Introduction
Abnormal body weight has been associated with poor mental health and poor quality of life (McLaren et al., Reference McLaren, Beck, Patten, Fick and Adair2008; Audureau et al., Reference Audureau, Pouchot and Coste2016). Obesity has also been associated with a high level of stigma (Wang et al., Reference Wang, Brownell and Wadden2004; Seacat et al., Reference Seacat, Dougal and Roy2016), which is a risk factor for poor mental health (Lillis et al., Reference Lillis, Levin and Hayes2011). Therefore a higher risk of suicidal behaviour for those with higher than normal body mass index (BMI) might be expected. However, evidence from prospective cohort studies is inconsistent with this hypothesis, with a lower risk of suicide or attempted suicide found for people who are overweight or obese in most (Magnusson et al., Reference Magnusson, Rasmussen, Lawlor, Tynelius and Gunnell2006; Kaplan et al., Reference Kaplan, McFarland and Huguet2007; Mukamal et al., Reference Mukamal, Kawachi, Miller and Rimm2007; Bjerkeset et al., Reference Bjerkeset, Romundstad, Evans and Gunnell2008; Batty et al., Reference Batty, Whitley, Kivimaki, Tynelius and Rasmussen2010; Mukamal and Miller, Reference Mukamal and Miller2010; Jee et al., Reference Jee, Kivimaki, Kang, Park, Samet and Batty2011; Gao et al., Reference Gao, Juhaeri, Reshef and Dai2013; McCarthy et al., Reference McCarthy, Ilgen, Austin, Blow and Katz2014; Sorberg et al., Reference Sorberg, Gunnell, Falkstedt, Allebeck, Aberg and Hemmingsson2014) but not all studies (Elovainio et al., Reference Elovainio, Shipley, Ferrie, Gimeno, Vahtera, Marmot and Kivimaki2009; Chang et al., Reference Chang, Wen, Tsai, Lawlor, Yang and Gunnell2012; Schneider et al., Reference Schneider, Lukaschek, Baumert, Meisinger, Erazo and Ladwig2014). Evidence from cohort studies on the risks of suicide or attempted suicide for underweight people suggests that they have a higher risk relative to individuals who have a healthy weight (Magnusson et al., Reference Magnusson, Rasmussen, Lawlor, Tynelius and Gunnell2006; Kaplan et al., Reference Kaplan, McFarland and Huguet2007; Batty et al., Reference Batty, Whitley, Kivimaki, Tynelius and Rasmussen2010; Jee et al., Reference Jee, Kivimaki, Kang, Park, Samet and Batty2011; Chang et al., Reference Chang, Wen, Tsai, Lawlor, Yang and Gunnell2012; Gao et al., Reference Gao, Juhaeri, Reshef and Dai2013; McCarthy et al., Reference McCarthy, Ilgen, Austin, Blow and Katz2014; Sorberg et al., Reference Sorberg, Gunnell, Falkstedt, Allebeck, Aberg and Hemmingsson2014). However, there have only been three studies that have separately estimated the risks of suicide or attempted suicide in women, and these studies have been underpowered to look at death by suicide (Kaplan et al., Reference Kaplan, McFarland and Huguet2007; Jee et al., Reference Jee, Kivimaki, Kang, Park, Samet and Batty2011; Gao et al., Reference Gao, Juhaeri, Reshef and Dai2013).
It is possible that the association between BMI and suicidal behaviour is confounded by psychiatric disorders such as depression, which is associated with weight loss and is implicated in suicidal behaviour (Mukamal and Miller, Reference Mukamal and Miller2010). An examination of the association with body size in early life, before the onset of most psychiatric disorders, could throw some light on this hypothesis, but to our knowledge no other study has used measures of body size at different stages of the lifecourse. The association could also be confounded by factors that may raise the risk of suicidal behaviour, such as deprivation and alcohol misuse, as well as by factors that may lower the risk, such as having a partner. Most previous studies have adjusted for these factors. However, having had children could also lower the risk of suicide (Kjaer et al., Reference Kjaer, Jensen, Dalton, Johansen, Schmiedel and Kjaer2011) and this has not been adjusted for in studies of individuals who are older at baseline. It has also been proposed that the lower risk of suicide or suicide attempt for overweight individuals could be due to their choice of less lethal methods of suicide, either because a highly lethal method such as hanging requires more physical exertion, or because a larger body mass would reduce the likelihood of dying from self-poisoning (Mukamal and Miller, Reference Mukamal and Miller2010; McCarthy et al., Reference McCarthy, Ilgen, Austin, Blow and Katz2014; Perera et al., Reference Perera, Eisen, Dennis, Bawor, Bhatt, Bhatnagar, Thabane, de Souza and Samaan2016). However, few studies have been able to test this hypothesis.
We investigated the association between BMI in midlife and suicidal behaviour separately for attempted suicide and suicide, after excluding major prior illnesses and adjusting for many potential confounders, in a large prospective cohort of women in the UK. This study has accrued about double the number of cases of attempted suicide and of suicide in women than any other prospective study. We were therefore also able to compare the method of suicide employed by women of different BMI and to examine the long-term association between BMI and attempted suicide using reported body size at younger ages.
Methods
Study population and definitions
In total, 1.3 million women, aged 50–64 years old, were recruited to the Million Women Study in 1996–2001 in England and Scotland at the time that they were invited for routine NHS breast screening. At that time, the NHS Breast Screening Programme routinely invited women aged 50–64 years for their mammographic screen, by sending each individual a letter offering them a date and time at a specific screening centre. In 66 (about half) of the screening programme's centres, Million Women Study questionnaires were included in the invitation letter, and approximately 50% of eligible women returned a questionnaire. Thus, the Million Women Study recruited one in four of UK women aged 50–64 years in 1996–2001, with most born between 1935 and 1950. The recruitment questionnaire contained questions on health conditions and socio-demographic, lifestyle and reproductive factors and forms the baseline for the main analysis in the current study. Participants were sent a re-survey questionnaire approximately 3 years later, with a response rate of 65%; this included questions on birth weight, relative body size at 10 years old and 20 years old, recent change in diet, happiness, self-rated health and marital status. The respondents gave written consent to follow-up and the study has ethical approval from the Multi-Centre Research Ethics Committee for Anglia and Oxford (REC 97/5/001). Questionnaires and further details of the data access policies can be viewed online at http://www.millionwomenstudy.org.
The records of study participants were linked electronically using their unique National Health Service (NHS) number and date of birth to routinely collected NHS databases for cause-specific deaths, emigrations, cancer registrations and hospital admissions. Information provided to the investigators included the date of each event, with diagnoses coded to the 9th and 10th Revisions of the World Health Organisation's International Classification of Diseases (ICD).
The outcomes for these analyses were: (i) attempted suicide: first hospital admission after baseline coded as ‘intentional self-harm’ (ICD-10 code X60–X84); and (ii) suicide: death with the underlying cause coded as ‘intentional self-harm’ (ICD-10 code X60–X84) or as ‘undetermined intent’ (ICD-10 code Y10–Y34). It is customary in the UK to combine deaths recorded as intentional self-harm and those of undetermined intent as suicides because it is recognised that the majority of deaths of undetermined intent are probable suicides (Office for National Statistics, 2017). ICD-9 codes (E950–59) were used to identify hospital admissions for attempted suicide prior to the study baseline for exclusion purposes, as participants were not asked about prior suicide attempts in the recruitment questionnaire. However hospital admission data were only available from 1st April 1997 for participants in England (Hospital Episode Statistics) and from 1st January 1981 for participants in Scotland (Scottish Morbidity Records).
At recruitment, women were asked for their current weight and height, and these were used to derive BMI in kg/m2, categorised as: <20, 20–24.9, 25–29.9, ⩾30 kg/m2. In a validation study where weight and height were measured in a sample of the cohort about 9 years after recruitment, the regression dilution ratio between BMI based on self-reported and measured weight was 0.95 (Wright et al., Reference Wright, Green, Reeves, Beral and Cairns2015). In order to minimise regression dilution bias, we adjusted for measurement error in analyses by taking the average BMI in each category as that measured nine years later, using data from 3564 women (MacMahon et al., Reference MacMahon, Peto, Cutler, Collins, Sorlie, Neaton, Abbott, Godwin, Dyer and Stamler1990).
Statistical analysis
Cox regression models, using time since recruitment as the underlying time variable, were used to estimate hazard ratios, described here as relative risks (RRs), of first suicide attempt and of suicide by categories of BMI at recruitment. Women with unknown BMI (n = 62 610) and women with a hospital record of prior suicide attempt (n = 2335) were excluded from the analyses. We also excluded women who had reported previous heart disease, stroke or cancer at recruitment (n = 98 818) as these conditions may affect weight. The remaining participants were followed from recruitment until 31st March 2015. For attempted suicide, person-years were calculated until the date of first hospital admission for attempted suicide, death, emigration or end of follow-up, whichever came first. For suicide, person-years were calculated until death, emigration or end of follow-up, whichever came first. Follow-up for this cohort is virtually complete: only 1% of participants has been lost to follow-up, mainly through migration, and those participants contributed person-years up to the date of loss.
The BMI category 20–24.9 kg/m2 was used as the reference group, and 95% group-specific confidence intervals (g-s CIs) were calculated using Plummer's method and presented in figures and tables (Plummer, Reference Plummer2004). This allows confidence intervals (CIs) to be estimated for each group, including the reference group, so that any two groups can be compared. Conventional 95% CIs are described in the text when comparisons are made only with the reference group.
The Cox models were stratified by single year of birth (⩽1930, 1931, 1932,…,1949, ⩾1950) and single year of recruitment, and adjusted for the following variables all reported at recruitment: geographical region (London and Southeast, South West, Midlands, Northern England, Scotland); area deprivation (tertiles, based on the Townsend index, a score incorporating census area data for employment, car ownership, home ownership and household overcrowding) (Townsend et al., Reference Townsend, Beattie and Phillimore1988); educational attainment [in three categories: tertiary qualifications (college or university), secondary qualifications (A levels or O levels usually obtained at 18 or 16 years of age, respectively) or technical qualifications (nursing, teaching, clerical or commercial), no educational qualifications] (Floud et al., Reference Floud, Balkwill, Moser, Reeves, Green, Beral and Cairns2016); parity (nulliparous, parous); smoking status (never, past, current); alcohol intake (<2, 2–6, ⩾7 drinks/week); strenuous exercise (rarely/never, once per week, 2–7 times per week); self-reported treatment for depression or anxiety (yes, no). Missing data for each adjustment variable (<6% for each variable) were assigned to a separate category.
Using χ2 tests, heterogeneity was examined across subgroups defined by length of follow-up, self-reported treatment for depression or anxiety, happiness, self-rated health, change of diet in past 5 years and marital status. The latter four variables were reported for the first time at the 3 year re-survey on average 3.3 (s.d. 1.1) years after recruitment.
We investigated whether BMI was related to method of suicide, by comparing the distribution of cases of attempted suicide and of suicide across BMI categories according to three types of methods: either (i) self-poisoning (ICD-10 X60–X69 and Y10–Y19), or (ii) self-injury using low lethality methods (ICD-10 self-cutting by sharp or blunt object X78, X79, Y28, Y29), or (iii) self-injury using high lethality methods (ICD-10 hanging/asphyxiation X70, Y20; drowning X71, Y21; firearms X72–X74, Y22–Y24; explosives X75, Y25; smoke, fire, flames X76, Y26; steam, hot vapours, hot objects X77, Y27; jumping X80, X81, Y30, Y31; crashing motor vehicle X82, Y32). Diagnoses codes X83–X84 and Y33–Y34 were excluded from this analysis because the method is not specified.
Using Cox regression, we also estimated RRs of attempted suicide and suicide in relation to birth weight and body size in childhood and early adulthood (10 years old and 20 years old) reported by participants at the 3-year re-survey, adjusted for the other variables reported at recruitment and with and without adjustment for BMI in midlife. Birth weight had two categories: <2.5 kg, ⩾2.5 kg. Relative body size at age 10 years had two categories: thinner than peers and plumper/average compared with peers. Clothes size at age 20 years used smallest UK clothes size reported among 8 or less, 10, 12, 14, 16 and 18+ to make two categories: small (size < 12), medium/large (size 12 or larger). These self-reported measures were found to be strongly correlated with birth weight and with BMI which had been measured at age 10 and at age 20 in a subgroup of women (n = 541) who also participated in the National Survey of Health and Development (1946 birth cohort) study (Cairns et al., Reference Cairns, Liu, Clennell, Cooper, Reeves, Beral and Kuh2011).
Three sensitivity analyses were conducted: using the WHO categories for BMI (<18.5, 18.5–24.9, 25–29.9, 30+ kg/m2), using finer categories for BMI (<20, 20–22.4, 22.5–24.9, 25–27.4, 27.5–29.9, 30–32.4, 32.5–34.9, 35+) and excluding deaths of undetermined intent (ICD-10 code Y10–Y34).
All analyses used Stata 14.1 (StataCorp, College Station, TX, USA) and figures were drawn in R (R Core Team, 2013).
Results
In total, 1 155 651 women with known BMI and without known prior suicide attempt, cancer, heart disease or stroke were eligible for analysis. Among these women, the average age at baseline was 56.0 (s.d. 4.8) years. The proportion with low BMI (<20 kg/m2) at baseline was 4% and the proportions of overweight (25–29.9 kg/m2) and obese (⩾30 kg/m2) women were 36% and 17%, respectively (Table 1). During a follow-up of 15.7 (s.d. 2.6) years on average, 4930 women attempted suicide and 642 died by suicide, at mean ages 63.4 (s.d. 7.0) years and 64.0 (s.d. 6.7) years, respectively.
Table 1. Characteristics of Million Women Study participants by BMI, and follow-up information
Percentages are calculated based on women with information for that specific variable.
a Denotes information collected 3.3 years after baseline.
Low weight women
Compared with women with BMI of 20–24.9 kg/m2, women with lower BMI at baseline were more likely to live in less affluent areas, be current smokers, be nulliparous and do no strenuous exercise (Table 1). They were also more likely to report being treated for depression or anxiety and, 3 years after baseline, they were more likely to rate their health as fair or poor and to report being unhappy (Table 1).
In prospective analyses, women with BMI lower than 20 kg/m2 had a higher RR of attempted suicide (adjusted RR = 1.38, 95% CI 1.23–1.56, p < 0.0001) and a higher RR of suicide (adjusted RR = 2.10, 95% CI 1.59–2.78, p < 0.0001) than women with a BMI of 20–24.9 kg/m2 (Fig. 1). The RR for suicide was significantly greater than for attempted suicide (test for heterogeneity between the two RRs p = 0.007). Adjustment for parity, deprivation, education, strenuous exercise and alcohol made little impact on the risk estimates, whereas there was some evidence of confounding by treatment for depression/anxiety and by smoking (online Supplementary Table S1). The excess risk either of attempted suicide or of suicide in women with a low BMI did not differ by duration of follow-up, self-reported treatment for depression or anxiety at recruitment, reported happiness, self-rated health, change of diet in past 5 years or marital status (Fig. 2).
Fig. 1. RR of attempted suicide and suicide by BMI.
Fig. 2. RRs of attempted suicide and suicide in various subgroups; in low weight women compared with women of BMI 20–24.9 kg/m2.
Having a smaller body size at 10 and 20 years old was associated with an increased risk of attempted suicide in middle age (adjusted RR = 1.20, 95% CI 1.10–1.30, p < 0.0001 and adjusted RR = 1.15, 95% CI 1.05–1.25, p = 0.002 respectively) compared with being larger at those ages; lower birth weight was also associated with an increased risk but the association was not statistically significant (adjusted RR = 1.14, 95% CI 0.99–1.31, p = 0.07; Table 2). Adjustment for BMI in midlife did not affect the estimates for the relation between smaller body size at younger ages and risk of attempted suicide. Numbers of deaths by suicide by body size in early life were too small for a reliable estimation of risk.
Table 2. RR of attempted suicide by birth weight and body size at age 10 and 20 years
a RRs and 95% CIs adjusted for geographical region, deprivation index, parity, education, strenuous exercise, alcohol, smoking, treatment for depression or anxiety. Stratified by year of birth and year of recruitment.
b RRs adjusted as in RRa with additional adjustment for BMI in midlife.
Where information about the methods of suicide or attempted suicide was available, 95% of women who attempted suicide used self-poisoning methods, and this did not vary greatly across categories of BMI (online Supplementary Table S2). For suicide, however, women with lower BMI were more likely to use self-poisoning as the method of suicide (53%, 28 out of 53) compared with women with BMI of 20–24.9 kg/m2 (37%, 100 out of 271; p = 0.03), and less likely to use methods of high lethality (43%, 23 out of 53, v. 61%, 164 out of 271, respectively; p = 0.02).
Overweight and obese women
Overweight and obese women were more likely to live in less affluent areas and do no strenuous exercise than women with BMI of 20–24.9 kg/m2, and were also more likely to report being treated for depression or anxiety (Table 1). Three years after baseline, they were more likely to rate their health as fair or poor and to have changed their diet in the past 5 years (Table 1).
In prospective analyses, women who were overweight or obese had lower RRs of attempted suicide and suicide compared with women with BMI of 24–24.9 kg/m2, but effect estimates were smaller in comparison with the increase in risk of suicide found for women with low BMI <20 kg/m2 (Fig. 1). For suicide, women who were overweight had an adjusted RR of 0.81 95% CI (0.68–0.98) and women who were obese had an adjusted RR of 0.74 (95% CI 0.58–0.95), with a statistically significant trend (test for trend p = 0.006). However the RRs of attempted suicide did not decrease with increasing BMI (test for trend p = 0.09): overweight women had an adjusted RR of 0.89 (95% CI 0.83–0.94) whereas obese women had an adjusted RR of 0.96 (95% CI 0.88–1.04). Adjustment for treatment for depression/anxiety had the greatest impact on the risk estimates (online Supplementary Table S1).
We combined overweight and obese women into one category and compared them with women with BMI 20–24.9 kg/m2 to examine differences in the associations in certain subgroups (Fig. 3). The RRs in women who were overweight and obese did not vary significantly by subgroup except that there was a lower risk of attempted suicide largely restricted to those who had not reported being treated for depression or anxiety at baseline (p = 0.005 for difference between those who reported and did not report being treated for depression or anxiety).
Fig. 3. RRs of attempted suicide and suicide in various subgroups; in overweight or obese women compared with women of BMI 20–24.9 kg/m2.
To enable comparison with other cohorts, we used the WHO classification of BMI in a sensitivity analysis (online Supplementary Table S3) and the results were similar to the main results except that the estimated risks for suicide attempt in the underweight group (<18.5 v. 18.5–24.9 kg/m2) were slightly higher than in the main analyses (where we used <20 v. 20–24.9 kg/m2). We also conducted a sensitivity analysis using finer categories of BMI to investigate the shape of the associations (online Supplementary Table S4). This analysis showed that the higher risks of attempted suicide and suicide were apparent in women with BMI of 20–22.4 kg/m2, and also showed that the lower risks for overweight and obese women were not consistently below the null as BMI increased. The results from the sensitivity analysis excluding deaths of undetermined intent did not differ markedly from the main results except that the estimated RR in the obese group was slightly lower than in the main results (online Supplementary Table S5).
Discussion
In this large prospective study of women in the UK, the small proportion (4%) who were of low BMI (<20 kg/m2) at baseline had a two-fold greater risk of suicide over the next 16 years than women with a BMI of 20–24.9 kg/m2 and a 38% greater risk of attempting suicide. Evidence suggests little, if any, bias due to reverse causation, as the higher risks did not differ by length of follow-up. There appeared to be some confounding by depression and smoking but not by any other factor adjusted for. Small body sizes at age 10 and at age 20 were also associated with increased risks of attempted suicide in later life, suggesting that this association was not a recent effect. The age at onset of eating disorders is 18 years old on average and therefore the effect of eating disorders, which are known to be risk factors for suicidal behaviour, could account for the higher risk of suicide in women who are of low weight in midlife (Kostro et al., Reference Kostro, Lerman and Attia2014; Volpe et al., Reference Volpe, Tortorella, Manchia, Monteleone, Albert and Monteleone2016). Unfortunately, we did not have information on eating disorders, although the association with body size at age 10 and, possibly with low birth weight, both before the onset of most eating disorders, suggest that other factors may be important.
We have compared our results with those of other studies that have reported results separately for women. Our finding that attempted suicide was associated with low BMI in adulthood agrees broadly with a study in the UK using The Health Improvement Network (THIN) database involving women aged 18 and over, of whom 782 had attempted suicide over a 7 year period (Gao et al., Reference Gao, Juhaeri, Reshef and Dai2013). The THIN study did not report the risk of suicide separately for women, but two other prospective studies, the US National Health Interview Survey (NHIS) and the Korean Cancer Prevention Study (KCPS) reported higher risks of death by suicide for women who were underweight, as we found (Kaplan et al., Reference Kaplan, McFarland and Huguet2007; Jee et al., Reference Jee, Kivimaki, Kang, Park, Samet and Batty2011). However both studies were underpowered, with only 261 and 83 suicides in women in the NHIS and KCPS, respectively, and the estimates from these studies were not statistically significant. We found a significant difference between the risk for low weight women of suicide and the risk of attempted suicide. Evidence from previous studies does not aid interpretation about the reason for this difference.
When we examined the method of suicide, women of low weight were more likely to die by self-poisoning than women in other BMI categories, suggesting that the excess risk in these women might perhaps be due to similar doses of drugs being more fatal in women of lower body mass (Perera et al., Reference Perera, Eisen, Dennis, Bawor, Bhatt, Bhatnagar, Thabane, de Souza and Samaan2016). There was no support for the hypothesis that women of low weight were choosing more lethal methods, such as hanging or jumping, which require more physical exertion (McCarthy et al., Reference McCarthy, Ilgen, Austin, Blow and Katz2014; Perera et al., Reference Perera, Eisen, Dennis, Bawor, Bhatt, Bhatnagar, Thabane, de Souza and Samaan2016). The finding regarding self-poisoning may have implications for clinical decision-making with respect to the dose of potentially toxic medications prescribed for women of low BMI, in whom self-poisoning may be particularly dangerous.
The large proportion of women (53%) who were overweight or obese was at a somewhat lower risk of attempting or dying from suicide than women with BMI of 20–24.9 kg/m2. This finding is difficult to interpret as the risk did not decrease consistently with increasing BMI, as could be seen when finer categories of BMI were used in the sensitivity analysis. The THIN study also reported that risk of attempted suicide did not decrease with increasing BMI but that for men and women combined the risk of suicide decreased with increasing BMI (Gao et al., Reference Gao, Juhaeri, Reshef and Dai2013), and both the NHIS and KCPS studies reported risk of suicide decreasing with increasing BMI. It is not clear why the effect of BMI should differ between attempted suicide and suicide. It is also not clear why the lower risks of attempted suicide were strongest in overweight and obese women who were not being treated for depression or anxiety, although this was also found by the THIN study (Gao et al., Reference Gao, Juhaeri, Reshef and Dai2013). Given our findings and the inconsistent results from studies of mixed and men-only cohorts (Magnusson et al., Reference Magnusson, Rasmussen, Lawlor, Tynelius and Gunnell2006; Kaplan et al., Reference Kaplan, McFarland and Huguet2007; Mukamal et al., Reference Mukamal, Kawachi, Miller and Rimm2007; Bjerkeset et al., Reference Bjerkeset, Romundstad, Evans and Gunnell2008; Elovainio et al., Reference Elovainio, Shipley, Ferrie, Gimeno, Vahtera, Marmot and Kivimaki2009; Batty et al., Reference Batty, Whitley, Kivimaki, Tynelius and Rasmussen2010; Mukamal et al., Reference Mukamal, Rimm, Kawachi, O'Reilly, Calle and Miller2010; Chang et al., Reference Chang, Wen, Tsai, Lawlor, Yang and Gunnell2012; Gao et al., Reference Gao, Juhaeri, Reshef and Dai2013; McCarthy et al., Reference McCarthy, Ilgen, Austin, Blow and Katz2014; Schneider et al., Reference Schneider, Lukaschek, Baumert, Meisinger, Erazo and Ladwig2014), it remains possible that residual confounding by unmeasured factors may explain the somewhat lower risks associated with being overweight and obese.
The strength of the current study is that it includes one in four of all UK women born in 1935–1950, making it the largest ever study of women's health. Other strengths are the prospective design, the long and virtually complete follow-up, and the ability to adjust for many potential confounding factors. The large number of events of both attempted suicide and of suicide has allowed an examination of the association within categories of BMI, rather than treating BMI as a continuous variable which can distort any non-linear associations. The size of the study also meant that we were able to exclude women with pre-existing diseases which may have affected weight. However, there are some limitations. Firstly, the measures of psychiatric morbidity available to us were self-reported treatment for depression or anxiety at baseline and self-rated health and happiness at the 3 year resurvey, and so not all forms of psychiatric morbidity would have been captured. Secondly, the self-reported body size variables at younger ages (birth weight, relative body size at age 10 years, and clothes size at age 20) were retrospective rather than prospective. However, they were validated against prospectively collected measurements of BMI in a subsample of the cohort, and so misclassification is unlikely (Cairns et al., Reference Cairns, Liu, Clennell, Cooper, Reeves, Beral and Kuh2011). Thirdly, use of hospital admission records for attempted suicide would have missed some cases since not all women who self-harm are admitted to hospital (Clements et al., Reference Clements, Turnbull, Hawton, Geulayov, Waters, Ness, Townsend, Khundakar and Kapur2016), but any misclassification would tend to dilute estimates of RRs. Lastly, within accidental deaths there could be some ‘hidden suicides’ and therefore some suicides may have been missed, but analyses of confirmed suicides with and without the inclusion of deaths of undetermined intent suggests that any misclassification of cases would not have had much effect on the results.
In conclusion, low weight is associated with a substantial increased risk of suicidal behaviour, particularly death by suicide, for women in their 50s and 60s. Smallbody size in childhood and early adulthood is also associated with increased risk of suicide attempt. For women who are overweight or obese, the risks of suicidal behaviour are somewhat reduced and inconsistently related to level of BMI and so residual confounding by unmeasured factors cannot be excluded.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291718003239.
Acknowledgements
The authors thank the women recruited to the Million Women Study and the staff of collaborating NHS Breast Screening Centres and General Practices. We also thank NHS Digital in England (http://www.digital.nhs.uk) and the Information Services Division in Scotland (http://www.isdscotland.org) for linked health data.
Million Women Study Collaborators
The Million Women Study Advisory Committee are: Emily Banks, Valerie Beral, Lucy Carpenter, Carol Dezateux, Jane Green, Julietta Patnick, Richard Peto and Cathie Sudlow.
The Million Women Study Co-ordinating Centre staff are as follows: Hayley Abbiss, Simon Abbott, Rupert Alison, Krys Baker, Angela Balkwill, Isobel Barnes, Valerie Beral, Judith Black, Roger Blanks, Kathryn Bradbury, Anna Brown, Benjamin Cairns, Andrew Chadwick, Dave Ewart, Sarah Floud, Kezia Gaitskell, Toral Gathani, Laura Gerrard, Adrian Goodill, Jane Green, Lynden Guiver, Carol Hermon, Darren Hogg, Isobel Lingard, Sau Wan Kan, Nicky Langston, Kirstin Pirie, Alison Price, Gillian Reeves, Keith Shaw, Emma Sherman, Rachel Simpson, Helena Strange, Sian Sweetland, Ruth Travis, Lyndsey Trickett, Anthony Webster, Clare Wotton, Lucy Wright, Owen Yang and Heather Young.
The following NHS breast screening centres took part in the recruitment and breast screening follow-up for the Million Women Study: Avon, Aylesbury, Barnsley, Basingstoke, Bedfordshire & Hertfordshire, Cambridge & Huntingdon, Chelmsford & Colchester, Chester, Cornwall, Crewe, Cumbria, Doncaster, Dorset, East Berkshire, East Cheshire, East Devon, East of Scotland, East Suffolk, East Sussex, Gateshead, Gloucestershire, Great Yarmouth, Hereford & Worcester, Kent (Canterbury, Rochester, Maidstone), Kings Lynn, Leicestershire, Liverpool, Manchester, Milton Keynes, Newcastle, North Birmingham, North East Scotland, North Lancashire, North Middlesex, North Nottingham, North of Scotland, North Tees, North Yorkshire, Nottingham, Oxford, Portsmouth, Rotherham, Sheffield, Shropshire, Somerset, South Birmingham, South East Scotland, South East Staffordshire, South Derbyshire, South Essex, South Lancashire, South West Scotland, Surrey, Warrington Halton St Helens & Knowsley, Warwickshire Solihull & Coventry, West Berkshire, West Devon, West London, West Suffolk, West Sussex, Wiltshire, Winchester, Wirral and Wycombe.
Financial support
The Million Women Study is funded by the UK Medical Research Council (grant no. MR/K02700X/1) and by Cancer Research UK (grant no. C570/A16491). GG is funded by the Department of Health. AF was funded by the National Institute for Health Research (NIHR), Programme Grant for Applied Research Programme (Grant Reference Number RP-PG-0610-10026). KH is an NIHR Senior Investigator and is funded by Oxford Health NHS Foundation Trust. Funders had no role in study design, data collection, analysis or interpretation, manuscript preparation, or the decision to publish. The corresponding author confirms that she had full access to the data in the study and had final responsibility for the decision to submit for publication.
Conflict of interest
The authors declare no competing interests.
