Articles. Funding UK Department of Health.

Transcription

1 Causes of death up to 1 years after admissions to hospitals for self-inflicted, drug-related or alcohol-related, or violent injury during adolescence: a retrospective, nationwide, cohort study Annie Herbert, Ruth Gilbert, David Cottrell, Leah Li Summary Background Emergency hospital admission with adversity-related injury (ie, self-inflicted, drug-related or alcohol-related, or violent injury) affects 4% of 1 19-year-olds. Their risk of death in the decade after hospital discharge is twice as high as that of adolescents admitted to hospitals for accident-related injury. We established how cause of death varied between these groups. Methods We did a retrospective, nationwide, cohort study comparing risks of death in five causal groups (suicide, drugrelated or alcohol-related, homicide, accidental, and other causes of death) up to 1 years after hospital discharge following adversity-related (self-inflicted, drug-related or alcohol-related, or violent injury) or accident-related (for which there was no recorded adversity) injury. We included adolescents (aged 1 19 years) who were admitted as an emergency for adversity-related or accident-related injury between April 1, 1997, and March 31, 212. We excluded adolescents who did not have their sex recorded, died during the index admission, had no valid discharge date, or were admitted with injury related to neither adversity nor accidents. We identified admissions for adversity-related or accident-related injury to the National Health Service in England with the International Classification of Diseases-1 codes in Hospital Episode Statistics data, linked to the Office for National Statistics mortality data for England, to establish cause-specific risks of death between the first day and 1 years after discharge, and to compare risks between adversity-related and accidentrelated index injury after adjustment for age group, socioeconomic status, and chronic conditions. Findings We identified adolescents ( [36 %] girls, [63 9%] boys, and 885 [ 1%] adolescents who did not have their sex recorded). Of these adolescents, we excluded (1 4%) girls, (8 1%) boys, and all 885 without their sex recorded. Of the (3 8%) adolescents admitted with adversity-related injury ( [54 6%] girls and [45 4%] boys) and (6 2%) admitted with accident-related injury ( [25 6%] girls and [74 4%] boys), 4782 ( 5%) died in the 1 years after discharge (1312 [27 4%] girls and 347 [72 6%] boys). Adolescents discharged after adversity-related injury had higher risks of suicide (adjusted subhazard ratio 4 54 [95% CI ] for girls, and 3 15 [ ] for boys) and of drug-related or alcohol-related death (4 71 [ ] for girls, and 3 53 [ ] for boys) in the next decade than they did after accident-related injury. Although we included homicides in our estimates of 1-year risks of adversity-related deaths, we did not explicitly present these risks because of small numbers and risks of statistical disclosure. There was insufficient evidence that girls discharged after adversityrelated injury had increased risks of accidental deaths compared with those discharged after accident-related injury (adjusted subhazard ratio 1 21 [95% CI ]), but there was evidence that this risk was increased for boys (1 26 [ ]). There was evidence of decreased risks of other causes of death in girls ( 64 [ 53 77]), but not in boys ( 99 [ ]). Risks of suicide were increased following self-inflicted injury (adjusted subhazard ratio 5 11 [95% CI ] for girls, and 6 2 [ ] for boys), drug-related or alcohol-related injury (4 55 [ ] for girls, and 4 51 [ ] for boys), and violent injury in boys (1 43 [ ]) versus accident-related injury. However, the increased risk of suicide in girls following violent injury versus accident-related injury was not significantly increased (adjusted subhazard ratio 1 48 [95% CI ]). Following each type of index injury, risks of suicide and risks of drugrelated or alcohol-related death were increased by similar magnitudes. Lancet 217; 39: Published Online May 25, S (17)3145- See Comment page 536 Population, Policy and Practice Programme, UCL Great Ormond Street Institute of Child Health (A Herbert PhD, Prof R Gilbert MD, L Li PhD) and Department of Behavioural Science and Health, Institute of Epidemiology and Healthcare (A Herbert), University College London, London, UK; and Leeds Institute of Health Science, Faculty of Medicine and Health, University of Leeds, Leeds, UK (Prof D Cottrell MA) Correspondence to: Dr Annie Herbert, Department of Behavioural Science and Health, Institute of Epidemiology and Healthcare, University College London, London WC1E 7HB, UK annie.herbert@ucl.ac.uk Interpretation Risks of suicide were significantly increased after all types of adversity-related injury except for girls who had violent injury. Risks of drug-related or alcohol-related death increased by a similar magnitude. Current practice to reduce risks of harm after self-inflicted injury should be extended to drug-related or alcohol-related and violent injury in adolescence. Prevention should address the substantial risks of drug-related or alcohol-related death alongside risks of suicide. Funding UK Department of Health. Introduction Evidence from population-based cohort studies suggests that different types of adversity-related injury (self-inflicted, including poisonings, drug-related or alcohol-related, or violent injury) during adolescence are associated with similar underlying psychosocial problems, including adverse experiences (eg, maltreatment), poor mental health (eg, anxiety and Vol 39 August 5,

2 Research in context Evidence before this study We searched for studies, including reviews, of cause-specific death after hospital attendance for any adversity-related injury published from Jan 1, 1995, to May 31, 216, restricted to either English or French. We searched Google Scholar, Scopus, PubMed, and Web of Science using the terms adolescents, injury, hospital, self-harm, drug or alcohol use, violence, and mortality. Our inclusion criteria were that the sample was collected in a high-income country, that the majority of the sample were 1 19 years old (as indicated by the mean, median, or proportions within age groups), and that adolescents were followed up after their hospital presentation for at least 1 month. There were no exclusion criteria. We found six studies (seven articles), but no relevant systematic review. Five European studies reported risks of death due to suicide, and some also reported risks of deaths due to drug or alcohol use (n=2), homicide (n=2), unestablished or accidental causes (n=3), and chronic conditions (n=3), up to 15 years after adolescents presented to hospital with self-inflicted injury. One of these studies compared risks of cause-specific deaths following selfinflicted injury with those in the general population. One US study reported deaths from homicide, drug overdose, and traffic accidents in the 2 years after discharge following violent injury in 559 adolescents. We did not identify any studies that reported cause-specific death following hospital presentation or admission for drug-related or alcohol-related injury, or compared risks of cause-specific deaths after discharge following any adversity-related injury with those following accident-related injury. Added value of this study Our study adds new evidence for the risks of cause-specific death up to 1 years after discharge following adversity-related and accident-related injury in young people. Our finding of elevated risks of suicide following all types of adversity-related injury versus accident-related injury (and versus risks in the general population) suggests that clinical and public health strategies need to be extended to reduce harm after all types of adversity-related injury, whether self-inflicted, drug-related or alcohol-related, or violent injuries. Similar risks of suicide and drug-related or alcohol-related deaths following discharge from any type of index injury found in our study also emphasise the need for preventive strategies, both within and outside the health-care sector, to reduce the public health burden of suicide and drug-related or alcohol-related deaths. Implications of all available evidence There are substantial risks of long-term mortality after hospital discharge following injury related to self-harm, drug or alcohol misuse, or violence through a range of causes. Current guidance focuses primarily on self-inflicted injuries: public health and clinical strategies need to address the full range of adversity-related injury. depression), and poor social circumstances (eg, poverty). 1 3 Among the 4% of adolescents (aged 1 19 years) who are admitted to hospital with one of these types of adversity-related injury in England, approximately three-quarters of girls and a third of boys are admitted with injuries related to multiple types of adversity. 4 Despite this apparent overlap between self-inflicted, drug-related or alcohol-related, and violent injury, most research in these adolescents has focused on specific types of adversity-related injury. A previous study 5 of adolescents admitted to hospital in England as an emergency with any adversity-related injury reported that one in 136 girls (7 3 per 1) and one in 64 boys (15 6 per 1) died within 1 years after hospital discharge, and that these risks were similar whether the initial injury was self-inflicted, drug or alcohol related, or violent. These 1-year risks were approximately twice the risks for adolescents discharged after accident-related injury (3 8 per 1 for girls and 6 per 1 for boys) or for the general population of adolescents (3 per 1 for girls and 3 per 1 for boys). Despite common underlying psychosocial problems and elevated mortality risks in adolescents with any of these three types of adversity-related injury, UK national clinical guidelines recommend different approaches to psychosocial assessment and intervention to reduce future harm. 6 8 For example, guidelines for management of self-inflicted injury presenting to hospital recommend admission of patients younger than 16 years and assessment of psychosocial circumstances and suicide risk at all ages. 6,7 Guidelines for drug-related or alcohol-related presentations do not specifically address the assessment of psychosocial needs of adolescents. 8 No UK guidelines exist for response to violent injury. A further issue is that clinical management to reduce the risk of further harm after self-inflicted injury focuses on risks of recurrent self-harm, despite evidence for increased risks of other adverse outcomes. 9 A cohort study of individuals aged years presenting to a hospital in Oxford, UK, with self-inflicted injury in reported increased mortality due to respiratory disorders, circulatory disorders, accidents, and suicide during the subsequent 2 years. 9 No comparable estimates have been published for risks of harm following drug-related or alcoholrelated or violent injury. In this study we aimed to inform preventive strategies for reducing risks of future harm for adolescents who are discharged from hospital after self-inflicted, drugrelated or alcohol-related, or violent injury. Given standard practice to reduce risks of repeated self-harm Vol 39 August 5, 217

3 or suicide after discharge following self-inflicted injury, we examined for girls and boys separately whether risks of suicide differ between adolescents discharged following self-inflicted injury, drug-related or alcoholrelated injury, and violent injury. Second, among girls and boys, we compared risks of cause-specific death (suicide, drug-related or alcohol-related, homicide, accidental, and other causes of death [ie, not suicide, drug-related or alcohol-related, homicide, or accidental]) up to 1 years from discharge after each type of index injury, including accident-related injury. Methods Study design and patients We did a retrospective, nationwide, cohort study using the same cohort 5 in Hospital Episode Statistics (HES) data, 1 which we have previously studied. The HES data contained all emergency (acute and unplanned) admissions to the National Health Service (NHS) in England, UK (April 1, 1997, to March 31, 212), including admissions to independent sector providers paid for by the NHS. 11 Approximately 98 99% of hospital activity in England is funded by the NHS, 12 and so these data captured nearly all admitted adolescents. We derived a cohort of adolescents (aged 1 19 years) who were admitted as an emergency for adversity-related or accident-related injury (the index injury). 5 We categorised them as adversity-related injury (comprising non-mutually exclusive groups of self-inflicted, drugrelated or alcohol-related, or violent injury, irrespective of whether the injury was also accident related) or accidentrelated injury (for which there was no recorded adversityrelated injury). Therefore, adversity-related injury and accident-related injury were two mutually exclusive groups. Deaths within the cohort were evaluated in five causal groups: suicide, drug-related or alcohol-related, homicide, accidental, or other causes of deaths. We compared risks of death (total and by cause) up to 1 years following discharge from hospital for admission for adversity-related injury (exposure) with risks after accident-related injury (comparator). On the basis of previous exclusion criteria, 5 we excluded adolescents who did not have their sex recorded, died during the index admission, had no valid discharge date, or were admitted with injury related to neither adversity nor accidents. Self-inflicted, drug-related or alcohol-related, violent, and accident-related injuries were identified with the International Classification of Diseases (ICD) 1 codes in HES data (ie, characteristics that were identified and recorded by clinicians). Details of classification of injury and descriptive statistics of the cohort have been reported elsewhere. 4,5 Because we used a standard, de-identified HES extract from NHS Digital (formerly known as the Health and Social Care Information Centre), ethics approval was not required. 13 Outcomes The primary outcome was cause-specific death between the first day and 1 years after discharge from the index injury admission. We identified deaths using the Office for National Statistics mortality data linked to HES data (within NHS Digital). We used any ICD-9 or ICD-1 codes in the mortality data (based on the underlying cause and up to 15 other contributing causes recorded in the death certificate) to categorise deaths into five causal groups ie, suicide, drug-related or alcoholrelated deaths, homicide, accidental deaths, or other deaths (appendix pp 2 4). Suicide, drug-related or alcohol-related deaths, and homicide were not mutually exclusive, but these three groups combined ie, adversity-related deaths, accidental deaths (no codes for adversity-related death, but codes for accidental causes), and other deaths (no codes for adversity-related or accidental deaths) were mutually exclusive. As advised by the Office for National Statistics, we classified undetermined causes of death (codes E98 89 and Y1 34) as suicide. 14 We categorised deaths with codes indicating an adjourned inquest (U5.9) as homicide (the appendix [pp 2 4] shows further details about codes for undetermined causes of death and adjourned inquests). Statistical analysis We first derived numbers and proportions of deaths (total and by cause) in the 1 years after discharge following adversity-related (self-inflicted, drug or alcohol related, or violent) or accident-related index injury. As per statistical disclosure rules (a condition of using the dataset) 15 required us not to publish counts of less than five, we do not present exact numbers of homicides for certain groups. We calculated unadjusted cumulative risks (and their 95% CIs) of deaths for each cause of death over the 1 years following discharge from the index injury admission, using the number of adolescents discharged alive after each type of index injury as the denominator. We estimated the cumulative risk of death by cause of death as model-based estimates of the cumulative incidence function, 16 which accounted for other competing causes (eg, for suicide, competing causes included homicide and drug-related or alcohol-related, accidental, and other deaths). For reference, we present unadjusted 1 year cumulative risks and 95% CIs by cause of death and type of index injury, sex, and age group. We estimated the 1 year cumulative risk of total death as a cumulative failure function (ie, within a standard survival and not a competing risks framework). We also estimated total and cause-specific risks of death for the general population aged 1 19 years in England using publicly available life-tables for total mortality and suicide from the Office for National Statistics, and bespoke life-tables for drug-related or alcohol-related and accidental deaths provided to us by the Office for See Online for appendix Vol 39 August 5,

4 Discharged Total deaths Numbers of deaths by cause Adversityrelated Suicide death* Drug or alcoholrelated death Homicide Accidental death Other causes of death Girls (1%) 63 (46 %) 361 (27 5%) 319 (24 3%) 28 (2 1%) 228 (17 4%) 481 (36 7%) Accident-related index injury (1%) 85 (19 4%) 47 (1 7%) 41 (9 3%) 5 ( 1%) 94 (21 4%) 26 (59 2%) Adversity-related index injury (1%) 518 (59 3%) 314 (36 %) 278 (31 8%) 21 (2 4%) 134 (15 4%) 221 (25 3%) Self-inflicted injury (1%) 48 (62 7%) 259 (39 8%) 21 (32 3%) 15 (2 3%) 93 (14 3%) 15 (23 %) Drug-related or alcohol-related (1%) 464 (59 8%) 283 (36 5%) 25 (32 2%) 17 (2 2%) 117 (15 1%) 195 (25 1%) injury Violent injury (1%) 25 (46 3%) 1 (18 5%) 16 (29 6%) 5 ( 9 3%) 11 (2 4%) 18 (33 3%) Boys (1%) 1736 (5 %) 93 (26 %) 861 (24 8%) 134 (3 9%) 891 (25 7%) 843 (24 3%) Accident-related index injury (1%) 711 (36 9%) 375 (19 5%) 311 (16 1%) 8 (4 1%) 6 (31 1%) 617 (32 %) Adversity-related index injury (1%) 125 (66 5%) 528 (34 2%) 55 (35 7%) 54 (3 5%) 291 (18 9%) 226 (14 7%) Self-inflicted injury (1%) 526 (74 7%) 34 (43 2%) 276 (39 2%) 9 (1 3%) 92 (13 1%) 86 (12 2%) Drug-related or alcohol-related (1%) 775 (69 7%) 418 (37 6%) 424 (38 1%) 17 (1 5%) 183 (16 5%) 154 (13 9%) injury Violent injury (1%) 268 (58 3%) 12 (26 1%) 135 (29 3%) 39 (8 5%) 122 (26 5%) 7 (15 2%) Data are n or n (%). *Suicides, drug-related or alcohol-related deaths, and homicides. These deaths were not mutually exclusive. Counts five or less have been suppressed as part of statistical disclosure control. Adolescents could be admitted with self-inflicted and drug-related or alcohol-related injury, or die from suicide and a drug-related or alcohol-related death. Table 1: Cause-specific deaths within 1 years after index injury admission, by sex and type of index injury National Statistics (according to ICD codes; appendix pp 2 4) We included covariates in the analyses on the basis of previous findings 5 of their relationship with adversityrelated injury and death, age, socioeconomic status, and chronic conditions. We grouped age (1 15 years, years, and years) to reflect different recommendations in the UK national guidelines for management of self-harm or alcohol misuse according to age and different stages of social development. 6 8 We categorised socioeconomic status according to the Index of Multiple Deprivation scores based on residential postcode, 2 with use of quintile cutoff values for England. We classified an adolescent as having an underlying chronic condition if HES records for the index injury admission or any admissions in the previous year included one of a cluster of ICD-1 codes for chronic conditions as defined by Hardelid and colleagues. 21 We fitted Fine and Gray s models 16 to estimate the relative risks of total and cause-specific mortality following adversity-related index injury, adjusted for covariates and taking into account competing risks of other causal groups. The exposure was type of index injury. We estimated subhazard ratios (SHRs) of each cause of death for adversity-related injury versus accident-related injury, age groups years and years versus 1 15 years, each level of socioeconomic status versus least deprived, and chronic condition versus none. To compare risks following each type of adversity-related injury, we also fitted these models when the exposure was self-inflicted, drug-related or alcohol-related, or violent injury, versus accidentrelated injury. We assessed whether the finding that increased risks of suicide and drug-related or alcohol-related deaths following self-inflicted or drug-related or alcoholrelated injury was due to the overlap between these two types of index injury, 5 or the overlap between suicide and drug-related or alcohol-related deaths. We fitted Fine and Gray s models to different combinations of exposures and outcomes. The exposures were selfinflicted (not drug-related or alcohol-related) injury, drug-related or alcohol-related (not self-inflicted) injury, or self-inflicted and drug-related or alcohol-related injury. The outcomes were any suicide, any drug-related or alcohol-related death, suicide (not drug-related or alcohol-related death), drug-related or alcohol-related death (not suicide), or suicide and drug-related or alcohol-related death (further details in the appendix; p 1). We checked model assumptions using log log plots of Kaplan-Meier estimates of the survival function and the link test, and assessed their goodness-of-fit using plots of the Nelson-Aalen estimate of the cumulative hazard function against Cox-Snell residuals. 16 We did all analyses separately for girls and boys in Stata SE (version 12). Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study, and all authors 58 Vol 39 August 5, 217

5 had final responsibility for the decision to submit for publication. Results There were adolescents ( [36 %] girls, [63 9%] boys, and 885 [ 1%] adolescents who did not have their sex recorded) who were admitted as an emergency to hospital with injury (appendix p 5). We excluded adolescents who did not have their sex recorded, died during the index admission (1877 [ 2%]; 588 [31 3%] girls and 1289 [68 7%] boys), were alive but had no valid discharge date or were not yet discharged (372 [< 1%]; 161 [43 3%] girls and 211 [56 7%] boys), and were admitted with injury related to neither adversity nor accidents (94 47 [8 7%]; 39 8 [42 2%] girls and [57 8%] boys) (57 4%) of adolescents who were admitted for neither of these injuries were admitted primarily for chronic conditions or complications of surgery. Therefore, our cohort comprised (3 8%) adolescents who had at least one adversity-related injury ( [54 6%] girls and [45 4%] boys; of which [2 3%] of girls and [24 %] of boys had an injury that was also accident related) and (6 2%) adolescents with at least one accident-related injury but no adversityrelated injury (girls [25 6%] and boys [74 4%]). 5 The most frequent type of injury in girls and boys was drug-related or alcohol-related injury followed by self-inflicted injury in girls and violent injury in boys (table 1). Of the adolescents with adversityrelated or accident-related injury who had a chronic condition, (79 7%) had a physical condition (data not shown). The most common physical condition was chronic respiratory disorder (eg, asthma, from 39 8% of boys with adversity-related injury to 55 4% of boys with accident-related injury by sex and type of injury 5 ). By 1 years after discharge from admission for the index injury, there were 2415 deaths (873 [36 1%] girls and 1542 [63 9%] boys) after adversity-related index injury, and 2367 deaths (439 [18 5%] girls and 1928 [81 5%] boys) after accident-related index injury (figure 1, table 1). After adversity-related index injury, 1543 (63 9%) deaths were related to suicide, drug or alcohol use, or homicide, compared with only 796 (33 6%) deaths after accident-related index injury. The proportions of deaths related to suicide, drug or alcohol use, or homicide were also higher after admission for adversityrelated injury than after accident-related injury. The proportions of deaths related to suicide, drug or alcohol use, or homicide were similar between girls and boys after adversity-related injury. The most frequent causes of death after accident-related index injury were other causes of death. 759 (67 8%) of all accidental deaths were recorded as transport accidents. This proportion did not differ Adversity-related index injury (all deaths 2415 [1 %]) Accident-related index injury (all deaths 2367 [1 %]) < 1% < 1% Homicide 2% 2 8% 26 7% 8 % 26 % Suicide Drug-related or alcohol-related 1543 (63 9%) Adversity-related deaths Homicide 3 7% 15 1% 2 7% 12 1% Suicide Drug-related or alcohol-related 796 (33 6%) Adversity-related deaths 425 (17 6%) Accident-related deaths 694 (29 3%) Accident-related deaths according to type of index admission (data not shown). In deaths due to other causes, the most common causes were related to neurological conditions (473 [35 7%] deaths), or cancer or blood disorders (384 [29 %] deaths; of nine possible groups of ICD codes relating to systems within the body, as reported by Hardelid and colleagues) year cumulative risks of total death after adversityrelated index injury were 7 3 per 1 girls (95% CI ) and 15 6 per 1 boys ( ; appendix pp 6 9). Cumulative risks of total death were lower after accident-related index injury (girls 3 8 per 1 [95% CI ]; boys 6 per 1 [ ]) than after adversity-related index injury. The increased risks of death after an adversity-related injury compared with accident-related injury were due to substantially higher risks of suicides and drug-related or alcohol-related deaths at all timepoints after the index injury (figure 2). After adjustment for other covariates, risks of suicides and drug-related or alcohol-related deaths were about three to five times higher following discharge from adversity-related injury admission than from accident-related injury admission (table 2). These increased risks equated to 175 (392 girls and 683 boys) excess suicides and drug-related or alcohol-related deaths (219 drug-related or alcohol-related deaths for girls and 394 drug-related or alcohol-related deaths for boys) in our cohort. 447 (18 5%) Other causes of death 877 (37 1%) Other causes of death Figure 1: Numbers and proportions of deaths for all adolescents by reported cause Circles represent proportions and are drawn to scale within each figure (ie, type of injury). Accidental death included ICD-1 codes for accidents and no ICD-1 codes for adversity in the death certificate. Other causes of death included no ICD-1 codes for accidents or adversity in the death certificate. ICD=International Classification of Diseases. Vol 39 August 5,

6 A 16 1 Adversity-related index injury Other causes of 1 Accident-related index injury death 8 Accidental death 14 Homicide* 8 Drug-related or 6 alcohol-related 6 death 12 4 Suicide 4 Cumulative risk of death (per 1) Time since index admission (6 months) Time since index admission (6 months) 4 2 B Cumulative risk of death (per 1) Time since index admission (6 months) Time since index admission (6 months) Time since index admission (years) Time since index admission (years) Figure 2: Cumulative risk of cause-specific death overtime for (A) girls and (B) boys Cumulative risks are cumulative incidence functions as estimated from Fine and Gray s competing risks models (per cause), in which the only covariate included was adversity-related (vs accident-related) injury. *Includes only deaths for which suicide or drug-related or alcohol-related death were not also implicated. Includes only drug-related or alcohol-related deaths for which suicide was not also implicated. Includes all suicides, whether homicide or drug-related or alcohol-related death were also implicated or not. 1 year risks of suicide were similar after hospital discharge following self-inflicted index injury versus drug-related or alcohol-related index injury (2 9 per 1 [95% CI ] vs 2 5 [ ] per 1 for girls; 9 8 per 1 [ ] vs 7 2 per 1 [6 5 8 ] for boys; figure 3, appendix pp 6 9). Compared with adolescents discharged after accident-related injury, risks of suicide were increased by about five to six times for adolescents discharged after self-inflicted or drug-related or alcohol-related injury (table 3). Risks of suicide were increased after self-inflicted and after drug-related or alcohol-related injury, whether the index injury was for Vol 39 August 5, 217

7 All deaths Suicide Drug-related or alcohol-related death Accidental death Other causes of death Girls Adversity-related (vs accident-related) injury 1 51 ( ) 4 54 ( ) 4 71 ( ) 1 21 ( ) 64 ( 53 77) Age group (vs 1 15 years) years 1 4 ( ) 2 3 ( ) 1 88 ( ) 1 13 ( ) 1 8 ( ) years 2 1 ( ) 4 34 ( ) 2 76 ( ) 1 6 ( ) 1 44 ( ) Socioeconomic status (vs least deprived) Second least deprived 1 17 ( ) 81 ( ) 1 13 ( ) 1 28 ( 8 2 4) 1 15 ( ) Middle deprived 1 19 ( ) 69 ( ) 1 29 ( ) 1 12 ( ) 1 28 ( ) Second most deprived 1 53 ( ) 89 ( ) 1 44 ( ) 97 ( ) 1 48 ( ) Most deprived 1 57 ( ) 78 ( ) 1 64 ( ) 1 2 ( ) 1 59 ( ) Chronic condition (vs none) 3 77 ( ) 1 91 ( ) 2 53 ( ) 2 35 ( ) 1 14 ( ) Boys Adversity-related (vs accident-related) injury 1 94 ( ) 3 15 ( ) 3 53 ( ) 1 26 ( ) 99 ( ) Age group (vs 1 15 years) years 1 73 ( ) 2 7 ( ) 3 5 ( ) 1 6 ( ) 1 14 ( ) years 2 23 ( ) 3 48 ( ) 5 4 ( ) 1 91 ( ) 1 22 ( ) Socioeconomic status (vs least deprived) Second least deprived 1 24 ( ) 1 17 ( ) 1 19 ( ) 1 62 ( ) 1 14 ( ) Middle deprived 1 28 ( ) 1 19 ( ) 1 66 ( ) 1 62 ( ) 1 2 ( ) Second most deprived 1 42 ( ) 1 53 ( ) 1 86 ( ) 1 29 (1 1 66) 1 33 ( ) Most deprived 1 63 ( ) 1 57 ( ) 2 17 ( ) 1 72 ( ) 1 26 ( ) Chronic condition (vs none) 2 63 ( ) 1 26 ( ) 1 81 ( ) 1 62 ( ) ( ) Data are adjusted subhazard ratios (95% CI). Data were adjusted for age group, socioeconomic status, and chronic condition status (multivariable analyses). Each column by sex represents a separate Fine and Gray s competing risks model. Adversity-related (vs accident-related) injury, age group, socioeconomic status, and chronic condition status were entered as covariates simultaneously per model. Table 2: Relative risk of cause-specific death within 1 years after adversity-related index injury vs accident-related index injury either one of these types of injury only or both (appendix p 1). 1 year risks of suicide and of drug-related or alcoholrelated death were similar after each type of index injury (figure 3, appendix pp 6 9). These risks were highest after self-inflicted or drug-related or alcoholrelated index injury. For example, after self-inflicted injury, the 1 year risk of suicide for girls was 2 9 per 1 (95% CI ), whereas the 1 year risk of drug-related or alcohol-related death was 2 7 per 1 ( ). After adjustment for covariates, the increased risks of suicide after self-inflicted and after drug-related or alcohol-related index injury versus accident-related injury were similar to the risks of drug-related or alcohol-related death (table 3). Risks were similar whether the death was related to suicide but not drugs or alcohol, drugs or alcohol but not suicide, or both causes (appendix p 1). Boys aged years who were discharged after selfinflicted injury or drug-related or alcohol-related injury had the highest risks of death due to any cause (1 year risk was 3 4 [95% CI ] per 1 after self-inflicted injury, and was 25 1 per 1 [ ] after drugrelated or alcohol-related injury; appendix pp 6 9). These risks were substantially higher than those after accidentrelated injury (8 8 per 1 [8 9 6]) or for the general population of boys aged years (8 9 per 1). These risks were driven by high risks of suicide and drug-related or alcohol-related death. Adolescents aged years had about twice the mortality risk of those aged 1 15 years because of increased risks of suicide and drug-related or alcoholrelated deaths in older girls and boys, and increased risks of accidental deaths in older boys (table 2). Low socioeconomic status (ie, those most deprived) was associated with increased risks of total and causespecific mortality in both girls and boys, apart from suicide and accidental deaths in girls, for which we noted no significant difference in risk. Adolescents with a chronic condition versus none had about a three to four times increased risk of death due to any cause, and about a ten to 12 times increased risk of death due to causes other than adversity or accidents, regardless of the type of index injury (table 2). For example, for boys aged years discharged after an adversity-related index injury, the 1 year risk of death due to any cause was 37 5 per 1 given a chronic condition and 14 8 per 1 given no chronic condition (data not shown). For boys aged years discharged after accident-related injury, the 1 year risk of death due Vol 39 August 5,

8 A 15 Cumulative risk Suicide Accidental death Drug-related or alcohol-related death Other causes of death Cumulative risk of death (per 1) 1 5 B 15 Cumulative risk of death (per 1) 1 5 Accident-related Self-inflicted Drug-related or alcohol-related Violent Type of index injury Figure 3: 1 year cumulative risk of cause-specific deaths for (A) girls and (B) boys Error bars are 95% CIs. All deaths Suicide Drug-related or alcohol-related death Accidental death Other causes of death Girls Self-inflicted injury 1 52 ( ) 5 11 ( ) 5 14 ( ) 1 17 ( ) 59 ( 48 72) Drug-related or alcohol-related injury 1 45 ( ) 4 55 ( ) 4 52 ( ) 1 2 ( ) 62 ( 51 75) Violent injury 1 24 ( ) 1 48 ( ) 2 75 ( ) 1 34 ( ) 76 ( ) Boys Self-inflicted injury 2 83 ( ) 6 2 ( ) 5 91 ( ) 1 31 ( ) 1 7 ( ) Drug-related or alcohol-related injury 2 46 ( ) 4 51 ( ) 4 91 ( ) 1 4 ( ) 1 11 ( ) Violent injury 1 25 ( ) 1 43 ( ) 1 78 ( ) 1 1 ( ) 76 ( 59 97) Data are adjusted subhazard ratios (95% CI). Data were adjusted for age group, socioeconomic status, and chronic condition status (in multivariable analyses). Each cell represents a separate Fine and Gray s competing risks model, for which the corresponding type of adversity-related index injury (vs accident-related injury), age group, socioeconomic status, and chronic condition status were entered as covariates simultaneously per model. Adjusted subhazard ratios for age group, socioeconomic status, and chronic condition status for each of the thirty models are not shown here (data available on request) but were very similar to those shown in table 2 (conditional on sex and cause of death). Table 3: Relative risk of cause-specific death within 1 years after each type of adversity-related index injury vs accident-related index injury Vol 39 August 5, 217

9 to any cause was 17 5 per 1 given a chronic condition and 8 8 per 1 given no chronic condition (data not shown). Discussion This retrospective cohort study established cause-specific risks of death up to 1 years after adolescents were discharged from the NHS in England following injury related to adversity or accidents. Within 1 years after discharge following adversity-related injury, 7 3 per 1 girls and 15 6 per 1 boys had died. 5 We found that suicide, drug-related or alcohol-related deaths, and homicides accounted for 63 9% of all deaths 1 years after adversity-related injury, but only 33 6% of deaths after accident-related injury. We showed that risks of suicide were all increased for adolescents following selfinflicted injury, drug-related or alcohol-related injury, and violent injury, except for girls following violent injury who did not have a significantly increased risk. These risks were highest for boys aged years. The risks of suicide were similar to those of drug-related or alcoholrelated deaths regardless of whether the adversity-related index injury was self-inflicted, drug or alcohol related, or violent. Adolescents with an underlying chronic condition at the index injury admission (1 15%) 5 were at increased risk of all causes of death, independent of the type of adversity-related or accident-related injury or age at admission. Our main finding of similar increases in risks of suicide following self-inflicted injury and following drug-related or alcohol-related injury has not been reported previously. We report lower 1 year risks of death after admission for self-inflicted injury (7 8 per 1 for girls and 24 2 per 1 for boys) than the 2 year mortality rates after presentation with selfinflicted injury reported by Hawton and Harriss (17 per 1 for girls and 5 per 1 for boys). 9 These differences might be because Hawton and Harriss studied young people aged years and established mortality after 2 years of follow-up. In their study, 6 % of deaths in girls and 45 6% of those in boys were from suicide (including deaths of unestablished intent and drug-related or alcohol-related suicides), 9 compared with 39 8% of girls and 43 2% of boys in our study. We have previously reported that the peak age group for adversity-related injury was years for girls (47%) and years for boys (46%), but for accident-related injury, the peak age group was 1 14 years for both girls (62%) and boys (54%). 5 Compared with adolescents admitted with accident-related injury, those admitted for adversity-related injury were more likely to be in the most deprived category of socioeconomic status, or to have a chronic condition recorded in the past year in hospital records. 5 The main strength of our study is the use of linked NHS emergency admissions and mortality data, which included all injury admissions in England linked to subsequent mortality records in England and Wales over 15 years. 14 The population-based cohort of about one million adolescents aged 1 19 years allowed us to compare risks of cause-specific mortality between different types of index injury admissions. We used time-to-event statistical methods to estimate risks while taking into account censoring of outcomes and competing risks of different causes of death. 16 Although we combined index injury admissions across a 15 year period, our conclusions were not sensitive to a calendar period. One limitation is that ICD codes used to define adversity-related injury and deaths are likely to have high specificity but low sensitivity The potential misclassification of exposure (ie, self-inflicted, drugrelated or alcohol-related, or violent injury, misclassified as accident-related injury) and outcomes (ie, suicide, drug-related or alcohol-related deaths, or homicides, misclassified as accidental or other deaths) might induce bias in the estimates of their associations, which is likely to underestimate the increased risks of suicide and drug-related or alcohol-related deaths after adversity-related injury relative to after accident-related injury. To minimise this potential bias we included codes for undetermined intent in the definition of suicide and adjourned inquests in the definition of homicide. The prevalence of chronic conditions recorded by codes at the index injury admission or at admissions to hospital during the previous year might be underascertained, particularly for the presence of chronic mental health conditions. A further limitation is potential linkage error between HES admissions data and the Office for National Statistics mortality data. One of the few studies that has investigated linkage errors in HES data showed high missed-match proportions (4 1%) that were higher for men than for women and for ethnic minorities than for white patients. 25 Linkage error between HES data and the Office for National Statistics data would favour underestimation of mortality rates. Lastly, our study was probably underpowered to detect differences in the risks of homicide between index injury groups. Our findings suggest that specialist psychosocial assessment by a child and adolescent mental health professional, which is part of recommended standard practice for self-inflicted injury in the UK, should be considered for adolescents presenting with drug-related or alcohol-related or violent injury. The need for a consistent approach targeting all three adversity-related injury groups is supported by previous evidence of their common underlying psychosocial problems, the overlap in the same admitted adolescents, 4 and similar patterns of risky behaviours into young adulthood, particularly relating to self-harm and drug or alcohol use Clinical and public health strategies need to be extended to include reduction of risks of death related Vol 39 August 5,

10 to drugs or alcohol, which are just as high as risks of suicide death. If complete eradication of the excess mortality risk associated with adversity-related injury in adolescents admitted to hospitals was possible, we could have expected 175 fewer suicides and drugrelated or alcohol-related deaths in our cohort (392 girls [219 drug-related or alcohol-related deaths] and 683 boys [394]; on the basis of the estimated relative risks in table 2). In adolescents aged years, the burden of suicides in the decade after adversity-related injury represented approximately 1 25% of suicides expected in the general population during the same follow-up (based on 1 year risks in the appendix, pp 6 9, and about 3 4% of the general population aged years admitted with adversity-related injury). 4 Findings from our study are likely to be generalisable to other UK countries, where rates of admissions to hospitals during adolescence for adversity-related injury and mortality through intentional injuries are similar. 21,29 Generalisation to non-uk settings requires further research. Investment is needed in interventions for reducing harm after all types of adversity-related injury, whether self-inflicted, drug or alcohol related, or violent. Risks of death are substantially increased in adolescents admitted with chronic conditions, and appropriate effective interventions might differ for this subgroup. There is a paucity of evidence to determine how public health bodies and services can reduce or ameliorate risks of long-term harm after adversity-related injury in adolescence Interventions need to be developed and assessed in randomised controlled trials to enable services to respond effectively and appropriately. Contributors AH, RG, and LL conceived and designed the study. AH analysed the data and drafted the first version of the Article. All authors interpreted the data, revised the article critically for important intellectual content, and approved the final version to be published. Declaration of interests We declare no competing interests. Acknowledgments This study was funded by the Policy Research Unit in the Health of Children, Young People and Families (reference 19/17), which is funded by the Department of Health Policy Research Programme. We thank members of the Policy Research Unit in the Health of Children, Young People and Families: Catherine Law, Russell Viner, Miranda Wolpert, Amanda Edwards, Steve Morris, Helen Roberts, Terence Stephenson, and Cathy Street. We thank Pia Hardelid for her advice on coding chronic conditions in Hospital Episode Statistics data, and Arturo González-Izquierdo for discussions of this work. We also acknowledge support from the Farr Institute of Health Informatics Research and the National Institute for Health Research Biomedical Research Centre at Great Ormond Street Hospital for Children NHS Foundation Trust and University College London. References 1 Hawton K, Rodham K, Evans E, Weatherall R. Deliberate self harm in adolescents: self report survey in schools in England. BMJ 22; 325: Wang RH, Hsu HY, Lin SY, Cheng CP, Lee SL. Risk behaviours among early adolescents: risk and protective factors. J Adv Nurs 21; 66: Viner RM, Ozer EM, Denny S, et al. Adolescence and the social determinants of health. Lancet 212; 379: Herbert A, Gilbert R, Gonzalez-Izquierdo A, Li L. Violence, self-harm and drug or alcohol misuse in adolescents admitted to hospitals in England for injury: a retrospective cohort study. BMJ Open 215; 5: e Herbert A, Gilbert R, Gonzalez-Izquierdo A, Pitman A, Li L. 1-y risks of death and emergency re-admission in adolescents hospitalised with violent, drug- or alcohol-related, or self-inflicted injury: a population-based cohort study. PLoS Med 215; 12: e NICE. Self-harm: longer term management (clinical guideline 133). London: National Institute for Health and Clinical Excellence, NICE. Self-harm in over 8s: short-term management and prevention of recurrence (clinical guideline 16). London: National Institute for Health and Clinical Excellence, Bekkering GE, Aertgeerts B, Asueta-Lorente JF, et al. Practitioner review: evidence-based practice guidelines on alcohol and drug misuse among adolescents: a systematic review. J Child Psychol Psychiatry 214; 55: Hawton K, Harriss L. Deliberate self-harm in young people: characteristics and subsequent mortality in a 2-year cohort of patients presenting to hospital. J Clin Psychiatry 27; 68: Herbert A, Wijlaars L, Zylbersztejn A, Cromwell D, Hardelid P. Data resource profile: Hospital Episode Statistics Admitted Patient Care (HES APC). Int J Epidemiol 217; published online March 15. DOI:1.193/ije/dyx Health and Social Care Information Centre. Hospital episode statistics (accessed Aug 28, 214). 12 National Audit Office. Healthcare across the UK: a comparison of the NHS in England, Scotland, Wales and Northern Ireland. London: The Stationery Office, Medical Research Council and NHS Health Research Authority. Do I need NHS REC approval? org.uk/ethics/ (accessed Sept 18, 215). 14 Office for National Statistics. Mortality metadata, (accessed Feb 18, 216). 15 Office for National Statistics. ONS policy on protecting confidentiality within birth and death statistics ons.gov.uk/file?uri=/methodology/methodologytopicsand statisticalconcepts/disclosurecontrol/uidanceforbirthand deathsstatistics/supersededdisclosurecontrolbriefingnotefor birthanddeathstatisticstcm pdf (accessed April 11, 217) 16 Cleves MA. An introduction to survival analysis using Stata, 3rd edn. College Station: Stata, Office for National Statistics. National life tables, England to peoplepopulationandcommunity/birthsdeathsandmarriages/ lifeexpectancies/bulletins/ nationallifetablesunitedkingdom/ (accessed Jan 11, 216). 18 Office for National Statistics. Suicide in the United Kingdom birthsdeathsandmarriages/deaths/datasets/ suicidesintheunitedkingdomreferencetables (accessed Sept 7, 216). 19 Office for National Statistics. Deaths from specific grouped causes, England, 1997 to peoplepopulationandcommunity/birthsdeathsandmarriages/ deaths/adhocs/6513deathsfromspecificgroupedcausesengland199 7to212 (accessed Feb 7, 217). 2 Health and Social Care Information Centre. Inpatient HES data dictionary. Leeds: Health and Social Care Information Centre, Hardelid P, Dattani N, Davey J, Pribramska I, Gilbert R. Overview of child deaths in the four UK countries. London: Royal College of Paediatrics and Child Health, McKenzie K, Harrison JE, McClure RJ. Identification of alcohol involvement in injury-related hospitalisations using routine data compared to medical record review. Aust N Z J Public Health 21; 34: Patrick AR, Miller M, Barber CW, Wang PS, Canning CF, Schneeweiss S. Identification of hospitalizations for intentional self-harm when E-codes are incompletely recorded. Pharmacoepidemiol Drug Saf 21; 19: Vol 39 August 5, 217

11 24 Wood DM, Conran P, Dargan PI. ICD-1 coding: poor identification of recreational drug presentations to a large emergency department. Emerg Med J 211; 28: Hagger-Johnson G, Harron K, Fleming T, et al. Data linkage errors in hospital administrative data when applying a pseudonymisation algorithm to paediatric intensive care records. BMJ Open 215; 5: e Mars B, Heron J, Crane C, et al. Clinical and social outcomes of adolescent self harm: population based birth cohort study. BMJ 214; 349: g Moran P, Coffey C, Romaniuk H, Degenhardt L, Borschmann R, Patton GC. Substance use in adulthood following adolescent self-harm: a population-based cohort study. Acta Psychiatr Scand 215; 131: Swanepoel A. Fifteen-minute consultation: safety assessment prior to discharge of patient admitted for self-harm. Arch Dis Child Educ Pract Ed 216; 11: Herbert A, Gonzalez-Izquierdo A, McGhee J, Li L, Gilbert R. Time-trends in rates of hospital admission of adolescents for violent, self-inflicted or drug/alcohol-related injury in England and Scotland, 25 11: population-based analysis. J Public Health 216; pubished online March 21. DOI:1.193/pubmed/fdw2. 3 Hawton K, Witt KG, Taylor Salisbury TL, et al. Interventions for self-harm in children and adolescents. Cochrane Database Syst Rev 215; 12: CD Snider C, Lee J. Youth violence secondary prevention initiatives in emergency departments: a systematic review. CJEM 29; 11: Newton AS, Dong K, Mabood N, et al. Brief emergency department interventions for youth who use alcohol and other drugs: a systematic review. Pediatr Emerg Care 213; 29: Vol 39 August 5,