Study Design and Population
We conducted a population-based retrospective cohort study in Ontario, Canada, using linked administrative databases. We identified individuals who resided in LTC facilities from April 1, 2014 to March 31, 2017 through the Continuing Care Reporting System (CCRS), which is a database that captures information on individuals receiving continuing care services in hospitals, as well as individuals in LTC facilities that offer 24-hour nursing services (2). Data are collected using the Resident Assessment Instrument – Minimum Data Set 2.0 (RAI-MDS 2.0). Complete assessments are administered within 14 days of admission into LTC; they are also repeated annually and after significant changes in functional or health status. Abbreviated assessments are administered quarterly (2). Residents who completed more than one assessment were indexed at the time of their first assessment in the study period and followed for 90 days. We excluded residents who: were younger than 65 or older than 105 at the time of the index assessment; or were not eligible for Ontario’s universal health insurance plan (i.e., the Ontario Health Insurance Plan (OHIP)) at any time during the study period.
Data Sources
Baseline characteristics, including functional and health characteristics as well as recent healthcare utilization, were obtained from the following databases: CCRS; OHIP claims database, which captures billing claims for healthcare services provided by physicians; Ontario Drug Benefit (ODB) database, which captures prescriptions dispensed to individuals who are eligible to receive publicly funded coverage of their prescription medications (i.e., residents over the age of 65 as well as residents living in LTC facilities); Registered Persons Database (RPDB), which captures personal information such as age, sex, and postal code. Finally, information related to hospitalizations were obtained from the Discharge Abstract Database (DAD), which captures information on all discharges from acute care hospitals. Chronic conditions were identified using algorithms validated by [removed for blinding] and applied in previous studies (Additional file 1) (32–41). The datasets were linked using unique encoded identifiers and analyzed at [removed for blinding].
Exposure
The following criteria were used to identify residents with dementia: 1) diagnosis of dementia during a previous hospitalization (obtained from the DAD), 2) three or more physician billing claims at least 30 days apart in a two-year period (obtained from the OHIP claims database), 3) prescription of a cholinesterase inhibitor (obtained from the ODB database), or 4) documentation of dementia or Alzheimer’s disease AND Cognitive Performance Scale score greater than or equal to 2 in index assessment or in any previous RAI assessment, including those administered for continuing care services (CCRS database) and home care services (RAI-HC database). The first three criteria of this algorithm have positive and negative predictive values of 80.4% and 99.0%, respectively, when applied to Ontario residents over the age of 65 (41). The last criterion was added to increase the sensitivity of the algorithm.
Resident language was obtained from the index RAI assessment. During these assessments, interviewers are instructed to record the primary language spoken by the resident, which is coded as English, French, or other. Anglophones and Francophones were defined as residents whose primary spoken language was English and French, respectively, the two official languages of Canada. We excluded residents whose primary language was other than English or French. Language of the LTC facility was defined using language designation status according to the French Language Services Act (42), which is a provincial law that requires government agencies to provide all or some of their services in both English and French. In Ontario, 16 LTC facilities are included in this law; we defined these facilities as French facilities (43). We defined the remaining 612 facilities, which are only required to provide services in English, as English facilities. Anglophones in French facilities and Francophones in English facilities were considered to be living in language-discordant facilities.
Outcome
All hospital admissions within 90 days of each resident’s index assessment were identified from the DAD. The primary outcome was binary; we compared residents with at least one hospitalization (i.e., hospitalized) to residents without any hospitalizations (i.e., not hospitalized) during the 90-day follow-up period.
Statistical analysis
We performed descriptive analyses to compare the characteristics of residents with and without dementia. We also compared outcomes after stratifying by dementia as well as language of the resident, language of the LTC facility, and resident-facility language discordance. Comparisons were performed using chi-square tests for categorical variables, analysis of variance (ANOVA) for normally distributed continuous variables, and the Mann-Whitney U test for ordinal variables.
The associations between dementia, resident language, facility language, and hospitalizations were further explored with multivariate logistic regression analyses. We ran three regression analyses; the first examined the relationship between dementia, resident language, and hospitalizations, while the second and third models considered resident language and resident-facility language discordance as effect modifiers of the association between dementia and hospitalization. In the first model, both dementia and resident language were included as independent covariates. The second model (dementia*language) included a categorical variable denoting the interaction between dementia status and resident language (i.e., Anglophone without dementia, Francophone without dementia, Anglophone with dementia, Francophone with dementia), and the third model (dementia*discordance) included a categorical variable denoting the interaction between dementia status and resident-facility language discordance (i.e., residents without dementia in language-concordant facilities, residents without dementia in language-discordant facilities, residents with dementia in language-concordant facilities, residents with dementia in language-discordant facilities). The reference categories consisted of Anglophones without dementia (model 2) and residents without dementia in language-concordant facilities (model 3). All regression models adjusted for potential confounders related to both resident characteristics (age, sex, education, urban/rural status of prior residence, number of prescription medications, Charlson Comorbidity Index (44), Changes in health, End stage disease, and Signs and Symptoms (CHESS) score (45), Activities of Daily Living scale (46)) as well as facility characteristics (income quintile of facility, urban/rural status of facility, total number of beds). Statistical tests were two-tailed, and the significance threshold was set at 0.05. We used SAS 9.3 (SAS Institute, Cary, NC) for all analyses.