Data

Availability of data information

2006 Census of Canadian population is collected and maintained by Statistics Canada. To prevent the confidentiality of the 2006 Census respondents, population-count data are released for areas with at least 40 persons and socio-economic (income) data are released for area with at least 250 persons. Census data are made available publicly and to libraries of Canadian universities in various aggregate databases. The 2006 Census data at forward sortation area (FSA) level used in our study were downloaded from the Data Library of University of Alberta.

DAD Metadata were requested and received from the Canadian Institute of Health Information. DAD records demographic, administrative and clinical data on all hospitalizations at acute care inpatient facilities. To prevent patient identification, each patient is assigned a unique scrambled patient identifier and has his/her address identifier restricted to the first three digits of his/her residential postal code. Ethics approval for secondary use of de-identified acute myocardial infarct data was received in 2008 and has been kept current from the Health Research Ethics Board – Health Panel (REB 3) of the University of Alberta.

Data on insolvency filings by Canadian individuals were provided by the Office of the Superintendent of Bankruptcy, Government of Canada. All data related to individual bankruptcy filings are publicly available because every bankruptcy filing is a legal proceeding before the courts. Thus no ethics approval, (e.g., related to issues of confidentiality), is required for the use of this data.

Analysis data can be made available by the authors, upon request.

Data aggregation

Our unit of analysis was the FSA. The FSA is a geographical region in which all postal codes (six-character strings) start with the same three characters. The average number of households served by an FSA is 8,000 but the number can range from 0 to 60,000. Our study included all FSAs with non-missing and non-zero population counts in all Canadian provinces, except Quebec, in effect throughout the entire study period Jan 1st, 2002 – Dec 31st, 2009 [6].

There were 1,155 FSAs included in our analysis. The FSAs selected in our study had in 2006 between 250 and 80,025 people aged 20 years or over. For each FSA i we extracted the count of population 20 years and over, P _i , in 2006, and several socio-economic characteristics (median household income, employment rate, education attainment, home ownership and being the head of a single parent family) from 2006 Census of Canadian population [3]. The selected socio-economic characteristics are factors incorporated in previous socio-deprivation indices [7, 8]. Hospital admissions of patients aged 20 and over residing within the selected 1155 FSAs were clustered into hospitalization episodes (i.e., uninterrupted time intervals of hospitalization days). Hospitalizations with a most responsible diagnosis of AMI with a discharge date between Jan 1st, 2002 and Dec 31st, 2009 were selected as indicators of AMI events. The calendar year of the episode discharge date was used as the calendar year of the AMI episode. Multiple AMI episodes of the same patient in the same year were counted only once. For each year [t] and each FSA i we identified the number of AMI events NM _[t]i . From Insolvency filing data we extracted for each year [t] and each FSA i the number of personal bankruptcy (BKC) filings NB _[t]i .

AMI and BKC counts were changed into rate estimates as follows: (1) \( {M}_i=\frac{{\displaystyle {\sum}_{\left[t\right]}}N{M}_{\left[t\right]i}}{8{P}_i} \) and \( {B}_i=\frac{{\displaystyle {\sum}_t}N{B}_{\left[t\right]i}}{8{P}_i} \) as rates over the 8-year time interval from 2002 to 2009 and (2) \( {M}_{\left[t\right]i}=\frac{N{M}_{\left[t\right]i}}{P_i} \) and \( {B}_{\left[t\right]i}=\frac{N{B}_{\left[t\right]i}}{P_i} \) as rates in year [t] for FSA i. Because FSA population counts were available for 2006 only, we used these as population counts in all years of the study period. We organized the data in two ways: (1) aggregate data that enclosed AMI rates and BKC rates over the 8-year time interval and (2) panel data that enclosed yearly AMI rates and BKC rates. Both 8-year and yearly rates were reported per 1000 person-years.

Statistical analysis

Descriptive statistics (mean, standard deviation) were calculated for the 8-year, as well as, yearly AMI and BKC rates and socio-economic factors. Pearson correlation coefficients were calculated to identify associations between AMI and BKC rates, as well as between the socio-economic factors and AMI or BKC rate, and to identify potential socio-economic confounders of the association between AMI and BKC.

Socio-economic variables were standardized to mean zero and variance one, across the 1,155 FSAs. Principal component analysis [9] was used to construct 3 new uncorrelated variables from the 8 standardized census variables. A VARIMAX rotation was applied on the uncorrelated variables to better align them with the census variable and improve their interpretability. We refer to the 3 principal components as education, material wealth and working-age population, and denote their values for FSA i by E _i , W _i , A _i .

We estimated the longitudinal association between AMI and BKC by fitting several longitudinal SEM on the panel data that incorporated both temporal relationships AMI × Following-year BKC and BKC × Following-year AMI.

The reference model was a cross-lagged model with measurement errors and time-independent covariates [10–14]. The first three waves of the reference model are shown in Fig. 1. The model used the observed yearly AMI and BKC rates as proxies for unobserved heart health and over-indebtedness levels, and further assumed that heart health and over-indebtedness levels incorporated effects of socio-economic variables and latent (unobserved) variables independent of socio-economic factors. Specifically, the system of AMI and BKC rates, measured across years, is explained by the model as follows: for each year [t] and each FSA i, the observed rates, M _[t]i and B _[t]i are functions of (a) the effects, α, of FSA’s time-invariant socio-economic characteristics, (b) unobserved true heart health and over-indebtedness components, m _[t]i and b _[t]i , and (c) measurement errors, e _m[t]i and e _b[t]i constrained to have equal variance σ _em ² and σ _eb ² across all years, except 2002. The equations with intercepts μ _M[t] and μ _B[t], to explain the measured rates are

$$ {M}_{\left[t\right]i}={\mu}_{M\left[t\right]}+{\alpha}_{me}{E}_i+{\alpha}_{mw}{W}_i+{\alpha}_{ma}{A}_i+{m}_{\left[t\right]i}+{e}_{m\left[t\right]i}, $$

$$ {B}_{\left[t\right]i}={\mu}_{B\left[t\right]}+{\alpha}_{be}{E}_i+{\alpha}_{bw}{W}_i+{\alpha}_{ba}{A}_i+{b}_{\left[t\right]i}+{e}_{b\left[t\right]i}. $$

Additional assumptions are that the socio-economic characteristics, latent heart health and over-indebtedness components, measurement errors and disturbances were mutually uncorrelated. The variances of measurement errors e _m[2002]i and e _b[2002]i in year 2002 were allowed to differ from the error variances in subsequent years and latent variables for year 2002 were assumed to be correlated, to adjust for the unavailability of data before 2002.

The time frame of the study incorporates two economically-distinct periods: (1) an economically-stable period from 2002 to 2007 and (2) the economic recession of 2008–09. SEMs were therefore first fitted using 2002–07 data only, and then using entire data from 2002 to 09. When 2002–09 data were used, the SEM coefficients of the last two years were allowed to be different than the coefficients of 2002–07. This SEM feature was added to explore how the longitudinal association between AMI and BKC was affected by an adverse global economic event.