DTP in low income countries: improved child survival or survival bias?BMJ 2005; 330 doi: https://doi.org/10.1136/bmj.330.7495.845-a (Published 07 April 2005) Cite this as: BMJ 2005;330:845
- Henrik Jensen, statistician (, )
- Christine S Benn, senior researcher,
- Peter Aaby, professor
- Bandim Health Project, Danish Epidemiology Science Centre, Statens Serum Institut, DK-2300 Copenhagen S, Denmark
EDITOR—We proposed that BCG and measles vaccine have non-specific beneficial effects, whereas diphtheria-tetanus-pertussis (DTP) vaccination might have no beneficial effect.1 In response, the World Health Organization commissioned several studies.2–4 A WHO expert task force found substantial evidence against a deleterious effect of DTP; all studies showed DTP to be associated with reduced mortality. A negative DTP effect was found only in Guinea-Bissau, and this was presumed to be due to a country specific issue or a peculiarity of the data.5
In our survival analysis, vaccination status was a time fixed variable, held constant from the initial visit to the next; without perfect information for all children, vaccinations during follow-up could not be accounted for, which means a potential source of bias.1 In the WHO sponsored analyses, vaccination status was a time varying variable changing status at the date of vaccination, based on information achieved at a subsequent visit.2–5 We used this approach to re-analyse our data (table).
The distribution of deaths was similar. When we used time varying variables, person years decreased for the unvaccinated and BCG groups, and mortality went up. Person years increased for the DTP groups and mortality decreased. Hence, DTP was associated with reductions in mortality (table), similar to results from WHO sponsored studies.2–4
Why this difference? Information on vaccinations is typically collected through periodic home visits. When a child dies, the vaccination card is usually thrown away; and information on vaccination is therefore collected conditionally on survival to the subsequent visit. If an unvaccinated child was vaccinated and died before the next visit the death would be classified as unvaccinated, in an analysis using time varying variables. If a vaccinated child survived then the follow up time as vaccinated would be moved to the new vaccination. This survival time is risk free—that is, we only know that the child was vaccinated because it survived. Such survival bias may turn a negative estimate into a positive one: our original 84% increase in mortality for one dose of DTP became a 32% reduction (table).
Survival bias can be avoided only if all vaccinations are provided by the researchers, or perfect vaccination information is obtained from all children. Nothing indicates that these conditions were met in the WHO commissioned studies.2–4 In contrast to our study, none of the WHO studies documented which children were unvaccinated; they do not distinguish between “unvaccinated” and “no information.” The contradiction between our study and the WHO sponsored studies might be due to methodological differences and not peculiarity of the data.
Details of the three other authors are on bmj.com