ESTIMATING NONPARTICIPATION BIAS IN A LONGITUDINAL-STUDY OF BEREAVEMENT

Nonparticipants in epidemiological studies may differ in important res pects from participants but the magnitude of this potential bias is ra rely quantified. This study estimates the effect of nonparticipation o n estimates of mental health problems following stillbirth, neonatal d eath or sudden infant death syndrome. Of 805 families approached, 512 (64 per cent) were recruited, of whom 77 per cent of mothers and 71 pe r cent of fathers completed four study interviews. Younger, unmarried, unemployed parents without private health insurance were less often r ecruited, and even if recruited, were less likely to complete the inte rview. By evaluating several possible scenarios, we estimated that had mothers lost to follow-up remained in the study, anxiety rates would have varied by no more than +/-4 per cent. Relative risks associated w ith bereaved-control comparisons would have differed little from the o bserved estimate of 2.33. Estimating the effects of initial nonrespons e is more difficult but the adoption of a worst-case scenario produced a relative risk of 3.47. Despite systematic nonparticipation suggesti ve of social disadvantage, attrition-related bias may have had only a modest effect on anxiety and depression rate estimates. However this m ay not be the case when sample loss is high, when associations between attrition and outcome are strong, and when attrition-related behaviou r differs across comparison groups.