Introduction

This concept will discuss the methodology used by Manitoba Center for Health Policy (MCHP) researchers to construct hospital and intensive care unit (ICU) episodes from administrative data ( Hospital Abstracts Data ) and the Winnipeg Intensive Care Unit Database (WICUDB) . The latter is also known as the Critical Care/Intensive Care Unit Database (CC/ICU DB). As there are different variables available for Manitobans and non-Manitobans in these databases, the methods used for both population groups are described.

Episodes of Care

An episode of care is defined as a continuous time in one or more hospitals, including all direct transfers between hospitals and/or ICUs.

Methodology

There are three major parts used by Garland et al. (2012) to create hospital and ICU episodes for Manitobans and non-Manitobans using the Hospital Abstracts and CC/ICU datasets, and multiple variables, including PHINs, Out of Province Unique Identifier (OOUPI), dates and times of entry and separation into hospitals and ICUs, and the location of patients before and after hospitalization.

In brief, Personal Health Identification Numbers (PHINs) from the Hospital Abstracts Data allow for identification of all hospital abstracts for all Manitobans. Garland et al. (2012) analyzed these abstracts to identify hospital episodes (which may be made up of one or more abstracts for an individual). Considering the entry and separation dates, ICU episodes were created from these hospital episodes (known as an "outside-in" approach).

For non-Manitobans, an Out of Province Unique Identifier (OOPUI) variable in the CC/ICU DB was used to specify non-Manitobans who have been in ICUs in Manitoba. These ICU records were specified as ICU episodes using specific criteria and then were used to identify hospital episodes based on entry and separation dates (known as an "inside-out" approach).

The following provides a description of the methodology used to create hospital and ICU episodes by Garland et al. (2012) . Please see Specific Aim 3: Creating Episodes of Hospital and ICU Care in The Epidemiology and Outcomes of Critical Illness in Manitoba deliverable for more details.

Part 1 - Constructing Hospital Episodes for Manitobans

Hospital abstracts from administrative data were used to identify variables that described hospital visits and care. Inter-hospital transfer was identified using the following variables:


• Timing of entry
• Timing of separation
• Location of patient before/after hospitalization (see Appendix Table A3.1 ) (using hospital separation codes and TransferTo and TransferFrom variables)

The process of combining hospital abstracts into ICU-containing hospital episodes began by identifying all abstracts for Manitobans with hospital entry date beginning on or after June 1, 1998 and separation date on or before March 31, 2008. For each individual these abstracts were arranged by ascending hospital entry dates, considered in chronological sequence, and then combination rules (see below) were applied.

When two abstracts were linked to the same Manitoban, the combination rules were followed to determine whether the two abstracts were from one hospital episode. We denote the first occurring abstract as "Abstract 1" and the second occurring abstract as "Abstract 2".

The final result for each patient was one or more hospital episodes. Each was constructed from one or multiple abstracts. Finally, only hospital episodes that contained at least one ICU record were retained for further analysis.

Two abstracts were considered to be one hospital episode if:

1. The hospital entry and hospital separation dates of Abstract 2 were entirely contained within those of Abstract 1 (i.e. the hospital entry date of Abstract 2 was the same or later than that of Abstract 1, AND the hospital separation date of Abstract 2 was the same or sooner than that of Abstract 2);
   
2. Abstract 1 had a separation code to another acute care facility AND/OR a TransferTo variable to another acute care facility, AND the hospital entry date of Abstract 2 was later than the hospital separation date of Abstract 1 by two days or fewer;
   
3. Abstract 1 did NOT have a separation code to another acute care facility NOR a TransferTo variable to another acute care facility, AND the hospital entry date of Abstract 2 was later than the hospital separation date of Abstract 1 by two days or fewer, AND the TransferFrom variable from Abstract 2 indicated an acute care facility.

Two abstracts were NOT considered to be one hospital episode if:

1. The hospital entry date of Abstract 2 was more than two days later than the hospital separation date of Abstract 1;
   
2. The hospital entry date of Abstract 2 was two days or fewer later than the hospital separation date of Abstract 1, AND Abstract 1 did NOT have a separation code to another acute care facility NOR a TransferTo variable to another acute care facility, AND the TransferFrom variable from Abstract 2 did NOT indicate an acute care facility.

Please see Appendix Table A3.2 for coding of acute care hospitals.

More information on this step can be found in Methods Part 1: Constructing Hospital Episodes for Manitobans in Garland et al. (2012) .

Part 2: Constructing ICU Episodes for Manitobans

The hospital episodes created by the methodology explained above were used in an "outside-in" approach to identify ICU episodes for Manitobans that occurred during its time period.

Note: In order for ICU episodes to be identified from hospital episodes, the ICU records had to be previously linked to the hospital abstracts. A detailed description of the method of record linkage used by Garland et al. (2012) can be found in the Record Linkage in Manitoba concept in the on-line Concept Dictionary or in Specific Aim 1: Linking the Datasets in Garland et al. (2012) .

Once ICU records were linked to their appropriate hospital abstract, ICU episodes were created by considering the ICU records associated with the specific abstracts comprising that hospital episode. The aim of combining ICU records from within a hospital episode was to identify conceptual episodes of ICU care, even if such an episode was interrupted by a reasonably brief period elsewhere, such as on a ward. This determination was motivated by data showing that ICU readmission after a shorter period is more likely to be for the same reason as the prior ICU stay (Rosenberg, Hofer, Hayward, Strachan, & Watts, 2001 as cited in Garland et al., 2012 ).

An episode of ICU can include time in multiple ICUs, either different ICUs at a single hospital and/or in different hospitals. If multiple ICU records were linked to the hospital episode, it was assessed whether they were of one or more ICU episode(s). The length of time between the two ICU records (first ICU separation and second ICU entry date and time) within a single hospital episode was used to identify ICU records that belonged to one hospital episode.

The date and time of ICU entry and separation derived from the CC/ICU DB were used to calculate the period of time between ICU records from a Database Hospital (DBH) (i.e. records from a hospital included in the CC/ICU DB, specifically the Health Sciences Centre, St. Boniface General Hospital, Seven Oaks General Hospital, Concordia Hospital, Victoria General Hospital, and Grace Hospital).

For ICU records from a non-DBH, the timing of ICU entry and separation, as recorded in the abstract, were used. ( Note: these included only dates pre-2004 and both date and time for post-2004 data. So, in order to calculate period of time in hours for pre-2004 ICU care in non-DBHs, an ICU entry or separation was assigned to have occurred at noon).

The following considerations were included when constructing ICU episodes:

• possible substantial delays in inter-ICU transfer, or from an ICU in a remote hospital
• inaccuracies in recording times/dates
• transfers of a patient to an ICU in another hospital, while the patient's bed is still held in the ICU of the original hospital
• lack of the timing of ICU entry and separation dates in hospital abstracts prior to 2004.

Using the distributions of the gap durations on our data (see Appendix Table A3.3 for time intervals separating successive ICU records within hospital episodes containing multiple ICU records) and the above considerations, successive ICU records contained within a hospital episode were considered part of the same ICU episode if the later ICU record began less than 48 hours after the earlier ICU record ended .

More information on this step can be found in Methods Part 2: Constructing ICU Episodes for Manitobans in Garland et al. (2012) .

Part 3: Constructing Hospital and ICU Episodes for Non-Manitobans

An "inside-out" approach was used to construct hospital episodes for Non-Manitobans. Hospital episodes were created from the hospital abstracts that linked to the ICU records making up the ICU episode.

When a patient was not a Manitoban, an Out of Province Unique Identifier (OOPUI) was used to track transfers of non-Manitobans between ICUs in DBHs. Each time a record is added to the CC/ICU DB for a non-Manitoban, a search is done by name and date of birth to identify prior CC/ICU DB records for that person. If any are found, the prior OOPUI is assigned to that new CC/ICU DB record. If no matches are found, then a new OOPUI is assigned.

Note: With the limitations of the CC/ICU DB data, only ICU-to-ICU transfers between two DBHs could be identified. The OOPUI does not indicate if a patient was admitted anywhere but an ICU in a DBH, thus if there was a time lapse between the entry date of one ICU record in a DBH and the entry date of a second ICU record in a DBH, it is unknown if the patient was admitted elsewhere between the two ICU records.

Successive CC/ICU DB records for a patient with the same OOPUI were considered to be part of the same ICU episode if the later ICU record began less than 48 hours after the earlier ICU record ended.

In all other situations for non-Manitobans, each hospital abstract was assumed to represent a full/entire hospital episode. When there were multiple ICU records within a single abstract, it was assessed whether those multiple ICU records contained one or multiple ICU episodes. The multiple ICU records within a given abstract were considered part of the same ICU episode if the later ICU record began less than 48 hours after the earlier ICU record ended.

More information on this step can be found in Methods Part 3: Constructing Hospital and ICU Episodes for Non-Manitobans in Garland et al. (2012) .

Results

Results of the methodology processes described above can be found in the Results section in Garland et al. (2012) , including a summary and discussion of the results related to:


• the use of hospital abstracts to identify ICU care in hospitals not included in the CC/ICU DB;
  
• the construction of hospital episodes from individual hospital abstracts for Manitobans;
  
• the construction of ICU episodes from individual ICU records for Manitobans; and
  
• the construction of ICU and hospital episodes from individual records for non-Manitobans

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.