Routine vaccinations and child survival: Authors' response
Rapid response to:
Papers
Routine vaccinations and child survival: follow up study in Guinea-Bissau, West AfricaCommentary: an unexpected finding that needs confirmation or rejection
Routine vaccinations and child survival: Authors' response
Both Tedd Koren and John P. Heptonstall did not understand our
statements that "Mortality was lower in the group vaccinated with any
vaccine compared with those not vaccinated…However, recipients of one dose
of diphtheria, tetanus, and pertussis or polio vaccines had higher
mortality than children who had received none of these vaccines". Note
that we wrote "none of these vaccines" rather than "no vaccine". Because
virtually all children having received diphtheria, tetanus, pertussis and
polio vaccines had already received BCG (see Table 4), the statement
implies that among children who had received BCG, those who had been
vaccinated with diphtheria, tetanus, pertussis and polio vaccines had a
higher mortality than those who had received only BCG. The trend was the
same among the few children who had received diphtheria, tetanus, and
pertussis and polio vaccines but not BCG. However, the main effect was
among children who had already received BCG as also pointed out in Paul
Fine's commentary.
Both Tedd Koren and the paper emphasised the possible selection
biases which may have influenced the mortality ratios between vaccinated
and unvaccinated children: these include better contact with the health
care system, and restriction of BCG vaccination to children with normal
birth weight. However, it is difficult to understand how selection bias
would work differently for different vaccines. If one believes that
mothers complying with the vaccination programme have children with lower
mortality, the beneficial effect of BCG and measles may have been
overestimated whereas the harmful effect of diphtheria, tetanus, pertussis
and polio vaccines would have been underestimated. Conversely, if one
believes that sick children are more likely to be vaccinated when their
mothers bring them to a health centre, the beneficial effect of BCG and
measles would have been underestimated and the harmful effect of
diphtheria, tetanus, and pertussis and polio vaccines overestimated. While
we discussed the likelihood of these interpretations in the paper, the
main point was that the estimated effects of different vaccines went in
different directions, and that the non-specific effects of the vaccines
are therefore likely to be important. Only further research can clarify
the relative importance of the different vaccines.
Children were not given vitamin A at the time they were immunised
against measles. Vitamin A has not been used for general supplementation
in Guinea-Bissau until the recent war in 1998-1999 (1).
John Heptonstall, Liratsopulos Georgios et al, and Lorna Barton et
al. raised questions about the causes of deaths. Virtually all the
children died from infectious disease, which presented with fever,
malaria, convulsions, diarrhoea, oedema, inflammation, malnutrition,
vomiting, cough, respiratory problems, tetanus, measles, or contaminated
maternal milk (that is the mother had broken the taboo against having
sexual relations while breastfeeding). Only one death may have been due to
an accident (burns). However, it should be recognised that the vast
majority were recorded only as having fever, so the precise cause of death
is unclear in most cases. As pointed out by Dr. Georgios et al, we have
done verbal autopsies for women of fertile age (2). It is, however, quite
a different matter to do verbal autopsies for child deaths in an animistic
culture, where many child deaths will be ascribed to malevolent spirits.
It is impolite to recall the death of a child in the presence of the
mother. It might have been possible to obtain better retrospective
information on the sequence of events leading to a child's death using a
medically qualified interviewer. However, such a study would be offensive
to the mothers, and would not provide definitive information about the
causes of death.
The study was large and complex and many potential details have not
been adequately covered even in the long version of the paper. Several
readers raised questions about the follow up, including John Heptonstall
and Liratsopulos Georgios et al. It is noted that 686 children had died
before the first visit, and that these deaths constitute such a large
proportion of the deaths in the study that the conclusions of the paper
are questionable. We have probably assumed too much familiarity with
mortality patterns in developing countries. We had indicated that there
were 437 stillbirths in the cohort, and one would normally expect a
similar number of neonatal deaths in a developing country. In the present
study there were 525 neonatal deaths few of which could have been affected
by immunisations, since most neonatal deaths occur on the day of birth.
The remaining 161 infant deaths occurred before we visited the village.
Using a Cox proportional hazard model and adjusting for cluster as in the
paper, the mortality rate ratio between 1 and 6 months of age was 0.88
(0.69-1.12) for children not yet visited for the first time compared with
mortality following the first visit. Liratsopulos Georgios et al are
saying "Not accounting for the likely differential effect of the
vaccination status of these 686 children introduces selection bias.." It
is not clear to us what differential effect there would be, and how it
would introduce a selection bias. It may be appropriate to look at the
study in the following way: Since all pregnancies were registered, we have
been able to report on total perinatal mortality in the cohort. However,
for immunization status we could only use the information collected at the
6-monthly visits. The visits can be seen as randomly distributed semi-
annual events where we recruited the children alive at the visit. By the
nature of this design, we would lose around 50% of the follow up time for
children between 1 and 6 months of age, and it is therefore not surprising
that many deaths occurred before we visited the village. For this
procedure to introduce a bias, there should be differential mortality of
vaccinated and unvaccinated children prior to the visit where the children
were recruited for the vaccine analysis, e.g. that frail children died
more quickly in the unvaccinated group, producing a higher proportion of
frail children in the vaccinated group. A similar pattern could have
produced higher mortality in the diphtheria, tetanus, pertussis vaccinated
group. If that were the case in the present analysis, the difference
between vaccinated and unvaccinated children should increase with age at
recruitment, as the selection process had had longer time to produce an
effect. The trend was actually in the opposite direction as can be seen
from Table 4 in the paper. Furthermore, we are unable to understand how a
selection bias related to inclusion-non-inclusion of certain children
could produce a beneficial effect for BCG but a negative effect for
diptheria, tetatnus, pertussis and polio vaccines.
Lorna Barton et al emphasise that we stated that the study initially
recruited 10,000 women, but that 15,351 were included in the analysis. As
explained in greater detail elsewhere (2), at each 6-monthly visit we
included young girls as they reached fertile age, as well as women moving
to the compound. There is no contradiction between the statements on "no
loss to follow up" and "information was unavailable", as "no loss to
follow up" refers to information about survival status, whereas
"information was unavailable" refers to vaccinations received between the
first and the second visits for children who had died, or moved or were
not present at the second visit.
John Heptonstall suggests that we may have made "subtle diversions,
calculated measures to limit the impact of the real findings". That was
certainly not intended. As all totally unexpected observations, the study
was not planned to prevent the major sources of potential bias. Therefore,
a single study cannot be conclusive. The important test of unexpected
observations is to see whether they are repeatable in subsequent studies -
and that is certainly the case with our findings.
(1) Aaby P, Gomes J, Fernandes M, Djana Q, Lisse I, Jensen H.
Nutritional status and mortality of refugees and residents in a non-camp
setting during conflict: follow up study. BMJ 1999;319:878-81
(2) Høj L, Stensballe J, Aaby P. Maternal mortality in Guinea-Bissau: the
use of verbal autopsy in a multi-ethnic population. Int J Epidemiol
1999;28:70-76
Rapid Response:
Routine vaccinations and child survival: Authors' response
Both Tedd Koren and John P. Heptonstall did not understand our
statements that "Mortality was lower in the group vaccinated with any
vaccine compared with those not vaccinated…However, recipients of one dose
of diphtheria, tetanus, and pertussis or polio vaccines had higher
mortality than children who had received none of these vaccines". Note
that we wrote "none of these vaccines" rather than "no vaccine". Because
virtually all children having received diphtheria, tetanus, pertussis and
polio vaccines had already received BCG (see Table 4), the statement
implies that among children who had received BCG, those who had been
vaccinated with diphtheria, tetanus, pertussis and polio vaccines had a
higher mortality than those who had received only BCG. The trend was the
same among the few children who had received diphtheria, tetanus, and
pertussis and polio vaccines but not BCG. However, the main effect was
among children who had already received BCG as also pointed out in Paul
Fine's commentary.
Both Tedd Koren and the paper emphasised the possible selection
biases which may have influenced the mortality ratios between vaccinated
and unvaccinated children: these include better contact with the health
care system, and restriction of BCG vaccination to children with normal
birth weight. However, it is difficult to understand how selection bias
would work differently for different vaccines. If one believes that
mothers complying with the vaccination programme have children with lower
mortality, the beneficial effect of BCG and measles may have been
overestimated whereas the harmful effect of diphtheria, tetanus, pertussis
and polio vaccines would have been underestimated. Conversely, if one
believes that sick children are more likely to be vaccinated when their
mothers bring them to a health centre, the beneficial effect of BCG and
measles would have been underestimated and the harmful effect of
diphtheria, tetanus, and pertussis and polio vaccines overestimated. While
we discussed the likelihood of these interpretations in the paper, the
main point was that the estimated effects of different vaccines went in
different directions, and that the non-specific effects of the vaccines
are therefore likely to be important. Only further research can clarify
the relative importance of the different vaccines.
Children were not given vitamin A at the time they were immunised
against measles. Vitamin A has not been used for general supplementation
in Guinea-Bissau until the recent war in 1998-1999 (1).
John Heptonstall, Liratsopulos Georgios et al, and Lorna Barton et
al. raised questions about the causes of deaths. Virtually all the
children died from infectious disease, which presented with fever,
malaria, convulsions, diarrhoea, oedema, inflammation, malnutrition,
vomiting, cough, respiratory problems, tetanus, measles, or contaminated
maternal milk (that is the mother had broken the taboo against having
sexual relations while breastfeeding). Only one death may have been due to
an accident (burns). However, it should be recognised that the vast
majority were recorded only as having fever, so the precise cause of death
is unclear in most cases. As pointed out by Dr. Georgios et al, we have
done verbal autopsies for women of fertile age (2). It is, however, quite
a different matter to do verbal autopsies for child deaths in an animistic
culture, where many child deaths will be ascribed to malevolent spirits.
It is impolite to recall the death of a child in the presence of the
mother. It might have been possible to obtain better retrospective
information on the sequence of events leading to a child's death using a
medically qualified interviewer. However, such a study would be offensive
to the mothers, and would not provide definitive information about the
causes of death.
The study was large and complex and many potential details have not
been adequately covered even in the long version of the paper. Several
readers raised questions about the follow up, including John Heptonstall
and Liratsopulos Georgios et al. It is noted that 686 children had died
before the first visit, and that these deaths constitute such a large
proportion of the deaths in the study that the conclusions of the paper
are questionable. We have probably assumed too much familiarity with
mortality patterns in developing countries. We had indicated that there
were 437 stillbirths in the cohort, and one would normally expect a
similar number of neonatal deaths in a developing country. In the present
study there were 525 neonatal deaths few of which could have been affected
by immunisations, since most neonatal deaths occur on the day of birth.
The remaining 161 infant deaths occurred before we visited the village.
Using a Cox proportional hazard model and adjusting for cluster as in the
paper, the mortality rate ratio between 1 and 6 months of age was 0.88
(0.69-1.12) for children not yet visited for the first time compared with
mortality following the first visit. Liratsopulos Georgios et al are
saying "Not accounting for the likely differential effect of the
vaccination status of these 686 children introduces selection bias.." It
is not clear to us what differential effect there would be, and how it
would introduce a selection bias. It may be appropriate to look at the
study in the following way: Since all pregnancies were registered, we have
been able to report on total perinatal mortality in the cohort. However,
for immunization status we could only use the information collected at the
6-monthly visits. The visits can be seen as randomly distributed semi-
annual events where we recruited the children alive at the visit. By the
nature of this design, we would lose around 50% of the follow up time for
children between 1 and 6 months of age, and it is therefore not surprising
that many deaths occurred before we visited the village. For this
procedure to introduce a bias, there should be differential mortality of
vaccinated and unvaccinated children prior to the visit where the children
were recruited for the vaccine analysis, e.g. that frail children died
more quickly in the unvaccinated group, producing a higher proportion of
frail children in the vaccinated group. A similar pattern could have
produced higher mortality in the diphtheria, tetanus, pertussis vaccinated
group. If that were the case in the present analysis, the difference
between vaccinated and unvaccinated children should increase with age at
recruitment, as the selection process had had longer time to produce an
effect. The trend was actually in the opposite direction as can be seen
from Table 4 in the paper. Furthermore, we are unable to understand how a
selection bias related to inclusion-non-inclusion of certain children
could produce a beneficial effect for BCG but a negative effect for
diptheria, tetatnus, pertussis and polio vaccines.
Lorna Barton et al emphasise that we stated that the study initially
recruited 10,000 women, but that 15,351 were included in the analysis. As
explained in greater detail elsewhere (2), at each 6-monthly visit we
included young girls as they reached fertile age, as well as women moving
to the compound. There is no contradiction between the statements on "no
loss to follow up" and "information was unavailable", as "no loss to
follow up" refers to information about survival status, whereas
"information was unavailable" refers to vaccinations received between the
first and the second visits for children who had died, or moved or were
not present at the second visit.
John Heptonstall suggests that we may have made "subtle diversions,
calculated measures to limit the impact of the real findings". That was
certainly not intended. As all totally unexpected observations, the study
was not planned to prevent the major sources of potential bias. Therefore,
a single study cannot be conclusive. The important test of unexpected
observations is to see whether they are repeatable in subsequent studies -
and that is certainly the case with our findings.
Peter Aaby,
Henrik Jensen
psb@sol.gtelecom.gw
(1) Aaby P, Gomes J, Fernandes M, Djana Q, Lisse I, Jensen H.
Nutritional status and mortality of refugees and residents in a non-camp
setting during conflict: follow up study. BMJ 1999;319:878-81
(2) Høj L, Stensballe J, Aaby P. Maternal mortality in Guinea-Bissau: the
use of verbal autopsy in a multi-ethnic population. Int J Epidemiol
1999;28:70-76
Competing interests: No competing interests