Studies in animals

Prospective controlled trials in animals have provided unequivocal evidence that previous infections and immunisations influence an individual's response to subsequent infections.1,–,6 Memory T cells that are specific for one pathogen can become activated during infection with an unrelated heterologous pathogen and alter the immune response.4 For example, in mouse models of respiratory viral infection, previous infection with lymphocytic choriomeningitis virus enhances Th1 cytokine clearance of pichinde virus and murine cytomegalovirus, and reduces mortality from vaccinia virus.41 On the other hand, previous infection with flu virus is associated with enhanced clearance of vaccinia virus, but with reduced clearance of lymphocytic choriomeninigits virus and murine cytomegalovirus with a Th2 cytotoxic response and increased lung injury.41 Useful reviews have been published that summarise the effects of sequential infections with viruses3 6 and the interactions between bacterial infections, parasitic infections and tumours.1

Measles vaccine

Dr Peter Aaby (pronounced Or-be) is a remarkable Danish anthropologist who has been working for more than 30 years in Guinea-Bissau, an impoverished ex-Portuguese colony in West Africa where approximately 15% of the children die before they are 5 years old. In early 1979, there was a severe measles epidemic with a case death rate of 25% among children less than 3 years old.42 Horrified by the deaths, Dr Aaby arranged for measles immunisation to be made available. He was surprised to see a substantial reduction in mortality among immunised children, even though there were few measles infections in the period following the epidemic.42 Subsequently, 10 cohort studies and two case–control studies from Bangladesh, Benin, Burundi, Haiti, Senegal and Zaire also suggested that measles vaccine may reduce mortality from diseases other than measles.25 The protective efficacy against death after measles immunisation ranged from 30% to 86%. Even though these studies adjusted for many of the important confounding variables, they were not randomised trials, and there may have been selection bias.

Table 1 presents an intention-to-treat analysis of randomised studies of the effect of measles vaccine given to African girls at 9–10 months of age.26,–,29 35 All the girls were immunised against measles at either 5 or 9–10 months of age, so any differences were not due to infection by measles virus. The girls who received measles vaccine at 9–10 months of age had a 47% (95% CI 23% to 63%) lower mortality than the girls who received inactivated polio, DTP or Neisseria meningitidis polysaccharide vaccine at 9–10 months. In the trial in Senegal,28 it is highly likely that the 54% (95% CI 22% to 74%) reduction in mortality was caused by the non-specific effects of measles vaccine because, apart from measles vaccine, all the girls received the same vaccines at 9–10 months and all were immunised against measles at either 5 or 9–10 months. Because this is an intention-to-treat analysis, the true difference is probably even greater than the observed difference: some girls in the measles vaccine group did not receive measles vaccine, and some girls in the control group did receive measles vaccine. Therefore, a conservative estimate is that measles vaccine given to African girls at 9–10 months of age reduced mortality from diseases other than measles by 47% (95% CI 23% to 63%). Measles vaccine has a stronger beneficial effect when it is given at 4–5 months of age.25 43

Bacillus Calmette–Guerin vaccine

There have been many studies of the non-specific effects of BCG in animals,11 21 and some of the mechanisms are understood.44 In the 1940s and 1950s, several controlled trials were performed in children in the USA and the UK to evaluate the effect of BCG vaccination on mortality. Table 2 summarises the results of these studies35 45,–,49: the Mantel–Haenszel combined estimate is that BCG reduced mortality from diseases other than tuberculosis by 25% (95% CI 6% to 41%).

There have been at least 14 case–control, cohort, hospital or community studies of the effect of BCG vaccination on all-cause mortality in children less than 2 years old in low-income countries.17 Tuberculosis is a rare cause of death in this age group. Six of the studies used the landmark method, which avoids survival bias by keeping vaccination status fixed between surveillance visits (see below)9,–,11 39; in these studies, the mortality ratio ranged from 0.11 to 0.68. BCG had a mortality ratio of less than 1.00 in seven of the eight studies that used retrospective updating, but this method is likely to overestimate the benefits of immunisation. In all 14 studies, the effect of BCG was adjusted for socioeconomic or health status; however, these were not controlled trials and there may have been selection bias.

Diphtheria–tetanus–pertussis vaccine

There are no randomised trials of the effect of DTP on mortality from all causes in children and very few cohort studies of what happened when DTP was first introduced. One such study is available from Guinea-Bissau: mortality was 11.3 per 100 person-years in children given DTP and 5.1 per 100 person-years in children who did not receive DTP because they were ill or travelling, or because DTP was not available (mortality risk ratio 2.03, with 95% CI 1.17% to 3.52%).31 There is nothing to suggest that the unvaccinated children were a low-risk group in this study: they tended to have a lower weight for age, and unvaccinated children are usually a high-risk group.38

In December 2000, the BMJ published a study of 15 351 children in Guinea-Bissau which suggested that BCG and measles vaccines lower mortality, but that one dose of DTP may increase mortality.50 Because of these controversial findings about DTP, several cohort studies were analysed using methods suggested by WHO: eight of the nine studies suggested that DTP reduces mortality.11 Unfortunately, these studies were affected by two problems: survival bias and the sequence of vaccinations. Survival bias can occur when data are collected in periodic surveys and the information about vaccinations is recorded at the time of the next survey; this is known as retrospective updating. Bias can occur because children who die between visits rarely have vaccinations recorded when they are given between the first visit and death, and these children are often incorrectly classified as having died unvaccinated.9,–,11 39 The sequence in which vaccines are given is also important. WHO recommends that BCG be given at birth and DTP at 6, 10 and 14 weeks. In eight of the nine studies that used retrospective updating, it is possible that a substantial proportion of the children received BCG and DTP at the same time; this factor in combination with survival bias may have obscured a harmful effect of DTP.11

Landmark analysis avoids survival bias by keeping vaccination status fixed between visits; it is less precise than retrospective updating, because it understates any benefit or harm from the vaccine, but it avoids the very large errors that can be caused by survival bias.9,–,11 39 There have been nine case–control or landmark cohort studies of the first dose of DTP (table 3)30,–,34 50,–,53; seven suggested that DTP may increase child mortality, especially in girls, and six of the seven are known to have followed the WHO immunisation schedule (with BCG given before DTP). Only two studies were large enough to be significant in their own right34 50; three studies were significant for any dose of DTP.31 34 50 Two of the landmark studies suggested that DTP may reduce child mortality.51 52 However, both studies assumed that children were unvaccinated if no information was available, which can cause serious bias,38 and BCG and DTP were given together to two-thirds of the children in one of the studies,52 and may have been given together in the other.51 All the retrospective and landmark observational studies suggest that BCG or DTP or both have very large non-specific effects on mortality;11 51 52 the contentious issue is whether DTP has harmful or beneficial non-specific effects. The discrepancy is likely to be due to differences in methodology rather than to differences in the biological effects of DTP.11

In 1992, WHO rescinded its recommendation that children in high-mortality areas be immunised with high-titre measles vaccine at 6 months of age, because the practice was associated with increased mortality in girls.27 Subsequent analysis has suggested that girls had an increased mortality only if they received a dose of DTP after they were immunised with high-titre measles vaccine at 6 months of age.27 Mortality increased when the adverse effects of DTP were not reversed by a subsequent dose of measles vaccine — the problem arose because measles vaccine was given earlier, not because a different (high-titre) vaccine was used.26,–,29 It is not enough to know that a vaccine protects against its target disease; we must know its effect on all-cause mortality in the context of the other vaccines in the schedule.

Dr Aaby and his colleagues in Guinea-Bissau and Senegal have conducted an elegant study of mortality in 626 sets of twins in which one twin was female and the other male.35 There was no difference in the mortality of girls and boys in the pre-vaccination era. The female–male mortality ratio was 0.25 (95% CI 0.05% to 0.93%) for pairs having received BCG as the last vaccine (table 2), 7.33 (95% CI 2.20% to 38.3%) for pairs having received DTP as the last vaccine (table 1) and 0.40 (95% CI 0.04% to 2.44%) for pairs having received measles vaccine as the last vaccine (table 1). The variation in the ratio was highly significant (exact test of homogeneity, p<0.001). This design allows an unbiased estimate of the mortality in girls relative to boys: girls had a lower mortality than boys when BCG or measles vaccine was the last vaccine received, but a higher mortality when DTP was the last vaccine received. It has been suggested that this may have occurred because DTP lowered mortality among boys rather than increased mortality among girls,52 but the mortality for girls was 1.1% after BCG, 13.4% after DTP and 1.3% after measles vaccination.35

In randomised trials of vitamin A conducted in Ghana and Guinea-Bissau, girls (but not boys) given vitamin A had a twofold increase in mortality if they subsequently received DTP.54 55

Several lines of evidence suggest that DTP may increase mortality from diseases other than diphtheria, tetanus and pertussis in high-mortality areas: the effects of introducing DTP in Guinea-Bissau,31 the case–control and landmark cohort studies of DTP (table 3), the studies of the interaction between DTP and early measles immunisation,26 27 29 the female–male twin pair study in Guinea-Bissau35 and the interaction between vitamin A and DTP.54 55 It is commonly held that it is unethical to withhold or delay the administration of DTP in randomised trials.38 56 However, the WHO Global Advisory Committee on Vaccine Safety recently endorsed the view that conclusive evidence about whether DTP increases mortality is likely to require randomised trials.57 Trials of DTP are urgently needed in low-income countries where a high proportion of children die from infectious diseases.40