Non-specific effects of vaccinations: lacking gradient, strength, and coherence?
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
Non-specific effects of vaccinations: lacking gradient, strength, and coherence?
In our study in Bissau, we found that different vaccines had opposite
effects on mortality: BCG and measles vaccines reduced mortality, and
diphtheria-tetanus-pertussis (DTP) and polio vaccines increased mortality.
It would not have been ethical for us to randomise children to receive
vaccine or placebo, so we had to perform a cohort study rather than a
controlled trial. Hence, as we stated in the Discussion section of the
paper, it is not possible to give precise estimates of the effect of the
individual vaccines on mortality. We were therefore surprised that
Professor Fine concluded that our study lacked the classical attributes of
causality: gradient, strength, and coherence (2). As should be evident
from the conclusion of the Abstract, we did not claim that our study had
these attributes for each vaccine - our main conclusion was that the
observed effects of the vaccines are unlikely to have been caused by
confounding because selection bias should not produce opposite mortality
trends for different vaccines. The strength of these trends should be the
subject of future studies.
Prof Fine gives several reasons for claiming lack of gradient,
strength, and coherence. Firstly, the estimates only just reach
conventional statistical significance. However, significance for each
vaccine was not the focus of the paper. The important finding was that
the trends in mortality were in opposite directions for BCG on the one
hand, and DTP-polio on the other (see Table 3 in reference 1) - with the
difference in trends being significant (p=0.005).
Secondly, Prof Fine suggests that the association of BCG and measles
vaccines with reduced mortality was surprising, and that it may have
occurred because the mothers of vaccinated children make greater use of
health services. He also suggested that the association of DTP with
increased mortality could be due to DTP coverage being higher in the
children of young mothers (who have increased mortality). In fact, Prof
Fine should not have been surprised at these findings, as they have all
been reported before: numerous studies have found that measles vaccine
reduces mortality from diseases other than measles (3), BCG was found to
be associated with reduced mortality from pneumonia in 1992 (4), and the
association of DTP with increased mortality was reported in 1991 and 1995
(3,5). As stated in the paper, adjustment for the background factors
listed in Table 2 had minimal effect on the estimates for the effect of
the vaccines (1). Given the low prevalence of young mothers (Table 2), it
is not surprising that the small increase in DTP coverage in the children
of young mothers has only a trivial effect: the mortality rate ratio
estimates for BCG, one dose of DTP, and two or more doses of DTP presented
in the paper were 0.55 (0.36 to 0.85), 1.84 (1.10 to 3.10), and 1.38 (0.73
to 2.61), respectively; adjustment for mother's age gave estimates of 0.54
(0.35 to 0.84), 1.82 (1.08 to 3.07), and 1.37 (0.72 to 2.59),
respectively.
Thirdly, Prof Fine considers that the results are anomalous because
mortality declines with age in BCG unvaccinated children, but increases
with age among vaccinated children. This is true in Table 3, which
compares BCG vaccinated and unvaccinated children, but not in Table 5,
where mortality is declining with age for both measles-vaccinated and
unvaccinated children (1). These findings are understandable if exposure
to DTP does increase mortality. As seen in Table A, among children who
survived to the second visit, exposure time to DTP and the number of DTP
doses increased markedly with age in the BCG vaccinated group, whereas
there was little change in the exposure time to DTP or number of DTP doses
for children who had not received BCG prior to the first visit. The
explanation for these patterns is presumably that children who have
reached several months of age without receiving BCG are less likely to
have contact with the health system, and are therefore less likely to
receive additional doses of DTP during follow-up. On the other hand, when
children have received BCG vaccine, the older they become, the longer they
have had to be exposed to DTP.
Table A.
Exposure to DTP according to age and initial BCG vaccination status
Average number of DTP doses received (total doses/children) and percent time exposed to DTP (time with DTP exposure*doses/person-years) §
BCG unvaccinated at 1st visit
BCG vaccinated at 1st visit
Age at first visit
No. DTP doses
Time with DTP exposure
No. DTP doses
Time with DTP exposure
0-1 months
1.14 (726/635)
56%(166.3/299.1)
2.02(872/432)
119%(244.0/204.5)
2-3 months
1.01 (323/321)
60%(90.4/151.0)
2.26(1614/714)
166%(563.6/339.4)
4-6 months
1.07 (235/219)
70%(71.4/101.6)
2.37(1975/832)
202%(785.3/389.0)
Note: § Time with exposure to DTP has been calculated as the sum of time exposed for each dose during the first follow-up period
Fourthly, Professor Fine finds it strange that the effect associated
with one dose of DTP is not significant for two or three doses. Because
fewer children received two or more doses of DTP, it is not surprising
that the effect is not significant in these children. However, Prof Fine
is presumably emphasising that the estimate of 1.84 (1.10 to 3.10) for one
dose of DTP declined to 1.38 (0.73 to 2.61) for two or more doses of DTP -
but the difference between these two estimates is not statistically
significant (p=0.250). We did not discuss this issue because the main
point of the paper was that BCG and measles vaccines on the one hand, and
DTP-polio on the other hand, have opposite effects on mortality. However,
we did mention in the Discussion that the mortality estimates were
conservative due to interference from subsequent vaccinations during the
period of follow-up (1). The children who received two or more doses of
DTP were much more likely to receive measles vaccine before or during
follow-up (51% - among children who survived to the second visit) than
children who received only one dose of DTP (28%) or no DTP (13%). If
follow-up is censored at 9 months of age to minimize the effect of
interference from subsequent measles vaccinations, there is minimal
difference between the estimates for one dose of DTP (mortality ratio
1.81, 95% CI 1.00-3.26) versus two or more doses of DTP (mortality ratio
1.75, 95% CI 0.86-3.59).
Although we did not claim this in the paper, the non-specific effects
of DTP-polio vaccination may well have had gradient, strength, and
coherence in our study (1). As pointed out by Prof Fine, the DTP-polio
effect was mainly measured among children who had already received BCG -
this is to be expected in a developing country, where virtually all
children get BCG as their first vaccine. We readily agree that a single
study is not convincing, and that further studies of the non-specific
effects of vaccines are needed - and this point was emphasized in the
paper (1). It will be particularly important to see what happens when DTP-
polio is used among children who have not received BCG.
Peter Aaby, Henrik Jensen
Bandim Health Project, Danish Epidemiology Science Centre, Apartado 861,
Bissau, Guinea-Bissau
1. Kristensen I, Aaby P, Jensen H. Routine vaccinations and child
survival: follow up study in Guinea-Bissau, West Africa. BMJ 2000;321:1435
-1438
2. Fine P. Commentary: an unexpected finding that needs confirmation
or rejection. BMJ 2000, 321; 1439
3. Aaby P, Samb B, Simondon F et al. Non-specific beneficial effects
of measles immunization: analysis of mortality studies from developing
countries. BMJ 1995;311:481-485
4. Niobey FML, Duchiade MP, Vasconcelos AGG, de Cavalho ML, Leal MC,
Valente JG. Fatores de risco para morte por pneumonia em menores de um ano
em uma região metropolitana do sudeste do Brasil. Un estudo tipo caso-
controle. Rev Saúde Públ, São Paolo 1992; 26: 229-238
5. Velema JP, Alihonou EM, Gandaho T, Hounye FH. Childhood mortality
among users and non-users of primary health care in a rural West African
community. Int J Epidemiol 1991;20:474-79
Rapid Response:
Non-specific effects of vaccinations: lacking gradient, strength, and coherence?
In our study in Bissau, we found that different vaccines had opposite
effects on mortality: BCG and measles vaccines reduced mortality, and
diphtheria-tetanus-pertussis (DTP) and polio vaccines increased mortality.
It would not have been ethical for us to randomise children to receive
vaccine or placebo, so we had to perform a cohort study rather than a
controlled trial. Hence, as we stated in the Discussion section of the
paper, it is not possible to give precise estimates of the effect of the
individual vaccines on mortality. We were therefore surprised that
Professor Fine concluded that our study lacked the classical attributes of
causality: gradient, strength, and coherence (2). As should be evident
from the conclusion of the Abstract, we did not claim that our study had
these attributes for each vaccine - our main conclusion was that the
observed effects of the vaccines are unlikely to have been caused by
confounding because selection bias should not produce opposite mortality
trends for different vaccines. The strength of these trends should be the
subject of future studies.
Prof Fine gives several reasons for claiming lack of gradient,
strength, and coherence. Firstly, the estimates only just reach
conventional statistical significance. However, significance for each
vaccine was not the focus of the paper. The important finding was that
the trends in mortality were in opposite directions for BCG on the one
hand, and DTP-polio on the other (see Table 3 in reference 1) - with the
difference in trends being significant (p=0.005).
Secondly, Prof Fine suggests that the association of BCG and measles
vaccines with reduced mortality was surprising, and that it may have
occurred because the mothers of vaccinated children make greater use of
health services. He also suggested that the association of DTP with
increased mortality could be due to DTP coverage being higher in the
children of young mothers (who have increased mortality). In fact, Prof
Fine should not have been surprised at these findings, as they have all
been reported before: numerous studies have found that measles vaccine
reduces mortality from diseases other than measles (3), BCG was found to
be associated with reduced mortality from pneumonia in 1992 (4), and the
association of DTP with increased mortality was reported in 1991 and 1995
(3,5). As stated in the paper, adjustment for the background factors
listed in Table 2 had minimal effect on the estimates for the effect of
the vaccines (1). Given the low prevalence of young mothers (Table 2), it
is not surprising that the small increase in DTP coverage in the children
of young mothers has only a trivial effect: the mortality rate ratio
estimates for BCG, one dose of DTP, and two or more doses of DTP presented
in the paper were 0.55 (0.36 to 0.85), 1.84 (1.10 to 3.10), and 1.38 (0.73
to 2.61), respectively; adjustment for mother's age gave estimates of 0.54
(0.35 to 0.84), 1.82 (1.08 to 3.07), and 1.37 (0.72 to 2.59),
respectively.
Thirdly, Prof Fine considers that the results are anomalous because
mortality declines with age in BCG unvaccinated children, but increases
with age among vaccinated children. This is true in Table 3, which
compares BCG vaccinated and unvaccinated children, but not in Table 5,
where mortality is declining with age for both measles-vaccinated and
unvaccinated children (1). These findings are understandable if exposure
to DTP does increase mortality. As seen in Table A, among children who
survived to the second visit, exposure time to DTP and the number of DTP
doses increased markedly with age in the BCG vaccinated group, whereas
there was little change in the exposure time to DTP or number of DTP doses
for children who had not received BCG prior to the first visit. The
explanation for these patterns is presumably that children who have
reached several months of age without receiving BCG are less likely to
have contact with the health system, and are therefore less likely to
receive additional doses of DTP during follow-up. On the other hand, when
children have received BCG vaccine, the older they become, the longer they
have had to be exposed to DTP.
Table A.
Exposure to DTP according to age and initial BCG vaccination status
Average number of DTP doses received (total doses/children) and percent time exposed to DTP (time with DTP exposure*doses/person-years) §
Fourthly, Professor Fine finds it strange that the effect associated
with one dose of DTP is not significant for two or three doses. Because
fewer children received two or more doses of DTP, it is not surprising
that the effect is not significant in these children. However, Prof Fine
is presumably emphasising that the estimate of 1.84 (1.10 to 3.10) for one
dose of DTP declined to 1.38 (0.73 to 2.61) for two or more doses of DTP -
but the difference between these two estimates is not statistically
significant (p=0.250). We did not discuss this issue because the main
point of the paper was that BCG and measles vaccines on the one hand, and
DTP-polio on the other hand, have opposite effects on mortality. However,
we did mention in the Discussion that the mortality estimates were
conservative due to interference from subsequent vaccinations during the
period of follow-up (1). The children who received two or more doses of
DTP were much more likely to receive measles vaccine before or during
follow-up (51% - among children who survived to the second visit) than
children who received only one dose of DTP (28%) or no DTP (13%). If
follow-up is censored at 9 months of age to minimize the effect of
interference from subsequent measles vaccinations, there is minimal
difference between the estimates for one dose of DTP (mortality ratio
1.81, 95% CI 1.00-3.26) versus two or more doses of DTP (mortality ratio
1.75, 95% CI 0.86-3.59).
Although we did not claim this in the paper, the non-specific effects
of DTP-polio vaccination may well have had gradient, strength, and
coherence in our study (1). As pointed out by Prof Fine, the DTP-polio
effect was mainly measured among children who had already received BCG -
this is to be expected in a developing country, where virtually all
children get BCG as their first vaccine. We readily agree that a single
study is not convincing, and that further studies of the non-specific
effects of vaccines are needed - and this point was emphasized in the
paper (1). It will be particularly important to see what happens when DTP-
polio is used among children who have not received BCG.
Peter Aaby, Henrik Jensen
Bandim Health Project, Danish Epidemiology Science Centre, Apartado 861,
Bissau, Guinea-Bissau
psb@sol.gtelecom.gw
References
1. Kristensen I, Aaby P, Jensen H. Routine vaccinations and child
survival: follow up study in Guinea-Bissau, West Africa. BMJ 2000;321:1435
-1438
2. Fine P. Commentary: an unexpected finding that needs confirmation
or rejection. BMJ 2000, 321; 1439
3. Aaby P, Samb B, Simondon F et al. Non-specific beneficial effects
of measles immunization: analysis of mortality studies from developing
countries. BMJ 1995;311:481-485
4. Niobey FML, Duchiade MP, Vasconcelos AGG, de Cavalho ML, Leal MC,
Valente JG. Fatores de risco para morte por pneumonia em menores de um ano
em uma região metropolitana do sudeste do Brasil. Un estudo tipo caso-
controle. Rev Saúde Públ, São Paolo 1992; 26: 229-238
5. Velema JP, Alihonou EM, Gandaho T, Hounye FH. Childhood mortality
among users and non-users of primary health care in a rural West African
community. Int J Epidemiol 1991;20:474-79
Competing interests: No competing interests