Selection bias: neighbourhood controls and controls selected from those presenting to a Health Unit in a case control study of efficacy of BCG revaccination

Odimariles Ms Dantas,Patrícia Varejão,Ricardo Aa Ximenes,Wayner V De Souza,Ulisses R Montarroyos,Maria De Fatima Pm De Albuquerque,Laura C Rodrigues

doi:10.1186/1471-2288-7-11

Abstract

BackgroundIn most case control studies the hardest decision is the choice of the control group, as in the ideal control group the proportion exposed is the same as in the population that produced the cases.MethodsA comparison of two control groups in a case control study of the efficacy of BCG revaccination. One group was selected from subjects presenting to the heath unit the case attended for routine prevention and care; the second group was selected from the neighbourhood of cases. All Health Units from which controls were selected offered BCG revaccination. Efficacy estimated in a randomized control trial of BCG revaccination was used to establish that the neighbourhood control group was the one that gave unbiased results.ResultsThe proportion of controls with scars indicating BCG revaccination was higher among the control group selected from Health Unit attenders than among neighbourhood controls. This excess was not removed after control for social variables and history of exposure to tuberculosis, and appears to have resulted from the fact that people attending the Health Unit were more likely to have been revaccinated than neighbourhood controls, although we can not exclude an effect of other unmeasured variables.ConclusionIn this study, controls selected from people presenting to a Health Unit overrepresented exposure to BCG revaccination. Had the results from the HU attenders control group been accepted this would have resulted in overestimation of vaccine efficacy. When the exposure of interest is offered in a health facility, selection of controls from attenders at the facility may result in over representation of exposure in controls and selection bias.

Highlights

In most case control studies the hardest decision is the choice of the control group, as in the ideal control group the proportion exposed is the same as in the population that produced the cases
Because hospital patients are in hospital, they are likely to have a higher frequency of hazardous exposures than the population in general; if hospital controls are used, it is necessary to exclude from the control groups people hospitalized for diseases caused by the exposure of interest
The objective of this paper is to examine the degree of bias for estimated vaccine efficacy (VE) using two control groups: neighbourhood controls and Health Unit attenders controls; and explore the extent to which this bias was caused by differences in the population, and how much caused by the fact that the HU control was a user of the HU

Summary

Introduction

In most case control studies the hardest decision is the choice of the control group, as in the ideal control group the proportion exposed is the same as in the population that produced the cases. Population controls are increasingly being used in case control studies because of the growing awareness of the limitations of hospital controls, first identified by Berkson (1946) [1]. Population controls, in contrast, are more likely to represent accurately the exposure state in the population that produced the cases. Neighbourhood controls are an alternative to population controls: they represent the exposure in the neighbourhood that produced each case, and tend to control for known and unknown confounding factors that clusters in neighbourhoods[2]. The logistics of data collection for population controls is often more difficult

Objectives

Methods

Results

Discussion

Conclusion