Publications
Health Research Working Paper Series
Mental comorbidity and increased use of acute care hospital services
Data sources, methods, and limitations
Archived Content
Information identified as archived is provided for reference, research or recordkeeping purposes. It is not subject to the Government of Canada Web Standards and has not been altered or updated since it was archived. Please "contact us" to request a format other than those available.
Data sources - Part 1
Hospitalization data
Statistics Canada's Hospital Person-oriented Information (HPOI) Database is a person-level dataset derived from discharge records of inpatients in most of the acute-care hospitals and some psychiatric, chronic and rehabilitation hospitals across Canada.30 The database excluded newborns, patients not resident in a Canadian province and those without a usable patient identification number. For 2003/2004, 13% of the records were excluded, of which 81% were for newborns. The database includes information on date of admission and separation, up to 25 International Classification of Diseases (ICD)-9 diagnoses identifying the reason(s) for the hospitalization and up to 20 procedure codes indicating interventions received during the course of the hospitalization based on ICD-9/10 codes.31,32,33,34 The discharge records contain demographic (gender, date of birth, postal code), administrative (health number, admission and separation dates) and clinical information derived from the Hospital Morbidity Database (HMDB) maintained by the Canadian Institute for Health Information (CIHI).35 During data processing at Statistics Canada, about 3% of HMDB records for patients aged 12 and older were excluded because of missing or invalid health numbers. The HPOI represent approximately 2.7 million acute-care hospitalizations each year.
Only records in the HPOI database that pertained to patients who were discharged during fiscal year 2003/2004 were used. Fifty-nine records with missing admission or separation dates were excluded, and the 23 records with missing sex were excluded. The number of records used for this analysis was 2,711,533.
Health region information was added to the HPOI from the 2006 Census. Postal codes were first converted to census geographic units using the Statistics Canada postal code conversion file, and then linked to health regions based on the correspondence file between health regions and their component census geographic units.36,37 Census linkages were created at the dissemination area (DA) level and block level for British Columbia, Alberta, Saskatchewan, Manitoba, and Ontario (Local Health Integration Networks, [LHINs]). Even these smaller geographic areas (DA/blocks) sometimes straddle health region boundaries. In those cases, the entire DA or block was assigned, in conjunction with the affected province, to just one health region and therefore represents a 'best fit' with census geography.
Data sources - Part 2
Canadian Community Health Survey
The Canadian Community Health Survey (CCHS) continuously gathers cross-sectional information about the health and health care use of Canadians. The survey covers the non-institutionalized household population aged 12 or older in all provinces and territories, except members of the regular Canadian Forces as well as residents of Indian reserves, of Canadian Forces bases (military and civilian) and of some remote areas. The methodology has been previously described.51 The overall response rate to the 2000/2001 CCHS was 85%; the total sample numbered 131,535.
Linking health survey and hospital data
Data from the CCHS (Cycle 1.1) were linked at the individual level to the Hospital Morbidity Database (HMDB) (2000/2001 to 2004/2005) using probabilistic linkage methodology. Unique identifying information including personal health number, postal code, date of birth and age were used in the linkage process. Insufficient information was available in the hospital data for Quebec residents' records to be linked; therefore, the 22,667 (17.2%) CCHS respondents in Quebec were dropped. Of the 108,868 respondents who remained, 90,450 had given permission for their survey data to be linked to administrative data. Check-digit algorithms were used to verify the plausibility of the health numbers they provided; 72,363 respondents provided a plausible health number required for data linkage and were used in this analysis. Thus, linkage was conducted only for CCHS survey respondents living outside Quebec who provided consent to link their survey information to other sources of health information and a usable personal health number. The linkage was conducted by Statistics Canada. Additional information regarding the linkage process is provided elsewhere.30 A recently published evaluation of the linkage between the CCHS and HPOI reported high coverage for the population younger than age 75.52
Methods - Part 1
Descriptive statistics (proportions, averages) were generated using SAS version 9.1.3 (Cary, NC), particularly PROC FREQ and PROC MEAN.
Methods - Part 2
Using the linked data, we compare the characteristics of individuals in the following four groups: individuals who within the next four years had
- at least one hospitalization with a most responsible mental diagnosis
- at least one hospitalization with a comorbid mental diagnosis (and no hospitalizations with a most responsible mental diagnosis)
- hospitalization without any mental diagnoses
- no hospitalization.
Variables were derived from the Canadian Community Health Survey (CCHS) survey data and grouped in the following categories: demographic (age, sex, rural/urban status, race), socioeconomic status (household income, education level, family/marital status, immigration status), health status (self-reported health, disability level, presence of comorbidity, self-reported daily stress, impact of health problems, pain), health behaviours / risk factors (smoking status, body mass index), and access to health care services (access to regular medical doctor, unmet healthcare needs).
Descriptive analysis conducted include weighted prevalence of various demographic, socioeconomic, health status and health care use indicators, which were calculated using SAS software (version 9.1). Individuals who were admitted to hospital with conditions not related to mental health were the reference group. Given the difference in age structure, the rates have been age/sex standardized.
The bootstrap technique was applied to all analyses to account for the complex survey design and to estimate the variance and confidence intervals. Survey weights were specifically produced by Statistics Canada for the linked file to adjust for non-response to the CCHS, as well as for the exclusion of records of respondents who did not provide plausible health numbers or give permission for linkage to administrative health data. These weights were applied to the analysis file. The weighted data were representative of the Canadian household population residing outside Quebec.
Limitations
Limitations of the hospital file
Like other work using administrative hospital data, this work has several general limitations. These data come from provincial reports, and linkages were conducted within each provincial dataset, so identifying patients who used hospital resources in more than one province is impossible. This could overestimate the number of patients and underestimate the total length of stay for some patients, and may be a significant factor for smaller provinces or areas near provincial borders. Deaths outside of hospital were not included, so the analysis dataset underestimates the number of patients who died subsequent to hospitalization. The validity of conclusions drawn from analyses of large administrative databases depends on the accuracy of case-defining diagnostic codes. Hospital administrative data do not include hospitalizations to psychiatric hospitals, so some of the most severe mental conditions are missing from analysis. Hospitalizations from psychiatric hospitals represent less than 15% of all mental related hospitalizations;22,23 nevertheless, their exclusion is unfortunate. The analyses presented here are limited to acute-care hospitalizations. Ideally, we would be able to consider the use of emergency rooms and hospitalizations to physicians and other providers as well.
Limitations of the linked survey-hospital file
The analyses here exclude respondents from Quebec, because that province provided Statistics Canada with scrambled health numbers, as well as postal codes with insufficient details, making it impossible to link administrative records and survey responses.
The linkage of CCHS and hospital data, while probabilistic, does not include names. This means that there is a strong dependence on health insurance number, and little likelihood of successfully linking records where the health insurance number is missing or incorrectly coded.
Censoring before the end of the four-year period because of events such as death or moving out of the province could not be accounted for, because information about these events was not available or incomplete.
- Date modified: