[Record linkage with cryptographic identification data in a population-based cancer registry. Development, implementation and error rates]

Gesundheitswesen. 2001 Jun;63(6):376-82. doi: 10.1055/s-2001-15686.
[Article in German]

Abstract

Background: Population-based cancer registries depend on the completeness of their case notification. At present, restrictive legal regulations in Germany requesting written informed consent of cancer patients hinder some health professionals, for example pathologists, from reporting to cancer registries. New cryptographic methods may be used to obtain record linkage based on anonymised data.

Objectives: To assess the feasibility of a record linkage in the Münster Cancer Registry (EKR-MS) using cryptographic methods as compared to traditional methods based on personal identifiers.

Methods: We generated so called control numbers--a sequence of 23 alphanumeric signs--for 19 name related characteristics employing consecutively two different chiffrations. Record linkage was carried out using a semi-automatic computer program (AUTOMATCH) that generates probabilities of identity for pairs of case notifiation based on the information of control numbers only. Probabilities exceeding present limits lead to automatic decisions whereas the remaining linkage has to be decided manually. Plausibility-based rules helped to considerably reduce the amount of manual decisions.

Results: We compared traditional and cryptographic record linkage for all new reports received in the EKR-MS during 1998 (n = 27,262) against the background of n = 101,880 known cases in the registry data base. Setting traditional, text-based record linkage results to be the reference the cryptographic method resulted in a synonomous error (false creation of a new case) of almost 2% and a homonymous error (false link to a known case) of less than 0.5%.

Conclusion: Cryptographic methods may be feasible procedures of record linkage in cancer registries. The size of the database of the EKR-MS prohibits extrapolation of findings to smaller registries. The error rates resulting in slight overestimation of disease rates that may seem acceptable. It should be noted though that cryptographic data are in many situations prohibitive for further epidemiologic research limiting the usefulness of the method exclusively to cancer registration under the given legal coercion.

MeSH terms

  • Bias
  • Computer Security / legislation & jurisprudence*
  • Databases, Factual / statistics & numerical data
  • Germany
  • Humans
  • Medical Record Linkage*
  • Neoplasms / epidemiology*
  • Population Surveillance*
  • Registries / statistics & numerical data*