Socioeconomic status and non-fatal injuries among Canadian adolescents: variations across SES and injury measures

The sample included 6967 Canadian adolescents aged 12 to 19 years who participated in the 1996–1997 cross-sectional National Population Health Survey (NPHS) [39], which sampled household residents in all Canadian provinces based on a complex probability sampling design. Specific Research Ethics Board approval was not required for this investigation, which used non-identifiable secondary data.

Demographic and health information was collected for every member of each surveyed household. One member of each household was selected for a more detailed health interview conducted by telephone. The overall household response rate was approximately 83%; over 95% of selected individuals within households responded to the detailed interview [39]. Adolescents in the present study were those selected for a detailed health interview. For a small minority (238 or 3.4%), responses were provided through a proxy interview with another household member.

Socioeconomic information for other members of the household as well as information on total household income was provided by the adolescent or another household member; interviewers were instructed to obtain these data from a knowledgeable household member [40].

Additional socioeconomic data were obtained from the 1996 Canadian census, by linking adolescents' postal codes to census tabulations at the enumeration area (EA) level [41]. EAs, which covered all of Canada, consisted of approximately 125 to 440 households [42].

Following descriptive and bivariate analysis, multivariable logistic regression was used to examine the relationship between each SES measure and each injury outcome in separate models, adjusted for demographic variables that, based on the literature, may be independently associated with injury (age, rural/urban status, geographic region, and whether the adolescent was living with parent/s). Since preliminary analyses revealed gender differences in the observed relationships, all final analyses were conducted separately for males and females. Because of the large number of statistical tests, the interpretation of results focused on identifying consistent trends or gradients across levels of SES, rather than solely on statistical significance.

Since the NPHS used a complex sampling design to yield a representative sample, sampling weights were incorporated in all analyses. Variance estimates for descriptive and regression analyses were adjusted using bootstrap replicate weights to account for clustering [39]. In the bootstrapping approach, the variance of the estimator for a parameter is approximated using a large number of subsamples drawn from the same original sample [44, 45]. Statistics Canada drew 500 such subsamples for the 1996–1997 NPHS, and provided researchers with a set of sampling weights (replicate weights) for each subsample [39]. To apply bootstrap variance estimation, we recalculated each parameter estimate using each set of replicate weights, and applied the variance of the replicate estimates to the statistical models.

The distribution of sampling weights had a wide range and was highly positively skewed, in part because there was a high variability in sampling fractions across provinces. This meant that a small number of adolescents with particularly high sampling weights had the potential to have a large influence on the parameter estimates in statistical models [46]. To reduce bias in the regression coefficients, therefore, influential outliers were identified using regression diagnostic statistics [47, 48]. These influential outliers were excluded from the final weighted multivariable models. Because the aim was to compare results across SES and injury variables, observations excluded from one model were also excluded from comparison models. Fewer than 1.5% of observations were excluded in total, and all final weighted models were free of influential outliers. No observations were identified as influential in any of the unweighted multivariable models, and all excluded outliers had sampling weights that exceeded the 90^th percentile for the overall sample, strongly suggesting that the large sampling weight relative to other observations was the primary reason for their influence.

There were 6826 adolescents in the final sample, after excluding 141 participants due to missing data on key variables. For the analyses including parental SES measures, the sample was restricted to adolescents who reportedly lived with one or more parents (N = 5667), and a further 122 participants missing information on parental education were excluded (N = 5545). A high proportion of adolescents (21.2%) were missing household income information and a substantial proportion of those living with parent/s were missing Blishen SES index data (10.8%). Thus, rather than excluding these participants, "missing" was a separate category for these variables.

Activity-limiting injury was reported by 18.8% of adolescents (Table 2). While 10.9% reported experiencing recreation injury, 9.8% reported non-recreation injury. A higher proportion of males reported injury (21.0%), relative to females (16.5%).

Among males, for the full sample SES indicators (household income, neighbourhood income quintiles, and neighbourhood education quintiles), some analyses yielded statistically significant differences in injury between specific categories of SES and the reference category (Table 3). None of these relationships represented clear gradients, though, and the direction of the association differed across measures of SES. Among males living with parent/s, there was little evidence for an association between parental education and injuries. The parental Blishen SES index, in contrast, was positively related to both total injuries and recreation injuries, although it was not clearly associated with non-recreation injuries.

Among females, there was evidence of a positive relationship between the full sample SES indicators and the injury outcomes (Table 4). The gradient was most apparent for neighbourhood education, particularly in relation to total and recreation injuries. There was also a statistically significant positive association for neighbourhood income and both total injuries and non-recreation injuries. Neither the parental Blishen SES index nor parental education was significantly related to injury among females living with parent/s.

The results of this study suggest that differences related to the measures of SES chosen and the causes of injury under study may both contribute to discrepancies in past research on SES and non-fatal injuries among adolescents.

The most striking result in our study was the variability in the observed relationship between SES and injuries across different measures of SES. This is not entirely surprising, given the discrepant findings in previous studies. No consistent trends in the SES-injury association are apparent in the literature when comparing studies incorporating parental and household SES indicators based on parental occupation [10, 20, 22, 24], parental education [18, 23], household income [23], or adolescent perceptions of family affluence [22, 24]. There is an intriguing pattern, though, whereby studies that have measured SES at the neighbourhood level have tended to observe an inverse relationship between SES and total (rather than cause-specific) non-fatal childhood or adolescent injuries [8, 9, 11‐13, 15‐17], while studies that have focused on individual SES indicators have been less consistent, with several finding no relationship or a positive association between SES and injuries [20, 22‐24]. Interpretation of this pattern is limited by other differences between the two groups of studies, though. For example, studies incorporating area-based indicators of SES have tended to be ecological in nature, comparing population-based rates of health care use for injuries across neighbourhoods or regions [8, 9, 11‐13, 15‐17], while studies relying on individual-level SES measures have tended to be based on injury self-reports among a defined sample of children or adolescents [20, 22‐24] (with some exceptions [10, 18]). These two types of studies (ecological and individual) likely differ in terms of the distribution of included injuries according to severity and cause as well as in their measures of SES, which may help to explain the differing findings (particularly in view of evidence that the observed SES-injury relationship likely depends on the cause of injury [10, 24, 49]). The two groups of studies may also be subject to different types of potential biases (for example, recall bias in studies based on self-reports, and possible biases related to health care system factors in studies based on medical records) [50]. Further, our study allowed for a comparison of the observed SES-injury relationship across both neighbourhood and individual-level SES measures within the same sample, and the findings did not follow the literature trend; any evidence of an association between SES and injury was positive.

Potential explanations for the heterogeneity we observed across SES indicators include conceptual differences in the SES measures (i.e., in terms of the underlying constructs captured), differences in measurement error, or chance. While chance may account for the few statistically significant gradients observed, it seems unlikely to account for the variability across measures.

Regarding measurement error, the level of missing data was relatively high for household income and for the Blishen SES index. In previous analyses with this sample, adolescents who were missing household income information were less likely to be living in higher income neighbourhoods; thus, "missingness" was not completely random [51]. The heterogeneity of the results is still apparent, though, excluding the findings related to household income. The household and parental SES measures were sometimes reported by the adolescent and sometimes by a household adult. Neither SES levels nor correlations among the SES measures differed systematically based on the reporter, however [51].

It seems likely that conceptual differences among the measures of SES account for some of the heterogeneity in our findings. This is supported by pairwise correlations among these measures in the sample, which were generally below 0.6 [51]. The results do not allow us to define which aspects of socioeconomic position are captured by the different SES indicators, or which indicators are most salient for adolescents. For example, at the individual level, income may be seen as an indicator of material resources, while occupation and education may reflect both economics and status or prestige [32]. When parental indicators are incorporated, there is the added complexity that any influence of occupation and education is indirect. It is possible that the positive relationship we observed between the Blishen SES index and injuries among males is a reflection that this indicator is more meaningful for male youth, relative to other SES measures; indeed parental education may have lower relevance for some adolescents because it reflects the more distant past [52]. If additional research supports the positive association, potential explanatory pathways could be explored. For example, we could explore whether the SES index reflects social status among adolescents, and if so, whether social status in turn is associated with injury risk behaviours among male youth.

Correlations between individual-level and neighbourhood indicators of SES for both genders were typically below 0.30 in our study [51], suggesting that they were measuring different constructs, or that the neighbourhood indicators were poor proxies for individual SES. Even when used as proxies for individual status, area-based indicators of SES may capture area-level characteristics [53]; neighbourhood SES indicators may be particularly relevant for injuries due to their potential association with characteristics such as physical hazards [30]. This does not seem to be a plausible explanation, though, for our observation of a positive relationship between the neighbourhood indicators of SES and injuries among females. Again, potential intermediate factors could be explored in future research. For example, neighbourhood SES measures may be proxies for aspects of material resources among adolescents, and female youth with higher material wealth may have greater access to injury risk activities.

There was some evidence that the relationship with SES was different for recreation versus non-recreation injuries. For females, the positive SES-injury gradient was generally more consistent for recreation injuries. Similarly, among males, the parental Blishen SES index was positively related to recreation injuries but not non-recreation injuries. This positive association is consistent with previous research [23‐25, 49]. Behaviours such as overall levels of physical activity could be explored as possible intermediate factors in the relationship between SES and recreation injuries; such behaviours may be associated with recreation injury risk among adolescents [54, 55], and adolescents with higher SES may be exposed to higher physical activity or sports participation levels [e.g., [56, 57]].

Although SES was not consistently related to non-recreation injuries, the grouping together of all such injuries may have limited our ability to detect associations with more specific injury circumstances. Inverse relationships have been observed for SES and both traffic injuries [10, 12, 15] and intentional injuries [10, 26, 27] among children and adolescents.

An important contribution of this study was the comparison of several different measures of SES in relation to injuries within the same adolescent sample. There were some important limitations, however. First, the sample size was not sufficient to examine injury causes beyond the breakdown into recreation and non-recreation injuries. Non-recreation injuries may be relatively heterogeneous with respect to causes and their relationship to SES (for example, as described, the literature suggests that some types of traffic and intentional injuries may be inversely related to SES). Related to this, while physical assault and suicide attempt were included as possible causes of injury in the data captured for the survey, given the interview format and the self-reported nature of these data, it is possible that intentional injuries may have been under-reported, and thus underrepresented in the analysis (approximately 2% of the most serious injuries in the unweighted sample were reported to be intentional). The sample size was also insufficient to explore age differences within the adolescent sample.

It was not possible to determine the severity of adolescent injuries in this study. Previous research has yielded mixed results in terms of SES and injury severity for children and adolescents [11, 12, 24], suggesting that severity could play a role in contributing to inconsistent study findings. The more consistent inverse relationship that has been documented between SES and injury mortality also supports the importance of considering severity [7]. This is also complicated by the notion that injury severity is likely related to injury cause (for example, recreation and non-recreation injuries may differ in severity).

A further potential limitation was that our measures of injury relied on adolescent self-reports. Recall errors have been identified in surveys of childhood injury [50, 58], and recall bias has been proposed as a potential explanation for positive relationships in studies of SES and injuries [23, 59]. Although this warrants exploration in terms of the positive relationships observed, it seems unlikely that such bias explains the heterogeneity in our findings.

Additional indicators of SES that may be worth exploring in relation to adolescent injury include measures of adolescents' own social position [2, 5], subjective evaluations of SES [e.g., [38]], and measures of deprivation [e.g., [60]]. Subjective SES measures were not available for this study; deprivation measures were excluded because they may not reflect a range of SES levels.

Finally, a multi-level modeling approach to the analysis of SES and injuries may help to distinguish between individual and neighbourhood-level associations. Our focus on separately exploring individual and neighbourhood SES indicators reflected our aim of identifying possible reasons for heterogeneity among the findings of previous studies. Our results reveal that additional development work is necessary at each level of analysis (individual and contextual), to further elucidate the meaning of indicators of adolescent SES and how they relate to injuries. Multi-level modeling approaches will be an important component of this work, and in particular for area-level SES measures, will be useful for distinguishing between aspects of the SES-injury association that are related to characteristics of neighbourhoods themselves, and those that are related to the aggregate characteristics of individuals living within neighbourhoods. In terms of neighbourhood characteristics, it may be informative to explore which particular aspects of the social and physical environment (for example, traffic hazards or crime) are contributors to injury risk, and how SES variables capture these characteristics. In light of recent attention being given to issues such as neighbourhood income distribution and health [61], it may also be worthwhile to explore whether the degree of SES homogeneity within neighbourhoods is related to injury, or whether the role of individual SES in contributing to injury risk is modified by neighbourhood characteristics. This may be especially interesting given that in our sample, household income and area-based income quintiles variables were only modestly correlated (Spearman rank correlation of approximately 0.3) [51].