 |
 |
|
|
|
In Edmonton and Calgary, health services are delivered under the auspices of regional health authorities (RHAs). For each year from 1986 to 1999 (except 1997), the age- and sex-standardized suicide rate for the group aged 15 years and over in the Edmonton RHA was greater than that in the Calgary RHA, with rate ratios ranging from 0.92 to 1.73 (mean 1.37). The cities of Edmonton and Calgary, which represent most of the population of the respective RHAs, are urban centres with many similarities. The question arises as to why their suicide rates should be so different. The term “parasuicide,” introduced by Kreitman and colleagues (1) as an alternative to “attempted suicide,” refers to an episode of intentional self-harm that is not necessarily aimed at a fatal outcome. It has been argued that parasuicide and (completed) suicide fall along a continuum of suicidal behaviour (2), although questions have been raised about the validity of this model in the case of children and adolescents (3). Of those who engage in parasuicide, 10% to 15% will eventually commit suicide, and in a given population, the ratio of parasuicide to suicide is approximately 10:1 (4,5). This suggests that by studying parasuicide we may better understand suicide. We undertook the present study to determine whether the parasuicide rate in the Edmonton RHA is elevated, compared with that in the Calgary RHA and, if so, whether the increase can be explained at the ecologic level by sociodemographic factors. MethodsCensus Data To conduct and report the national census, Statistics Canada partitions the country into several geographic units. Cities are located in census metropolitan areas (CMAs), each of which includes an urban area and, possibly, adjacent suburban and rural areas. Each CMA is made up of census tracts that are in turn divided into enumeration areas (EAs). According to the 1996 national census, the cities of Edmonton and Calgary had populations of 616 306 and 768 082, respectively, while the corresponding CMAs numbered 862 597 and 821 628. By design, the borders of the Edmonton and Calgary RHAs are nearly coterminous with the corresponding CMAs. We selected the Edmonton and Calgary CMAs as the population bases for our study. According to the 1996 national census, the Edmonton and Calgary CMAs each had over 1000 EAs. We obtained sociodemographic data from the 1996 national census from Statistics Canada’s public-use files. First, we extracted data at the level of EAs and then aggregated across EAs as described below. The following sociodemographic variables were available for EAs and relevant to the study: population counts by age and sex, average income among those who received an income from any source (average income), proportion of those in the labour force who were unemployed (percentage unemployed), proportion of those in the group aged 15 years and over who received an education beyond high school (percentage postsecondary), proportion of those in the group aged 15 years and over who were married (percentage married), and proportion of the population who were, in Statistics Canada’s terminology, “Native” (percentage Native). For each CMA, EAs were aggregated in the following manner: first, EAs were ranked according to average income; next, EAs with the lowest average income were combined until (approximately) 10% of the CMA’s population was reached. We used a similar procedure for the next 10%, and so on, thereby creating 10 geographic units, which we refer to as “income regions.” The EAs making up an income region were not necessarily contiguous, since EAs with similar average incomes might be distributed throughout a CMA. Once we created income regions, we obtained sociodemographic variables for each income region by forming weighted averages across constituent EAs, with weights proportional to population size. Parasuicide Data When we conducted the study, there were 5 hospitals with an emergency department in Edmonton and 4 such hospitals in Calgary. We defined a parasuicide as an episode of nonfatal intentional self-harm that resulted in an emergency department visit and for which the diagnostic code assigned by a medical records technician was in the ICD-9-CM range of E950 to E959 (“suicide and self-inflicted injury”) (6). With this approach, the study did not include episodes of intentional self-harm for which the individual was not treated in an emergency department. Data on parasuicides for 1997 were kindly provided by the Edmonton and Calgary RHAs; this was the first year for which such data were available in electronic form. We considered only parasuicides where the individual was a resident of either the Edmonton or Calgary RHA. For each episode of parasuicide, we obtained information on age, sex, date of parasuicide, method used, and residence postal code. Data Linkage We linked census and parasuicide data by assigning the residence postal code of the person who engaged in parasuicide to the corresponding EA, using the Postal Code Conversion File (PCCF) created by Statistics Canada. A limitation of this approach is that some postal codes cover more than one EA, especially in rural areas. To resolve such conflicts, we used an indicator provided in the PCCF that identifies the EA that best matches the postal code. A second limitation of the linkage process was that we could not link postal codes created after the PCCF was developed, such as those for newly developed residential areas. Of the 4388 parasuicides in the combined Edmonton and Calgary files, 4125 (94%) could be assigned to an EA. The remaining 263 entries had either a missing or invalid postal code or a postal code that did not appear in the PCCF. Some individuals were treated for multiple parasuicides during the study period, making it necessary to distinguish between person-based and event-based parasuicide rates: the former considers only the first parasuicide by each person during the study period, while the latter includes all parasuicides. In what follows, we consider only person-based rates. Restricting the data to individuals in the group aged 15 years and over and considering only the first of possibly several events, we found 2899 parasuicides during 1997 (Edmonton 1663, Calgary 1236). Statistical Analysis We calculated age- and sex-standardized parasuicide rates and 95% confidence intervals (CIs), using the combined population of the Edmonton and Calgary CMAs as the standard population (7). We performed an ecologic analysis using weighted linear regression: the dependent variable was the logarithmic transformation of the age- and sex-standardized parasuicide rate in each income region; the independent variables were the sociodemographic variables listed above, along with an indicator variable for CMA. We defined statistical weights as the reciprocals of estimated variances of log-transformed rates (7). For this analysis, there were a total of 20 observations corresponding to the 10 income regions in each of the Edmonton and Calgary CMAs. Weighted linear regression is a classical method of ecologic analysis, but it has the drawback of failing to account for the multilevel (that is, hierarchical or nested) nature of such data as are considered here, where there is information at 3 levels: the individual treated for parasuicide (level 1), the income region wherein the individual resides (level 2), and the CMA within which the income region is located (level 3). We performed a multilevel Poisson regression analysis (8), using the statistical package MLwiN (9). The dependent variable was the logarithmic transformation of age-and sex-specific parasuicide rates for income regions; the independent variables were as above; and, in MLwiN terminology, the “link” was the logarithmic function. ResultsTable 1 gives the characteristics of the Edmonton and Calgary CMAs for the variables under consideration according to the 1996 census. The age and sex distributions are almost identical, but the economic indicators show that the Edmonton CMA was somewhat disadvantaged, compared with the Calgary CMA. In particular, the average income for the Edmonton CMA was 11.4% less than that for the Calgary CMA. In both CMAs, the vast majority of parasuicides involved an overdose (81%), followed by cutting or piercing (13%) (Table 2).
Tables 3 and 4 show the average incomes and age- and sex-standardized parasuicide rates for the Edmonton and Calgary CMAs, by income region. When we compare the Edmonton CMA with the Calgary CMA, the overall rate ratio for parasuicide is 1.30 (2.49/1.91), which is very close to the mean rate ratio for completed suicide (1.37) mentioned in the introduction. For each CMA, there is a clear pattern of decreasing rates with increasing average income (Figure 1). The curves have a somewhat exponential appearance, with a tendency to slope upward at larger average incomes (especially for the Edmonton CMA). At lower average incomes, the rate for the Edmonton CMA is greater than that for the Calgary CMA, with a crossing-over of curves at an intermediate income.
Figure 1 Age- and sex-standardized parasuicide rates
(with 95%CI bars) in the Edmonton and Calgary CMAs,
by average income 
Table 5 presents the best-fitting linear regression model, where average incomes have been centred at $15 000 (that is, $15 000 was subtracted from each average income). The model includes terms for average income, CMA (with the Calgary CMA as the reference), and their interaction term. With R2 = 0.82, much of the variation in the data is accounted for by these few variables. The parameter estimates in Table 5, which are presented as rate ratios, are consistent with Tables 3 and 4 and Figure 1. In particular, the rate ratio from Table 5, comparing the Edmonton and Calgary CMAs, is 2.54, which is precisely equal to the rate ratio for the first income region based on Tables 3 and 4 (8.71/3.43).
The sociodemographic variables for the study (average income, percentage unemployed, percentage postsecondary, percentage married, percentage Native) were all highly positively or negatively correlated. In absolute value, the Pearson correlation coefficients fell in the ranges 0.72 to 0.98 and 0.78 to 0.98 for the Edmonton and Calgary CMAs, respectively. Linear regression models using percentage unemployed, percentage postsecondary, percentage married, or percentage Native as the independent variable in place of average income produced models similar in nature to the one shown in Table 5, so our choice of average income as the independent variable is largely a matter of convenience.
Table 6 gives the best-fitting multilevel Poisson regression model. The parasuicide rate decreases steadily as age increases, and rates are greater for women than for men. Both of these findings are consistent with well-known patterns of parasuicide (10). The rate ratios for CMA, average income, and their interaction term are almost identical to their counterparts in Table 5, showing a striking consistency between the linear regression and multilevel Poisson regression models. In fact, with only 2 CMAs and 10 income regions per CMA, the multilevel Poisson regression model in Table 6 was identical to the standard Poisson regression model in which CMA of residence and average income were imputed to the level of individuals.
DiscussionWe conducted a study comparing parasuicide rates in the Edmonton and Calgary CMAs. The main findings are that the rate ratio for parasuicide is similar to the mean rate ratio for suicide and that, at the ecologic level, much of the variation in the parasuicide rate is explained by average income, CMA, and their interaction term. We found that for each CMA, rates exhibit a decreasing trend as average income increases. In addition, owing to an interaction, rates in the Edmonton CMA are higher than those in the Calgary CMA when average income is low, whereas the reverse is true when average income is high. To our knowledge, the only other ecological study of parasuicide that specifically considered income (as opposed to labour force participation rate, proportion unemployed, or some other economic indicator) was conducted in London, Ontario (11). In that study, a factor analysis identified 16 sociodemographic variables, one of which was “income more than $10 000,” which loaded on the factor “blue-collar familism.” The authors of that paper concluded that “socioeconomic status has a strong negative relationship with self-injury” (11, p 99). We need to emphasize that the observed association between average income and the parasuicide rate in our study is at the ecologic (that is, aggregate or group) level, not at the level of the individual. Accordingly, we can assert that the parasuicide rate is elevated in those areas where average income is low, but we cannot conclude that this finding applies to individuals. That is, it does not necessarily follow that individuals with low income face an increased risk of parasuicide (12). However, such an association is entirely plausible. Based on an extensive review of the literature on unemployment as a risk factor for parasuicide, Platt and Hawton (13) found evidence of an association at both the individual and ecologic levels, although it must be added that not all studies were consistent on this point. We noted above that sociodemographic factors were highly correlated in the CMAs. For average income and percentage unemployed, the Pearson correlation coefficients were 0.80 and 0.78 for the Edmonton and Calgary CMAs, respectively. Platt and Hawton’s conclusions for unemployment (13) suggest that what we observed at the ecologic level for income likely translates to the level of the individual. As we remarked earlier, the choice of average income as an independent variable in the regression analyses was somewhat arbitrary. Morgenstern pointed out that sociodemographic variables tend to be more highly correlated with each other at the ecologic level than at the individual level (12). The implication is that a study corresponding to the present one, but based on individual-level data, might have identified a broader range of sociodemographic variables as being important in a multivariate analysis. Nevertheless, the high degree of ecologic correlation among the sociodemographic variables suggests that it may not be low income per se that underlies an increased risk of parasuicide but, rather, the consequences of belonging to a socially disadvantaged stratum of society. With terms only for average income, CMA, and their interaction, the linear regression model accounted for 82% of the variability in the data. It would have been desirable to have ecologic data on psychiatric variables, but this information is not collected as part of the national census. However, results for major depression from the Canadian Community Health Survey, a national survey conducted during 2000–2001, are available from the Statistics Canada Web site (www.statcan.ca). For “possible” and “probable” major depression, the prevalence rates for the Edmonton RHA are 3.4% and 9.9%, while the corresponding rates for the Calgary RHA are 3.7% and 9.3%. The similarity of prevalence rates suggests that major depression may not play a large role at the ecologic level in explaining differences in parasuicide rates in the Edmonton and Calgary RHAs. Funding and SupportThis study was funded by the Medical Research Council of Canada. AcknowledgementThe authors thank Arif Alibhai for data management and record linkage. References1. Kreitman N, Philip AE, Greer S, Bagley CR. Parasuicide. Br J Psychiatry 1969;115:746–7. 2. Diekstra RFW. The epidemiology of suicide and parasuicide. Acta Psychiatr Scand 1993;71(Suppl 3):9–20. 3. Gould MS, King R, Greenwald S, Fisher P, Schabe-Stone M, Kramer R, and others. Psychopathology associated with suicidal ideation and attempts among children and adolescents. J Am Acad Child Adolesc Psychiatry 1998;37:915–23. 4. Buglass D, Duffy JC. The ecological pattern of suicide and parasuicide in Edinburgh. Soc Sci Med 1978;12:241–53. 5. Maris RW. The relationship of nonfatal suicide attempts to completed suicide. In: Maris RW, Berman AL, Maltsberger JT, Yufit RI, editors. Assessment and prediction of suicide. New York: Guilford Press; 1992. p 362–80. 6. Commission on Professional and Hospital Activities. ICD 9 CM: international classification of diseases, 9th revision, clinical modification. Ann Arbor (MI): Commission on Professional and Hospital Activities; 1978. 7. Newman S. Biostatistical methods in epidemiology. New York: Wiley; 2001. 8. Langford IH, Day RJ. Poisson regression. In: Leyland AH, Goldstein H, editors. Multilevel modelling of health statistics. Chichester (UK): Wiley; 2001. p 45–57. 9. Rasbash J, Browne W, Goldstein H, Yang M, Plewis I, Healy M, and others. A user’s guide to MLwiN, version 2.1. London: Institute of Education, University of London; 2000. 10. Bland RC, Dyck RJ, Newman SC, Orn H. Attempted suicide in Edmonton. In: Leenaars AA, Wenckstern S, Sakinofsky I, Dyck RJ, Kral MJ, Bland RC, editors. Suicide in Canada. Toronto: University of Toronto Press; 1998. p 136–50. 11. Jarvis GK, Ferrence RG, Whitehead PC, Johnson FG. The ecology of self-injury: a multivariate approach. Suicide Life Threat Behav 1982;12:90–102. 12. Morgenstern H. Ecologic studies. In: Rothman KJ, Greenland S, editors. Modern epidemiology. 2nd ed. Philadelphia (PA): Lippincott-Raven; 1998. p 459–80. 13. Platt S, Hawton K. Suicidal behaviour and the labour market. In: Hawton K, van Heeringen K, editors. The international handbook of suicide and attempted suicide. Chichester (UK): Wiley; 2000. p 309–84. Author(s)Manuscript received April 2004, revised, and accepted July 2004. 1. Professor, Department of Psychiatry, University of Alberta, Edmonton, Alberta. 2. Associate Professor, Department of Community Health and Epidemiology, Queen’s University, Kingston, Ontario. Address for correspondence: Dr SC Newman, Department of Psychiatry, Mackenzie Centre, University of Alberta, Edmonton, AB T6G 2B7 e-mail: stephen.newman@ualberta.ca
1 | 2
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||