Survival bias in observational studies of the impact of routine immunizations on childhood survival.

OBJECTIVES: In situations with vaccination coverage high enough to control pertussis, observational studies have reported divergent effects of diphtheria-tetanus-pertussis (DTP) vaccination on childhood survival. We examined whether this could be because of methodological differences. METHODS: Some studies of the impact of DTP updated information on vaccination retrospectively (retrospective updating approach) while others kept vaccination status fixed for the time between follow-up visits (landmark approach). First, we conducted simulations with these approaches to investigate the impact of different mortality levels, vaccination incidence rates, intervals between data collection visits, and the proportion of children whose vaccination card had not been seen after death. Second, we re-analysed data from Guinea-Bissau using the retrospective updating approach. RESULTS: In simulations, the retrospective updating approach produced rate ratio (vaccinated/unvaccinated) estimates biased towards zero because of a differential misclassification which created survival bias as risk-free observation time was allocated to the vaccinated group. The landmark approach produced rate ratio estimates biased towards one. Biases increased with interval between data collection visits and incidence of vaccination, but were unaffected by the underlying mortality level. Survival bias increased with the proportion of dead children whose vaccination status could not be updated. The re-analysis of data from Guinea-Bissau changed the estimated impact of DTP from an 84% higher mortality using the landmark approach to a 37% lower mortality using the retrospective updating approach. CONCLUSIONS: The apparent contradiction between DTP studies could be because of methodological differences. To assess child survival associated with routine vaccinations, studies minimizing the effect of biases are warranted.