Abstract
Record linkage techniques can help identify the same patient for matching diverse files (hospital discharge abstracts, insurance claims, registries, Vital Statistics data) which contain similar identifiers. Prior knowledge of whether a linkage is feasible is important to prevent wasted effort (additional data collection or data manipulation), which decreases the cost-effectiveness of the linkage. Using examples generated by linking the Manitoba Health Services Commission data with Vital Statistics files, a method of estimating the information in each data set is presented first. Further, the feasibility of several different record linkage strategies is described and tested, given varying amounts of information. At the margin, relatively small amounts of information (having just one more variable to match with) can make a great difference. Probabilistic linkage’s great advantage was found in those situations where only a moderate amount of extra information was available.
By using the above techniques when working with one or both files in a proposed record linkage project, a much more informed judgement can now be made as to whether a linkage will or will not work. In facilitating record linkage, flexibility of both software and the strategy for matching is very important.
Key-Words
Record Linkage - Probabilistic Linkage - Deterministic Linkage - Administrative Data - Information Theory
