Introduction

This concept describes how record linkage is performed in Manitoba for data stored at MCHP. The concept lists the key points of record linkage; how record linkage is performed with health and non-health data sets; and provides two examples of record linkage methods used at MCHP, including:


• record linkage of hospital abstracts with ICU records from the Winnipeg Intensive Care Unit Database (WICUDB) ( Garland et al., 2012 )
  
• record linkage of family members using the Manitoba Health Insurance Registry and hospital abstracts data ( Chartier et al., 2012 )

For a brief definition of record linkage, see the glossary term Record Linkage available in our on-line glossary. For a more detailed description of the record linkage process, see the concept Record Linkage / Data Linkage available in our on-line Concept Dictionary.

Record Linkage - Key Points

The following key points identify the objectives and general process of record linkage in Manitoba for data held in the MCHP Data Repository:


• to maintain privacy and confidentiality of the data;
  
• record linkage is performed at Manitoba Health, not at MCHP;
  
• each data file is stored separately and unlinked at MCHP;
  
• only authorized persons can access the data files. In order to access the data, Ethics and Data Provider approvals must be obtained and a signed Researcher Agreement must be completed;
  
• files are merged during the analysis stage at MCHP using an encrypted Personal Health Identification Number (PHIN).

Record Linkage of Health Data

The key points of performing record linkage with health files include:


• record linkage occurs at Manitoba Health;
  
• an encrypted Personal Health Identification Number (PHIN) is assigned to each record consistently across different health data sets;
  
• identifying data, such as names and address are removed from the file(s) prior to being sent to MCHP ( Note: most health related data files are updated on an annual basis); and
  
• the encrypted PHIN in each of the files allows merging of data across different data sets.

Record Linkage with Non-Health Data

The key points of performing record linkage with non-health files involve a "multistage de-identification process". This process, described in Roos et al. (2008), contains the following steps:


1. the organization who maintains the data prepares a data file containing a scrambled organizational identification number for each individual, as well as identifying information, such as name, address, and date of birth. This file contains no program/service data from the organization.
   
2. Manitoba Health uses the information in this file to find the corresponding Personal Health Identification Number (PHIN) for each individual in the Manitoba Health Insurance Registry Data file. For records that match, an encrypted PHIN is attached to each record in the file received from the organization.
   
3. all identifying information is removed from the file, so that this "crosswalk" file contains only an encrypted PHIN and the scrambled organizational identification number for each individual. This file is then sent to MCHP.
   
4. the organization then sends the corresponding program/service data file to MCHP that contains the scrambled organizational identification number. There is no identifying information in this file.
   
5. each data file is stored separately and unlinked at MCHP. Only authorized persons can access the organizational data file(s) and the corresponding "crosswalk" file for approved projects.
   
6. files are merged during the analysis stage at MCHP using either the encrypted PHIN and/or the scrambled organizational identification number.

Examples of Record Linkage at MCHP

The following is a description of the methods used in two MCHP deliverables to link data. Please see the provided links for more information.

1. Garland et al. (2012)

To evaluate the epidemiology and outcomes of critical illness in Manitoba, Garland et al. (2012) linked two sources of health data in the Manitoba Population Research Data Repository (Repository) : (1) the Winnipeg Intensive Care Unit database (WICUDB) and, (2) the Hospital Abstracts Data. The following is a description of the six parts to the methodology of creating appropriate one-to-one links between individual ICU records in the WICUDB and individual hospital abstract records for patients who were aged 17 and older at the time of hospital entry.

This is followed by a description of the validation process for this data, and then a discussion of the SAS® code example available for working with the ICU data.

1) Preliminary Processing of the Hospital Discharge Abstracts

Prior to linking the data, two preliminary processing steps were completed to exclude the following:
• hospital abstracts that did not represent inpatient hospitalizations (e.g.: outpatient surgery); and
• duplicate hospital abstracts, identified as identical on five items - PHIN, hospital, hospital chart number, hospital entry date, and hospital separation date.

2) Preliminary Identification of Linkage between the WICUDB and the Hospital Discharge Abstracts

Linkage of WICUDB records and hospital abstracts was completed by linking:


1. the combination of the hospital and the hospital chart number
   AND
2. a. the ICU entry date that occurs on or after hospital entry and on or before hospital separation
   OR
   b. the ICU separation date that occurs on or after hospital entry and on or before hospital separation

For records that did not link with this initial step, a second step was completed to link the records using:


1. PHIN
   AND
   
2. a. the ICU entry date that occurs on or after hospital entry and on or before hospital separation
   OR
   b. the ICU separation date that occurs on or after hospital entry and on or before hospital separation

In some instances there were records that were linked to more than one hospital abstract. Steps 3 and 4 address this issue.

The results of linking the WICUDB records and the hospital abstracts can be found in Table 1.1: Interim Summary of Linkage between the Winnipeg ICU Database (WICUDB) and the Hospital Abstracts after Preliminary Processing on p.6 of The Epidemiology and Outcomes of Critical Illness in Manitoba deliverable by Garland et al. (2012) .

3) Addressing the WICUDB Records that Linked to Two Hospital Discharge Abstracts

If a single hospital abstract (or two or more abstracts) could be identified to link the WICUDB record to, it was included in the project analyses. For example, if a WICUDB record was linked to two abstracts, the abstract that contains the ICU entry and separation dates within its hospital entry and separation dates would be selected as the abstract to link to. Cases where a WICUDB record could not be linked to a single abstract were eliminated from the project.

The following six patterns were identified when two abstracts were linked to one WICUDB record (the first occurring abstract is denoted as "Abstract 1", and the second as "Abstract 2"):


1. The date of hospital separation for Abstract 1 was the same as the date of hospital entry for Abstract 2.
2. The hospital entry date for Abstract 1 was the same as Abstract 2, but both abstracts had different separation dates.
3. The hospital entry and separation dates were identical for Abstract 1 and Abstract 2.
4. The separation date of Abstract 1 was at least one day earlier than the entry date of Abstract 2.
5. The separation date of Abstract 1 was later than the entry date of Abstract 2, but before the separation date of Abstract 2.
6. The separation date of Abstract 1 was later than the entry date of Abstract 2, and both abstracts had the identical separation date.

In all of the above situations, entry and separation dates for the ICU records and the hospital abstracts were compared in order to determine which were appropriately linked, and which records should be eliminated. Please see Results Part 3: Addressing the 180 WICUDB Records that Linked to Two Hospital Abstracts for more information on addressing the WICUDB records that were linked to two abstracts.

4) Addressing the WICUDB Records that Linked to Three Hospital Discharge Abstracts

When three hospital abstracts were linked to one WICUDB record, the one abstract that included the ICU entry and separation dates within its entry and separation dates was used. Please see Results Part 4: Addressing the 12 WICUDB Records that Linked to Three Hospital Abstracts for more information on addressing the WICUDB records that were linked to three abstracts.

The results of steps 3 and 4 can be found in Table 1.2: Summary of Preliminary Linkage between the Winnipeg ICU Database (WICUDB) Records and the Hospital Abstracts Based on Entry and Separation Dates on p.9 of the Epidemiology and Outcomes of Critical Illness in Manitoba deliverable by Garland et al. (2012) .

5) Assessing How the Linked Records from Parts 2-4 Match on Seven Identifying Variables

Garland et al. (2012) assessed record linkage by matching seven identifying variables between the two records including:

1. Hospital
2. Hospital chart number
3. ICU entry date
4. ICU separation date
5. Sex
6. Birth year
7. Birth month

Records were accepted if there were matches for six of the seven identifiers.

The results of assessing the record linkage can be found in Table 1.3: Linkage between the Winnipeg ICU Database (WICUDB) and Hospital Abstracts Using Primary Identifying Variables and Table 1.4: Final Summary of Linkage between the Winnipeg ICU Database (WICUDB) and the Hospital Abstracts on p.9 of the Epidemiology and Outcomes of Critical Illness in Manitoba deliverable by Garland et al. (2012) .

6) Clarifying Manitoba Residency Status

Manitoba residency status was determined by identifying the presence of a PHIN in the hospital abstract.

The results of clarifying Manitoba residency status can be found in Table 1.5: Provincial Residency Status of Linked Winnipeg ICU Database Records and Hospital Abstracts on p.10 of the Epidemiology and Outcomes of Critical Illness in Manitoba deliverable by Garland et al. (2012) .

Validation Process

Health Information Management (HIM) at Manitoba Health validates records of newly admitted patients to WRHA ICUs every four to six weeks. The validation process involves the crosschecking of names, PHIN, and date of birth data in the WICUDB with the Manitoba Health Insurance Registry. The WRHA makes changes in the WICUDB for any inconsistencies HIM indicates in the crosschecking. Fields in the file from the WRHA used for linkages are PHIN, surname, given name, date of birth, and sex. Four steps are completed sequentially to link files in the WICUDB and the Manitoba Health Insurance Registry. These linked files (matches) were given a match indicator of 1 or 2, specifying how the files were matched, where "1" = PHIN used in linkage, and "2" = PHIN not used in linkage.

The remaining linkages are visually verified using a 'Proc Print' in SAS in three additional steps. These linked files are also granted a match indicator of 1, 2, or 0, specifying how and whether the files were matched, where "1" = PHIN used in linkage, "2" = PHIN not used in linkage, and "0" = non match.

At MCHP, the WICUDB database is updated with new files or missing information using the WICUDB records, a Medical Patient Registry Manitoba Health Insurance Registry snapshot, and Manitoba Health hospital abstracts with ICU admission. The following seven steps are followed in order to update missing data on new WICUDB records since previous linkage:


• Step 1. Exclude all records previously checked by HIM with valid PHINs
• Step 2. By surname, given (including initials), full birth date
• Step 3. By surname, full birth date, sex
• Step 4. By SOUNDEX (surname), full birth date, sex, COMPARE (given)
• Step 5. By SOUNDEX (surname),sex, COMPARE (given), birth year, birth month,
• Step 6. By full birth date, sex, COMPARE (given)
• Step 7. By sex, full birth date, facility, admit and separation date (against hospital abstracts)

NOTE: ALL matches found in steps 2 through 7 were visually verified.

More information on the steps of these validation processes can be found in Appendix 1.1:WRHA Critical Care Validation Process-Linkages to Manitoba Health Insurance Registry Performed Regularly and in Appendix 1.2:WRHA Critical Care Validation Process-Linkages to Manitoba Health Insurance Registry when Transferring WICUDB Data to MCHP in Garland et al. (2012) .

SAS Code for ICU Linkage / ICU Episodes

Example SAS® code for working with the ICU data and identifying ICU episodes of care in the MCHP data is available (internal access only) in the SAS code and formats section below.

2. Chartier et al. (2012)

Chartier et al. (2012) used record linkage techniques to connect first-order family members (also known as First Degree Relatives : parents, children, siblings, and spouses) together in the Health and Health Care Utilization of Francophones in Manitoba deliverable. The Population Registry contains variables, including the family registration number (REGNO) / (REGNO_CODE) , date of birth, sex, and Personal Health Identification Number (PHIN), that were used to specify de-identified individuals of a family.

The following additional characteristics were used to specify the type of family members:

• Mothers - had a Manitoba birth occurring in 1978 or later, as specified in the hospital birth records
  
• Fathers - family head, as specified by the REGNO
  
• Children - dependents of the female and male family head
• Siblings - dependents of the same family head at birth
• Spouses - "spouse pair flag" indicating spouses and common-law spouses reported to Manitoba Health

More information can be found in the Appendix for Chapter 3: Methods of Chartier et al. (2012)