Invited commentary: using vital statistics databases for perinatal epidemiology: does the quality go in before the name goes on?

Most epidemiologic research using birth certificate records utilizes cross-sectional or case-control study designs. These records are typically made available to qualified researchers as computerized files of vital events for a state or the entire United States by year of occurrence, either for events occurring within the state or to residents of the state. This structure clearly lends itself to the cross-sectional approach. Occasionally, researchers step outside the box and conduct studies utilizing prospective study designs, generally with one of two approaches: 1) intergenerational studies based on record linkage between the birth certificate for the index case and the birth certificate for the mother of the index case (1–3) or 2) maternally linked pregnancy outcomes based on record linkage of the birth certificate for a firstborn infant of the index mother and her subsequent birth outcomes (4). The intergen-erational approach is somewhat limited due to the smaller number of statistical data fields collected on birth certificates in the past (5). Datasets with maternally linked pregnancy outcomes can be used to study birth outcomes within sibships, especially for conditions in which a prior adverse outcome (for example, low birth weight or fetal death) may be associated with an adverse outcome in a subsequent pregnancy (6). Over the past half-century, the US standard certificate of livebirth has gradually expanded to include numerous questions concerning reproductive history, characteristics of the infant at birth, weight gain, use of tobacco and alcohol during pregnancy, and a variety of clinical data elements ranging from risk factors associated with complications of pregnancy to intrapartum events to birth defects and neonatal therapies (5, 7, 8). This expansion of the range of data elements collected through vital registration has, unfortunately, not been accompanied by careful evaluation of the quality of the data reported on birth certificate documents. Most published birth certificate data quality studies have limitations. First, many studies are secondary methods analyses conducted by using available data collected for another research project (9–11). The generalizability of the results of some of these analyses is limited, and it is difficult to extrapolate study results to a comprehensive statewide birth certificate data quality improvement initiative. Second, many studies are too small to yield valid and reliable results to guide epidemiologists interested in using these data; either the samples are too small (12), or the study is based on data from one or, at most, a few hospitals (13, 14). Third, the …