Rapid Responses to:
|
|
Rapid Responses: Rapid Responses published:
-
![[Read Rapid Response]](/icons/misc/sectionDOWN.gif)
Replicating findings from Guinea-Bissau?
- Peter Aaby, Christine S. Benn, Ida M. Lisse, Henrik Jensen (25 November 2004)
-
Non-specific beneficial effects of vaccinations in Burkina Faso: Survival bias?
- Henrik Jensen, Christine S. Benn, Ida M. Lisse, Peter Aaby (25 November 2004)
-
Re: Replicating findings from Guinea-Bissau?
- Jayne LM Donegan (1 December 2004)
|
|
|||
|
Peter Aaby, Head Bandim Health Project, Guinea-Bissau, Christine S. Benn, Ida M. Lisse, Henrik Jensen
Send response to journal:
|
Editor – In the paper on non-specific effects of vaccination on child survival in Burkina Faso (1), Vaugelade and colleagues conducted the main analyses with a Cox proportional hazards model with vaccination dates as time-varying covariates but also claimed to replicate the survey method we used in rural Guinea-Bissau (2). In the latter analysis, follow-up began from the first visit (before 7 months) and ended at the earliest among: second visit, 6 months after first visit, out-migration, or death. The effects of BCG and diphtheria, tetanus, pertussis (DTP) vaccines were analysed jointly by classifying children into four groups: BCG only, DTP only, DTP and BCG, or neither. In contrast to our study (2), they found 50% lower mortality for BCG plus DTP-vaccinated children. However, there are several important differences from the method we used (2) which may not be immediately apparent. First, the Burkina Faso study registered vaccination status from the vaccination card, but did not document who were unvaccinated. In Guinea-Bissau, 35% of the children were not included in the analysis because it was not possible to get information on their vaccination status (card not seen, travelling, lost card). In Burkina Faso, no child was excluded from the analyses, and when the card was not seen it was assumed that the child had not been vaccinated. Hence, the “unvaccinated” group in Burkina Faso was a mixture of unvaccinated with no card and children for whom information was not available because the card was not inspected. Children whose family was not home to present a vaccination card may be a special group with higher mortality for a variety of reasons (e.g. sick, seeking treatment, travelling, living far away from the health centre). Many children classified as “unvaccinated” may have been high-risk vaccinated children seeking treatment elsewhere. For instance, in rural Malawi (unpublished data), mortality was 3-fold higher for absent children than for children examined anthropometrically at home. If such differences exist in Burkina Faso, the mortality level in the “unvaccinated” comparison group would be inflated, and the estimated beneficial impact of BCG and DTP would be irrelevant. That this may be the case is supported by the fact that the mortality rate ratio (MR) between unvaccinated and vaccinated (any vaccine) children in Guinea-Bissau was 1.35 (0.97-1.89). In Burkina Faso, this MR was 2.29 (1.76-2.98) for the children followed from the first visit (see Table 3 - 281 deaths for 28,128 person-months for unvaccinated children and 64 deaths for 14,662 person-months for vaccinated children). Second, whereas we compared BCG only with BCG+DTP1, the Burkina Faso analysis included children who have received more than one dose of DTP in the BCG plus DTP group. Those with two doses of DTP tended have slightly lower mortality than those with one dose of DTP (Table 5) and may therefore have reduced mortality inappropriately in the BCG plus DTP group. Third, in the analysis from Guinea-Bissau, we controlled for village since vaccination coverage, access to health care, and exposure to potential fatal epidemic diseases is likely to vary by village (2). Had we not controlled for village, DTP1 would have been associated with a MR of 1.52 (0.94-2.45) rather than 1.84 (1.10-3.10), suggesting that control for village was important. In the present analysis from Burkina Faso, control was only made for whether there was a dispensary in the village. Fourth, the age structure of vaccination was quite different in Burkina Faso and Guinea-Bissau. Burkina Faso reported the mean ages of BCG and DTP to be 4.8 and 6.3 months, respectively, whereas the mean ages were 2.3 and 4.3 months, respectively, in Guinea-Bissau. Our original study design focusing on the age group 0-6 months was partly constructed to fit the interval of 6 months between visits and partly to minimise the influence of subsequent measles vaccinations. We subsequently showed that if follow-up was censored at 9 months to reduce the effect of MV, the negative effect of DTP was strengthened (3). However, if vaccinated children in Burkina Faso were considerably older, it would mean that a higher proportion of the vaccinated children, particularly the BCG plus DTP- vaccinated children, would have received measles vaccine during follow-up and benefited from this vaccine (3). This difference in age structure could easily have distorted the comparisons both of vaccinated versus unvaccinated and of BCG plus DTP versus BCG only. If DTP as used in the global immunisation programme has a beneficial effect it should have reduced the mortality of children who received DTP after BCG compared with children who received only BCG. As presented in the Burkina Faso analysis, there is no indication that mortality was lower among children who had received both BCG and DTP compared with children who had received only BCG. This is surprising since the additional DTP vaccination should have a positive effect on survival and, furthermore, one would expect that children receiving DTP vaccination were healthier than other children (2) and presumably had lower mortality. To document that there is no problem with DTP outside Guinea-Bissau, Vaugelade and colleagues mention that DTP was not associated with mortality in Senegal with reference to one of our previous studies (4). This is a surprising use of our data. In this study from a high-titre measles vaccination trial, children who had been called for vaccination at 5 months of age and received DTP-IPV tended to have higher mortality than children who had not turned up for vaccination between 5 and 10 months of age (MR=1.60 (0.76-3.37)). If anything, this estimate suggests a negative effect of DTP-IPV which is compatible with our previous estimate from Guinea-Bissau of 1.84 (1.10-3.10) (2) but clearly incompatible with the effect of DTP reported by Vaugelade (1). In conclusion, the analysis claimed to replicate our study from Guinea-Bissau (2) and to document BCG plus DTP to be associated with a 50% reduction in mortality; however, we believe that it has not shown such a reduction. The beneficial estimate for BCG plus DTP is obtained by comparing with a non-comparable control group. We suggest that control should be done for village, subsequent DTP vaccinations, and measles vaccination. Peter Aaby, Christine Stabell Benn, Ida Maria Lisse, Henrik Jensen Bandim Health Project, Danish Epidemiology Science Centre/Statens Serum Institute, Apartado 861, Bissau, Guinea-Bissau References 1) Vaugelade J, Pinchinat S, Guielle G, Elguero E, Simondon F. Lower mortality in vaccinated children: follow up study in Burkina Faso. BMJ, doi.:10.1136/bmj.38261.496366.82 (published 18 November 2004) 2) Kristensen I, Aaby P, Jensen H. Routine vaccinations and child survival: follow up study in Guinea-Bissau, West Africa. BMJ 2000;321:1435 -8 3) Aaby P, Jensen H. Routine vaccination and child survival in Guinea -Bissau. Authors’ reply to commentary. BMJ 2001;322:360 Competing interests: None declared |
|||
|
|
|||
|
Henrik Jensen, Statistician Bandim Health Project, Guinea-Bissau, Christine S. Benn, Ida M. Lisse, Peter Aaby
Send response to journal:
|
Editor – There has been growing recognition of the non-specific effects of vaccinations in areas with high-mortality – that is, the possibility that vaccinations may affect mortality in ways that cannot be explained by the prevention of specific diseases. These effects may be both positive and negative (1-3). Controversy has surrounded our observation that diphtheria-tetanus-pertussis (DTP) might be associated with increased infant mortality (2). Since such effects can only be examined in observational studies, it is important that bias and confounding are not leading to wrong conclusions. We cannot be sure to remove all possible confounding and in general bias is by nature impossible to control for. However, we can assure that the statistical method of analysis is not introducing further bias. In their paper on vaccination status and subsequent mortality in rural Burkina Faso (3), Vaugelade and colleagues indicate that both BCG and DTP vaccinations were associated with reductions in mortality greater than can be expected based on prevention of specific diseases. As indicated by the authors, vaccination dates were used as time-varying covariates in the statistical analysis. The median interval between the data collection visits was 8 months. Hence, a major part of the vaccinations would only be known long after they had actually been administered. This would be no problem if perfect information was available on all vaccinations provided in the area. As admitted by the authors, there was a possibility of misclassification of vaccination status, particularly for dead children whose vaccination card had been destroyed (3). This would result in dead children vaccinated with a particular vaccine after the last visit being incorrectly classified as unvaccinated for that vaccine. The effect of such misclassification is survival bias: children who die between two visits will remain in the unvaccinated group because their vaccination card was not examined, whereas children who survived to the next visit will have observation time from the date of vaccination and until the date of the visit allocated to the vaccinated group. This latter survival time will be risk-free survival time because had the child died, the death as well as follow-up time would have remained in the unvaccinated group. Hence, misclassified vaccinated deaths remain in the unvaccinated group and the mortality rate in this group becomes much too high. At the same time, mortality is much too low in the vaccinated group because the time at risk is inflated with risk- free survival time. There are clear indications of the magnitude of these problems in the study from Burkina Faso. In the cohort of 9,085 children, 45% had obtained a vaccination before 6 months of age but only 12 vaccinated children had died before 6 months of age compared with 435 deaths among unvaccinated children (see Table 1). This corresponds to 10-15-fold higher mortality among unvaccinated than vaccinated children in the 0-5 month age group. For comparison, the mortality rate ratio between unvaccinated and vaccinated children in Guinea-Bissau was 1.35 (0.97-1.89) for 0-12 months of age (2) and 1.94 (1.20-3.14) for children aged 0-5 months of age. To document the differential mortality for unvaccinated and vaccinated children from six to 24 months age, the authors use a Kaplan- Meier survival curve (see Figure in (3)) starting at six months of age, which means that children are included only if they survived until six months of age. Subsequently 20% of the unvaccinated and almost 15% of the BCG-vaccinated children died before two years of age. These are high rates taking into consideration the children survived until six month of age. However, the classification as vaccinated or not at six months of age depends on data collection visits after six months of age and is therefore also influenced by survival bias. For example, a child seen at 3 months of age as unvaccinated and subsequently visited at 11 months of age but having died at 9 months would be classified as unvaccinated even if it had been vaccinated between 3 and 9 months of age. Vaugelade and colleagues recognised the possibility of misclassification but must have underestimated the potential impact of survival bias, since they concluded that BCG and DTP have beneficial effects. They mention that in order to control for the bias introduced by misclassification of dead vaccinated children as dead unvaccinated children they repeated all analyses by vaccination status recorded at the first visit. This change in method of analysis changed the mortality of vaccinated children from being 3-fold lower to 2-fold lower than the mortality of unvaccinated children (Table 3). This indicates that the main mortality ratio estimates using vaccination status as time-varying variable have been inflated by survival bias. However, as argued elsewhere, the 2-fold estimate is also potentially misleading because the comparison group is incorrect (5). Statistical methods of analysis that introduce survival bias cannot be used to document non-specific effects of vaccination. Only future studies can clarify whether BCG and DTP have non -specific effects on infant mortality in low-income countries with high mortality. Henrik Jensen, Christine Stabell Benn, Ida Maria Lisse, Peter Aaby Bandim Health Project, Danish Epidemiology Science Centre/Statens Serum Institute, Apartado 861, Bissau, Guinea-Bissau References 1) Aaby P, Samb B, Simondon F, Coll Seck AM, Knudsen K, Whittle H. Non-specific beneficial effect of measles immunisation: analysis of mortality studies from developing countries.Br Med J 1995;311:481-485 2) Kristensen I, Aaby P, Jensen H. Routine vaccinations and child survival: follow up study in Guinea-Bissau, West Africa. BMJ 2000;321:1435 -8 3) Vaugelade J, Pinchinat S, Guielle G, Elguero E, Simondon F. Non- specific effects of vaccination on child survival: prospective cohort study in Burkina Faso. BMJ, doi.:10.1136/bmj.38261.496366.82 (published 18 November 2004) 4) Aaby P, Jensen H. Routine vaccination and child survival in Guinea -Bissau. Authors’ reply to commentary. BMJ 2001;322:360 5) Aaby P, Benn CS, Lisse IM, Jensen H. Replicating findings from Guinea-Bissau? BMJ Rapid Responses (submitted) Competing interests: None declared |
|||
|
|
|||
|
Jayne LM Donegan, GP & Homoeopath London NW4
Send response to journal:
|
Peter Aaby, Henrik Jensen et al admirably document how unintended confounders in study method and data collection can lead to inappropriate conclusions – such as those made by Vaugelade et al who conclude: “vaccination with diphtheria, tetanus and pertussis as well as BCG is associated with better survival of children up to two years of age.” Peter Aaby is also head of the team which observed the increased female mortality rate associated with administration of the high titre Edmonston-Zagreb measles vaccine (1) and then tragically confirmed this effect in a subsequent trial with the same vaccine in Senegal (2). The Edmonston-Zagreb vaccine was eventually withdrawn. It had also been administered to the children of 900 mostly black and Hispanic families in southern California whose parents had not been notified that it was unlicensed or experimental when it was given to their children between 1989 and 1991. The parents received an apology in 1996. (3) However Aaby et al are not immune to some questionable conclusions of their own. In their study to investigate the effect of measles on subsequent cell mediated immunity, they concluded: “there is a significant reduction in cell mediated immunity three years after infection in children who had had measles compared with vaccinated controls.” (4) In this study, forty per cent of the children who contracted measles had been vaccinated against measles twenty one days or more before the onset of the rash, the percentage would have been higher if those who were vaccinated less that twenty one days before the onset of rash were also included. Eleven per cent of the vaccinated ‘controls’ were excluded from the final analysis because they had contracted measles in the time between selection for the study and follow up. This study is one of those cited as showing that measles the disease, but not the vaccine, significantly decreases levels of cell mediated immunity in its aftermath but no such conclusion can be deduced from the presented data as such a large proportion of measles ‘cases’ had been vaccinated. Cell mediated immunity is an important factor in the host response to viral infection. We continue to give measles vaccine, whose effect on cell mediated immunity is not fully known, at the same time as two other live viruses in the MMR (mumps, measles, rubella) vaccine. I think that this can only be described as ‘risky’. References (1) Aaby P et al ‘Long-term survival after Edmonston-Zagreb measles vaccination in Guinea-Bissau: Increased female mortality rate’ The Journal of Pediatrics 1993;122:904-8 (2) Aaby P et al, ‘Sex-specific differences in mortality after high- titre measles immunisation in rural Senegal’ Bulletin of the World Health Organisation 1994;72:761-770 (3) ‘Parents get apology for vaccine tests’ BMJ;312:1628 (4) Shaheen SO, Aaby P et al, ‘Cell mediated immunity after measles in Guinea-Bissau: historical cohort study’ BMJ 1996;313:969-74 Competing interests: None declared |
|||