Discriminating Powers of Partial Agreements of Names for Linking Personal Records

Challenges in administrative data linkage for research

Linkage of population-based administrative data is a valuable tool for combining detailed individual-level information from different sources for research. While not a substitute for classical studies based on primary data collection, analyses of linked administrative data can answer questions that require large sample sizes or detailed data on hard-to-reach populations, and generate evidence with a high level of external validity and applicability for policy making. There are unique challenges in the appropriate research use of linked administrative data, for example with respect to bias from linkage errors where records cannot be linked or are linked together incorrectly. For confidentiality and other reasons, the separation of data linkage processes and analysis of linked data is generally regarded as best practice. However, the 'black box' of data linkage can make it difficult for researchers to judge the reliability of the resulting linked data for their required purposes. This article aims to provide an overview of challenges in linking administrative data for research. We aim to increase understanding of the implications of (i) the data linkage environment and privacy preservation; (ii) the linkage process itself (including data preparation, and deterministic and probabilistic linkage methods) and (iii) linkage quality and potential bias in linked data. We draw on examples from a number of countries to illustrate a range of approaches for data linkage in different contexts.

Record linkage in Scotland and its applications to health research

AIMS AND OBJECTIVES

  This paper will focus on the key concepts behind record linkage and describe how probability matching of Scottish health records can be used for national health research.

BACKGROUND

  Record linkage can bring together two or more records relating to the same individual. This allows information from multiple sources to be joined together to produce richer data sets for research purposes and has wide applicability in public health and epidemiological research. The probability matching techniques underpinning record linkage bring together records on a patient basis using key identifying information on each record. Scotland has a strong track record for performing linkage for research purposes owing to routinely collected and well-maintained national administrative health data sets, the emergence of the Scottish record linkage system and organisations like the Information Services Division of NHS National Services Scotland who centrally hold permanently linked patient-based databases. Design.  A record linkage retrospective population cohort study is described within this paper.

METHODS

  The paper will describe current linkage methodology before discussing typical applications in the setting of Information Services Division and focusing on a particular linkage study investigating rates and risk factors for gastroschisis.

RESULTS

  Conclusions from the gastroschisis study are typical of the types of important findings drawn from analysing linked health data.

CONCLUSIONS

  Scotland's good track record for linking records for health research is evidenced by the high volume of research projects, publications and findings resulting from probability matching of national health data. Relevance to clinical practice.  Record linkage allows information relating to the same person held across different data sources to be brought together. Probabilistic record linkage can overcome data quality issues, producing accurate matches. This allows linked, analysable, patient-based databases, capable of answering complex research questions, to be produced from several data sources with wide applications in the field of health research.

Results from simulated data sets: probabilistic record linkage outperforms deterministic record linkage

OBJECTIVE

To gain insight into the performance of deterministic record linkage (DRL) vs. probabilistic record linkage (PRL) strategies under different conditions by varying the frequency of registration errors and the amount of discriminating power.

STUDY DESIGN AND SETTING

A simulation study in which data characteristics were varied to create a range of realistic linkage scenarios. For each scenario, we compared the number of misclassifications (number of false nonlinks and false links) made by the different linking strategies: deterministic full, deterministic N-1, and probabilistic.

RESULTS

The full deterministic strategy produced the lowest number of false positive links but at the expense of missing considerable numbers of matches dependent on the error rate of the linking variables. The probabilistic strategy outperformed the deterministic strategy (full or N-1) across all scenarios. A deterministic strategy can match the performance of a probabilistic approach providing that the decision about which disagreements should be tolerated is made correctly. This requires a priori knowledge about the quality of all linking variables, whereas this information is inherently generated by a probabilistic strategy.

CONCLUSION

PRL is more flexible and provides data about the quality of the linkage process that in turn can minimize the degree of linking errors, given the data provided.

Enhancing the power of record linkage involving low quality personal identifiers: use of the best link principle and cause of death prior likelihoods

The Heartstart Scotland study collects details of all resuscitation attempts carried out by the Scottish Ambulance Service. The linkage between records for Heartstart study subjects who died before admission to a hospital and the national file of death records maintained by the Registrar General for Scotland is described. The conditions under which the Heartstart data is collected make it inevitable that the personal identifying information on which linkage must rely tends to be relatively incomplete and of low accuracy. The linkage process was able to use the best-link principle to take maximum advantage of the fact that, because the Heartstart subjects involved had died, there was an extremely high a priori probability that they would be represented on the national deaths file. In addition, although no cause of death information was recorded on the Heartstart records, a priori expectations of the distribution of causes of death among linked death records were used. Despite these enhancements, however, clerical resolution of a proportion of the potential links generated by the automatic algorithm significantly improved the accuracy of the linkage.

Evaluating the quality of anonymous record linkage using deterministic procedures with the New York State AIDS registry and a hospital discharge file

Linkage of same-person records across multiple databases relies on high-quality, uniformly available identifying information. These data quality issues become increasingly important when personal names are not available for record linkage. Using deterministic decision criteria, we linked records from two population-based files in the absence of personal names. The sensitivity of anonymous record linkage procedures ranged from 32 to 85 per cent for the two years studied, and the positive predictive value (PPV) ranged from 14 to 99 per cent. Decreasing sensitivity and PPV were primarily attributed to (1) errors in computerized identifying information and (2) the deterministic decision criteria specified for record linkage. An evaluation of the contribution of personal names to the quality of record linkage found no measurable impact.

Pharmaco-morbidity linkage: a feasibility study comparing morbidity in two pharmacy based exposure cohorts

STUDY OBJECTIVES

The aims were (1) to compare discharge diagnoses and concurrent medication in a pharmacy based cohort of users of H2 receptor antagonists to those in a population of users of other drugs in the same period, who did not use H2 receptor antagonists; (2) to compare these results to those of a similar study performed with the Tayside record linkage scheme.

DESIGN AND SETTING

The study was a retrospective cohort study. The morbidity data from the only hospital in one medium sized city (62,000 inhabitants) were linked to the dispensing data of all five community pharmacies on an individual basis (April 1, 1986-December 31, 1989). In the absence of a unique patient identification number, data from pharmacies and hospital were linked by the combination of date of birth, gender, and general practitioner code. For every user of H2 receptor antagonists two controls were obtained from all patients who had not used these drugs, and matched for age (within 5 years), gender, and general practitioner. All discharge diagnoses which followed this first prescription up to December 31, 1989, in a patient in the index cohort, and during the same period in his or her matched controls, were included in the study.

MAIN RESULTS

In the index cohort (n = 2174) 341 persons were admitted (526 admissions) as against 398 persons (527 admissions) in the control cohort (n = 4348). There was increased morbidity in the index cohort, especially concerning the gastrointestinal system (peptic ulcers and malignancies, abdominal pain, gastrointestinal haemorrhage), but also concerning the musculoskeletal, respiratory, and circulatory systems. The morbidity in the last three groups corresponded with drugs used concomitantly by patients in the index cohort, so it was probably not causally related to the intake of H2 receptor antagonists but was rather an indicator of higher levels of morbidity in the index cohort.

CONCLUSIONS

The figures were grossly comparable to those of the Tayside record linkage scheme. Probabilistic linking with the patient characteristics of gender, date of birth, and general practitioner code can facilitate the undertaking of postmarketing surveillance studies.

Application of exact ODDS for partial agreements of names in record linkage

Automated methods for linking records pertaining to the same individuals have in the past made only crude use of the name information. A perceptive filing clerk is more sophisticated because humans retain a lifetime memory of instances in which variant forms of names were employed interchangeably, and of synonyms that sometimes did not even resemble one another. This limitation of the machine can be rectified, but the body of knowledge required to serve as its memory must be large. The needed data have now been brought together on a suitable scale, from many past searches of Canada's Mortality Data Base. Described here are the development and use of the resulting tables of essentially exact discriminating powers (or ODDS) to do with comparisons of male given names. The aim is to reduce the proportion of ambiguously linked pairs of records requiring labor intensive clerical resolution. The tables are intended for general use in this country, and as a model for similar facilities appropriate to other populations.