Self-harm and rurality in Canada: an analysis of hospitalization data from 2015 to 2019

The Canadian Institute for Health Information’s Discharge Abstract Database (DAD) was the main source of data in this study. The DAD is comprised of demographic, administrative, and clinical information about patients discharged from acute care. Hospital-based care is part of the universal, publicly funded healthcare system in Canada. Because acute care facilities in all provinces and territories except Quebec are required to submit data to the DAD, it is a comprehensive and robust dataset of hospital discharges. The DAD does not include patients who visited the emergency department but were not admitted to hospital. That is, a patient with an episode of self-harm who presented to the emergency department, received an assessment and treatment, and was then discharged, would not be included in the DAD. Rather, the DAD captures the less frequent, but typically more medically serious self-harm events that may require interventions such as trauma management or involuntary admission for safety or for psychiatric care.

For the present study, we used a subnational version of the DAD that covered 77% of the 2016 population of Canada. Data from two northern territories, Northwest Territories and Yukon, were excluded because population estimates by sex and age group at the census subdivision level (CSD; municipalities or small areas equivalent to a municipality) [32], which were needed for stratified analysis by level of rurality, were not available.

The DAD is coded with the International Statistical Classification of Diseases and Related Health Problems, 10th Revision, Canada (ICD-10-CA) [33]. The primary outcome in this study was self-harm hospitalizations, identified by ICD-10 diagnosis codes for intentional self-injuries (X60-X84 and Y87.0). These codes include suicide attempts and non-suicidal self-injuries because it is not possible to distinguish suicidal from non-suicidal forms of self-harm in the discharge diagnosis fields. Patients who died in hospital because of their injuries (n = 897) were excluded from all analyses; cases from jurisdictions that provided data but lacked record-level CSD information (n = 2137) were excluded from the rurality analysis.

We analyzed data on all patients aged 10 or older who were discharged from a hospital with a diagnosis of self-harm between April 1, 2015 and March 31, 2020. DAD uses fiscal years (April 1 to March 31), rather than calendar years. The fivefiscal-year study period was chosen to establish pre-COVID-19 baseline hospitalization rates.

Rurality was measured using the Index of Remoteness [34] developed by Statistics Canada. Unlike other measures of rurality [35], this index captures geographic remoteness by considering population size and distance from points of service provision for each census subdivision [36]. By using population size and accessibility to other areas, the Index of Remoteness attempts to differentiate between urban, rural, and remote regions. The Index is based on 2016 Census data and geographic boundaries.

Each census subdivision is assigned a remoteness index (RI) value ranging from 0 to 1, with 0 being the least remote (and most urban) and 1, the most remote. The most remote areas are those with the smallest populations and low accessibility to population centers. A previous study categorized RI scores according to five natural breaks, while also considering the number of census subdivisions and the population distribution in each category [22, 37]. The categories are: easily accessible (RI score < 0.1500), accessible (0.1500 to 0.2888), less accessible (0.2889 to 0.3898), remote (0.3899 to 0.5532), and very remote (> 0.5532). The Index of Remoteness is mapped to DAD data via the census subdivision of each discharged patient’s place of residence.

We calculated crude and age-standardized self-harm hospitalization rates overall, and by year, sex, age group, and rurality. We used a binary variable for sex (male/female) because the number of hospitalizations with sex coded as non-binary (including transgender, intersex, sex unknown, or other) accounted for less than 0.1% of all cases; these cases could not be disaggregated and so were excluded from the analyses. We used 5-year age groups for the overall and sex-stratified analysis. To avoid small cell counts for the rurality analysis, we used four life course age groups (10–19, 20–34, 35–64, 65 +). Population data were drawn from Statistics Canada’s annual demographic estimates for the denominators in rate calculations; the 2011 Canadian Census population was used as the standard population. A Poisson regression was fit to estimate rate differences and rate ratios with 95% confidence intervals for the effect of rurality on self-harm hospitalization, using the least remote area (easily accessible) as the reference group. Analyses were conducted in SAS Enterprise Guide version 7.1 (SAS Institute, Cary, NC, USA).

We identified 68,378 self-harm hospitalizations between 2015 and 2019 (Table 1). For the five-year period, the overall age-standardized self-harm hospitalization rate was 54.4 hospitalizations per 100,000. The annual rate fell slightly from 55.4 per 100,000 in 2015 to 52.0 per 100,000 in 2019, though confidence intervals between consecutive years often overlapped, which suggested that rates were not always significantly different. The five-year age-standardized rate for females was 70.2 per 100,000 (95% CI 69.5–70.8); the rate for males was 39.9 per 100,000 (95% CI 39.4–40.4). Each year during the period, rates were higher among females than males (Table 2).

Five-year self-harm hospitalization rates by sex and age group are reported in Fig. 1. For both males and females, self-harm hospitalization rates peaked in the 15–19-year age group. Rates were significantly higher for females than males for age groups under 65 years old. In the two oldest age groups (85–89 and 90 years and older), males had higher rates than females.

We identified 66,241 self-harm hospitalizations between 2015 and 2019 that had census subdivision information, and so could be used for analyses by rurality. Age-standardized self-harm hospitalization rates were highest for males and females in very remote areas (Fig. 2). Across all levels of rurality, self-harm hospitalization rates were significantly higher for females than males. For each additional level of rurality, there was a significant increase in the rate of self-harm hospitalization for both sexes (Fig. 2).

Populations in very remote areas had a 6.5 times higher rate (an additional 215.9 self-harm hospitalizations per 100,000), compared with populations in easily accessible areas (Table 3). Compared with easily accessible areas, all other levels of rurality had higher rates of self-harm hospitalization for females, males, and both sexes combined (Table 3). Among females, the rate ratio was 7.5 times higher in very remote areas, compared with easily accessible areas; the rate among males in very remote areas was 5.0 times higher.

Figure 3 shows self-harm hospitalization rates for females and males, by age group and rurality. Rates increased with each level of rurality for all age groups except 65 or older. For both sexes, the largest rate differences were between very remote and easily accessible areas among 10–19 year-olds, followed by 20–34-year-olds (Fig. 3). For females, rates decreased with age across all levels of rurality. A similar pattern was found for males, except in the 10–19-year age group, who had similar or lower rates compared to males aged 20–35 years.

A strength of this study is that we used a definition of self-harm based on an internationally standardized classification system and a multifaceted, scaled definition of rurality. We chose the Index of Remoteness because it is based on population size and accessibility to other regions. Thus, we were able to distinguish between rural and remote regions, unlike studies of self-harm and suicide using binary measures of rurality [14, 17, 19, 66], which can obscure differences between diverse rural contexts. This is an important consideration in Canada, one of the least densely populated countries in the world, where residents of remote communities may experience place-specific risks. Another strength of our study is that the dataset had comprehensive coverage for most provinces and territories.

Several limitations should be considered when interpreting the results. Not all cases of self-harm hospitalization were included in our analyses. We did not use data from Quebec, Northwest Territories, or Yukon. These three jurisdictions have proportionately large Indigenous populations in remote regions [67], and some rural and northern Indigenous communities have high rates of self-harm compared with the general population [68]. Consequently, our estimate of the national self-harm hospitalization rate for the most remote areas may be less representative than our estimate of the national rate for urban areas.

The DAD does not include patients who visited the emergency department because of self-harm but were not hospitalized. Therefore, we likely captured the more medically serious non-fatal incidents, but not all episodes of hospital-treated self-harm. In the suicide surveillance indicator framework used in Canada, self-harm hospitalizations and suicide deaths are recognized as less frequent, but more harmful, outcomes [5, 68]. Other datasets based on emergency department visits [17] and population surveys [69] provide evidence about more prevalent but less severe outcomes such as suicidal ideation [70]. This “iceberg” model [68, 71] requires indicators across a continuum of lethality in order to describe the scope and scale of suicidality in the population. Our study provides evidence on only one outcome in this model.

Owing to the nature of the data, we were unable to analyze co-variates such as income, marginalization, or deprivation, either at the individual or area level. This meant that we could not adjust for the confounding effects of these social determinants, which are associated with self-harm, especially in high-income countries [38, 39]. Examining relationships between self-harm and chronic diseases or multimorbidity of physical and mental illnesses was also not within the scope of our study. These factors are known to be associated with suicidal behaviour [44], increase the odds of seeking hospital care [72], and may be more prevalent among females and rural populations [43]. Even without adjusting for these confounders, our results were consistent with recent evidence from Canada’s largest province (Ontario) which indicated that, after controlling for income, immigration status, marginalization, and chronic conditions, males and females in rural areas were significantly more likely to attempt suicide than were their urban counterparts, [21]. Future research with national data could clarify relationships between rurality, socio-economic deprivation, chronic disease status, and sex differences in self-harm.

Another limitation was the difficulty of ascertaining injury intent, which may have resulted in misclassifications of those who presented to the hospital [73]. This may have led to an underestimation of self-harm in the population, particularly for self-harm by poisoning, where the intent of an overdose may be difficult to determine [74]. To minimize the effects of the pandemic on our results, DAD data collected after March 31, 2020 were not included in this study. Additional analyses to understand the impact of the pandemic on self-harm in Canada is needed. Early evidence on self-harm-related emergency department visits suggests that the pandemic may have had a differential impact on urban and rural populations [70].