Results

CRUDE NUMBERS.

At KCH 1341, and at LH 1011 persons attended for DSH. Of these 2352 consecutive attenders, 244 had no address recorded and 38 lacked information on ethnicity and address. Of the 2070 with recorded address, 178 lived outside the study area. Of the remaining 1892, 188 had ethnicity unrecorded, leaving 1704 subjects. Overall 97% of study area residents presented at their own A&E. This did not differ between ethnic groups. Given the randomness of inter area migration with respect to ethnicity, the main analysis used those only whose ethnicity was recorded and who presented at their own A&E (n=1643) (table1).

ECOLOGICAL ASSOCIATIONS

Crude DSH attendance rates were lower in wards with higher deprivation levels (RR per SD deprivation 0.84 (0.78 to 0.91); p<0.001). This association did not differ between areas (LRI χ² (1)=0.3; p=0.600), age groups (LRI χ²(2)= 1.2; p=0.550), genders (LRI χ² (1)=0.5; p=0.542) and ethnic groups (LRI χ² (2)=4.0; p=0.138). After adjustment for these variables the RR was 0.87 (0.80 to 0.95); p<0.001.

Crude DSH rates were lower in wards with relatively more Asian residents. There were no crude associations between wards' DSH rates and their minority and African-Caribbean populations densities respectively.

DENSITY EFFECTS

Linear associations of DSH attendance rates with minority, African-Caribbean and Asian population densities respectively were adjusted for confounding and effect modification by deprivation, area, and attenders' gender, age and ethnicity. Rates were higher in area 1 than area 2, but, given this difference, patterns were similar in both areas. Links of local minority and ethnic densities with DSH rates differed between White and non-White groups, the regular pattern being that, as local minority and ethnic population sizes increased, White rates rose more or declined less than of members of those respective minority or ethnic groups. However, the fit of these fully adjusted linear models improved significantly by inclusion of squared and cubed density indices, suggesting curvilinear, rather than straight linear relations between ethnic/minority density and the (relative) risk of DSH in the respective groups (table 2).

As the same exposure (local minority or ethnic density) held different associations with minority and White DSH, it follows that a given ethnicity may appear a risk factor in one and a protective factor in another context. Figures 1 to 3 plot DSH rates in specific groups relative to Whites using the ratio of relative rates. The linear model yields a steady decline of DSH rates in given ethnic groups relative to the White group with increasing ethnic density, while the better fitting curvilinear model yields an inverted U shaped curve with maximum relative rates in wards with minority/ethnic densities around a standard deviation below the study area mean.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

DSH rates in non-Whites relative to Whites by local minority density.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

DSH rates in African-Caribbeans (AC) relative to Whites by local AC density.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

DSH rates in Asians relative to Whites by local Asian density.

SENSITIVITY OF RESULTS TO MISSING DATA AND INTERCATCHMENT AREA MIGRATION

Similar proportions of the White (207 of 1717=12%), African-Caribbean (35 of 353= 10%), Asian (2 of 42=5%) and undetermined ethnic groups (38 of 240=16%) lacked information on address (including the undetermined group χ² (3)=6.9; p=0.080. Excluding undetermined group χ² (2)=3.2; p=0.197).

Analyses, rerun under the assumption that those who had been excluded because of unrecorded ethnicity (n=188) had all been White or non-White respectively, did not change results, RRRs being comparable, in size and significance to those given in figs 1-3. The same applied when data were reanalysed including those who attended the other A&E (n=61) (results of sensitivity analyses available from JN).

Discussion

Risk factor research is fraught with contradictory findings.1 In case of ethnicity this is often attributed to imprecise classification of ethnic group.21 This study of associations between DSH and ethnicity, identifies a more general explanation for the confusion. A given ethnicity implies membership of a group and, separately, a degree of minority status dependent on that group's local size (density). When these are considered separately, ethnic DSH rates, relative to White groups, range from values larger than 1—indicating risk—, to smaller than 1—suggesting protection. The manner in which ethnic density at the small area level modifies the link between ethnicity ands DSH risk is only partly compatible with the density9 and status integration10 hypotheses; as reported elsewhere in connection with divorce as suicide risk factor,11 risk does not decline steadily as the risk factor's prevalence increases, but rather curvilinearly.

Absolute DSH rates were not calculated as a sizable proportion of DSH in the community escapes medical attention.22 The study's focus was on comparing relative rates. The diverging pattern of associations between ethnic/minority densities and DSH rates of White and non-White groups suggests that ethnic individuals' local minority status modifies their DSH risk. However, the results could also have arisen if rates of help seeking following DSH were lower in larger ethnic groups, for instance because of better developed lay referral systems in them.23 24 Such help seeking bias is unlikely to fully explain the results, particularly as the density effect not only applies to minority groups but to White groups also. Moreover, the results fit a previous study, conducted in the same area, documenting density effects on risk of completed suicide, an outcome that surfaces irrespective of helpseeking.12 Some other circumstances support the validity of the findings. The focus was on calculations of RRs within, and their comparison between ethnic groups. Missing data and inter area migration were random with respect to the variables of interest. Findings applied in both areas despite independent data collection. More importantly, the calculation of ethnic specific relative rates and ratios of relative rates between ethnic groups, cancels out error factors because of missing data, unbalanced migration and ethnic differences in help seeking.

In small area analyses, individual instances of the outcome frequently concentrate in adjoining neighbourhoods.25 Negative binomial regression adjusts for this phenomenon to which two factors may have contributed. Firstly, DSH clusters not only because of high prevalence of risk factors in communities but also because of imitation.26 Moreover, attempters living near the casualties may have been more likely to present than those living elsewhere so that clustering may have arisen centrally in the areas.

Electoral wards, the base for indexing ethnic density, varied in size from around 5000 to 15 000 residents. Especially in larger wards, the ethnic mix may have varied between neighbourhoods. Analyses at a finer level, such as that of enumeration districts, would have reduced this heterogeneity but were not feasible as deprivation and census data by ethnic group, age and gender, are unavailable at levels lower than that of wards. However, the fact that density effects were apparent despite residual ethnic heterogeneity within wards, emphasises this phenomenon's robustness.

Attenders were asked to describe their own ethnic group following census procedure.17 The extent of missing data illustrates the difficulties of this in busy inner city A&E departments. The decision to analyse White, African-Caribbean and Asian groups will attract criticism given the large heterogeneity within these groups and the practical difficulties of taking mixed ethnic origin into account. However, our aim was not to obtain ethnic DSH rates but to examine density effects, which are based more on the notion of (dis)similarity than exact matching of ethnic descriptors.

Factors other than ethnicity and suicidal behaviour have been examined in the literature, supporting the notion that one's fit with one's social environment modifies one's risk of adverse outcomes. Associations of religious belief with depression27 or suicide acceptance28 depend on the local dominance of religion. Suicide risk associated with alcohol use is lower when overall consumption is high.29 DSH rates in the unemployed are lower when unemployment is common.30 Relative deprivation, which sets one apart from one's neighbours, may contribute more to suicide risk than absolute poverty levels.31 This may explain our failure to find positive associations between deprivation and DSH rates, in contrast with previous reports.32 The study area is one of the most deprived in the UK. When a risk factor becomes more prevalent it explains fewer individual cases,33 because, in such situations, personal exposure, for instance to deprivation, is less likely to be aggravated by marginalisation or stigmatisation. This possibility, another example of a density effect, is suggested by the data but could not be confirmed, as information on personal deprivation was unavailable.

The density/status integration hypotheses predict a steady decline of risk as the risk factor in question becomes more prevalent and thus less associated with role strain. The inverted U shaped curves found in this study suggest a more complicated process. Not only social causation—increased strain associated with minority status—but also social selection may have contributed to the effect described. For instance, migration of ethnic minority people into more affluent, predominantly White, may indicate increased resilience; this may outbalance the effects of being isolated from one's own ethnic group in the lowest ethnic density wards. Longitudinal, individual level studies are needed to disentangle how opposing processes such as these contribute to variation of ethnic DSH risk according to the demography of small neighbourhoods.

The RR is widely used as a measure of aetiological force. The expectation that it should be stable across populations34has caused neglect of the phenomenon described here. At most, density effects tend to be considered as nuisance factors in ecological studies.8 Our study suggests that the extent to which exposure to either risk or protective factors is a minority experience, strongly modifies RRs. These therefore cannot be interpreted without knowledge of the exposure's prevalence in original study areas. Health policy decisions based on assumptions of invariant relative risks across large regions, may be misinformed.