Missing hospital records: a confounding variable in retrospective studies.

In retrospective chart reviews there are often a certain number of missing hospital records. To elucidate this variable we compared the outcomes of very low birth weight breech infants with respect to the method of collecting data. A prospective sampling, during the hospital stay, of data was performed in 1979 to 1980 and in 1983 to 1984, and the frequencies of very low birth weight were 1.89 and 1.90, respectively, per 1000 live births. For 1981 to 1982 a retrospective record search was performed with the use of the ordinary medical record search system at this institution. For this period 39 of 52 (75%) hospital records were recovered, giving an apparent frequency of 1.32/1000 live births, which differed significantly from either period studied prospectively. An analysis of demographic data of the three groups revealed that the mean gestational age and the mean birth weights were higher in the period studied retrospectively compared with both periods studied prospectively and that the neonatal mortality rates were higher in the periods studied prospectively (74.1% and 57.1%, respectively) than in the period studied retrospectively (28.2%). It is concluded that the more complicated a clinical case is, the more likely the record will not be found for retrospective chart review. This problem should be kept in mind, and it ought to be a requirement that the number of missing charts be stated in retrospective observational studies.