Background Up until the mid–late 2000s, the national suicide rate in Scotland was the highest among all the UK countries, but the reasons for this phenomenon are poorly understood.
Methods In a multilevel study of suicide risk in Scotland and England during 2001–2006, the authors examined a range of social, cultural and health-related factors at small area level: postcode sector and Health Board in Scotland and ward and Primary Care Organisation in England.
Results Scotland's national suicide rate was 79% higher than in England (rate ratio 1.79, 95% CI 1.62 to 1.98), with younger male and female Scots aged 15–44 years having double the risk compared with their English peers. Overall, 57% of the excess suicide risk in Scotland was explained by a range of area-level measures, including prescriptions for psychotropic drugs, alcohol and drug use, socioeconomic deprivation, social fragmentation, and other health-related indices. The use of psychotropic drugs, acting as a proxy measure for mental ill health, was the variable most strongly associated with the between-country differences in suicide risk. Alcohol misuse also made an important contribution to the differentials. Overall, the contribution of socioeconomic deprivation and social fragmentation was relatively small.
Conclusions Any attempt to reverse the divergent trend in suicide between Scotland and England will require initiatives to prevent and treat mental ill health and to tackle alcohol and drug misuse. Differences in prescribing rates, however, may also be explained by differences in illness behaviour or the availability of psychosocial interventions, and addressing these may also reduce Scotland's excess risk.
- mental health
- drug misuse
- public health
Statistics from Altmetric.com
Until the late 1960s, national suicide rates in Scotland were lower than in England and Wales.1 ,2 The rate in Scotland has, however, risen markedly since then, and up until the mid–late 2000s was the highest of the constituent UK countries. During the same period, there was an overall decline in suicide risk in England and Wales.3–6 The reasons why Scotland has a much higher rate than England are poorly understood. Artefactual factors, arising from between-country differences in ascertainment procedures, are unlikely to be the predominant cause, as any impact would be expected to be fairly constant over time and would not explain the diverging trends in rates between the two countries, especially since the early 1990s.6 ,7 National differences in levels of socioeconomic deprivation have been proposed.8 However, Hanlon et al 9 found that, having adjusted for age, sex and deprivation, excess mortality from suicide in Scotland versus England and Wales rose from 1.2% in 1981 to 15.1% in 1991 and 41.3% in 2001; the equivalent between-country trend in excess mortality from all causes was 4.7% to 7.9% to 8.2%, across the same three annual time points. The mortality gap between Scotland and England, over and above that which is explained by deprivation alone, is commonly labelled the ‘Scottish effect’. Several hypotheses, including socioeconomic and cultural factors arising from deindustrialisation and substance misuse, have been proposed, but there is little robust evidence to explain the underlying causes of this phenomenon.10
We aimed to explain the excess suicide risk in Scotland compared with England during 2001–2006, using a range of area-level indicators and to estimate their relative importance to the intercountry differential in risk. The area-level indicators were grouped under one of six domains: socioeconomic deprivation, social fragmentation, mental ill health, alcohol, drugs and ‘other’. Variables were selected based on evidence from the literature of their relationship with suicide, either at ecological or individual level. Previous research has indicated that area-level socioeconomic deprivation is linked with higher suicide risk,11 ,12 and significant associations have been reported between unemployment and suicide at age 15–44.13 Social fragmentation was also found to be a strong predictor independent of deprivation.12 ,14 ,15 Other reported ecological relationships with suicide include psychiatric morbidity,16 ethnic mix of local population,17 population density,18 population with limiting long-term illness,15 lone parent household15 and healthcare provision.19 People with substance misuse disorders are also found to have higher suicide risk versus the general population.20 ,21 Indicators relating to these themes were explored in relation to their effect on explaining the difference in suicide risk between Scotland and England.
Suicide counts, geographical areas and population denominators
All suicides in England and Scotland were identified using the National Confidential Inquiry into Suicide and Homicide by People with Mental Illness (‘the Inquiry’) database, compiled using data from the Office for National Statistics and the General Register Office for Scotland (GROS).22 All cases for persons aged 15 years and over who died between 2001 and 2006, with a valid usual residential postcode in either country, and whose deaths were classified as intentional self-harm (ICD10 codes X60-X84 and Y87.0) or undetermined intent (Y10-Y34 and Y87.2, excluding Y33.9 (verdict pending)) were included.
The smallest level geographical area units assessed were Census Area Statistics (CAS) wards in England and postcode sectors in Scotland. These are nested within Primary Care Organisations (PCOs) and Health Boards, respectively. In 2001, there were 7969 CAS wards in England (average population =6166), nested within 303 PCOs (average population =163 200). In Scotland, there were 937 whole postcode sectors in 2001 (average population =5402) nested within 15 Health Boards (average population =337 613). In 2006, the numbers of PCOs and Health Boards were reduced due to changes in administrative boundaries; hence, 2006 was the final year of our study period and data were matched to the old boundaries only. Those wards or postcode sectors lying across more than one PCO or Health Board were treated as separate areas under each health administrative boundary, with their populations split accordingly, which yielded 9104 small areas (8093 in England and 1011 in Scotland) for final analysis.
Each suicide with a usual residential postcode in England was assigned to a CAS ward and PCO and in Scotland, to a postcode sector and Health Board. Midyear population estimates for 2001–2006 for CAS wards, PCOs and Health Boards were obtained from Office for National Statistics and GROS. Population data for Scottish postcode sectors were available for 2001 only from the Census. In estimating the postcode sector population for 2002–2006, annual rates of change in the national Scottish population by 5-year age group and sex were calculated for 2001–2006, using midyear population estimates from GROS. These annual rates of change were then applied to the 2001 postcode sector population to project the population for years 2002–2006, inclusive.
A full list of the variables are shown in table 1. Summary statistics and further details can be found in boxes A and B in the online appendix.
Directly age-standardised suicide rates were calculated from population estimates for age 15 and over, using the European Standard Population. To explore the association between the variables and suicide risk, Pearson's correlation coefficients between the variables and standardised mortality ratios (SMRs) were calculated, weighted by ward and postcode sector, or PCO and Health Board, population. SMRs were standardised using age-specific rates calculated from both countries combined.
To assess which combinations of variables best explained the between-country difference in suicide risk, multilevel Poisson regression models were fitted. The aggregated suicide counts for wards or postcode sectors (n=9104) were nested within PCOs and Health Boards (n=318). The multilevel model was implemented using a three-level structure, with wards or postcode sectors at levels 1 and 2 and PCOs at level 3. The use of a pseudo-level, such that each level 2 unit contains exactly one level one unit, has previously been employed as a means of allowing the underlying mortality rate to vary across areas (versus the constant rate assumed by a standard Poisson regression) and, in this manner, to take heterogeneity or overdispersion into account.26 ,27 With only two countries, the country variable was added as a fixed effect instead of a level, using England as the reference category for estimating the rate ratio for Scotland. In addition to modelling all suicides, analyses were carried out separately for men and women and for age groups 15–44, 45–64 and 65 years and over.6 For each model, the expected number of deaths was used as the offset. All multilevel modelling analyses were conducted using MLwiN V.2.22. In selecting which explanatory variables (from the list in table 1) were explored in the multilevel models, single-level Poisson regression analysis was first carried out, using intercooled Stata V.11.1. Using the data for all suicides, each domain was tested individually using backward elimination procedures and a two-sided significance level set initially at p=0.1. Variables that were significant were then tested along with those meeting the same criteria from the other five domains. After checking for multicollinearity, variables that were significant at the p=0.05 level in the fully adjusted single-level ‘all suicides’ model were carried forward to the multilevel models. For all subsequent analyses, the conventional two-sided significance level of p=0.05 was used, and the explanatory variables selected for the all suicides model were also fitted in each of the sex- and age-specific models.
For each significant variable identified from the single-level model, we investigated its contribution in explaining the between-country differences in suicide risk, by fitting univariate multilevel Poisson regression models and assessing the effect of the variable on the rate ratio of the country variable. This was conducted for all suicides and for sex and age subgroups. Similarly, each domain was entered into the multilevel model with the country variable only to assess the contribution of each domain separately. After investigating these effects, fully adjusted models with all significant variables carried forward from the all suicides single-level model were generated. All models were fitted with random intercepts at the ward/postcode sector and PCO/Health Board levels. Analyses using multilevel negative binominal models showed no evidence of overdispersion in the multilevel Poisson models.
The Inquiry was notified of a total of 31 648 suicides during 2001–2006 with a valid residential postcode in England or Scotland. Table 2 shows the number of cases and age-standardised rates by country, sex and age group. Of the 9104 small areas, 85% had at least one suicide. Figure 1 shows the rate ratios for Scotland versus England, stratified by sex and age. Overall, suicide risk in Scotland was 79% higher than that in England (rate ratio 1.79, 95% CI 1.62 to 1.98), with men and women aged 15–44 years showing double the risk. The between-country differences were less pronounced among those aged 65 years and over, with the risks being 40% higher for men and 20% higher for women in Scotland compared with England, the latter difference not being statistically significant (p=0.13).
Table A in the online appendix shows the weighted Pearson's correlation coefficients between SMRs and the assessed explanatory variables for the two countries combined as well as separately. With the exceptions of number of patients per general practitioner (GP) and population who were non-white or born outside the UK, all variables showed a positive correlation with suicide rates.
Factors contributing to the between-country differences in suicide risk
Table 3 shows the rate ratios of the binary country variable in multilevel Poisson regression models with one explanatory variable and country only and in models mutually adjusted for all explanatory variables, for: all suicides, all male suicides and all female suicides. We also investigated the age subgroups; these results are discussed below (data not shown). Table 4 shows the effect of each domain on explaining between-country differentials for all suicides and for the age and sex subgroups (excluding women aged 65 years and over).
In the absence of other explanatory variables, psychotropic medication prescribing patterns had the greatest impact on explaining the excess risk in Scotland, accounting for 42% of the differential (country rate ratio decreased to 1.46). Rate of hospital inpatient discharge with an alcohol-related main diagnosis, a proxy for alcohol misuse, contributed 34% of the differential (country rate ratio decreased to 1.52). Overall, the between-country suicide differential could be most effectively explained by a higher rate of psychotropic drug prescription, included as a proxy for mental ill health, and ‘other’ domains. Much of the differential attributable to the ‘other’ domain was accounted for by variation in the number of patients per GP, used as an indicator for primary care provision. In addition, although socioeconomic deprivation and social fragmentation showed positive associations with suicide, the contribution of these two factors to the between-country differentials was relatively small.
When all variables had been taken into account in the fully adjusted models, 57% of the between-country difference in suicide risk could be explained by the factors that we could examine.
Sex and age subgroups
In the absence of other variables, psychotropic drug prescription accounted for 49% of the between-country risk differential in men and 25% in women. Alcohol misuse, as represented by hospital inpatient discharge with an alcohol-related main diagnosis and liver cirrhosis deaths,28 accounted for 37% of the difference in male suicide risks and 24% for women, while drug misuse (measured by drug-related deaths) accounted for 17% and 19%, respectively. With all variables accounted for in the fully adjusted models, 63% of the between-country differential for men and 43% for women could be explained.
When each variable was investigated separately in the age-specific analyses, psychotropic drug prescription accounted for 59% of the between-country risk differential among men aged 15–44 years and 33% among these younger women. Similar to what we found for all male and female suicides, 38% and 28%, respectively, of the between-country difference among younger men and women was accounted for by alcohol misuse. Among those aged 45–64 years, alcohol misuse explained 47% of the between-country differential among men and 36% of the female differential, while drug misuse accounted for 34% and 32%, respectively. Similarly, 37% and 31%, respectively, of the male and female between-country differences in risk was accounted for by psychotropic drug prescribing patterns. As with all suicides, the ‘other’ domain, in particular variations in the number of patients per GP, accounted for a large percentage of the between-country differentials in almost all sex and age strata.
In the fully adjusted models, 64% of the between-country differential among men aged 15–44 years and 75% among men aged 45–64 years could be explained by the variables investigated compared with 46% and 49%, respectively, among women of the same age groups. The between-country difference was no longer significant for men aged 45–64 years when all variables had been included. Similarly, among those aged 65 years and over, 48% of the excess male suicide risk in Scotland over England was accounted for, and this differential was also no longer significant.
To test whether the effect of variables differed between urban and rural areas, interactions terms between all the variables and population density were fitted for the all suicides and for the men and women all ages suicides models. The great majority of these interaction terms were non-significant, including interaction with the country variable. The addition of these terms also made little difference to the between-country rate ratios.
During 2001–2006, the national suicide risk in Scotland was 79% higher than that in England, with men and women aged 15–44 years having double the risk. Overall, 57% of the between-country differentials could be explained by the area-level factors we could examine. Our models accounted for more of the between-country differential in male than in female risk. The factor most strongly associated with the between-country differences in suicide was psychotropic medication prescribing patterns, included as a proxy measure for population mental ill health, although other factors may also influence prescribing practice (see below). Alcohol misuse was another major contributor to the differentials.
Strengths and limitations
To our knowledge, this is the only published study seeking to explain the marked difference in suicide rates between these two neighbouring countries in recent years, by examining a broad range of area-level indices. Previous studies have tended to focus on England and Wales or Scotland alone, and the few that have made between-country comparisons have inferred, but not tested empirically, explanations for the national differences in suicide risk.2 ,7 ,8 Middleton et al 15 reported that associations between area-level characteristics and suicide rates in England and Wales were generally stronger at the ward versus constituency level. We examined CAS wards and postcode sectors as the lower geographical analytical unit, and we aggregated 6 years of data to maximise statistical power.
Using data from the 2001 Census provided a wide range of area-level measures with good between-country comparability. By utilising data from 1 year only, however, we assumed that these variables stayed constant across the 6-year period. In addition, as population estimates for Scottish postcode sectors were only available for 2001, figures for 2002–2006 had to be projected from the national age- and sex-specific rates of change that occurred across this period, thus making the assumption that rates of change were the same across all postcode sectors.
Overall, we believe that these assumptions were unlikely to have impacted markedly on our main findings and conclusions. Some of the variables used in the study, however, may have measured additional related area-level phenomena unintentionally. For example, psychotropic drug prescribing may be an indicator of the propensity to demand (and receive) medication rather than, or in addition to, serving as a measure of population mental ill health. It may also reflect prescribing behaviour, which tends to vary between individual GPs, or indicate the availability or use of more appropriate psychosocial interventions for common mental health problems. Estimates of the number of people being prescribed psychotropic drugs at small area level, however, have been used as one of the proxy indicators for population mental ill health within the health deprivation domain in both the English and Scottish indices of multiple deprivation.29 ,30 Middleton et al 31 have also previously found that the patterns seen in antidepressant drug prescribing, which constitute the majority of psychotropic drugs prescribed, concurred with temporal changes in other population measures of mental ill health, such as psychiatric admissions and self-harm. In the absence of other robust indicators, we therefore considered prescription for psychotropic drugs and psychiatric inpatient admissions as being reasonable proxy measures for population mental ill health. Due to the issue of comorbidity and because psychotropic medications are not always prescribed according to indication,32 we have also used the combined grouping of psychotropic drugs BNF 4.1, 4.2 and 4.3 in all analyses. Another example is the number of patients per GP, a proxy measure for primary care provision, which may also act as a marker for other unknown area characteristics, and as such, causal inference cannot be made. In addition, all the variables from the mental ill health, alcohol and drugs domains, as well as data on the number of patients per GP, were available only at the health administrative area levels. There may be more sources of heterogeneity within the larger Health Boards (average population 337 613) than the smaller PCOs (163 200) that obscure the observed relationships between suicide and the explanatory factors.33 We do not know exactly how this issue could have impacted on our findings. If there was a degree of measurement error, however, it is likely that attenuation occurred, so that some of the effect sizes reported may be underestimates.
Overall, 57% of the excess suicide risk in Scotland was accounted for by the area-level indices investigated. This leaves over two-fifths of the intercountry difference in risk unexplained. As reported earlier, Hanlon et al 9 reported that, after adjusting for age, sex and deprivation (in the form of Carstairs scores), 41.3% of the excess suicide mortality in Scotland relative to England and Wales remained. This is similar to what we found here. If individual-level risk factors had been available for examination, we believe that the combination of these and the area-level indices would have explained a much higher percentage of the between-country differentials.
We found that the higher suicide rate in Scotland over England is most strongly associated with a greater rate of psychotropic drug prescription. This suggests that in Scotland: (1) there may be a higher prevalence of mental ill health, (2) people with mental ill health may seek medical help more readily or (3) they may be more likely to receive pharmacological treatment when they seek help. Results from the Psychiatric Morbidity Survey in 2000 suggest that prevalence of neurotic symptoms might be slightly higher in England than in Scotland, but this difference was not marked or statistically significant.34 The survey also covered a wide range of disorders, some of which might not be severe enough to require psychotropic medication, and participation and response might have been strongly biased, given the study's sensitive nature. As such, we doubt it could be used as a valid tool to compare population mental ill health in Scotland versus England. Psychiatric inpatient admission, the other proxy for mental ill health we examined, was a non-significant predictor. Although this variable may intuitively be considered as being a stronger measure of population mental illness than psychotropic medication prescription, only a very small proportion of all detected and undetected mental illness results in a psychiatric admission. Admission threshold and availability of beds may also vary widely between health administrative areas. In addition, as indicated in table 1, we have doubts regarding the quality of the English psychiatric inpatient admission data, and any relationship with suicides may therefore have been masked by measurement error.
Alcohol misuse has become another key contributor to the excess suicide risk in Scotland. Liver cirrhosis deaths in Scottish men doubled between 1987–91 and 1997–2001 compared with a rise of 67% in England and Wales.35 Similarly, it has been estimated that the prevalence of problem drug use was 1.8% for Scotland and 1.1% for England, accounting for a third of Scotland's excess mortality over England.36 In our study, drug misuse accounted for 17% of the overall between-country difference in suicide risk. Although it appeared that both the mental ill health and alcohol domains attenuated the country effect to a greater extent for men than for women, we are uncertain as to whether this difference is significant. In agreement with other studies, our research has also found that socioeconomic deprivation and social fragmentation are positively associated with suicide risks,11–15 particularly in Scotland. The contribution of these two factors to the between-country suicide differentials over the study period, however, was relatively small. In addition, further analyses of interaction terms fitted between the explanatory variables with population density (as a proxy measure of rurality vs urbanity) made little difference to our results, and these minor variations did not explain the residual between-country difference in suicide risk.
Attempting to reverse the divergent trend in suicide between England and Scotland will require concerted efforts to prevent and treat mental ill health and promote mental well-being and to tackle alcohol and drug misuse. Prescribing patterns are also likely to reflect illness behaviour or the availability of psychosocial interventions, and addressing these may also contribute to reducing Scotland's excess suicide risk.
What is already known on this subject
The national suicide rate in Scotland has risen markedly since the late 1960s. Up until the mid–late 2000s, it was the highest among the constituent UK countries, while there was an overall decline in suicide in England and Wales during the same period. The reasons why Scotland has a much higher national suicide rate than England are poorly understood.
What this study adds
Greater use of psychotropic drugs, used as a proxy for mental ill health, was the factor most strongly associated with the between-country differences in suicide risk in recent years. Alcohol misuse was another important contributor to the differentials. Any attempt to reverse the divergent trend in suicide between England and Scotland will require concerted efforts to prevent and treat mental ill health and to tackle alcohol and drug misuse.
We are grateful to Miss Cathy Rodway (Centre for Suicide Prevention, Centre for Mental Health and Risk, University of Manchester) for carrying out the pilot study and to Dr Frank Popham (School of Geography & Geosciences, University of St Andrews) for providing advice on the study methodology.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online Appendix
Funding The study was funded by the Chief Scientist Office (CSO) of the Scottish government. The CSO has no role in the study design, collection, analysis and interpretation of data, and in the writing and in the decision to submit the article for publication.
Competing interests None.
Ethical approval The National Confidential Inquiry into Suicide and Homicide by People with Mental Illness had ethical approval from South Manchester Medical Research Ethics Committee and latterly the North West Research Ethics Committee as well as approval under Section 60 (now Section 251) of the Mental Health and Social Care Act.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.