Erosion of representativeness in a cohort study

BackgroundThe National Child and Development Study (1958 British Birth Cohort) follows the lives of over 17 000 people born in a single week in England, Scotland, and Wales. Since the initial recruitment there have been nine sweeps to gather subsequent life-course data, and a Biomedical Sweep (Age 44) - between Sweeps 6 (Age 42) and 7 (Age 46) - that has found widespread application in genetic studies. Due to its non-selective recruitment, the survey is frequently used as a representative proxy for the British population in demographic, epidemiological, and medical studies. We examine the effect of attrition on representativeness of female fertility and education length. MethodsWe compare numbers and timings of fertility-related events of female cohort members with national estimates. Spline approximation was used to link records with different aggregation intervals. Participants present in the Biomedical Sweep (Age 44) were compared to those who were not. ResultsWe established that both timings and counts of maternities and terminations in the cohort diverge from the patterns of their contemporaries. For women who participated in the Biomedical Sweep (Age 44), we noted positive correlations of study continuation with years spent in full time education, and with age at first birth. We determined that women who did not participate in the Biomedical Sweep (Age 44) reported different fertility patterns from those who did. ConclusionsIt is possible to use National Statistics to quantify various forms of selection bias that inevitably creep into even the most meticulously sampled longitudinal study, and the misreporting that affects particular questions. While the 1958 Birth Cohort has reasonably been described as "broadly representative" of the British population, such a characterisation becomes questionable when the data are to be applied to questions about female fertility. KEY MESSAGESO_LIFertility patterns of women reported in later sweeps of the 1958 British Birth Cohort diverge from those of their contemporaries as estimated from national statistics using spline approximation. Both maternities and terminations are underreported. C_LIO_LIFemale participants who dropped out earlier follow national maternity trends more closely than those who participated in the study longer, but they still report slightly fewer maternities than national statistics suggest. C_LIO_LIFemale participants who persisted through later sweeps experienced first births later and left education at a later age than those who dropped out earlier. C_LIO_LIAlthough the 1958 British Birth Cohort has been judged as representative of the British population for some research questions, the cohort population presents increasingly biased fertility patterns in female participants over time. Studies related to fertility using data from this cohort may require adjustment. C_LI