The survival benefit of measles immunization may not be explained entirely by the prevention of measles disease: a community study from rural Bangladesh

Peter Aaby1, Abbas Bhuiya2, Lutfun Nahar2, Kim Knudsen1, Andres de Francisco2 and Michael Strong2

1 Department of Epidemiology Research, Danish Epidemiology Science Centre, Statens Seruminstitut, Copenhagen, Denmark.
2 International Centre for Diarrhoeal Disease Research, Dhaka, Bangladesh.

Peter Aaby, Department of Epidemiology Research, Statens Serum Institut, Artillerivej 5, 2300 Copenhagen S, Denmark. E-mail: psb{at}sol.gtelecom.gw


    Abstract
 Top
 Abstract
 Subject and Methods
 Results
 Discussion
 Conclusion
 References
 
Objective To examine whether the reduction in childhood mortality after immunization can be explained by the prevention of measles and its long-term effects.

Methods and Data We re-analysed an existing data set from Matlab, Bangladesh. During 1982–1985, measles immunization was used from 9 months of age in half of the study area, and the other half was used as an unvaccinated control area. A total of 8134 immunized children had been matched by age with 8134 non-immunized children; 578 children died during the follow-up period of 3 years. Using these data, we calculated the vaccine effectiveness against death (VED) controlling for significant factors in a matched analysis. In the absence of measles, there should be no difference in mortality between immunized, uninfected children and non-immunized, uninfected children. We therefore calculated VED after the exclusion of all measles cases in the survival analysis. To assess the long-term effects of measles, we compared survival of unvaccinated children after measles disease with children who had not yet contracted measles.

Results Prior to immunization and again after 1985, childhood mortality rates were 10% lower in the area that had received immunization. Though measles deaths only constituted 12.4% of the non-accidental deaths, the VED controlling for significant factors was 49% (95% CI: 38–58%). The vaccine was protective against measles death throughout the study, but it also had a marked effect against other causes of death, particularly diarrhoea and oedema. This effect may have been particularly strong in the first 6 months after immunization (VED = 74, 95% CI: 57–84%). The VED was only reduced from 49% to 43% (95% CI: 31–54%) when measles cases were excluded in the survival analysis. Controlling for background factors, mortality among measles cases was increased during the acute phase (0–45 days) (mortality ratio [MR] = 17.35, 95% CI: 11.9–25.3) and in the following 11/2 months (MR = 2.35, 95% CI: 0.95–5.84). However, post-measles cases had significantly lower mortality than uninfected, non-immunized children in the following 9 months (MR = 0.40, 95% CI: 0.16, 0.98).

Conclusions The non-randomized character of the original study and the possibility of uncontrolled confounding between the two areas prevent a precise estimate of the effectiveness of measles vaccine, but it is likely to have been substantial. Though there may have been some underreporting of cases of measles, the prevention of measles infection can only explain a limited part of the observed impact of measles immunization in Bangladesh. Furthermore, mortality may be reduced after the acute phase of measles infection. The observations from Bangladesh are consistent with recent research from Africa suggesting that measles immunization may have non-specific beneficial effects on survival.


Keywords Measles, measles immunization, non-specific effects of vaccination, post measles mortality

Accepted 23 July 2002

Community studies from India and Bangladesh have reported relatively low case fatality rates (CFR) in measles disease1 and measles control has not been a priority in this part of the world.2 For example, in Matlab, Bangladesh, the CFR has been 1–3% in different studies,3–5 and measles deaths only represented a small proportion of all deaths.6 Nonetheless, two studies from Matlab have estimated that measles immunization reduced mortality against all causes by 36% and 46%, respectively.6,7 Similarly, investigations from Guinea-Bissau,8–10 Senegal,11,12 Benin,13 Zaire,14 Nigeria,15 Burundi,16 and Haiti,17 have reported much greater reductions in mortality than should be expected.18 It has therefore been speculated that measles disease was associated with post-measles excess mortality that would also be prevented by measles vaccination.5,6,8

Few studies have analysed the long-term survival impact of measles disease. Earlier studies from West Africa8,19 reported increased mortality among post-measles cases. However, these studies did not control for potential confounding factors, particularly measles vaccination status, and they compared survival of children after measles infection with children who had been immunized against measles.18 Recent studies from Guinea-Bissau,20 Senegal,21 Burundi16 and Ghana22 have found no difference or lower mortality for post-measles cases when the immunization status of controls was taken into consideration. If the prevention of acute and post-measles mortality does not explain the reduction in mortality associated with measles immunization, measles immunization may have non-specific beneficial effects.18

The Matlab field station of the International Centre for Diarrhoeal Disease Research, Bangladesh (ICDDR,B), has maintained a demographic and health surveillance system since 1966 in the Matlab rural district. The world’s largest study on the impact of measles immunization was conducted in this area between 1982 and 19857 when the population under surveillance counted 190 000 individuals (1985).23 Since the understanding of the impact of measles immunization and the long-term effects of measles infection may have major implications for disease control programmes, we requested permission to re-analyse the existing data set to examine to what extent the prevention of measles infection may explain the impact of measles vaccination.


    Subject and Methods
 Top
 Abstract
 Subject and Methods
 Results
 Discussion
 Conclusion
 References
 
The original study7 was carried out to examine the effect of measles vaccination on childhood mortality. We re-analysed the existing data set7 introducing only minor modifications as described below. We combined this data set with routine surveillance data on measles infection and causes of death from the Demographic Surveillance System (DSS).

Design of original study and modifications
The data set has been described in detail elsewhere.7 The introduction of measles vaccine in the Matlab area was phased; children aged 9–60 months living in two of the four sub-areas (blocks A and C) were offered Schwarz measles vaccine in March 1982 and onwards, the coverage reaching more than 80% in 1983–1984.6 In November 1985, measles vaccination was extended to blocks B and D and reached a similar coverage. In blocks A and C, a total of 9133 children between 9 and 60 months of age were immunized against measles during the period of the study. These vaccinated children were matched individually with non-immunized children from blocks B and D with date of birth (+/– 1 month) as the only matching criterion.7 Unvaccinated children were alive at the day of immunization of the matching case. Unvaccinated children from blocks B and D could only be found for 8135 of the immunized children. Only eight children in B and D had been immunized, and thus no real selection was involved among the unvaccinated children. The 998 vaccinated children for whom a matched child could not be found were not included in the analysis.

The present analysis is restricted by the available data previously presented7 and we are therefore subject to the same potential confounding factors.7second pair with this child was excluded and our analysis has therefore 8134 pairs instead of 8135 pairs.

Definitions and concepts
The identification of measles cases depended on the parent’s perception. This had been found to be adequate when the study started.4 Measles onset has been defined by the date of the rash. In the present study, deaths within 45 days of the rash have been ascribed to measles unless they were due to accidents. Long-term effects of measles refer to factors affecting mortality after the acute phase of infection and could be due to persistent measles virus infection as occurs in some individuals, permanent damage like blindness, or changes in susceptibility to other infections due to nutritional or immunological changes initiated during measles infection.

Measles surveillance
Surveillance of measles disease in the study area has been conducted by two different and partly overlapping systems. Firstly, an epidemiological study of all measles cases was initiated in October 1979 and continued until December 1984. Data from the first year of this study have been analysed previously.4 Cases of measles were reported to field assistants of the Demographic Surveillance System (DSS) through community health workers who visited all households in the study area every second week. Field assistants had been trained to fill in questionnaires containing simple information on onset of measles rash and symptoms.4 Cases were defined on the basis of the simultaneous presence of rash and fever. Initially, a random sample of 108 cases were examined by a medical doctor and verified in all but two cases.

Secondly, community health workers visited all households in the study area twice a month to provide free health services and maintain records of demographic events, immunizations, and morbidity episodes, including measles, for children <5 years of age.5 The two systems were not cross-referenced at the time, and they sometimes differed with respect to date of measles disease. We have assumed the first date to be correct. For the period when both systems were in operation, the majority of cases were reported by the morbidity surveillance system.

Causes of death and death from measles
Information on causes of death was obtained from the DSS where field supervisors complete a death form based on information from parents. In previous studies of measles in Bangladesh, deaths within either 45 days or 3 months of a measles rash have been considered to be due to measles.4,5 However, death reports in the DSS have not necessarily followed these definitions. Deaths classified as measles that occurred more than 45 days after measles disease (N = 7) have been reclassified as ‘other causes’. Likewise deaths for which there was no report of measles disease (N = 10) have been reclassified as ‘other causes’ since these deaths are either misclassified or correspond to surviving measles cases which have remained unreported.

Statistical methods
The main analyses are based on the original paired design; 16 268 children were included, 8431 boys and 7837 girls. Boys were not paired more frequently with boys than with girls or vice versa ({chi}2 = 3.24,1 d.f.; P = 0.072). Using a proportional hazards model for paired survival data, the mortality ratio (MR) between vaccinated and unvaccinated children is the ratio of the number of pairs in which the vaccinated child dies first to the number of pairs in which the unvaccinated child dies first. The standard error and the test statistic for no effect can also be calculated from the two numbers.24

The data have been presented previously in an unmatched analysis7 and statistical efficiency can be improved by considering the matching variable, month and year of birth, as a confounding factor25 in a proportional hazards model with age as the time scale. Follow-up time after the first death or withdrawal of a pair can be included in the analysis while still only comparing children of the same age. By this procedure, the data can be analysed ‘unpaired’, but still controlling for month and year of birth and age. Immunized children were evaluated from age at 1 October 1982, or from the age at vaccination, if occurring later (left truncation), and until death, moving, or 31 October 1985, respectively (right censoring). The unvaccinated children’s date of entry is set from their matched vaccinated child’s date of entry. Using similar catagorizations, the background factors used in previous analyses7 were examined for confounding as time-independent variables (e.g. sex, number of siblings, maternal education, size of dwelling) or as time-dependent variables (e.g. measles, season). In these analyses, 635 children were excluded due to missing information regarding number of siblings, maternal education, or size of dwelling.

The analyses of measles incidence and acute measles mortality were limited to cases occurring from 1 October 1982, and after entry into the study. Within the unvaccinated areas (B + D), the age-adjusted mortality of children who contracted measles during the study was compared with that of all uninfected, unvaccinated children using an unmatched multivariate proportional hazards model with age as the underlying time. In this analysis children who had measles before the study were excluded. Children with measles were included in the measles group from the day of the rash. Uninfected children were included in this analysis from the day of entry into the study and were censored when they got measles, moved, or died, or on 31 October 1985, whichever came first.

Survival information for individuals in the vaccination study was available until December 1991 and this information has been used to compare mortality after the measles vaccination had been introduced in the previously unvaccinated area.


    Results
 Top
 Abstract
 Subject and Methods
 Results
 Discussion
 Conclusion
 References
 
Comparison of study populations: mortality, vaccination coverage, and measles disease
Prior to the introduction of measles immunization in blocks A and C, there may have been a difference in childhood mortality of around 5–10% between blocks A and C versus blocks B and D (Table 1Go). A similar level was reached again in 1987–1988 after the introduction of immunization in blocks B and D. The difference in measles immunization coverage may have been associated with other differences in heath service programmes; for example, the coverage for maternal tetanus immunization in blocks A and C was 75%, 87%, 87%, 89%, and 84% in January of 1982, 1983, 1984, 1985, and 1986, respectively, whereas the coverage was 49%, 54%, 57%, 59%, and 61% on these dates in Block B and D. However, there were no major differences in coverage levels for the childhood vaccines; the proportion of DPT immunized children was 50% in Block A and C and 51% in Block B and D at the end of 1985, and the coverage for BCG was 46% and 39% in the two areas, respectively.


View this table:
[in this window]
[in a new window]
 
Table 1 Child mortality rates among 1–4 year-olds. Matlab, Bangladesh, 1981–1988
 
Between October 1982 and October 1985, there were 578 deaths among the 16 268 children participating in the study, 209 immunized and 369 non-immunized children; 534 deaths among children under 5 years of age7 and 44 deaths after 5 years of age. Six months after immunization had been introduced in blocks B and D there was no longer any difference between the areas; from May 1986 to December 1991, 91 deaths occurred in the cohort of 7631 immunized children of blocks A and C and 86 deaths among the 7317 previously unvaccinated children in blocks B and D (relative risk [RR] = 1.01, 95% CI: 0.76–1.36).

Measles prior to immunization or matching was reported for similar numbers of individuals in both groups (Table 2Go). During the study (Figure 1Go), vaccine effectiveness against measles disease was 93.7% (95% CI: 91.8–95.1%) in a paired analysis. The cumulative incidence of reported measles was 25.3% (95% CI: 22.4–28.2%) between 9 and 48 months of age for unvaccinated children in blocks B and D, and 3.9% had had measles prior to entry at 9 months of age. The incidence among immunized children was 2.3% (95% CI: 0.7–5.7%). Measles accounted for 7.4% of all deaths (43/578) and constituted 12.4% of the non-accidental deaths in the non-immunized area (Table 3Go).


View this table:
[in this window]
[in a new window]
 
Table 2 Number of reported measles cases according to study group and time of measles immunization. Matlab, Bangladesh, 1980–1991
 


View larger version (13K):
[in this window]
[in a new window]
 
Figure 1 Number of measles cases by month of onset in the unvaccinated and the vaccinated group. October 1982 to October 1985. Matlab, Bangladesh

 

View this table:
[in this window]
[in a new window]
 
Table 3 Measles deaths as a proportion of all non-accidental deaths in the non-immunized area. Matlab, Bangladesh, 1982–1985
 
Vaccine effectiveness against death
In the paired analysis, there were 197 deaths in the vaccinated group while both children in a pair were under observation, and 357 corresponding deaths among the unvaccinated children, giving an MR of 0.55 (95% CI: 0.46–0.66), i.e. a vaccine effectiveness against death (VED) of 45% (95% CI: 34–54%) ({chi}2 = 46.21; P < 0.0001). In the unpaired multivariate Cox analysis controlling for age, sex, size of dwelling, number of siblings, and maternal education, VED was 47% (95% CI: 36–55%). There was no difference by sex (P = 0.93).

Figure 2Go indicates the number of deaths over time in the two groups. During the first year, there were 60 immunized and 90 non-immunized deaths (VED = 33.3%, 95% CI: 7.6–51.9%). A marked excess of deaths was noted in the second year (68 immunized versus 178 non-immunized deaths) when epidemic outbreaks of shigella dysentery and measles occurred in the community (VED = 61.8%, 95% CI: 49.5–71.1%). In the last 13 months from October 1984 to October 1985, 69 immunized and 89 non-immunized children died (VED = 22.5% (95% CI: –6.2%, 43.4%).



View larger version (19K):
[in this window]
[in a new window]
 
Figure 2 Number of deaths by month in the unvaccinated and the vaccinated group. October 1982 to October 1985. Matlab, Bangladesh

 
Measles immunization had a major protective effect against dying of measles, diarrhoea and dysentery, and dropsy, whereas it had little impact on fevers, accidents, and other causes (Table 4Go). Accidents for children in the Matlab are mainly drowning and may constitute as much as 10% of the child deaths (Tables 3 and 4GoGo). If accidents were excluded from the analysis, the VED was 47% (95% CI: 36–56%) ({chi}2 = 46.63, d.f. = 1; P < 0.0001). In the multivariate analysis controlling for background factors, VED was 49% (95% CI: 38–58%). Deaths due to accidents have been excluded in the following analyses.


View this table:
[in this window]
[in a new window]
 
Table 4 Vaccine efficacy against specific causes of death. Paired analysis. Matlab, Bangladesh, 1982–1985
 
Vaccine effectiveness against death and time since immunization
The vaccine had a marked effect on measles mortality in all years following immunization. However, for non-measles related mortality, the impact may have been much stronger in the first 6 months after immunization, the VED being 74% (95% CI: 57–84%) (Table 5Go). The reduction was consistent over the 3 years of the study, being 68% (95% CI: 29–86%) during the first year, 81% (95% CI: 54–92%) during the second year, and 71% (95% CI: 20–89%) in the third year. There was little effect between 6 and 11 months after vaccination. However, there was a marked effect also in the second year following immunization. This effect was due to the large number of children recruited in the initial vaccination campaign and a high mortality due to shigella and measles epidemics in the second year of the study (Figures 1 and 2GoGo). Two years after immunization, there was no longer any impact on non-measles related deaths.


View this table:
[in this window]
[in a new window]
 
Table 5 Vaccine effectiveness against non-accidental death (VED) in relation to time since vaccination. Paired analysis. Matlab, 1982–1985
 
Vaccine effectiveness against death in the absence of measles disease
In the absence of measles disease, there should be no difference in mortality between immunized, uninfected children and non-immunized, uninfected children.18 We therefore carried out an analysis in which measles cases were censored from the onset of measles disease, thus excluding both acute and delayed measles deaths from the survival analysis. In the paired analysis, there were 136 non-accidental deaths among vaccinated children, and 221 deaths among unvaccinated children while the other child in the pair was still under observation (VED = 38%, 95% CI: 24–50%). In the unpaired analysis adjusting for age, sex, size of dwelling, maternal education, and number of siblings, VED was 43% (95% CI: 31–54%).

Studies from a later period in Matlab26 have suggested that the validity of measles diagnoses is lower among infants (53%) than among children aged 1–4 years (88%). Even though this may not apply to the period under study, we conducted a separate, paired analysis in which reported cases among infants were not taken into consideration. With 142 deaths among vaccinated and 247 deaths among unvaccinated children, VED was still 43% (95% CI: 29–53%).

Vaccine effectiveness against death after immunization of previous measles cases
There were 136 pairs in which both had had measles prior to immunization or matching. Six of the 136 non-immunized children died a non-accidental death, whereas none of the immunized children died while both children in a pair were under observation (P < 0.05). In an unpaired, adjusted analysis including 924 immunized (13 died) and 891 non-immunized children (27 died) who had measles prior to entry into the study, VED was 55% (95% CI: 10–77%).

Post-measles mortality
High post-measles mortality has been suggested as a possible explanation for the marked effect of measles immunization.5–8 Data from the present study suggested rather the opposite; among measles cases occurring during the study and surviving the first 3 months, there were only 4 deaths compared with 11 deaths among their immunized, paired controls (MR = 0.36, 95% CI: 0.12–1.14, exact two-sided P = 0.118).

In the two non-immunized blocks B and D, we therefore compared post-measles non-accidental mortality of the 1133 children who had measles during the study period (Table 2Go) with the mortality of all the 8134 children who had not yet had measles. The CFR within 45 days of measles disease was 3.6% (40/1120) and 4.0% (45/1120) within 3 months. Controlled for age, sex, number of siblings, maternal education, and housing area, measles cases had higher mortality during the first 45 days after measles and even in the following 11/2 months (Table 6Go) than children who had not yet had measles. However, in the following 9 months, the measles cases had significantly lower mortality than the children who had not yet had measles (MR = 0.40, 95% CI: 0.16–0.98).


View this table:
[in this window]
[in a new window]
 
Table 6 Results of the multivariate proportional hazards analysis of post-measles mortality. Only unvaccinated children. Matlab, Bangladesh, 1982–1985
 

    Discussion
 Top
 Abstract
 Subject and Methods
 Results
 Discussion
 Conclusion
 References
 
We re-analysed an existing data set7 to determine to what extent the impact of measles immunization on child survival could be explained by the prevention of measles and its long-term effects. The comparison of immunized and non-immunized children suggested a reduction in non-accidental mortality of 49% (95% CI: 38–58%) after measles immunization. If measles cases were excluded from the analysis, VED was still as high as 43%. The main impact of measles immunization may therefore have little to do with the prevention of the acute and long-term consequences of measles. The re-analysis of data from Matlab provided several other noteworthy observations; the non-specific impact of immunization may have been strongest in the first 6 months following immunization; measles immunization apparently reduced mortality also for children who had already had measles; and after the acute phase, post-measles cases had a temporary reduction in mortality compared with uninfected children.

It is essential that these tendencies are not the result of methodological flaws. None of the studies comparing immunized and non-immunized children have been double-blind placebo studies, nor could they have been. Authors have therefore tried to control for potential confounding factors and selection bias in different ways.18 The Matlab study may be the most satisfactory because it compared the effect of introducing measles immunization in one of two areas which had similar mortality levels prior to the intervention, and it is clearly the largest of the studies.18 It is important that its potential limitations are considered.

Limitations of the Matlab study
Comparability of the two areas
Prior to immunization, there was only a small difference of 3–10% in mortality of 1–4 year old children and no difference in the number of reported measles cases in the two populations being compared. There was a slightly bigger difference of 10–15% in the years after the introduction of vaccine in B and D (Table 1Go) but the vaccination programme was somewhat less effective in B and D since measles continued to be more frequent in B and D (Table 2Go). The BCG coverage was also lower in B + D and we have found BCG to be associated with reduced mortality.27 The likelihood of the two areas being relatively similar was strengthened by the fact that there was no difference in mortality between the two cohorts after the introduction of immunization in the control area. Likewise, it mitigates against an inherently higher mortality in the unvaccinated area that there was no difference between vaccinated and unvaccinated children 2 years after vaccination (Table 5Go), and that unvaccinated children who had survived measles tended to have lower mortality than their matched vaccinated control.

Localized epidemics not related to measles could have had an impact on the area-to-area comparison, for example, the shigellosis epidemic (Figure 2Go). What precise role these epidemics may have played is impossible to assess as there is no incidence data by area. Some of the effect during the second year of the study can probably be ascribed to the shigellosis epidemic. However, it should be noted that measles vaccinations were associated with beneficial effects in all three years and there was a particularly strong effect during the first 6 months after vaccination. It would not be possible to explain this pattern with reference to localized epidemics.

Differences in compliance
Since the children immunized in areas A and C were a compliant group accepting immunization, whereas controls from areas B and D were unselected, a systematic bias cannot be totally excluded. However, acceptance for all children <5 years of age was high in areas A and C, with 82% of the children reaching 9 months being vaccinated.7 In a previous analysis including the 18% non-immunized children in areas A and C, the difference between A + C and B + D only changed from 46% to 40%.7 Furthermore, adjustment for risk factors for child mortality did not lead to any modification of the VED. It is therefore unlikely that a selection bias is a major cause of the observed difference in mortality between immunized and non-immunized children from blocks A + C and B + D, respectively.

Reporting of measles infection
The measles surveillance system was not perfect. With a cumulative incidence of 28.2% before 4 years of age, measles infection may have been underreported. There is no previous study of cumulative incidence in a non-immunized population in the Matlab area with which to compare. However, in a previous study3 of a major epidemic in the pre-vaccination period, the age-specific annual attack rates suggested that 50% would have had measles between 4 and 5 years of age. In the present study, there had been a strong epidemic just before the vaccinations were initiated6 and there was therefore very little measles during the first year of the study (Figure 1Go). The high immunization coverage in Blocks A and C may also have contributed to herd immunity and lower incidence in Blocks B and D. The number of measles deaths among children not reported to have had measles compared with the number of measles deaths among known cases suggests that there may have been 23% (10/43) more cases not detected by the surveillance system. Given the degree of herd immunity when the study started, a 20–25% underreporting of measles cases may be a fair estimate.

Exclusion of the known cases of measles explained 12% ([49% – 43%]/49%) of the reduction in mortality associated with measles immunization. Adding 25–50% more cases would presumably only increase the proportion of vaccine efficacy explained to 15–18%. There may also have been some invalid reports of measles cases and the fact that measles vaccination of children with previous measles infection reduced mortality is likely to raise such concerns. However, misreporting is unlikely to have been substantial. A subsequent study from the immunization era, when knowledge of measles would have been less, indicated an 88% validity of reported measles cases aged 1–4 years.26 Major misreporting seems unlikely given the mortality profile of post-measles cases; were many case reports false, the decline in mortality among post measles cases would be even more extreme.

Potential impact of study limitations
We have pointed out a number of limitations of the data collected and of differences between the areas; for example, we found differences in coverage for maternal tetanus immunization. The possibility that uncontrolled differences exist questions the absolute estimate of VED, but hardly the foundation for the study. Assuming for example a 15% effect of differences in compliance and mortality levels as may have existed prior to 1982 (Table 1Go), measles immunized children would still have a reduction in mortality considerably larger than the proportion of deaths due to measles infection (12%) in the period 1982–1985. Given that measles cases had lower mortality after the acute phase, the 12% would actually overestimate the proportion of deaths due to measles.

Had incomplete and inaccurate detection of measles cases been a major factor behind the strong effect of measles immunization on non-measles mortality, the VED should have been much lower in the first year of the study when there was little measles (Figure 1Go) due to the strong epidemic in 1982. There was no such effect, the VED being as strong in 1982–1983 as in later years with a high incidence of measles (data not shown).

Furthermore, it should be noted that even if further uncontrolled differences exist between the two areas, these would not invalidate the conclusions that the impact of measles immunization6,7 has little to do with the specific prevention of measles, that the non-specific impact of immunization was temporary, and that post-measles cases had a reduction in mortality compared with uninfected children in blocks B and D where children had not been immunized. These tendencies correspond closely to observations made in West Africa.18,20,21

Post-measles mortality: selection bias or immune activation?
In contrast to previous hypotheses,5,6,8 mortality was increased only in the second and third months after measles, mainly due to diarrhoea. After the third month, post-measles cases had significantly lower mortality than uninfected individuals. This is contrary to usual assumptions1,6 about the impact of measles disease on long-term mortality. A small underreporting of measles cases as may well have existed in the present study would have contributed minimally to exaggerate the mortality level in the large group of non-immunized, uninfected children since less than half of the deaths among measles cases were excess deaths (Table 6Go). Furthermore, underreporting would not affect the time-dependent mortality profile of previous measles cases and it seems therefore unlikely it would change the conclusion that previous measles cases may have lower mortality after measles. Better post-measles survival could be due to ‘selection’ of stronger individuals among measles survivors. In both Bangladesh and West Africa,4,28 intensive exposure as a secondary case is a major risk factor for acute measles mortality and if ‘selection’ was the main reason, post-measles mortality should be particularly low among secondary cases. In studies from Senegal and Guinea-Bissau, index cases of measles had both lower acute and significantly lower post-measles mortality than secondary cases (ref. 21, unpublished data). As low post-measles mortality follows low acute mortality, ‘selection’ of stronger individuals is not the best explanation. Instead a better survival after measles infection could be due to a beneficial immune activation as suggested for measles immunization.18,29

Studies from Guinea-Bissau20 and Senegal21 have shown reduced mortality among post-measles cases and larger than expected mortality reduction after measles immunization.18,27 The reduction in mortality has been temporary, being strongest in the first period after immunization18 or infection. The Bangladesh study was consistent with previous studies from Bissau, Senegal, and Burundi, showing that exclusion of measles disease had very little impact on the estimated VED.18,27 Data do not support a disease-specific interpretation. The simplest explanation would be that measles vaccination and measles disease stimulate the immune system, providing protection against other infections. Though both measles and measles immunization are associated with immune suppression in the acute phase, there is growing evidence that in the longer term they stimulate a Th1 cytokine profile.29–32 For example, measles immunization stimulates interferon production, providing protection against subsequent challenge with vaccinia virus.30 In animal studies a Th1 profile induced by live-vaccine leads to mild infection whereas a Th2 profile due to inactivated vaccination leads to severe disease when challenged with respiratory syncitial virus or influenza.33–35


    Conclusion
 Top
 Abstract
 Subject and Methods
 Results
 Discussion
 Conclusion
 References
 
The observations from Bangladesh and Africa18 question the disease-specific perspective underlying the planning of immunization programmes. In areas with high mortality, it is necessary to measure the impact on total mortality.36 Immunization policies for the optimal age at immunization and the correct number of doses cannot be determined merely by investigating antibody responses and vaccine effectiveness against the specific disease. For example, we showed that children who were immunized against measles at 6–8 months of age and then offered re-immunization had lower mortality than children immunized at the recommended age of 9–11 months of age.37

The suggestion that measles disease could lead to immune stimulation reducing mortality in the post-acute phase may raise concerns about ‘natural’ infection and ‘unnatural’ immunizations. However, measles has a high acute mortality and there is no doubt that measles immunization has a more beneficial effect on survival than measles disease18 as it can be distributed to everybody at an earlier and specific age and with less risk of adverse effects. Nevertheless, there may be reasons to question the ‘culture of eradication’ which assumes that vaccines have only specific effects and that they can be withdrawn later to save production and distribution costs once diseases such as smallpox, measles, and polio have been eradicated. If measles vaccine does have non-specific beneficial effects, child mortality may increase again should measles vaccine be withdrawn after the eradication of measles.


KEY MESSAGES

  • We re-analysed a study of 16 268 measles-vaccinated and measles-unvaccinated children in Matlab, Bangladesh, to examine whether the prevention of measles and its long-term consequences can explain the reduction in mortality associated with measles vaccine.
  • The introduction of measles vaccination in half of the study area was associated with a 49% reduction in childhood deaths after 9 months of age.
  • When measles cases were excluded from the survival analysis, the reduction associated with measles vaccination was still 43%, suggesting that the effect of measles vaccination cannot be explained by the prevention of measles infection.
  • The reduction in mortality affected measles, diarrhoea, and oedema deaths and was particularly strong in the first 6 months after vaccination.
  • Unvaccinated children surviving acute measles had lower mortality compared with unvaccinated, measles-uninfected children.
  • These observations which are consistent with observations from West Africa support the possibility that measles immunization may have non-specific beneficial effects.

 


    Acknowledgments
 
This research was supported by the European Union (TS3*CT91*0002) and the International Centre for Diarrhoeal Disease Research, Bangladesh (ICDDR,B). Peter Aaby holds a research professorship funded by Novo Nordisk Foundation. The ICDDR,B is supported by countries and agencies which share its concern for the health problems of developing countries. Current donors include: the aid agencies of the Governments of Australia, Bangladesh, Belgium, Canada, China, Germany, Japan, The Netherlands, Norway, Republic of Korea, Saudi Arabia, Sweden, Switzerland, the UK, and the US; international organizations including the Arab Gulf Fond, Asian Development Bank, European Union, International Atomic Energy Centre, the United Nations Children’s Fund (UNICEF), the United Nations Development Programme (UNDP), the United Nations Population Fund (UNFPA), and the World Health Organization (WHO); private foundations including Child Health Foundation, Ford Foundation, Population Council, Rockefeller Foundation and the Sasakawa Foundation; and private organizations including America Express Bank, Bayer AG, CARE, Family Health International, Hellen Keller International, the Johns Hopkins University, Procter Gamble, RAND Corporation, SANDOZ, Swiss Red Cross, The University of California Davis, and others. We are grateful to Dr Michael A Koenig, who organized the original data set, and to professor Per Kragh Andersen, University of Copenhagen, who provided statistical advice.


    References
 Top
 Abstract
 Subject and Methods
 Results
 Discussion
 Conclusion
 References
 
1 Aaby P, Clements J, Orinda V. Mortality from Measles: Measuring the Impact. Geneva: Expanded Programme on Immunization, World Health Organization, 1991.

2 Dave KH. Measles in India. Rev Infect Dis 1983;5:406–10.[ISI][Medline]

3 Koster FT, Curlin GC, Aziz KMA, Haque A. Synergistic impact of measles and diarrhoea on nutrition and mortality in Bangladesh. Bull World Health Organ 1981;59:901–08.[ISI][Medline]

4 Bhuiya A, Wojtyniak B, D’Souza S, Nahar L, Shaikh K. Measles case fatality among under-fives: a multivariate analysis of risk factors in a rural area of Bangladesh. Soc Sci Med 1987;24:439–43.[CrossRef][ISI][Medline]

5 Fauveau V, Chakraborty J, Sarder AM, Khan MA, Koenig MA. Measles under-9-month-old in rural Bangladesh: its significance for age at immunization. Bull World Health Organ 1991;69:67–72.[ISI][Medline]

6 Clemens JD, Stanton BF, Chakraborty J et al. Measles vaccination and childhood mortality in rural Bangladesh. Am J Epidemiol 1988;128: 1330–39.[Abstract]

7 Koenig MA, Khan MA, Wojtyniak B et al. The impact of measles vaccination upon childhood mortality in Matlab, Bangladesh. Bull World Health Organ 1990;68:441–47.[ISI][Medline]

8 Aaby P, Bukh J, Lisse IM, Smits AJ. Measles vaccination and reduction in child mortality: a community study from Guinea-Bissau. J Infect 1984;8:13–21.[ISI][Medline]

9 Aaby P, Pedersen IR, Knudsen K et al. Child mortality related to seroconversion or lack of seroconversion after measles vaccination. Pediatr Infect Dis J 1989;8:197–200.[ISI][Medline]

10 Aaby P, Knudsen K, Jensen TG et al. Measles incidence, vaccine efficacy and mortality in two urban African areas with high vaccination coverage. J Infect Dis 1990;162:1043–48.[ISI][Medline]

11 Garenne M, Cantrelle P. Rougeole et mortalité au Senegal: ensi de l’impact de la vaccination effectué a Khombole 1965–1968 sur la survie des enfants. In: P Cantrelle, S Dormont, P Farques, J Goujard, J Guignard, C Rumeau-Rouquette (eds). Estimation de la Mortalité du Jeune Enfant (0–5 ans) pour Guider les Actions de Santé dans les Pays en Developpement. Paris: INSERM, 1986, pp. 515–32.

12 Aaby P, Samb B, Simondon F et al. Divergent mortality for male and female recipients of low-titre and high-titre measles vaccines in rural Senegal. Am J Epidemiol 1993;138:746–55.[Abstract]

13 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.[Abstract]

14 The Kasongo Project Team. Influence of measles vaccination on survival pattern of 7–35-month-old children in Kasongo, Zaire. Lancet 1981;i:764–67.

15 Hartfield J, Morley D. Efficacy of measles vaccine. Journal of Hygiene (Cambridge) 1963;61:143–47.[ISI][Medline]

16 Chen RT, Weierbach R, Bisoffi Z et al. A ‘Post-honeymoon period’ measles outbreak in Muyinga Sector, Burundi. Int J Epidemiol 1994; 23:185–93.[Abstract]

17 Holt EA, Boulos R, Halsey NA, Boulos IM, Boulos C. Childhood survival in Haiti: Protective effect of measles vaccination. Pediatrics 1990;85:188–94.[Abstract]

18 Aaby P, Samb B, Simondon F, Coll Seck AM, Knudsen K, Whittle H. Non-specific beneficial effect of measles immunization: analysis of mortality studies from developing countries. BMJ 1995;311:481–85.[Abstract/Free Full Text]

19 Hull HF, Williams PJ, Oldfield F. Measles mortality and vaccine efficacy in rural West Africa. Lancet 1983;i:972–75.

20 Aaby P, Lisse I, Mølbak K, Knudsen K, Whittle H. No persistent T lymphocyte immunosuppression or increased mortality after measles infection: a community study from Guinea-Bissau. Pediatr Inf Dis J 1996;15:39–44.[CrossRef][ISI][Medline]

21 Aaby P, Samb B, Andersen M, Simondon F. No long-term excess mortality after measles infection: A community study from Senegal. Am J Epidemiol 1996;143:1035–41.[Abstract]

22 Dollimore N, Cutts F, Binka FN, Ross DA, Morris SS, Smith PG. Measles incidence, case fatality, and delayed mortality in children with and with vitamin A supplementation in rural Ghana. Am J Epidemiol 1997;146:646–54.[Abstract]

23 Demographic surveillance system—Matlab. Registration of Demographic Events. Scientific Reports no. 64, 67, 68, 69, 70, 71. Dhaka: International Centre for Diarrhoeal Disease Research, Bangladesh, 1985–1993.

24 Holt JD, Prentice RL. Survival analyses in twin studies and matched pair experiments. Biometrika 1974;61:17–30.[ISI]

25 Cox DR. Regression models and life tables (with discussion). J R Stat Soc B 1972;34:187–220.[ISI]

26 de Francisco A, Fauveau V, Sarder AM, Chowdhury HR, Chakraborty J, Yunus MD. Measles in rural Bangladesh: Issues of validation and age distribution. Int J Epidemiol 1994;23:393–99.[Abstract]

27 Kristensen I, Aaby P, Jensen H. Routine vaccinations and child survival: follow up study in Guinea-Bissau. BMJ 2000;321:1435–38.[Abstract/Free Full Text]

28 Aaby P, Bukh J, Lisse IM, Smits AJ. Measles mortality, state of nutrition, and family structure: A community study from Guinea-Bissau. J Infect Dis 1983;147:693–701.[ISI][Medline]

29 Schnorr JJ, Cutts FT, Wheeler JG et al. Immune modulation after measles vaccination of 6–9 months old Bangladeshi infants. Vaccine 2001;19:1503–10.[CrossRef][ISI][Medline]

30 Petralli JK, Merigan TC, Wilbur JR. Action of endogenous interferon against vaccinia infection in children. Lancet 1965;ii:401–05.[CrossRef]

31 Pabst HF, Spady DW, Carson MM, Stelfox HT, Beeler JA, Krezolik MP. Kinetics of immunological responses after primary MMR vaccination. Vaccine 1997;15:10–14.[CrossRef][ISI][Medline]

32 Ito M, Watanabe M, Karniya H, Sakurai M. Changes in intracellular levels in lymphocytes induced by measles virus. Clin Immunol Immunopathol 1997;83:281–86.[CrossRef][ISI][Medline]

33 Fischer JE, Johnson JE, Johnson TR, Graham BS. Pertussis toxin sensitisation alters the pathogenesis of subsequent respiratory syncytial virus infection. J Inf Dis 2000;182:1029–38.[CrossRef][ISI][Medline]

34 Graham BS, Henderson GS, Tang YW, Lu X, Neuzil KM, Colley DC. Priming immunization determines T helper cytokine mRNA expression patterns in lungs of mice challenged with respiratory syncytial virus. J Immunol 1993;151:2032–40.[Abstract/Free Full Text]

35 Moran TM, Park H, Fernandez-Sesma A, Schulman JL. Th2 responses to inactivated influenza vaccine can be converted to Th1 responses and facilitate recovery from heterosubtypic virus infection. J Inf Dis 1999;180:579–85.[CrossRef][ISI][Medline]

36 Hall A, Aaby P. Tropical trials and tribulations. Int J Epidemiol 1990; 19:777–81.[ISI][Medline]

37 Aaby P, Andersen M, Sodemann M, Jakobsen M, Gomes J, Fernandes M. Reduced childhood mortality following standard measles vaccination at 4–8 months compared to 9–11 months of age. BMJ 1993; 307:1308–11.[ISI][Medline]