Lost to follow-up in the Norwegian mother, father and child cohort study

Abstract

Background: The aim of pregnancy cohorts was to understand causes and development of health and disease throughout the life course. A major challenge in cohort studies is to avoid selection bias from loss to follow-up. Objective: The aim of this study was to describe what characterises drop out from the Norwegian Mother, Father and Child Cohort Study (MoBa), and provide a resource to inform the interpretation of results from analysis of cohort data. Methods: We estimated loss to follow-up in subsets of participants that responded to questionnaire waves in MoBa through an eight-year period and described characteristics of participants who responded to follow-ups. Within each wave of questionnaires, we estimated two exposure-outcome associations: the relationship between maternal smoking during pregnancy and offspring birthweight, and between educational level and pre-pregnancy body mass index (BMI). We explored the use of inverse probability weighting to correct the bias due to loss to follow-up. Results: Participants who continued to respond were older, higher educated, less likely to smoke and had lower BMI. We observed a decline in participation of current smokers from 22.3% to 17.5%, and participants who reported an unplanned pregnancy dropped from 19.2% to 16.4%. There was a gradual decline in the inverse relationship between maternal smoking during pregnancy and offspring birthweight with increasing follow-up information, indicating that selection bias due to drop out resulted in lower effect estimates. For the relationship between parental educational level and BMI, the inverse association increased with amount of follow-up information, indicating that the selection bias resulted in higher effect estimates. Inverse probability weighting did not completely correct the estimates for bias due to loss to follow-up. Conclusions: Participants who remain cohort members are different from subjects who drop out. Users of large cohorts should be aware of selective loss to follow-up and consider imputation or weighting to account for loss to follow-up when analysing questionnaire responses.