The Scottish Government

« Previous | Contents | Next »

Listen

2. METHODS

2.1 Data

2.1.1 Deaths data

An anonymised dataset of intentional self harm and undetermined deaths occurring over the period 1989 to 2004 was provided by the General Registrar Office for Scotland ( GROS) .

For the purposes of this study, a suicide death is defined to comprise both those deaths which are officially classified as suicide/intentional self harm (ICD9 E950-959; ICD10 X60-84) and also 'undetermined' deaths (ICD9 E980-989; ICD10 Y10-34) . The inclusion of 'undetermined' deaths is in line with accepted international research practice: 'undetermined' deaths are often considered to be probable suicides, whereas suicide/intentional self harm deaths are labelled definite suicides. Appendix A1 includes a discussion of the possible impact of using two different ICD systems on the reliability of the classification of suicide.

The dataset is confined to deaths among adults aged 15+ years. There were 13185 deaths recorded over the period 1989-2004, of which 74% percent occurred in males (n=9759) and 26% percent in females (n=3426) . (An additional 42 deaths among adults were excluded from the analysis: 41 cases did not have council codes and one death did not have an age code.)

Deaths among children under 15 years of age are omitted from the analysis because, first, suicide is only rarely recorded in this age group, and, second, some of the deaths labelled undetermined will result from uncertainty between accident and homicide (rather than between accident and suicide) .

Data have been provided for each death on a range of variables, as listed in table 2.1. The dataset permits description of this sample of deaths by key socio-demographic characteristics.

With the exception of 189 deaths which did not have Output Area codes, each death record was assigned by the research team to a 2001 Census Output Area and also to a Consistent Area Through Time ( CATT) . The CATT enables small areas to be reliably compared using data from the 1981, 1991 and/or the 2001 Censuses (Exeter et al. 2005) . Pseudo health boards were derived from these CATTs ( see appendix A2 for technical details) , thus permitting comparison between health board areas.

On the basis of residential address (postcode sector) , the research team assigned the Carstairs deprivation score to each death ( see appendix A3 for technical details) . The original Carstairs index of deprivation was divided into 7 disproportionate categories, based on a normal distribution curve, in order to make comparisons between the least and most deprived areas in Scotland. In this study, we also divided the continuous score into categories for comparative analyses. At national level we used population weighted ¹⁷ quintiles (five categories) , while at local level population weighted terciles (three categories) were used. Thus, in 1991 and 2001, each quintile comprised approximately 800,000 adults aged 15 years and older , while each tercile comprised approximately 1.3 million adults aged 15 years and older .

Table 2.1 Socio-demographic variables included in the suicide dataset

2.1.2 Population data

In order to conduct the epidemiological analysis, population (denominator) data were required for all key variables, at both national and local (council and health board) area level. Population data were obtained from the mid-year population estimates (from GROS) and from the 1991 and 2001 censuses. Mid-year population estimates provide denominators by sex and age for all suicide rates and standardised mortality ratios ( SMRs ¹⁸) , while the census data provide denominators used to calculate suicide rates by social class. Denominators for social class were obtained from the census data: the 1991 census provides population denominators for the period 1989-1995 and the 2001 census provides population denominators for the period 1996-2002.

2.2 Data analyses

The software package STATA was used for all data management and analysis. Graphs were prepared using Microsoft Excel.

2.2.1 Three year moving average of suicide rates

A moving average is a useful indicator to show the temporal trend when data are subject to marked fluctuations from one time period to another. The underlying trend becomes more discernible through the 'smoothing' process. A three-year moving average of crude suicide death rates was computed separately for males and females in each local area and displayed against the national moving average. The moving average was constructed by centring on the mid-year. For example, from 1990 to 1992, the crude rates for 1990, 1991, and 1992 were calculated and the average was taken as the moving average suicide death rate of 1991. Crude rates were used because they give a more readily interpretable measure of the burden of mortality than standardised rates and the population structure did not change significantly over the relatively short time period covered by the study.

2.2.2 Suicide rates by social class

National suicide rates by social class (at death) were computed for males during two time periods (separately) : 1989-1995 (based on 1991 census) and 1996-2002 (based on 2001 census) . This analysis was not undertaken for women due to the high proportion of female suicide deaths that were not assigned to a substantive social class category. The main reason appears to be non-involvement in the labour market.

The Registrar General's Social Class ( SC) has been used as the measure of socio-economic classification in this study. There are five SC categories, with one divided into two subgroups. These categories are:

I Professional etc occupations
II Managerial and technical occupations
IIIN Skilled non-manual occupations
IIIM Skilled manual occupations
IV Partly-skilled occupations
V Unskilled occupations.

Technical details relating to the use of the SC scheme can be found in appendix A4.

2.2.3 Ratio of male: female age-adjusted rates

The indirect standardisation ¹⁹ method was employed to adjust male and female suicide death rates. Age (over 15 years) was assigned to a ten-year age group (15-24, 25-34, 35-44, 45-54, 55-64, 65-74) or an open-ended category (75 and over) , as appropriate.

The expected numbers of deaths were calculated for each local area using the death rates at the national level. The standardised mortality ratio ( SMR) was calculated by dividing the observed number of deaths by the expected number of deaths. SMRs permit the comparison of suicide incidence in a given local area against a common standard (the whole of Scotland) . The age-adjusted rate was obtained by multiplying SMR and crude death rate. In order to identify the ratio of male: female suicide rates, it is necessary to adjust for differences in the age structure of male and female populations.

2.2.4 Methods of suicide

Methods of suicide were aggregated into eight categories (Platt, et al, 1988) from detailed categories of ICD9 and ICD10, which were attached to death records.

The methods are:

• Hanging
• Firearms/shooting
• Jumping
• Cutting
• Poisoning/overdose
• Gassing
• Drowning
• Others (not elsewhere classified) .

Rates were calculated at both national and local area level, for the whole time period and two separate periods (1989-95, 1996-2002) .

2.2.5 SMR analysis

SMRs and 95% confidence intervals ²⁰ were calculated (Gardner and Altman, 1989) for each local area, over the whole period and for two separate time periods. SMRs were standardised around population structure in 1996-2002 in order to make comparisons between the two periods.

2.2.6 Age analysis

At national level and local level, suicide death rates by age were calculated. At national level, seven age groups (15-24, 25-34, 35-44, 45-54, 55-64, 65-74, 75 and over) and three age groups (15-44, 45-64, 65+) were used. At local level only the three age groups were used. Rates were computed for the whole period and also for 1989-1995 and 1996-2002. In order to indicate the uncertainty (imprecision) around these rates, 95% confidence intervals were computed. Analyses were conducted separately for males and females.

2.2.7 Area deprivation analysis

At national and local (health board and council) levels, we used age and sex specific deaths from 1996-2002 as the numerator while the denominator was obtained from the 2001 Census. This standard population was used for the calculation of suicide SMRs for the 1989-1995 and 1996-2002 periods, to enable comparisons of the suicide trends over time. Note that, if the observed deaths were equal to the expected deaths, the SMR would equal 100.

An established literature shows that suicide is increasing among young adults, and decreasing among older adults. Evidence also suggests that temporal trends in suicide differ among males and females. Therefore, in the analysis of Scottish trends we calculated SMRs and 95% confidence intervals for males, females and all persons, aged 15 years and older , 15 to 44 years, and 45 years and older . At local level, we calculated SMRs and 95% confidence intervals for all persons, aged 15 years and older , 15 to 44 years, and 45 years and older . (More fine-grained analysis was ruled out as a result of small numbers.) SMRs were calculated for population weighted quintiles for Scotland, while SMRs were calculated for population weighted terciles (three categories) for local areas.

« Previous | Contents | Next »