Risk factors for nonfatal self‐harm and suicide among adolescents: two nested case–control studies conducted in the UK Clinical Practice Research Datalink

Introduction

Self-harm and suicide in young people represent major public health concerns (Hawton, Saunders, & O'Connor, 2012). Individuals who self-harm are at much greater risk of harming themselves again and of subsequently dying by suicide, which is the leading cause of death in the UK between 5 and 19 years of age (Office of National Statistics, 2020). Self-harm and suicide occur as a result of complex interactions among genetic, biological and environmental factors, including sex, socioeconomic position, family dynamics (e.g. parental divorce), adverse childhood experiences (e.g. physical and emotional abuse), interpersonal difficulties, psychological and personality factors (e.g. impulsivity, hopelessness, low self-esteem, etc.) and drug and alcohol misuse (Hawton et al., 2012). Evidence from the United Kingdom, United States of America, Australia and Canada indicate that an increasing number of adolescents have been harming themselves and dying from suicide in recent years, but data from Germany, Japan, Italy and France do not follow this trend (Bould, Mars, Moran, Biddle, & Gunnell, 2019; Morgan et al., 2017; Padmanathan, Bould, Winstone, Moran, & Gunnell, 2020). In common with previous research and consistent with clinical practice in the United Kingdom and elsewhere, when referring to self-harm we have not used terms that imply degree of suicidal intent because self-reported intent is oftentimes transient and unreliable (National Collaborating Centre for Mental Health, 2012). We use the term nonfatal self-harm to describe any self-harm event that had a nonfatal outcome, and vice versa with fatal self-harm, without making judgements about the degree of suicidal intent in the episode.

Whilst every suicide is an individual tragedy, in absolute terms it is a rare outcome in young people, which renders the study of its aetiology challenging. Studies that have examined risk factors in young persons who engage in nonfatal self-harm, a far more common behaviour, are by contrast abundant in the literature (Chen et al., 2017; Chou, Lin, Sung, & Kao, 2014). Most previous studies that have examined the antecedents of suicide in young people are so-called psychological autopsy studies, in which the prevalence of antecedents (e.g. prior history of bullying) are estimated post-mortem among ascertained suicide cases (Rodway et al., 2016). Although informative to a degree, such studies typically lacked a matched control group in which these characteristics were also examined, which means that relative risks for putative risk factors were not reported. Existing studies that have included matched controls (Brent et al., 1994; Gould et al., 1998; Portzky, Audenaert, & van Heeringen, 2009) have tended to be very small and have, therefore, not been able to establish risks associated with individual psychiatric disorders with an adequate degree of statistical precision. It is therefore unclear how strongly psychiatric illnesses are associated with suicide in young individuals, and whether the strength of these relationships is similar or different between young people who harm themselves nonfatally and those who die by suicide.

We addressed this gap in the evidence base by conducting two separate population-based nested case–control studies in which we examined risks of nonfatal self-harm and suicide associated with psychiatric illnesses diagnosed among adolescents aged 10–19 years who were registered with a general practitioner (GP) in the United Kingdom. We also examined risks associated with clinical and sociodemographic contextual information that is available in routinely collected primary care records, including psychotropic medication prescribed by GPs or practice nurses, frequency of primary care clinical contact and varying neighbourhood deprivation levels.

Method Data source

We delineated two nested case–control studies from the Clinical Practice Research Datalink (CPRD) Aurum and GOLD datasets. These large longitudinal primary care datasets draw routinely collected information from UK general practices and are broadly representative of the national population in terms of its distributions of sex, age and ethnicity (Herrett et al., 2015; Wolf et al., 2019). In the United Kingdom, approximately 98% of the population is registered with a GP, whose purpose is to provide the first point of contact in the UK healthcare system. Information about treatments received in hospitals and in other parts of the National Health Service (NHS) and private healthcare providers are fed back to patients’ primary care records. The Aurum dataset covers approximately 13.0% of the population of England, and GOLD covers roughly 6.9% of the UK population (England, Northern Ireland, Scotland and Wales). We used the April 2020 CPRD GOLD and Aurum release, which contained 18,782,246 and 31,745,393 patients respectively.

The information recorded is similar in the two CPRD datasets, although each uses a different electronic patient record system (Aurum: EMIS Web®; GOLD: Vision®). The datasets contain anonymised patient information pertaining to consultations, diagnoses, drug prescriptions and referrals to other National Health Service (NHS) providers. Clinical information is captured using Read or SNOMED codes which are clinical classification systems developed for primary care (Benson, 2012). All Aurum records, and GOLD records for a subset of patients registered at general practices in England, were linked routinely to Office for National Statistics (ONS) mortality registration records, which contain information about cause of death, and to Hospital Episode Statistics (HES), which provide information regarding patient admission or attendance at hospitals, and to the Index of Multiple Deprivation (IMD) – a composite area-level deprivation measure (Smith et al., 2015).

Population

We identified adolescents aged 10–19 years who had harmed themselves nonfatally or died during the study’s observation period, 1st January 2003 to 31st December 2018. Individuals were censored on leaving the study practice or from death from another cause, whichever occurred first (Figure 1). As data are not transferred from patients migrating from one practice to another, we required that individuals had to have been registered at a CPRD practice for a minimum of 12 months before entering the study. This requirement was imposed to reduce the likelihood of including individuals with no exposure information and also to minimise the likelihood of misclassifying prevalent exposure episodes as incident. We identified suicide cases in the ONS mortality register using the following International Classification of Disease version 10 (ICD-10) codes: X60-84, Y10-34 (excluding Y33.9), Y87.0, Y87.2. X-codes are used when a coroner has ruled that the cause of death was suicide; y-codes are applied when the intent is undetermined. We included open verdicts (i.e. y-codes) in order to not underestimate the number of suicides (Linsley, Schapira, & Kelly, 2001; Neeleman & Wessely, 1997) but we did not include Y33.9 because it pertains to adjourned inquests in alleged homicide cases. Index self-harm episodes were identified in CPRD or HES, whichever was the earliest indicated date of occurrence, through clinical Read codes (Appendix S1) and ICD-10 codes respectively. We included HES data pertaining to inpatient admissions and to accident and emergency (A&E) department presentations. For the accident and emergency (A&E) data in the HES dataset, we also used the “aepatgroup” field to identify additional self-harm cases through the “Deliberate self-harm” code. The same set of ICD codes that were used to identify suicide cases were used to identify self-harm episodes in HES.

image Graphical representation of nested case-control study design. aCensoring occurred on leaving the study practice or on the end of the study period, whichever occurred first. bSelf-harm was only included as an exposure in the suicide case–control study, but not in the self-harm case–control study as incident self-harm episodes (‘cases’) were examined. cCases were risk-set matched to up to 25 controls on age (±1 year), sex and registered practice. Figure based on template provided in Schneeweiss et al. (2019) Measures

We used Read codes to identify diagnoses of attention-deficit hyperactivity disorder (ADHD), anxiety disorders, autism spectrum disorder (ASD), depression and eating disorders. Because there is evidence that general practitioners increasingly use symptom codes (e.g. “low mood”) to classify depression and anxiety disorders (John et al., 2015; Sarginson, Webb, Stocks, Esmail, & Garg, 2017), we included symptom codes as well as diagnostic codes (e.g. “depression”) when collecting information on exposures. We also used the following ICD-10 codes to identify these psychiatric illnesses in hospital settings: F32-F34, F38 and F39 (depression); F40–43 and F93 (anxiety disorders); F50 (eating disorders); F84 (ASD); F90 (ADHD). In the suicide case–control study, we included self-harm as an exposure using the same ICD-10 and Read codes that we used to identify nonfatal self-harm episodes in the other case–control study. We created code lists for three categories of psychotropic medication: antidepressants, ‘other’ non-antidepressant types of psychotropic medication, including antipsychotics, anxiolytics, hypnotics, mood stabilisers and stimulants, and a miscellaneous ‘any psychotropic medication’ category which included all psychotropic medications. Finally, we measured deprivation through the IMD, which provides a composite indicator of deprivation based on information regarding the following seven domains: income, employment, education, skills and training, crime, barriers to housing and services, health and disability and living environment. All code lists were reviewed by three experienced clinical academics in our group: NK: a psychiatrist with specific expertise in self-harm and suicide; SG: a child and adolescent psychiatrist and CCG: a GP with specific expertise in mental health. A full list of Read codes is available in the Supporting Information.

Study design and statistical analyses

In the Aurum and GOLD datasets, we created two nested case–control studies: one in which individuals who had died by suicide were the cases, and another in which individuals with an index nonfatal self-harm episode constituted the cases. The self-harm case–control dataset that was created in Aurum was then appended to the one that was created in GOLD to produce a single case–control study dataset; this step was repeated in building the suicide case–control study dataset. This approach was taken to maximise the number of cases available for analysis and to thereby optimise statistical power. We used a ‘bridging’ file to identify practices that had migrated between the GOLD and Aurum datasets and removed those practices from the GOLD dataset. Each case was matched with up to 25 controls on sex, age (±1 year) and registered general practice. All suicide cases were matched to 25 controls; 55,883 of 56,008 (98.8%) self-harm cases were matched to 25 controls. The remaining 1.2% (n = 125) of self-harm cases were registered with small general practices, and therefore could not be matched to 25 controls; the mean number of matched controls per case in this small subset was 19 (range: 4–24). Through incidence density sampling, control patients were randomly sampled from the risk-set for each case (Clayton & Hills, 1993). In matching controls to cases on registered general practice, by design we accounted for the potential confounding influences of practice-level deprivation and of local and regional differences in service provision.

We fitted conditional logistic regression models to estimate relative risks as exposure odds ratios (ORs) that were inherently adjusted for by age, sex and practice-level effects in the matched design. These ORs are interpretable as hazard ratios as would have been estimated by a survival analysis conducted on the whole study cohort, in which the nested case–control study was delineated (Clayton & Hills, 1993). We examined the associations among psychiatric diagnoses, type of psychotropic medication prescribed by a GP, patterns of primary care clinical contact and area-level of deprivation and risks for nonfatal self-harm and suicide. As potential confounding influences linked with practice-level deprivation were accounted for in the matched design, the relative risks that we have reported by IMD quintile indicate the independent association with the varying deprivation levels of patients’ neighbourhoods.

Results Suicide

We identified a total of 324 Individuals who had died by suicide, of which two thirds (67%) were boys. Suicide frequency increased linearly with age, with two thirds of suicides occurring at 17–19 years of age (Table 1). One third of cases were diagnosed with at least one of the examined psychiatric illnesses, compared to 10% of controls. The most common method of suicide was asphyxiation by hanging (68%), followed by poisoning and exposure to a substance (11%) (Table S1). Depression was by far the commonest of the examined conditions among suicide cases, accounting for over a half (54%) of all recorded diagnoses. As can be seen in Table 2, each of the diagnostic categories that we examined was associated with an elevated suicide risk, except for ASD. Among psychiatric illnesses, the association between depression and suicide was the strongest, with a 7-fold elevation in risk observed. However, prior history of self-harm was more strongly associated with suicide than any psychiatric illness: OR 19.8 95% CI 14.8–26.5. Having two or more diagnosed conditions was associated with an approximately doubled suicide risk than having a single diagnosis. Being prescribed any psychotropic drug strongly predicted suicide, especially so within a year of suicide, and particularly if the drug was an antidepressant. Most (70%) individuals presented at least once in the year preceding their death, and risk increased with rising consultation frequency. We did not observe an association between varying neighbourhood deprivation levels and suicide risk. Please see Table S1 for a distribution of suicide cases by ICD10 code.

Table 1. Distribution of sociodemographic matching variables in the suicide and nonfatal self-harm nested case–control study datasets Suicide case–control study Self-harm case–control study Case (N = 324) % Control (N = 8,100) % Case (N = 56,008) % Control (N = 1,399,356) % Age in years 10–12 7 2 169 2 4,795 8 120,368 8 13–16 107 33 2,674 33 30,097 54 751,354 54 17–19 210 65 5,221 65 21,116 38 526,402 38 Gender Girls 108 33 2,700 33 38,558 69 963,326 69 Boys 216 67 5,400 67 17,450 31 436,030 31 Practice-level deprivation quintile 1 (least deprived) 56 17 1,400 17 8,075 14 201,764 14 2 68 21 1,700 21 9,083 16 227,017 16 3 58 18 1,450 18 10,015 18 250,163 18 4 59 18 1,475 18 13,327 24 333,033 24 5 (most deprived) 83 26 2,075 26 15,508 28 387,379 28 Table 2. Risk factors for suicide estimated as exposure odds ratios (relative risks) Case (n = 324) % Controls (n = 8,100) % Odds ratios 95% CIs Diagnostic categories ADHD 11 3 140 2 2.0 1.1–3.8 Anxiety disorder 38 12 324 4 3.3 2.3–4.8 ASD 7 2 96 1 1.9 0.9–4.1 Depression 77 24 365 5 7.4 5.5–9.9 Eating disorder 9 2 74 1 3.1 1.5–6.3 Any psychiatric illness 104 32 809 10 4.6 3.5–5.9 No. of diagnostic categories 1 70 22 643 8 3.8 2.9–5.1 2 or more 34 10 166 2 7.7 5.1–11.5 History of self-harm 104 32 210 3 19.8 14.8–26.5 Any psychotropic drug Ever prescribed 105 32 751 9 5.1 4.0–6.6 Prescribed in past year 77 24 295 4 9.0 6.7–12.1 Antidepressant Ever prescribed 78 24 302 4 9.9 7.3–13.4 Prescribed in past year 64 20 152 2 16.5 11.5–23.6 Other psychotropic drug Ever prescribed 56 17 544 7 3.0 2.2–4.1 Prescribed in past year 33 10 173 2 5.3 3.6–7.9 Clinical contacts in past year None 97 30 3,853 48 1 (reference) 1 49 15 1,308 16 1.5 1.1–2.2 2 41 13 889 11 1.9 1.3–2.8 3 27 8 575 7 2.0 1.3–3.1 4 31 10 424 5 3.2 2.1–4.9 5 or more 79 24 1,051 13 3.4 2.4–4.7 Deprivation quintile (Neighbourhood level) 1 (least deprived) 71 22 1,679 21 1 (reference) 2 58 18 1,689 21 0.8 0.6–1.2 3 47 14 1,474 18 0.8 0.5–1.2 4 70 22 1,586 20 1.1 0.8–1.7 5 (most deprived) 78 24 1,666 20 1.3 0.8–2.0 Nonfatal self-harm

We identified 56,008 adolescents with a nonfatal self-harm episode. Compared to suicide, the sex ratio was reversed; just over two thirds (69%) of these episodes were among girls (Table 1). Individuals who self-harmed also tended to be younger, with almost two thirds (65%) of these cases having their first recorded episode below age 17 (Table 1). The most common method of nonfatal self-harm was self-poisoning (42%) followed by cutting (12%). However, many episodes were identified via clinical codes that did describe the method (e.g. “[x]intentional self-harm – U2…”) (Table S2). The proportion of self-harm cases decreased incrementally as practice-level deprivation decreased; we observed the largest proportion (28%) in the most deprived quintile and the smallest (14%) in the least deprived quintile.

As with suicide, only a third of patients had received a psychiatric diagnosis prior to their index self-harm episode. The associations between the psychiatric illnesses were of a similar magnitude (Table 3) to those that we observed in the suicide case–control set (Table 2). Thus, the rank order for strength of association (magnitude of ORs) across the array of examined conditions was almost the same for self-harm as it was for suicide, although the association between ASD and self-harm risk was statistically significant. Self-harm risk increased incrementally with rising number of prior diagnoses recorded (Table 3); risk was three times greater among adolescents with three or more diagnostic categories compared to one (OR 2.96; 95% CI 2.7–3.2). Self-harm risks among adolescents who had been prescribed psychotropic medication were substantially elevated, but these ORs were somewhat lower than the equivalent values generated from the suicide case–control study. Most (85%) individuals presented in primary care at least once in the preceding year, and self-harm risk increased with rising consultation frequency, but the increase in risk associated with each additional visit was greater compared to suicide. A positive linear relationship was observed between heightened levels of neighbourhood deprivation, independent of practice-level deprivation confounding influences. We examined the associations with diagnostic categories stratified by practice-level deprivation quintile, which revealed a considerably stronger relationship between depression and eating disorders and self-harm risk among adolescents registered at practices in more affluent localities (Figure 2); depression -–least deprived quintile: OR 9.6, 95% 9.1–10.2; most deprived quintile: OR 6.9 95% CI 6.6–7.2; eating disorders – least deprived quintile: OR 4.1, 95% CI 3.6–4.6; most deprived quintile: OR 2.5 95% CI 2.3–2.8. For the other three diagnostic categories examined, there was no evidence of risk being modified by practice-level deprivation. Side-by-side comparison of the ORs for the associations between the diagnostic categories and self-harm risk, generated in the GOLD and Aurum datasets separately, showed that these two sets of estimates were similar in their magnitude and equal in rank order (Table S3).

Table 3. Risk factors for self-harm estimated as exposure odds ratios (relative risks) Case (n = 56,008) % Controls (n = 1,399,356) % Odds ratios 95% CIs Diagnostic categories ADHD 2,082 4 16,995 1 3.3 3.1–3.4 Anxiety disorder 7,203 13 54,269 4 3.8 3.7–3.9 ASD 1,516 3 16,058 1 2.4 2.3–2.6 Depression 12,366 22 51,731 4 7.9 7.8–8.2 Eating disorder 1,820 3 15,160 1 3.1 3.0–3.2 Any psychiatric illness 19,000 34 131,208 9 5.2 5.1–5.3 No. of diagnostic categories 1 13,790 25 110,378 8 4.5 4.4–4.6 2 4,502 8 18,811 1 8.9 8.6–9.2 3 or more 708 1 2,019 0.1 13.4 12.3–14.6 Any psychotropic drug Ever prescribed 14,285 26 118,454 9 4.0 3.9–4.1 Prescribed in past year 10,288 18 52,578 4 6.1 6.0–6.2 Antidepressant Ever prescribed 9,368 17 44,937 3 6.8 6.6–6.9 Prescribed in past year 7,553 14 27,259 2 8.7 8.4–8.9 Other psychotropic drug Ever prescribed 7,889 14 85,682 6 2.6 2.5–2.7 Prescribed in past year

留言 (0)

沒有登入
gif