Comment on the paper: “Child survival and BCG vaccination: a community based prospective cohort study in Uganda”.
Ane Fisker, Bandim Health Project
13 July 2015
We have with interest read the paper by Nankabirwa et al[1] and would like to commend the authors for having made the great effort to collect information about vaccination status at several home visits between 3 weeks and 7 years of age. At the Bandim Health Project (www.bandim.org), we are accessing the real-life effects of vaccines, including BCG, on child health. In concordance with our studies, Nankabirwa et al. present data supporting that BCG vaccination reduces all-cause mortality in early life.
As pointed out by the authors, good data on early life mortality in low-income countries is scarce, and birth cohorts like the one followed in the PROMISE-EBF study, provides important information to fill the gap. In such birth cohorts, we urge that the opportunity to collect vaccination information is used. However, assessing the effects of vaccines on child survival using vaccination data collected at regular interview rounds poses some challenges[2] which any researcher making similar efforts should take into account.
In the present paper, the authors have in their primary analysis classified children as BCG vaccinated from their date of vaccination. Hence, a child contributes risk time prior to date of interview where it was established, that a child was vaccinated. This does not pose a problem if all vaccinations in the studied cohort are known, but when they are not, it may introduce a bias. Imagine a situation where a child had its vaccination information collected at 6 weeks of age and at that time the child was un-vaccinated. At the following interview at 12 weeks of age it is established, that the child had died when it was 8 weeks – but was it BCG vaccinated? The authors report that “Vaccination data in this study was collected prospectively and validated with vaccination cards before almost all the deaths happened”. However, to ensure that the methodology applied does not bias the results, the vaccination status should be assessed after all deaths and for survivors at the end of follow-up.
In situations where the vaccination information is not available up to the end of follow-up, landmark analyses should be applied[3]. In a landmark analysis, children are followed prospectively from the day the vaccination status is assessed, and remain in the same vaccination-status group until the next visit where the vaccination status is updated. Children are only allowed to shift from one vaccination group to another at the time the information is obtained, because information on subsequent vaccinations are rarely available for children who moved or died. This implies, that some of the unvaccinated children are vaccinated during follow up, but retained in the analysis as unvaccinated, hence moving the estimate towards no effect of a vaccine.
To be able to conduct landmark analyses, the date on which the vaccination information is obtained must be registered[2]. It is furthermore important that “no vaccine” and “no information” can be distinguished: It will rarely be possible to obtain information for all children if the information is collected at interview rounds, and therefore there will be children and/or observation periods excluded from the analyses. If children with no information on vaccination status are grouped together with unvaccinated children, the “unvaccinated group” will have an artificially high mortality, since children who die are more likely not to have their vaccination status assessed. In the present paper, the 18 neonatal deaths all belong to the unvaccinated group. This period is not included in the BCG analyses and therefore has no impact on the estimates, however if a child was classified as unvaccinated until a vaccination record was seen it would have an effect.
Hence, though a landmark analysis does not use all the information available and provides conservative estimates, we would recommend and place more confidence in those results.
Ane Bærent Fisker and Sanne Marie Thysen
Bandim Health Project, Guinea-Bissau and Research Center for Vitamins and Vaccines (CVIVA), Bandim Health Project, Statens Serum Institut, Denmark
References
1. Nankabirwa V, Tumwine JK, Mugaba PM, Tylleskär T, Sommerfelt H: Child survival and BCG vaccination: a community based prospective cohort study in Uganda. BMC Public Health 2015, 2015(15):175.
3. Jensen H, Benn CS, Lisse IM, Rodrigues A, Andersen PK, Aaby P: Survival bias in observational studies of the impact of routine immunizations on childhood survival. Trop Med Int Health 2007, 12(1):5-14.
Competing interests
We work at the Bandim Health Project, in the research field of non-specific effects. In Denmark Bandim Health Project is based at the Statens Serum Institut (SSI) which is a vaccine producer- However, neither Bandim Health Project nor any of the research conducted by BHP employees is funded by the SSI, but rely on external grants.
Comment on the paper: “Child survival and BCG vaccination: a community based prospective cohort study in Uganda”.
13 July 2015
We have with interest read the paper by Nankabirwa et al[1] and would like to commend the authors for having made the great effort to collect information about vaccination status at several home visits between 3 weeks and 7 years of age. At the Bandim Health Project (www.bandim.org), we are accessing the real-life effects of vaccines, including BCG, on child health. In concordance with our studies, Nankabirwa et al. present data supporting that BCG vaccination reduces all-cause mortality in early life.
As pointed out by the authors, good data on early life mortality in low-income countries is scarce, and birth cohorts like the one followed in the PROMISE-EBF study, provides important information to fill the gap. In such birth cohorts, we urge that the opportunity to collect vaccination information is used. However, assessing the effects of vaccines on child survival using vaccination data collected at regular interview rounds poses some challenges[2] which any researcher making similar efforts should take into account.
In the present paper, the authors have in their primary analysis classified children as BCG vaccinated from their date of vaccination. Hence, a child contributes risk time prior to date of interview where it was established, that a child was vaccinated. This does not pose a problem if all vaccinations in the studied cohort are known, but when they are not, it may introduce a bias. Imagine a situation where a child had its vaccination information collected at 6 weeks of age and at that time the child was un-vaccinated. At the following interview at 12 weeks of age it is established, that the child had died when it was 8 weeks – but was it BCG vaccinated? The authors report that “Vaccination data in this study was collected prospectively and validated with vaccination cards before almost all the deaths happened”. However, to ensure that the methodology applied does not bias the results, the vaccination status should be assessed after all deaths and for survivors at the end of follow-up.
In situations where the vaccination information is not available up to the end of follow-up, landmark analyses should be applied[3]. In a landmark analysis, children are followed prospectively from the day the vaccination status is assessed, and remain in the same vaccination-status group until the next visit where the vaccination status is updated. Children are only allowed to shift from one vaccination group to another at the time the information is obtained, because information on subsequent vaccinations are rarely available for children who moved or died. This implies, that some of the unvaccinated children are vaccinated during follow up, but retained in the analysis as unvaccinated, hence moving the estimate towards no effect of a vaccine.
To be able to conduct landmark analyses, the date on which the vaccination information is obtained must be registered[2]. It is furthermore important that “no vaccine” and “no information” can be distinguished: It will rarely be possible to obtain information for all children if the information is collected at interview rounds, and therefore there will be children and/or observation periods excluded from the analyses. If children with no information on vaccination status are grouped together with unvaccinated children, the “unvaccinated group” will have an artificially high mortality, since children who die are more likely not to have their vaccination status assessed. In the present paper, the 18 neonatal deaths all belong to the unvaccinated group. This period is not included in the BCG analyses and therefore has no impact on the estimates, however if a child was classified as unvaccinated until a vaccination record was seen it would have an effect.
Hence, though a landmark analysis does not use all the information available and provides conservative estimates, we would recommend and place more confidence in those results.
Ane Bærent Fisker and Sanne Marie Thysen
Bandim Health Project, Guinea-Bissau and Research Center for Vitamins and Vaccines (CVIVA), Bandim Health Project, Statens Serum Institut, Denmark
References
1. Nankabirwa V, Tumwine JK, Mugaba PM, Tylleskär T, Sommerfelt H: Child survival and BCG vaccination: a community based prospective cohort study in Uganda. BMC Public Health 2015, 2015(15):175.
2. Optimunise Project, Indepth network: Guide: field studies to monitor and assess the impact of vaccinations and other childhood interventions on morbidity and mortality. In. http://www.indepth-network.org/optimunise/key_activities.html; 2013.
3. Jensen H, Benn CS, Lisse IM, Rodrigues A, Andersen PK, Aaby P: Survival bias in observational studies of the impact of routine immunizations on childhood survival. Trop Med Int Health 2007, 12(1):5-14.
Competing interests
We work at the Bandim Health Project, in the research field of non-specific effects. In Denmark Bandim Health Project is based at the Statens Serum Institut (SSI) which is a vaccine producer- However, neither Bandim Health Project nor any of the research conducted by BHP employees is funded by the SSI, but rely on external grants.