The Avon Longitudinal Study of Parents and Children (ALSPAC) (http://www.alspac.bris.ac.uk) is an ongoing population-based study investigating the effect of a wide range of environmental and other influences on the health and development of children 16. Pregnant women resident in the former Avon Health Authority (which included the city of Bristol), in south-west England, who had an estimated date of delivery between 1 April 1991 and 31 December 1992, were invited to take part, resulting in a cohort of 14 541 pregnancies, 13,796 of whom were singletons or first-born of twins, and who were alive at one year of age. The current manuscript is focussed on the remaining offspring who received the questionnaire aged 16 (n = 9,384), that is they were not lost to follow up. Additional file 1: Appendix A shows the numbers present at each stage. The former County of Avon includes both urban and rural areas, and the extent to which the cohort is representative of children in the UK is discussed elsewhere 16. The parents have completed regular postal questionnaires from pregnancy onwards, and the children have completed questionnaires from approximately 5 years.

Questions about self-harm and suicidal thoughts were included in a self-completion postal questionnaire, sent to study participants when they were aged 16 years (see Additional file 2: Appendix B for the full list of self-harm questions asked). Participants were asked “have you ever hurt yourself on purpose in any way (e.g. by taking an overdose of pills or by cutting yourself)?”, wording which was used in the Childhood Interview for DSM-IV Borderline Personality Disorder (CI-BDP) 17, asked during clinic interviews with the ALSPAC sample aged 11. Those who answered yes were asked further closed response questions regarding frequency (once, 2–5 times. 6–10 times, >10 times), what they did the last time they hurt themselves on purpose (4 response categories) and why they did it that time (6 response categories) - response options were a modified version of those in the CASE questionnaire 18. Participants were also asked whether they sought medical help and how they felt following the most recent time, and whether they had ever seriously wanted to kill themselves when self-harming. The whole sample were then asked whether they had ever felt life was not worth living, wished they were dead and away from it all, thought of killing themselves, or made plans to kill themselves; questions which were drawn from a study of suicidal feelings in the USA 19.

Those who selected ‘other’ for what they did when they self-harmed and why they did it were invited to give further details (see Additional file 2: Appendix B). These free text responses were independently coded by JK and DG, based on the themes emerging from the data 20, for example ‘head butting a wall’ and ‘pulling hair’ were classified into the theme ‘self-battery’ for what was done, and ‘because I was grieving and it made me feel better’ was classified as ‘response to difficulty’ for why it was done. Where appropriate, themes arising from our data were classified using the categories from Hawton et al. in their coding of open responses 15. Where more than one code applied, the response was given as many codes as needed. If the raters did not deem a described action to be self-harm then it was given a code of ‘not self-harm’. All cases that had received discrepant codes were examined and a consensus reached for each. Initial inter-rater agreement was 88% for ‘what they did’ and 72% for ‘why they did it’, but consensus was easily reached once the final coding frame was agreed. The coding frames and discrepant cases were then discussed with JE and GL, and a final coding frame and codes for each case agreed. All categories that emerged in our study had equivalents in Hawton et al’s study 15; there were no discrepancies. In total nine out of 147 free text responses concerning method/reason for self-harm were excluded from the self-harm group following this process (this included six responses that referred to not eating).

Demographic data, collected in previous ALSPAC questionnaires, were included in the analysis, to enable comparison of responders and non-responders, and to identify key psychosocial characteristics associated with self-harm. The following variables were used: i) mother’s social class (manual/intermediate/non-manual), ii) mother’s highest educational level (A level or ‘advanced level’ which are post compulsory schooling qualifications or higher education degree/O level or ‘ordinary level’ which are now defunct examinations taken at the end of compulsory schooling by students deemed more academically able/lower than O level which includes any other qualifications of a lower academic standard or no qualifications at all) as reported when the participant was first born, iii) mother’s score on the Edinburgh Postnatal Depression Scale (EPDS) (which has been validated for use in non-postnatal women 21, and was introduced in ALSPAC during pregnancy then retained for consistency) - using the cut-off point of 12/13 recommended for identifying a depressive illness 22 - when the participant was 11 years old, iv) participant’s gender, v) participant’s ethnicity (white/non-white) as reported by the mother prior to birth, vi) participant’s performance in national examinations undertaken aged 15/16 years in the final year of compulsory schooling (GCSEs/GNVQs) gained from national records (only available for those attending state schools), and vii) participant’s score on the short Moods and Feelings Questionnaire (SMFQ), included within the same questionnaire as the self-harm data. Following Patton et al., a score of 11 or more on the SMFQ was taken as indicative of depressive symptoms 23.

Ethical approval was obtained from ALSPAC’s Law and Ethics Committee, a registered Institutional Review Board and LREC. Information was included at the end of the questionnaire about help sources that participants could contact. Due to ALSAC’s rigorous anonymity rules, the research team could not follow up any participants with regard to their responses. However, participants could request on the questionnaire that a member of the ALSPAC team contact them with regard to the issues raised.

The association between rates of self-harm and demographic data was assessed using χ² statistical tests and univariable logistic regression models. As information on self-harm was not available for the whole ALSPAC cohort (due to those who were lost to follow up, and those who received the questionnaire but did not return it) there was the potential for bias both in the estimated prevalence and also in the association with demographic factors. This was explored using a multiple imputation, more details of which can be found in the section below. A secondary analysis looked in more detail at the reported cases of self-harm. The proportion of respondents who fell into each category for frequency of self-harm, reasons for it, and consequences in terms of subsequent feelings and the need to seek medical help were tabulated overall, and comparisons were made by gender using χ² statistical tests. The characteristics of those who self-harmed with desire to die and those who self-harmed without desire to die the most recent time were compared using χ² statistical tests. Finally, prevalence of suicidal thoughts and plans were examined across the whole sample, comparing those who had ever self-harmed with those who had not, again using χ² statistical tests. This second set of analyses did not involve data imputation as respondents typically had either all or none of the measures of interest. All analyses were carried out using Stata version 10.0.

We assessed the impact of non-response and missing data on our findings using MICE (Multivariate Imputation by Chained Equations) 24 implemented using the ice routine 25 in Stata. This procedure creates multiple copies of the dataset and in each dataset replaces missing data with imputed values, sampled from their predictive distribution 26. The use of this method is based on the Missing At Random (MAR) assumption, namely that conditional on the other data included in the imputation model, there should not be systematic differences between observed and missing values for a given variable. A number of variables were included to assist with the imputation. These included indicators of family adversity at enrolment such as home overcrowding, financial problems, and lack of social support; earlier (more complete) measures of the predictive factors considered in this manuscript such as mother’s social class, and other measures more proximal to the outcome such as maternal and young person’s substance use behaviour in early adolescence, and reported self-harm from an earlier clinic at 11 years. Missing data for the binary measure of self-harm was imputed using logistic models and additional data on demographics employed binary and multinomial logistic models as appropriate. 100 imputed datasets were derived, each entailing 20 cycles of regression switching.

Imputation was carried out on two different samples: i) to estimate values for those among the sample of 4,810 who had information of self-harm but an incomplete set of the demographic predictors described above and ii) to impute missing data for the 4,529 participants (32.8% of the initial cohort) that were sent the relevant questionnaire but did not return it to give a useable sample of 9,268 to enable us to estimate prevalence amongst the cohort members who were sent questionnaires.

It has been suggested that there may be little benefit to imputing data for individuals who lack outcome data as their inclusion in one’s regression model may add nothing but noise 27. To address this as far as possible we included auxiliary variables strongly related to the outcome (as described above), and we paid close attention to the Monte Carlo error associated with both regression parameters and their standard errors.

Almost one in five (18.8%) 16–17 year olds in the ALSPAC cohort reported that they had self-harmed on one or more occasions, with an almost threefold higher prevalence in females than males. Altogether, 5.7% of responders had ever wanted to kill themselves while self-harming. Self-harm was more likely among those with higher depressive symptoms and lower educational qualifications, and those whose mother was in a manual social class and had suffered from depression. When the most recent episode of self-harm was considered, those whose self-harm was associated with a desire to die had higher levels of depressive symptoms and lower educational qualifications compared to those who self-harmed with no desire to die.

The majority who had ever self-harmed had done so at least twice in the last year, with one quarter having self-harmed 6 times or more in that time period. Self-cutting was the most common method in both genders, but girls were more likely to have taken an overdose compared to boys. Those who self-harmed with a desire to die were also more likely to have taken an overdose, compared to those who did not express a desire to die. A range of reasons were given for the most recent self-harm, the commonest being to gain relief from terrible feelings followed by a desire to punish oneself, with one quarter wanting to die. Just under half of those who had ever self-harmed felt better after the last time, although this percentage was lower among those who had wanted to die. Feeling better subsequently was associated linearly with frequency of self-harm; the increasing rewards in terms of positive affect may help explain why many individuals self-harm with greater and greater frequency. Nevertheless, those who frequently self-harmed were a high risk group and almost half had made plans to kill themselves at some point in their lives. Only 12% had ever sought medical help following an episode of self-harm; even amongst those who self-harmed with a desire to die during that episode only one third had sought medical help subsequently, indicating that a high number of suicide attempts are likely to remain hidden from medical services among this age group.

Just under a quarter of respondents had ever thought life was not worth living, approximately 19% had ever wished they were dead, 15% had ever thought of killing themselves and 4% had ever made plans to do so. Females were more likely to have experienced all these levels of suicidal thoughts compared to males. Unsurprisingly, those who had wanted to die the most recent time they self-harmed were the most likely to have thought about and made plans to kill themselves (90.4% and 52.8% respectively), but those who had self-harmed without desire to die the most recent time were at increased risk of suicidal thoughts and plans compared to those who had never self-harmed. The association between self-harm and suicidal thoughts was stronger the more frequent the self-harming behaviour.

This is the first UK-based birth cohort study to look at teenage self-harm and suicidal thoughts and plans, and one of the largest studies internationally to examine these interrelationships in detail. Our results confirm previous findings that much adolescent self-harm, even where there was a desire to die, does not receive medical attention, emphasising the importance of community-based studies for providing information about prevalence. A key advantage of this study is the size and representativeness of the sample at the outset and the detailed prospectively recorded measures of socio-economic circumstances, development and mental health. The loss to follow up over the years may have led to selection bias, as those from the original sample who did not receive the questionnaire were more likely to be male and non-white, to have less than five GCSE/GNVQs at grades A*-C, and to have a mother in a manual social class, with lower educational qualifications, and with depressive symptoms when the individual was aged 11. Further, those who received the questionnaire but did not return it differed from respondents in similar ways. The results from the imputation models indicate that the differences between respondents and non-responders did not have a large impact on prevalence figures for self-harm or risk factor associations. This may be because males were more likely to have missing data yet were less likely to self-harm, whereas those with low educational performance were more likely to have missing data and were more likely to self-harm, so these features of the missing data may have cancelled each other out.

The use of self-report may have encouraged more honest answers than an interview situation 28, although the stigma of self-harm may still have led to underreporting and therefore an underestimation of prevalence. ALSPAC has very strict procedures in terms of anonymity and confidentiality, with which study participants are familiar, which it is hoped will have minimised this possibility. A second problem with using self-report is that participants are defining self-harm for themselves. The free text responses regarding what they did gave some insight into the behaviours that participants included - of the 147 who gave such a response only 9 were excluded. However, for those who did not include a free text response, we were reliant on their own decisions as to what constituted self-harm. Relatedly, respondents who had depressed mood may have interpreted their own behaviour and underlying motivations in more negative ways, which may have introduced bias in the association seen between depressed mood and self-harm, particularly self-harm with intent to die. A third problem, relating to the face validity of the questionnaire, is that approximately 10% of those who had wanted to die during the most recent episode of self-harm had never had thoughts of killing themselves. This indicates that for some respondents ticking ‘I wanted to die’ as a reason for their most recent self-harm did not necessarily mean they were trying to kill themselves - it may be that for this group the self-harm was a reaction to the misery they were experiencing, but was not an attempt to fulfil their desire to die. Therefore the distinction made in this paper between those who wanted to die and those who did not during the most recent self-harm act, although a useful indication of the most serious cases of self-harm, is not exactly equivalent to a distinction between suicide attempts and NSSI.

The prevalence of self-harm in ALSPAC is higher than that reported in the only other community-based study in England of similar size (13.2%) 4. This may be partially explained by a number of methodological differences. Our sample was approximately one year older, and we had a higher proportion of girls (58.9% compared to 42.7% 7), although this has been adjusted for in the imputation models, with little change to prevalence. Further, we used the term ‘hurt yourself on purpose’ – to remain consistent with questions asked of the ALSPAC sample aged 11 during clinic interviews - whereas Hawton et al. used ‘harm yourself’ which may have implied more serious self-harm and led to inclusion of fewer episodes. The earlier study collected data through schools which may have led to greater concerns about confidentiality and higher levels of under reporting, and also to an underestimation of prevalence due to absent pupils 4. However, the increase in prevalence may also reflect a genuine rise in self-harm rates over time as the data were collected approximately eight years apart. There is some evidence that self-harm rates among teenagers and young adults have increased over recent decades so this rise may have continued 329. Further, increases in psychological distress among teenagers in the UK – strongly associated with self-harm in this and other research 5 - have also been reported in some studies 30, although not in others 31.

The prevalence of suicidal thoughts and plans is similar to that in smaller studies from other countries 1314, and indicates that thoughts of killing oneself are as common among teenagers as self-harm. Although the two are associated in this sample, the relationship is complex. The majority of self-harm behaviour appears to be non-suicidal in intent, with approximately three quarters not wanting to die the last time they self-harmed. Conversely, a minority of the sample had never self-harmed but had had suicidal thoughts, indicating that self-harm is not a necessary precursor to or outcome of suicidal thoughts. There was evidence that self-harm behaviour motivated by a desire to die was different from other self-harm, in terms of method used – taking an overdose – and consequences, in that the individual was less likely to feel better, and more likely to seek medical help. Such findings support the distinction between NSSI, which tends to refer to actions involving tissue damage such as self-cutting 26, and self-harm with suicidal intent, which is most closely associated with overdosing. However, a strong association was observed between self-harm without a desire to die the most recent time and suicidal thoughts and plans, with over a third having had suicidal thoughts, and approximately 9% having made plans to kill themselves. This resonates with previous findings that, although individuals who self-harm with suicidal intent are at the highest risk of suicidal thoughts and attempts, those who engage in NSSI are also at significantly higher risk than those who do not self-harm at all 269. Further, this elevated risk of suicidal thoughts and plans increases as the self-harm becomes more frequent, thus those who self-harm the most frequently are potentially at the most risk for suicide 32. The nature and direction of the relationship between self-harm and suicidal thoughts remains unclear. It may be that suicidal thoughts lead the individual to self-harm – either to enact or to reduce the urge 11 - or that self-harm causes psychological distress which then contributes to suicidal thoughts 33, or that the two are co-occurring phenomena caused by a third factor, for example low self-esteem. Furthermore, it has been speculated that NSSI may make suicide/suicidal self-harm more “accessible” to high risk individuals as suicide is seen as a frightening and extreme action, but repetitive engagement in NSSI may make people more courageous and prepared to make suicide attempts 34.

The gender difference in prevalence of self-harm is striking, and replicates other studies of this age group 418. However, these extreme differences reduce in later periods of the life course, resulting in a female prevalence that is only slightly higher than the prevalence for males 3536. This suggests that the gap observed during adolescence may partly be due to males and females being at different developmental stages.