Choosing the best estimate of gestational age from routinely collected population-based perinatal data

Obtaining gestational data of acceptable validity on whole populations is a considerable challenge, which must be met in order to further epidemiological investigation involving perinatal factors. As the means of estimating gestational duration multiply, routinely available population data pertaining to gestational duration may be conflicting. This exacerbates its reputation for unreliability, which is due primarily to the generally occult nature of conception and secondly to the propensity for data entry errors of the data from which gestational duration is estimated. However, the key to improving reliability may paradoxically lie in the increasing variety of methods for estimating gestational duration, because agreement between independent observations increases the reliability that can be placed on estimates of factors that are inherently difficult to measure. This paper demonstrates that the acceptability of population data for gestational duration can be improved by simple rectification of recurring data entry errors and demanding compatibility of two independent estimates of gestational duration. Compared with the previous algorithm that relied primarily on last menstrual period data, the new algorithm, that considers both antenatal and neonatal indicators of gestational duration, decreased the proportion of births in a geographically defined birth cohort 1986-99 (n=354,216) without a gestational estimate, from 0.5% to 0.03%, halved the number of births of >400 g estimated to occur before 20 weeks, and almost eliminated gestational estimates of >45 weeks.