Background: It should no longer be necessary to demonstrate the importance of knowing the completeness of ascertainment for a morbidity register, particularly with respect to the interpretation of prevalence rates and their trends, but also when using register data for etiological studies.
Method: The study covered 9 generations of children born between 1980 and 1988. All of these children lived in the Isère department in SouthEast France, and each of them had at least one major deficiency, according to the inclusion criteria laid down by the RHEOP ("Registre des Handicaps de l'Enfant et Observatoire Périnatal", in French, or Childhood Handicap Register and Perinatal Observatory). These children were recruited from four different data sources. The completeness of ascertainment of the register was estimated first by means of the capturerecapture method, based on two sources that were shown to be independent by the Wittes method. Following this, loglinear models were used. The advantage of this was the absence of restrictions involved in adhering to the necessary validity conditions before applying the capture-recapture method, and the possibility of introducing heterogeneity variables, such as the number of deficiencies per child, for example.
Results: The applied capturerecapture method, with two main sources that have been found to be independent, gives an overall completeness of ascertainment of 86% CI(95%)[8291], with a variation of between 76% CI(95%)[6787] and 97% CI(95%)[93100] when the number of deficiencies per child is taken into account. After application of the loglinear models, the results obtained are very close to those obtained with the capture-recapture method, both in the case of estimation of the overall completeness of ascertainment and in the case of the completeness of ascertainment that is estimated according to the number of deficiencies variable. The similarity of the results obtained by the two methods appears to support our empirical study, but is only possible because of the validity of certain conditions (the interactions of the order of three were not significant) which can only be verified using statistical tests in the linear log models.
Conclusion: If the application conditions of the capture-recapture method are carefully adhered to, it becomes possible, without the help of software, to produce a correct estimate of the number of missing cases. Nevertheless, it would be unreasonable to continue using this method alone since log linear models have been found to be independent of these validity conditions.