Record Linkage Process Research Articles

Das Zentrum für Krebsregisterdaten (ZfKD) im Robert Koch-Institut (RKI) in Berlin

Cancer represents the second most common cause of death in Germany. The country's federal states operate regional population-based cancer registries that collect and analyze data on cancer patients. This provides an essential basis for describing the cancer burden in the German population. In order to obtain valid and reliable information on cancer incidence at the national level, the Robert Koch Institute (RKI) set up the Federal Cancer Surveillance Unit in 1983 as a central institution for evaluating this cancer registry data. In August 2009, when the Federal Cancer Registry Data Act (BKRG) came into force, the Center for Cancer Registry Data (ZfKD) at the RKI took over the work of the Cancer Surveillance Unit with a broader remit. In the future, it will also regularly publish findings on survival, prevalence, and tumor stage distribution. A newly established record linkage process will help identify multiple submissions from the federal states. Further innovations and new tasks of the ZfKD include expanding an interactive Internet platform and encouraging a more intensive use of cancer registry data for epidemiological research by providing datasets to external scientists. The range of information available to the interested public is also to be expanded.

Kd-trees and the real disclosure risks of large statistical databases

Estimating the disclosure risk of a Statistical Disclosure Control (SDC) protection method by means of (distance-based) record linkage techniques is a very popular approach to analyze the privacy level offered by such a method. When databases are very large, some particular record linkage techniques such as blocking or partitioning are usually applied to make this process reasonably efficient. However, in this case the record linkage process is not exact, which means that the disclosure risk of a SDC protection method may be underestimated.In this paper we propose the use of kd-trees techniques to apply exact yet very efficient record linkage when (protected) datasets are very large. We describe some experiments showing that this approach achieves better results, in terms of both accuracy and running time, than more classical approaches such as record linkage based on a sliding window.We also discuss and experiment on the use of these techniques not to link a whole protected record with its original one, but just to guess the value of some confidential attribute(s) of the record(s). This fact leads to concepts such as k-neighbor l-diversity or k-neighbor p-sensitivity, a generalization (to any SDC protection method) of l-diversity or p-sensitivity, which have been defined for SDC protection methods ensuring k-anonymity, such as microaggregation.

Methodological issues in linking study participants to Australian cancer registries using different methods: lessons from a cohort study

Objective: To compare results of concurrent linkage of records from the same occupational cohort to cancer registries at both a State and national level. Methodological issues affecting the record registry linkage process will be explored in cases of discordant results between the State and national cancer linkages.Methods: The number of incident invasive cancers in an occupational cohort of more than 11,000 workers was determined by linkage to the National Cancer Statistics Clearing House (NCSCH) in 2003. The results were then compared with linkages by the cancer registries of Victoria and Western Australia in 2004, and also with a previous NCSCH linkage in 1998.Results: Our analyses show an underascertainment of confirmed cancer cases by the NCSCH of about 13% (26/205) in Victoria and 11% (32/297) in Western Australia. In addition, 14 cases (5%) identified in a previous NCSCH record linkage in 1998 were not matched in 2003.Conclusions: These findings strongly indicate that record linkage to the State cancer registries was essential to maximise the ascertainment of cancer cases in our cohort. We attribute the discordant linkage results to differences in the quality of the record linkage process by the cancer registries.Implications: The record linkage methods of the national and State cancer registries need further standardisation. At present it is advisable to perform cancer record linkages through the State registries in addition to or instead of the NCSCH.

Estimation de la valeur discriminante des traits d’identification utilisés pour le rapprochement des données d’un patient

To reconstruct a patient's medical history, one has often to combine information from different sources, whatever the context of this reconstitution: epidemiological studies or health care. As a linkage using less informative identifiers could lead to linkage errors, it is essential to quantify the information associated to each identifier. The aim of this study was to estimate the discriminating power of different identifiers which could be used in a record linkage process based on the means of the likelihood ratio proposed by Jaro as a probabilistic record linkage method. Six identifiers were considered: date of birth, maiden name, usual last name, first and second christian names and the gender. Two types of phonetic treatment were taken into account: the Soundex and a phonetic treatment adapted to the French language. Three situations were considered: 1) and 2) linkage of the data collected during two consecutive years in a university hospital (CHU de Dijon; 100000x100000 records) and a Paris hospital (50000x50000 records), 3) linkage of two files obtained through a healthcare network (Burgundy Perinatal Network; 200x2500 records). Whatever the situation, this work showed the interest of three identifiers when linking data concerning a same patient. The date of birth had the best discriminating power followed by the first and the last names. Including a poorly discriminating identifier like gender did not improve the results. Moreover, adding a second Christian name, often missing, increased linkage errors. On the contrary, it seemed that using a phonetic treatment adapted to the French language could slightly improve the results of linkage in comparison to Soundex. Whatever the method used, it seems necessary to improve the quality of identifier collection, in particular of the date of birth and of the first and last names as it could make the linkage of data obtained from different sources easier. Further research is needed to estimate the discriminating power of other identifiers (birth place and parents identifiers).

Which are the best identifiers for record linkage?

As a linkage using less informative identifiers could lead to linkage errors, it is essential to quantify the information associated to each identifier. The aim of this study was to estimate the discriminating power of different identifiers susceptible to be used in a record linkage process. This work showed the interest of three identifiers when linking data concerning a same patient using an automatic procedure based on the method proposed by Jaro; the date of birth, the first and the last names seemed to be the more appropriate identifiers. Including a poorly discriminating identifier like gender did not improve the results. Moreover, adding a second christian name, often missing, increased linkage errors. On the contrary, it seemed that using a phonetic treatment adapted to the French language could improve the results of linkage in comparison to the Soundex. However, whatever, the method used it seems necessary to improve the quality of identifier collection as it could greatly influence linkage results.

Regional repositories, reintermediation and the new GMS contract: cardiovascular disease in Tayside.

The new contract for general medical practitioners will make increasing demands on the primary care informatics community. There are a number of potential ways to provide reports which meet the requirements for data on the quality of care being provided by practices. In Scotland there are four components of the national information technology strategy which make meaningful comparisons of data possible. Using cardiovascular data as an example, to describe how the community health index number, managed clinical networks (MCNs), increasing consistency of Read codes, and regional repositories of data make the acquisition, processing and use of data more straightforward. The cardiovascular MCN collects the majority of its data electronically and four properties are crucial to its success: automatic collection of electronic data from many sources, prioritisation of data derived from multiple sources, record linkage processes, and manual validation of electronic data. Clinicians in primary and secondary care enter data during consultations and see the results of consultations recorded elsewhere. Because all data from the region are able to be Read coded according to prespecified templates, we are able to indicate to practices where they are in relation to the new contract targets and indicate which patients need to be seen, or excluded from, calculations. Effectively integrated management is facilitated by provision of regular prompted recall and review of people with chronic disease by multidisciplinary teams collaborating across the health service and into the community. In Scotland, use of newer informatics tools are proving to be useful contributions from primary care computing to equitable, evidence-based care.

Social deprivation and prevalence of epilepsy and associated health usage

OBJECTIVESTo examine the relation between social deprivation and the prevalence of epilepsy and associated morbidity using hospital activity data as a proxy.METHODSThe study was conducted in the health district of...

RISK OF CANCER OTHER THAN KS AND NHL IN PERSONS WITH AIDS IN ITALY.

15 Background: Uncertainty persists on whether cancers other than KS and NHL are in excess among HIV-seropositive persons. Italy offers a special opportunity for research, having a large HIV epidemic that includes many women. Methods: A record linkage process between the national Registry of AIDS and 13 Cancer Registries (covering about 15% of the Italian population) was carried out. Ratios between observed and expected numbers of cancers in 6,067 persons with AIDS living in an area covered by cancer registration, for a total of 25,759 person-years, were computed. Results: Significantly increased risks were found for invasive carcinoma of the cervix uteri (17.2; 95% confidence interval -CI-: 4.5-44.4), and for Hodgkin's disease (HD) (9.3; 95% CI: 4.6-16.6), mostly of the mixed cellularity type (7/11). In addition, an excess was also found for non-melanomatous skin cancer (3.1, 95% CI: 1.3-6.1), mostly basal-cell carcinoma (5/8). An excess was also seen for brain tumors, but potential misdiagnoses with brain masses of non-neoplastic origin could not be excluded. Overall cancer risk was approximately twice that in the general population. Conclusion: The excess of HD observed in this study in persons with AIDS is consistent with findings from similar studies, and a positive association with invasive carcinoma of the cervix was showed. The excess of non-melanomatous skin cancers was much lower than that reported among transplant recipients, perhaps due to surveillance bias.

Historical Demography, Life-cycle Reconstruction and Family Reconstitution: New Perspectives

This article explores the nature of record linkage in conventional family reconstitution studies, building upon previous pieces in this journal. It argues that the record linkage process on which reconstitution is based compromises the results obtained in two ways. First, it generates bias in the types of life-cycle detected. Second, because it assumes that potential links without obvious competition should be made, it generates a range of incorrect linkages. This article suggests that a better approach to life and demography in early modern communities might lie in the reconstruction of enriched life-cycles from which demographic data can be culled, rather than the explicit creation of demographic sub-samples.

Mortality in DSM-IIIR schizophrenia

Mortality was investigated in 356 DSM-IIIR schizophrenics admitted to a university psychiatric hospital over a 12-year period. Determination of death was made through a record-linkage process and observed death was compared with the mortality experience of the general population of the State of Iowa. Schizophrenics had nearly a three-fold increase in overall mortality. Mortality was primarily attributable to unnatural causes of death, particularly suicide, which was more than twenty-three times greater than expected. Mortality was greater in schizophrenic patients younger than 40 years and during the early portion of follow-up. Studies of mortality in psychiatric patients continue to be important, particularly as diagnostic criteria become better refined.

Mortality in patients with primary unipolar depression, secondary unipolar depression, and bipolar affective disorder: a comparison with general population mortality.

Mortality data are presented from a two to fourteen year follow-up of 705 primary unipolar depressives, 302 secondary unipolar depressives, and 586 patients with bipolar affective disorder (BAD) hospitalized at a tertiary care facility. Death ascertainment was made through a record-linkage process. Using sex- and age-standardized mortality ratios (SMRs), the mortality experience of the study population was compared with that of Iowa, the geographical area served by the admitting medical facility for this study group. Results show that risk for all-cause mortality was most pronounced during the first two years following hospital discharge, although secondary unipolar depressives continued to show a significant excess of deaths throughout the entire follow-up period. Deaths occurring from natural causes were significantly excessive only during the initial portion of the follow-up. Deaths from unnatural causes were significantly excessive throughout follow-up except for patients with bipolar affective disorder.

Age, mortality and chronic schizophrenia

Mortality data are presented from a 0 to 10 year follow-up of 636 chronic schizophrenic patients. Death ascertainment was made through a record-linkage process and comparisons made with the general population of the State of Iowa using sex and age standardized mortality ratios (SMRs). Risk for mortality was most pronounced among men younger than 40 and women younger than 70. No suicides occurred in chronic schizophrenic patients over age 40 years. The data demonstrate that patients at greatest risk for premature death are those younger than 40 years. Reasons for these findings are discussed.

Mortality in Schizophrenia—The Iowa Record-Linkage Study: A Comparison with General Population Mortality

Mortality data from 0- to 10-year follow-up of 688 schizophrenic patients were obtained through a record-linkage process and compared with that of the general population of the State of Iowa. Results show that the risk for mortality was most pronounced during the first two years following hospital discharge and among men and women younger than 40 years of age, although women continued to show excess risk after age 40 as well. Deaths occurring from cancer, heart disease, other natural causes, and accidents were not excessive, but mortality from suicide was significantly excessive for both men and women. Total mortality for schizophrenic patients was > 2.5 times expectation (standardized mortality ratio = 2.53, χ2 = 25.00, p < .001). Distribution of method of suicide did not differ significantly from that in suicides by nonschizophrenic patients.