Background
Suicidal
ideation (or suicidal ‘thoughts’) and
attempts are known to be strongly associated with completed suicide [
1] and contribute to a large burden of disease in terms of adverse psychosocial impacts, long-term disability and increased health care costs [
2,
3]. In high income countries, men outnumber women in suicide deaths [
3] whereas suicide ideation and attempts are found to be far more common in women [
4,
5].
The prevalence of attempted suicide can be studied either by the use of register studies which include all suicide attempts treated in the health-care system, or by population based surveys involving questionnaires or interviews [
6]. Most previous research focusing on suicide attempts in Sweden has involved register data [
7,
8]. These studies show that between the mid 1990’s and the early 2000s, there was a substantial increase of attempted suicides among women aged 15 to 24 years whereas no increase was observed among women 25 years or older [
7,
8]. In their study, Bogdanovica and her colleagues [
8] found that among women aged 15–24 years, suicide attempts rates rose from 324.9 per 100,000 in 1989 to 369.4 per 100,000 in 2003. Since 2007, suicide attempt rates have decreased among young women, although they are still higher than in 1987 [
9].
While national registers provide extremely valuable data, they lack information about suicide attempts that do not lead to the use of health care services. It has previously been observed that only 50–60% of all suicide attempts are known to the health care systems [
6]. Thus, far too little attention has been paid to potential period changes of self-reported suicide attempts among women in general, and in young women outside the health care setting in particular. In addition, no previous study in Sweden has investigated whether the prevalence of self-reported suicidal ideation among young women has increased during the last decades. Studying trends in suicidal ideation is important due to its association with suicide attempts and completed suicide [
1], but also in the light of an increasing concern that the mental health of young women is deteriorating [
7]. To our knowledge, only one previous population based study in Sweden made a comparison of the prevalence of self-reported suicidal ideation over time [
10]. This study presented prevalence figures for men and women as a composite group and showed a decrease in lifetime suicidal ideation between 1986 and 1996 from 33.3 to 21.1% [
10]. Although research performed in other high-income countries used repeated measures of suicidal ideation over time, most of these studies have used prevalence figures for men and women as a composite group [
11,
12]. These studies did not find any significant change in suicidal ideation between 1990/92 and 2001/03 [
12] and 2001/02 and 2011/12 respectively [
11]. However, one previous U.S study reported a decrease in lifetime suicidal ideation among women between 1991 and 1992 and 2001–2002 (from 8.2 to 7.1%), while lifetime prevalence of self-reported suicide attempts remained unchanged during this period [
13].
The etiology of suicidal ideation is multifactorial and a wide range of sociodemographic factors seem to be of importance, either directly or indirectly by influencing individual’s susceptibility to suicidal ideation [
3,
14]. Since the effect of sociodemographic factors on suicidal ideation may be altered through policies and public health interventions [
14], it is important to include these factors in research in this field. Earlier population-based studies focusing on women, have demonstrated associations between suicidal ideation and younger age [
15], lack of a stable relationship [
16,
17], lack of a stable employment [
16] and lower educational attainment [
16]. Several studies also suggest that university students are a potential risk group [
18‐
20]. However, while these studies have reported on the association between sociodemographic factors and suicidal ideation, only a few studies have investigated the association between sociodemographic factors and suicidal ideation across time among women from the general population [
10]. Temporal trends of disease occurrence and risks among different social groups in a population are likely to be dependent on time and place [
21]. Thus, historical or social events that occur in a given time period, such as an economic recession may alter the prevalence of suicidal ideation within specific risk groups [
14]. This indicates the need to conduct stratified analyses in order to identify temporal changes in the relationship between sociodemographic factors and suicidal ideation in specific risk groups.
The aim of this study was to assess the prevalence of self-reported suicidal ideation and attempts over a 26 year period (1989–2015) in two groups of women from the general population aged 20–30 and 31–49 years. A further aim was to investigate associations between sociodemographic factors and suicidal ideation over this period.
Methods
Study design and data collection
This study is part of the four-wave, longitudinal population-based project titled “Women and alcohol in Gothenburg” (WAG) initiated in 1986. The project involved a two-stage stratification procedure with a) an initial postal screening questionnaire (Screening Women and Alcohol in Gothenburg, SWAG), followed by b) structured personal interviews with a stratified sample in the second stage. This two-stage procedure is relevant in epidemiological studies focusing on uncommon conditions such as alcohol-related disorders in women, for which this study was initially designed [
22]. The main purpose of this two-stage procedure was to increase the number of individuals with alcohol-related problems, while keeping the numbers of interviews at a reasonable level [
22].
The SWAG questionnaire was developed for screening of alcohol-related problems, containing 13 items with a four choice answering mode. This screening questionnaire has been published and described more in detail in a previous article by Spak and Hällström [
22]. For the stratification procedure, each question was dichotomized with a negative statement scoring 0 points and a positive statement scoring 1 point, making the maximum total score 13 points. Based on the scores obtained in the SWAG, women were selected for face-to-face interviews by a three category, stratified random sampling. In the first wave, the answers were grouped into three categories: 0 point (no alcohol related problem), 1–3 points (possible alcohol related problems) and ≥ 4 points (probable alcohol related problems) [
23]. Out of these groups, all respondents with a SWAG score of ≥4 points, a random quarter of those with 1–3 points and a random one-fifteenth of those who scored 0 points were invited for interview [
23]. In the second and third wave, the stratification groups consisted of ≥5, 1–4 and 0 points [
24]. The increased cut-off level of ≥5 points was based on an increased alcohol consumption observed among younger women [
25]. In the two first waves, a random selection of those who had not responded to the SWAG questionnaire were invited for interview in order to increase the numbers and statistical power. In the fourth and latest wave, due to low response rate in SWAG, all women born in 1993 who returned the SWAG questionnaire were invited to the interview and no stratification procedure was applied.
Those included in the stratified sample were invited by letter to take part in an interview, followed by written reminders and if necessary, telephone calls to non-responders. The same procedure was applied for participants invited to follow-up interviews. Respondents who preferred not to participate in a long face-to-face interview were offered either a shorter version of the interview, a telephone interview or in earlier waves (1989–1991, 1994–1998, 2000–2002), a postal questionnaire. Since questions about suicidal ideation and attempts were not included in the shorter version of the interview, this study includes data from long interviews only.
Study population
The study population includes women born in 1965, 1970, 1975, 1980 and 1993 who were at the time of the respective examination registered in Central and Western districts in Gothenburg, Sweden’s second largest city. In order to increase the number of participants in the fourth wave of the study, women born in 1993 who were registered residents in Northern and Eastern districts were also invited to participate.
The overall participation rates at each screening and interview wave are presented in Table
1. In 1986, the first SWAG questionnaire was mailed out to all women (
n = 673) born in 1965, with a response rate of 67.9%. A stratified random sample of 128 respondents were invited for interview and between 1989 and 1991, 74.2% (
n = 95) of the women completed a long baseline interview. In the second wave in 1994–1998, 2910 screening questionnaires were sent out to all women born in 1970 and 1975 with a response rate of 77.2%. Out of the stratified sample, 543 women participated in their first long interview between 1994 and 1998. In 2000, the screening questionnaire was mailed to 1103 women born in 1980 with a response rate of 75.2%. Out of the stratified sample, 284 long baseline interviews were performed between 2000 and 2002. Finally, in the fourth wave in 2013, the screening questionnaire was sent out to 1687 women born in 1993, out of which 33.9% responded to the questionnaire. After excluding those who refrained further participation (
n = 4), all 568 women who had responded to the SWAG were invited to take part in the interview and 171 women completed a long baseline interview between 2013 and 2015. All women who took part in a baseline interview were invited to follow-up interviews in subsequent waves. In total 2072 baseline and follow-up interviews were performed between 1989 and 2015. The four waves of data collection are henceforth labelled as W1 (performed during 1989–1991), W2 (1994–1998), W3 (2000–2002) and W4 (2013–2015).
Table 1
Screening and interview stages of WAG 1986 to 2015: study population, weighted and unweighted numbers and prevalence (%)
1965 | 1986 | 673 | 457 (67.9) | 128 | 95 (74.2)a | 352 | 85 | 349 | 53 | 216 | 31 | 165 |
1970, 1975 | 1995 | 2910 | 2247 (77.2) | 829 | NA | | 543 (66.0)a | 1891 | 415 | 1604 | 244 | 920 |
1980 | 2000 | 1103 | 829 (75.2) | 491 | NA | | NA | | 284 (57.8)a | 485 | 151 | 259 |
1993 | 2013 | 1687 | 572 (33.9) | 568 | NA | | NA | | NA | | 171 (30.1)a | NAb |
Analysis on differences between long and short interviews
An analysis on the difference between those who completed the long and short interviews showed that women who had completed the long interviews were older, had a higher educational attainment and had a somewhat higher alcohol consumption compared to those who had completed the short interviews.
Attrition analysis
An analysis performed in a previous study [
26] showed no difference in sociodemographic variables (age, marital status, number of children, education and employment status for women and her partner) between those who did and did not respond to SWAG. This type of attrition analysis was possible to perform in the first to waves, when women from the stage 1 attrition group were invited for interview in stage 2, thus answering to a range of background factors during the interview. Telephone interviews with non-responders in previous waves indicated shortage of time as a common reason for declining participation [
27].
The interview
Interviews were conducted after obtaining oral, informed consent in W1-W3 and written informed consent in W4. The interviews were performed by health care professionals and social workers with several years of work experience. A psychiatrist trained the interviewers in the use of the interview questionnaire, as well as classification of psychiatric conditions in accordance with the Diagnostic and Statistical Manual of Mental Disorders (DSM III-R and DSM-IV). A psychiatrist with extensive clinical experience was available for consultation regarding the diagnostic procedures. The interview included questions about sociodemographic factors, childhood conditions, family relations, sexual abuse, intimate partner violence, work-related questions, alcohol consumption, mental health problems and suicide-related behaviors. Structured interviews were conducted either face-to-face at the respondent’s home or at the University of Gothenburg, or by telephone. Interviews lasted for about 1.5 to 3 h.
Variables
Suicidal ideation and attempts
Questions about suicidal ideation and attempts were based on Paykel et al. [
28] and Meehan et al. [
29] using the following three questions: (1) ‘Have you
ever had thoughts of taking your life, even if you would not really do it? (suicidal ideation), (2) Have you
ever reached the point that you seriously considered taking your life, and perhaps made plans how you would go about doing it? (suicidal ideation), (3) Have you
ever made an attempt to take your life? (suicide attempt). During W2 and W3, the women in the follow-up interviews were asked if they had experienced such thoughts or attempts
during the past 5 years (instead of
ever). This time frame was changed in the follow-up interviews in W4 to
‘ever’, to be consistent with the baseline interviews. All women were asked at both baseline and follow-up if they had experienced questions 1 to 3 above
during the past 12 months.
In this study, a positive response to either question 1 or 2 during the past 12 months was considered as experiencing suicidal ideation during this time period. A positive response to either question 1 or 2 for ever/ during past 5 years was considered as experiencing suicidal ideation earlier in life. In the next step, changes in prevalence were analyzed using lifetime experience of suicidal ideation which was considered when a woman gave a positive response to either of the two merged variables past 12 months or earlier in life. The same procedure was applied for question 3 (suicide attempts). If values were missing on both past 12 months and earlier in life, the value was coded as missing. If a negative response (‘no’) on one of the questions and a missing value on another, the answer was coded as ‘no’.
Sociodemographic factors
The following sociodemographic factors were included at the four waves of data collection: education level (≤ 9 years, 10–12 years (high school) and > 12 years). Relationship status was divided into three groups: (1) married, cohabiting, registered partnership (2) widowed, single, never married, non-cohabiting partner and (3) divorced, separated. For the purpose of this paper, all women in category 1 were categorized as ‘married/ cohabiting’ and those in category 2 were categorized as ‘single’. Current occupation was tricotomized: (1) those who were working half-time or more (‘employed’), (2) homeworkers, unemployed, women who responded ‘not working because of other reasons’ as well as those on disability pension or sickness absence exceeding 3 months (‘unemployed’), (3) women answering that they were studying half-time or more (‘students’). Participants on parental leave were categorized based on their occupation prior to parental leave.
Statistical analysis
Analyses were carried out in SPSS 24 using the Complex Samples Plan. This type of analysis adjusts for weights and
“…stratification of the sampling design to produce unbiased national estimates of population means and frequencies from the sample after taking into account weights for over-or undersampling of specific groups” pp. 232 [
30]. Descriptive statistics were presented with unweighted and weighted prevalence (%) and 95% confidence intervals (CI) regarding the sampling fractions according to the scores obtained in SWAG. Since no randomized selection based on the SWAG scores was performed with the cohort born in 1993, no weights for oversampling of alcohol related problems were applied. To test for significant differences in prevalences (Δ) of lifetime suicidal ideation and attempts, 95% CIs were computed in W1 and W4 for women aged 20–30 years, and in W3 and W4 for women aged 31–49 years. [
31]. Bivariable associations between each sociodemographic factor and lifetime suicidal ideation were estimated using logistic regression with weighted odds ratios (OR) and 95% CI.