Deterministic Record Linkage versus Similarity Functions: a Study in Health Databases from Brazil

The record linkage is a strategy that allows linking different databases of information from patient records. Adopting the deterministic method and similarity functions (Dice, Jaro, Jaro-Winkler and Levenshtein) for the integration of heterogeneous databases aimed at different levels of health care Brazilian (primary, secondary and tertiary). The sensitivity of deterministic method was 54.5% (95% CI: 50.4 to 58.5). The best result obtained with the dissent of only one variable (mother's name) was 80.6% (95% CI: 77.2 to 83.6) and the best result obtained using the similarity function Jaro-Winkler was 91.8% (95% CI: 89.4 to 93.9). The deterministic method has high specificity but sensitivity can be reduced by the existence of spellings and typing errors in the databases. Thus, the step-by-step approach where there was disagreement in at least one of the relationship variable can increase the sensitivity of the method and the use of similarity functions.