Validation of Registries: A Neglected, but Indispensable Investment

The national registries of the Nordic countries have a high international reputation. The medical birth registries are considered a ‘goldmine for clinical research’ for registry-based epidemiological studies especially when they are combined with analyses of biological material from the population. A specific example is when data from registries form the medical background for biochemical analyses in large cohorts such as the Danish National Birth Cohort (DNBC) and Norwegian Mother and Child Cohort Study (MoBA) birth cohorts of 100 000 pregnant women each that include meticulous, prospectively collected data from questionnaires and blood samples. In this issue of Paediatric and Perinatal Epidemiology, Klungsøyr and colleagues have produced a noteworthy work – a validation of the diagnosis of preeclampsia in the Medical Birth Registry of Norway (MBRN) for use in MoBa. The validation showed that less than half of women with the diagnosis of preeclampsia based on data from antenatal records or discharge diagnosis were recorded with preeclampsia in the MBRN. However, the diagnosis was confirmed for nearly all women recorded with preeclampsia in the registry. This study shows that the MBRN can be relied upon for the identification of cases with preeclampsia for further analysis, but is less useful for monitoring the incidence of the condition. Validation of these data for use in MoBa means that blood analyses can be carried out in blood samples collected during pregnancy from women with a validated diagnosis of preeclampsia and matched to validated controls. Validation is an integral component of scientific research, but validation of registry-based data is rarely published with epidemiological studies. When we publish experimental or clinical studies that include biological material such as blood, we often describe the collection procedure, storage, and kits used for biochemical analyses. In epidemiology, these descriptions are not available or, when they are, validated variables are not always relevant for the study or relate to a different time period. Validation studies of registries are expensive to implement and not easy to publish, as editors know that the appeal of these studies remains limited. Furthermore, researchers may have a conflict of interest because they are reluctant to question the validity of their own data. Yet routine validation of data from registries is necessary to ensure the quality of epidemiological studies. Primary validation should be done at the time of reporting to the registries, by ascertaining that values fall within an acceptable range and that data inconsistencies are corrected. These explicit criteria for routine validation as well as the proportion of missing values for each variable in the register should be freely accessible. Secondary validation should be performed on an ad-hoc basis by matching of variables within the registry and by linkage of data to other registries, as well as with primary sources, usually medical records. This type of validation – such as reported in this issue of the journal – is very resource demanding and should be carried out for a specific purpose, in selected materials and for limited periods. In particular, validation studies should be carried out when new variables are introduced as this can be associated with missing and erroneous values. In general, validation studies shows that clinically significant conditions and interventions such as placenta previa and caesarean section are more valid than less significant and more common events such as cystitis in pregnancy. Also, interventions are more valid than diagnosis and a clear and clinically relevant definition improves validity. Unfortunately, definitions of clinical conditions change over time. For instance, pregnancy oedema was originally included Correspondence: Lone Krebs, Department of Obstetrics and Gynaecology, Holbaek Hospital, Smedelundsgade 60, Holbaek, DK-2700, Denmark. E-mail: lone.krebs@dadlnet.dk bs_bs_banner