This issue of the TMIH journal carries two papers by Aaby and his colleagues on the ‘non-specific effects’ of vaccines. They discuss a methodological issue that might explain inconsistent findings in papers published in recent years with respect to non-specific effects of DPT vaccination on child mortality. In particular they point out that some commonly applied methods of analysis may introduce substantial bias in the estimation of such effects.
It has long been known that vaccines can have effects other than reducing the risk of the disease against which they are targeted. Perhaps most obvious are the adverse reactions that, albeit rarely, can occur as a consequence of vaccination – ranging from anaphylaxis attributable to particular components (e.g. egg albumen) to pericarditis from smallpox vaccines. The pathogenesis of many of these adverse effects is reasonably well understood, and warnings regarding the risk of such effects are described in package inserts of all vaccines. Less widely appreciated is the fact that vaccines may have unanticipated beneficial effects, other than on the target disease. For example, BCG vaccines, originally given for their protective effect against tuberculosis, also protect against a variety of mycobacterial infections (including Mycobacterium leprae and Mycobacterium intracellulare). Beyond these examples is the possibility that some vaccines may have less precisely defined effects, perhaps mediated through the immune system, that manifest as a general increase, or decrease, in morbidity or mortality attributable to a variety of causes or diseases. It is with reference to such effects of vaccines that the term ‘non-specific’ is increasingly used.
A striking example of an apparent non-specific effect was described 15 years ago in the context of studies evaluating the efficacy of high titre measles vaccines. These vaccines, carrying more than 105 measles virus particles, were evaluated during the 1980s in trials in Senegal and elsewhere for use in early infancy in an effort to overcome maternal antibodies and hence to provide protection to young infants in high-risk populations. Analyses in the early 1990s, after efficacy against measles had been demonstrated, revealed an unexpected excess in overall child mortality in the high titre recipients in the Senegal trial occurring many months after the vaccines had been administered. The excess mortality was particularly apparent among females (Aaby et al. 1994). Subsequent analyses of all available data indicated similar trends in Haiti and Guinea Bissau, and led to the withdrawal of high titre measles vaccines from use (Knudsen et al. 1996). The mechanism underlying the apparent mortality excess has never been fully explained, and the effect has remained controversial.
Subsequent efforts to study the possible non-specific effects of routinely administered vaccines on infant and child mortality have been hampered by considerable methodological difficulties. Because the vaccines are in routine use, randomized-controlled trials have been considered unethical. Such historic trial data as are available have generally not been informative as most evaluations have concentrated upon specific target conditions, and long-term follow-up has been limited. Most of the evidence to date for non-specific effects on mortality has come from observational studies in populations with high infant and child mortality rates, in which the health infrastructure has been poor and where morbidity and mortality data are of generally poor quality. In such populations the take-up of vaccines is often highly selective and this may confound efforts to identify vaccine-attributable outcomes (e.g. BCG at birth given only to infants born in hospital, differential reporting to vaccination centres by mothers according to their socio-economic status).
Despite these difficulties, Aaby and his colleagues have persisted in their investigation of non-specific effects of common vaccines on mortality. They have argued that BCG and measles vaccines can enhance survival in some populations by amounts greater than can be explained by the prevention of tuberculosis or measles alone (Aaby et al. 1995; Roth et al. 2006). Even more controversially, they suggested, initially on the basis of data from Guinea Bissau, that DTP vaccination could, under some circumstances (e.g. absence of pertussis) be associated with increases in overall mortality, at least until children received measles vaccine (Kristensen et al. 2000). This observation was based upon an unorthodox analysis of a complicated data set. It led to considerable discussion, and WHO sponsored the analysis of a variety of data sets in other populations to test the hypothesis. None of these studies have replicated the observation of increased mortality associated with DTP vaccination (Nyarko et al. 2001; Breiman et al. 2004; Vaugelade et al. 2004; WHO Task Force on Routine Infant Vaccination and Child Survival 2004; Elguero et al. 2005; Lehmann et al. 2005; Moulton et al. 2005).
Aaby and his colleagues now point out that the studies which failed to show any mortality increase associated with DTP vaccination used methods of analysis that can introduce a bias against finding such an effect. In these studies, data on childhood vaccinations were typically collected in periodic surveys, and the information on vaccinations, which occurred between successive home visits, was updated at the time of the second visit. The person-time at risk in unvaccinated and vaccinated states was then divided up according to the date of vaccination during the time interval between visits. This method opens up a potential bias, insofar as the updating of person time at risk from unvaccinated to vaccinated is only possible for children who survive to the second follow-up. Those who die between visits typically do not have vaccinations between the first visit and death recorded, and thus they will tend to be allocated as deaths in unvaccinated children – thus incorrectly inflating the mortality rate among unvaccinated children.
This bias has been described before, but in different contexts, as the distinction between ‘landmark’ and ‘retrospective updating’ analysis of cohort data. The two papers in this issue explore this problem, both by simulation and by reanalysis of available data, and show that the retrospective updating method can lead to a considerable bias in vaccine studies, biasing observed mortality rate ratios towards zero, whereas the landmark method leads to a non-specific misclassification and biases the rate ratio towards unity.
An additional problem with the literature on non-specific effects of vaccines has been the variety and unexpected nature of the hypotheses which have appeared (in particular relating to sex-specific effects), which has meant that it has not always been clear whether some apparent ‘effects’ were the result of post hoc analyses or whether they were reflections of a priori hypotheses. This was discussed at length at a review of the work of Aaby and his colleagues in Copenhagen in 2005 in which we participateda. The review was convened by the Danish National Research Foundation and the Novo Nordisk Foundation who have sponsored much of the work of Aaby and his colleagues. An outcome of the review was the explicit formulation of a series of testable hypotheses, agreed by the Aaby group, and set out in an Annex to this commentary. It is hoped independent investigators will design and conduct studies powered to confirm or refute these hypotheses.
This discussion comes at an important time, when WHO's Strategic Advisory Group of Experts (SAGE) has recommended a review of the common EPI strategy of administering most infant vaccines according to the 6, 10 and 14 week and 9-month schedule that was set 30 years ago, when both the epidemiological situation and the available vaccines were very different from what they are today (WHO 2006). The dramatic global decreases in frequency of most of the diseases targeted by EPI vaccines (e.g. measles, polio, pertussis, tetanus and diphtheria) are measures of the success of that programme, but mean that any potential other (e.g. non-specific) effects of the vaccines will assume greater relative importance than in the past. As an increasing number of new vaccines is introduced into all populations, the possibility of changing the basic vaccine schedule provides a window of opportunity for making controlled changes that could be designed not only to evaluate the immunological benefits of different schedules, but also to evaluate possible longer-term beneficial or adverse non-specific effects. The methodological points raised in this issue of TMIH will be important in this enterprise.