptibility to infection.44–46
The representativeness of the target population in the TND studies is guaranteed since the controls come from the same population as the cases because the same selection process is used for controls as for tested positives, and the selection biases that case‒control studies often incur are prevented.44 Thus, for Franke et al, the TND may minimise bias due to differential health-seeking behaviour and recall.36
The mpox outbreak has predominantly affected MSM; thus, the control measures4 34 aimed at the rapid detection of cases in the risk group entail an association between the performance of the test and the characteristics of the tested persons. Given this potential association, the classic control group in case‒control studies may carry a bias given that tested persons might have different characteristics than the general population.44 45
The potential differential selection bias between cases and controls, which can occur in all case‒control studies, is not believed to have affected the results for three reasons. First, because of the exhaustive recording of all tests performed and the lack of differentiation between cases and controls, there was no prior knowledge of the vaccination history of any of them. Second, the suspected cases eliminated from the analysis for missing information represent 1.8% of the total finally analysed, thus minimising possible selection bias on this basis. Finally, the sensitivity analysis of the results using matched case‒control groups leads to the same results. Other authors36 38 have obtained similar results in VE using the TND and other study designs.
In addition, given the lack of vaccines, vaccination was prioritised for the groups at the highest risk of getting sick given their behavioural practices related to mpox exposure. Therefore, if the controls did not reflect this population at risk, there would not have been a higher proportion of vaccinated controls than vaccinated cases.
Potential disease or exposure misclassification was minimised by restricting the suspected cases to residents in the CM. The availability of electronic records (linkage data) of both diagnostic tests and vaccination status also prevented differential errors between cases and controls. Additionally, the extensive use of diagnostic tests with high sensitivity and specificity11 12 on samples with a high positivity rate ensured that false negatives were also reduced to a minimum.
There is a possibility that in vaccinated cases with suspicious symptoms, doctors tend to discard the diagnostic test because they consider the presence of mpox more unlikely than other differential diagnosis assessments. This would lead to a reduction in the number of vaccine recipients, especially among controls (there are 11% more controls than cases in our study) and would mean that our VE is underestimated. Narrowing down to severe cases in order to reduce this possibility would not be appropriate, as the majority of symptoms were mild: only 2.9% of cases (23/799) and 0.56% of controls (5/891) required hospital admission, mostly due to the occurrence of comorbidities.
The study accounted for possible differences or changes in exposure risk by controlling for proxy variables, although there may be some residual confounding, both in individual variables and factors derived from the evolution of the outbreak.
We do not carry out a sensitivity analysis with gender because it was not included in the sanitary registry.
Despite the above, among the possible limitations of the study is the generalisability of the results. The individuals studied may have had a lower intensity of exposure to risk factors than the individuals who became ill at the beginning of the outbreak. In other words, the individuals most likely to be exposed would have become ill at the beginning of the outbreak and therefore before the start of the vaccination campaign. These possible differences would not influence the VE estimate, provided that the likelihood of being vaccinated among the confirmed cases and the likelihood of being vaccinated among controls does not vary according to risk factors associated with higher exposure since higher exposure, dependent on the evolution of the outbreak, has been temporospatially controlled. If the presence of risk factors corresponded to a higher probability of immunisation among case subjects, our VE would be overestimated. However, if subjects with higher exposure to risk factors were less likely to be vaccinated, our VE would be underestimated.
In summary, appropriate vaccination strategies, measures aimed at reducing risk behaviours,47 the notification of risks and community engagement efforts, early diagnosis, isolation and effective contact traceability continues to be essential to control this outbreak.10
