Study design and setting
This is a retrospective cohort study in which we used data from four cycles of the Canadian Community Health Survey (CCHS) linked with health administrative data from Ontario, Canada’s largest province. Ontario has a population of approximately 14 million residents with the vast majority receiving provincial health insurance coverage for acute care services. Health administrative databases were used to obtain the one-year health service use from the index date of each of the four CCHS cycles. Since the distribution of health service use and other variables were similar over the time spanning the four CCHS cycles, we pooled the results for the CCHS cycles.
Data Sources
The CCHS is a national cross-sectional survey that collects information related to health status, health care utilization, and health determinants for the Canadian population. CCHS cycles 2005-2006 (27), 2007-2008 (28), 2009-2010 (29), and 2011-2112 (30) were chosen to maximize sample size and ensure consistency in the framing of questions relating to CCHS items used in this study. The four CCHS cycles were administered in participants’ homes using computer-assisted personal interviewing and participants in Ontario were asked if they would consent to have their CCHS data linked to provincial administrative data holdings. The index date for linkage was the participant’s CCHS interview date. Administrative databases used in the study were the: Registered Persons Database (demographics); Ontario Health Insurance Plan (OHIP) (physician visits); Discharge Abstract Database (inpatient hospitalizations); National Ambulatory Care Reporting System (emergency department other ambulatory contacts); Same Day Surgery (same-day surgeries, procedures); and Ontario Drug Benefit (outpatient prescription claims). Two additional data sources were accessed for specific diagnostic information on chronic conditions: the Ontario Mental Health Reporting System and the Ontario Cancer Registry. More information on theses databases is provided in Additional File 1. All data are held at ICES, where they were linked using encoded identifiers and analyzed. ICES is an independent, non-profit research institute funded by an annual grant from the Ontario Ministry of Health and Long-Term Care. As a prescribed entity under Ontario’s privacy legislation, ICES is authorized to collect and use health care data for the purposes of health system analysis, evaluation and decision support. Secure access to these data is governed by policies and procedures that are approved by the Information and Privacy Commissioner of Ontario. The study received approval from the Hamilton Integrated Research Ethics Board at McMaster University (certificate #13-590) and renewed yearly as required.
Study sample
We included Ontario CCHS participants who responded to any of the included CCHS cycles and who agreed to have their data linked to the health administrative data. We excluded those who were under 65 or over 85 years of age (n=103,377) because those under 65 to have health service use substantially different from older adults (31) and there was only a small number of CCHS participants over age 85. We excluded people who could not be identified as Ontario residents (n=94), who did not have health system contact within the 5 years prior to their survey date (n=161), who resided in long-term care (n=158) or received hospice or palliative care services (n=322), who participated in more than one CCHS cycle (we chose only the first cycle to avoid duplicate participants in the pooled data, n=580), who did not report their chronic disease status (n=1,016) and who were ineligible for OHIP coverage at index (n=71). The final sample included 28,361 individuals (see Figure 1).
Chronic conditions and multimorbidity
We identified 12 chronic conditions: Alzheimer’s diseases/dementia, anxiety/depression, arthritis, cancer, asthma, chronic obstructive pulmonary disease (COPD), diabetes, heart disease, hypertension, inflammatory bowel disease, stomach or intestinal ulcers, and stroke. These conditions were chosen because they could be identified in the administrative data, are prevalent in older adults, and are frequently reported in the literature on multiple chronic conditions (7, 32, 33). Chronic conditions were identified as either entry into a disease-specific database created at ICES or using an algorithm that searched for specific diagnostic codes and/or outpatient prescription claims within the 5 years prior to baseline. More details about each diagnostic definition can be found in Additional File 2.
Multimorbidity was operationalized as a count of individual chronic conditions (0, 1, 2, 3, 4, or 5 or more).
Acute Care Service Use
The outcome in this study was acute care service use, hospital admissions and emergency department visits that occurred in the 12 months after the CCHS interview date. The analyses included two measures for each service: a dichotomous variable (any visit/admission over one year) and a count variable (number of visits/admissions over one year).
Socio-demographic & Health Status Variables
The socio-demographic and health status variables selected for the study were determined by those available from the CCHS, guided by Anderson and Newman’s Behavioural Model of Health Care Utilization which identifies the following 3 determinants of health service use: need, enabling, and predisposing factors (34, 35). Need is determined by a person’s perceived need for health services, which in turn is a function of self-perceived health, activities of daily living (ADLs) or restricted activity, self-reported symptoms, quality of life, etc. Enabling factors include a person’s income, health insurance status, and access to regular care. Predisposing factors include demographic variables, attitudes, and beliefs. Anderson and Newman’s model has been primarily used for explaining health care utilization in the general population, with strong evidence of a socio-economic gradient in both developed and developing countries (36-40). Our goal in using the Anderson and Newman model was to ensure that we were comprehensive in capturing the main determinants that have been hypothesized to shape health service use. Predisposing variables in our study included sex, age, marital status and living arrangement; enabling factors included education, household income, and rurality; and needs-based variables were captured by the need for any assistance with daily activities, and two health-status variables (self-perceived physical and mental health). Additional File 3 provides further details on how these variables were defined and operationalized in the study.
Multiple imputation using a discriminant function method (41) was employed to address missing data, which only occurred for certain CCHS variables. All CCHS data and chronic conditions and a flag indicating death during a five-year prospective observation period from administrative data were used to impute missing data. Supplemental Table B1 shows the amount of missing data for each variable, with household income have the highest level of missing data (7.56%), followed by education (3.22%) and self-reported mental health (2.25%).
Statistical Analysis
Our main interest was the role of socio-demographic and health status factors in moderating the association between multimorbidity and acute care service use, in other words, the interaction of these variables with multimorbidity to influence service use. We began by conducting bivariate analyses exploring the association between multimorbidity and acute care service use stratified by each socio-demographic and health status variable. Logistic regression was employed in the bivariate analysis, modelling any vs. no acute care service over one year. All odds ratios (ORs) used a reference of 0 chronic conditions for one of the subgroups, e.g., the OR for any hospitalization during the past year for females and males with 1 chronic condition represent the odds in those with 1 chronic condition compared to females with no chronic conditions. Analyses showing differences in the multimorbidity-service use relationship across strata, were followed up with more detailed stratified analyses to determine whether the patterns were similar across demographic subgroups. For example, a difference in the relationship between multimorbidity and service use between males and females was followed with further stratified analyses (e.g., age and sex) to determine if the sex difference was similar across age groups. Given the large number of socio-demographic and health status variables in the study, it was not feasible to explore the interactions of multimorbidity with all variables. Therefore, we examined the patterns in the stratified analyses, and identified a subset of variables for which the relationship between multimorbidity and acute care service use showed evidence of significant variability across subgroups. Because of the significant role of age and sex in healthcare analyses, the regression models that explored interaction effects were run separately for each age/sex stratum (i.e., males 65-74 years, males 75-84 years, females 65-74 years, females 75-84 years). We also performed multivariable regressions, which included multimorbidity and all the socio-demographic and health status variables, to adjust for these factors in exploring the association between multimorbidity and acute care service use. Logistic regression was used for the dichotomous dependent variable (any versus no health service use) and Poisson regression was used to analyze the number of health service encounters. Interactions between multimorbidity and the covariates were explored in the regressions. Regressions did not include the CCHS cycle as a variable because early analyses showed no evidence of changes in sociodemographic or service use variables over time.
SAS version 9.4 was used for all statistical analyses, and the level of significance used throughout the study was alpha=0.05.