Healthcare utilization, psychiatric disorders, and physical illnesses shortly before suicide mortality in adolescents in Taiwan

Abstract

Background

This study examined the pattern of medical utilization and the distribution of comorbidities shortly before death among adolescents who died from suicide and compared these data with those of living controls.

Methods

From Taiwan's National Health Insurance Research Database, this study identified adolescents aged 10–19 years who died from suicide (n = 935) between 1 January 2000, and 31 December 2016, by linking each patient with the national mortality database. The researchers conducted a nested case–control study through risk set sampling, and for each case, 20 age- and sex-matched controls (n = 18 700) were selected from the general population. The researchers applied conditional logistic regression to investigate differences in medical utilization and physical and psychiatric comorbidities between cases and controls.

Results

Cases had a higher proportion of contact with the psychiatric department but a similar proportion of contact with any non-psychiatric medical department within 1 year before suicide compared with controls. There were 18.6% of adolescent suicide victims who only had contacted with a psychiatric department 3 months before suicide. Moreover, cases had a higher proportion of contact with non-psychiatric services within 3 months before suicide, particularly with emergency, surgery, and internal medicine departments. Cases had higher risks of several psychiatric disorders and physical illnesses, including heart diseases, pneumonia, and ulcer disease, than did controls.

Conclusions

The findings of increased medical utilization and higher risks of physical and psychiatric comorbidities in adolescent suicide victims are crucial for developing specific interventions to prevent suicide in this population.

Introduction

Adolescent suicide is a significant public health concern. Suicide is a leading cause of death among adolescents (World Health Organization, 2019), with the pooled rate of suicide being 3.77 per 100 000 inhabitants among adolescents aged 10–19 years worldwide (Glenn et al., 2020). In many countries, including Taiwan, the suicide rate among young people has increased in the past decade (Bould, Mars, Moran, Biddle, & Gunnell, 2019; Curtin, 2020; Ministry of Health and Welfare (Taiwan), 2020), making youth suicide prevention a public health priority. Visiting medical providers before suicide presents a key opportunity for suicide prevention. The identification of high-risk youth, their referral, and treatment for risk factor reduction are primary prevention goals (Gould, Greenberg, Velting, & Shaffer, 2003).

Currently, even with the advancement of neuroscience and improved understanding regarding human behavior pathophysiology, understanding suicide behavior remains a challenge (De Berardis et al., 2018). A recent review study (Orsolini et al., 2020) reported several pathophysiological factors that potentially exert synergistic effects on suicide, including genetics, epigenetics, the hypothalamic–pituitary–adrenal stress response system, exogenous and endogenous stressors, the lipid profile, neuroimmunological biomarkers, monoaminergic neurotransmitter systems, and other neuromodulators.

Psychiatric disorders are one of the most prominent and remediable risk factors for suicide in young people (Agerbo, Nordentoft, & Mortensen, 2002; Gould et al., 2003; Stenager & Qin, 2008). The estimated prevalence of psychiatric disorders among suicide victims has ranged from 48% to 94% in psychological autopsy studies (Fleischmann, Bertolote, Belfer, & Beautrais, 2005; Renaud, Berlim, McGirr, Tousignant, & Turecki, 2008; Zhang, Xiao, & Zhou, 2010). Although physical illness is an established risk factor for suicide in the general population (Ahmedani et al., 2017), it has not been adequately investigated as a risk factor for adolescent suicide. Only one study demonstrated that nearly 70% of young suicide victims had a lifetime history of physical illness and that a substantial proportion of them developed mental disorders concurrently (Viilo et al., 2005). Despite the high prevalence of psychiatric and physical illness among young suicide victims, a large treatment gap existed in this population (Renaud et al., 2014). Factors such as stigma associated with mental illness and a lack of understanding regarding the effectiveness of psychiatric treatment are predominant barriers to mental health treatment; these factors also make suicide prevention in youth more complex and challenging (Moskos, Olson, Halbern, & Gray, 2007). Therefore, characterizing adolescents’ healthcare utilization patterns before suicide can increase our understanding regarding youth suicide.

To date, few studies have explored the health care utilization of adolescents before suicide. A review published in 2002 indicated that suicide decedents aged <35 years tended to have a lower health care utilization rate than did those aged >35 years (Luoma, Martin, & Pearson, 2002). Nevertheless, the studies included in that review have mainly used methods such as psychological autopsy and medical record review; these methods are limited by a small sample size and recall or self-reporting bias. More recent studies including larger sample sizes have reported that a majority (78%) of adolescents who died by suicide had a health care encounter in the previous 1 year (Farand, Renaud, & Chagnon, 2004) and that their lifetime contact with mental health services reached 44% (Rodway et al., 2020). A retrospective matched case–control study (Ruch et al., 2021) including youth aged from 5 to 21 years derived from Ohio's Statewide Automated Child Welfare Information System found that more than half of the suicide decedents (59.2%) were diagnosed as having a mental health condition compared with less than one-third of controls (31.3%). A population-based case–control study (Fontanella et al., 2020) including Medicaid-enrolled youth found that 41.3% of them were diagnosed as having a mental health condition within 6 months before suicide death compared with 17.5% of controls. In addition, more mental health visits within 30 days before the index date were associated with decreased odds of suicide (Fontanella et al., 2020). This finding is consistent with those of previous studies indicating that the rate of service utilization, particularly mental health services, decreased during the period shortly before suicide (Chang et al., 2012; Renaud et al., 2009). These studies collectively indicate low levels of recognition and treatment referral for mental health problems in adolescent suicide victims. Differences in patterns of medical utilization between adolescent suicide victims and healthy living individuals remain unclear. Such information is crucial to develop preventive strategies for adolescent suicide, thus warranting further exploration.

To fill the aforementioned gaps in the relevant literature, we conducted a nested case–control study by enrolling an adolescent cohort in Taiwan. The objectives of this study were as follows: (1) to evaluate differences in the health-care utilization pattern between adolescents who died by suicide and controls and (2) to determine the role of psychiatric and physical comorbidities in the risk of suicide. The results of this study may help optimize health-care intervention strategies to prevent adolescent suicide.

Method

Study population

Taiwan introduced its single-payer National Health Insurance (NHI) program in 1995, and it covered approximately 99.9% of all residents in Taiwan at the end of 2014. Beneficiaries of the NHI program can visit any physician among NHI-contracted facilities throughout Taiwan. The National Health Insurance Research Database (NHIRD), established by the National Health Research Institutes, contains the medical claims data of all beneficiaries of the NHI program and includes demographic information, clinical diagnoses, and medical expenditures. The entire general population (approximately 23 000 000) who lived in Taiwan during the study period (1 January 2001 to 31 December 2016) was identified from the NHIRD.

Researchers can access these data after signing an agreement guaranteeing patient confidentiality. The requirement for informed consent was waived because the identifying information of beneficiaries and medical care providers in the database is deidentified by the Bureau of NHI. This study was approved by the Taipei City Hospital Research Ethics Committee (TCHIRB – 10911005-E).

Suicide ascertainment

Because every Taiwanese citizen has a unique national identification (ID) number that is routinely recorded in the NHIRD, each individual registered in the NHIRD can be electronically linked with the National Death Certification System to obtain one's mortality status. We searched for all deceased individuals by matching their national IDs with computerized data files from the National Death Certification System issued from 2000 to 2016. We identified 2 443 169 decedents during 2000–2016 in the entire population of Taiwan; of these decedents, 16 159 were aged between 10 and 19 years at the time of their death. For each decedent, the underlying cause of death was recorded in the National Death Certification System according to the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes. A total of 935 suicide decedents aged between 10 and 19 years with the ICD-9-CM codes E950–E959 were identified. Suicide methods were categorized as drug overdose (ICD-9-CM code E950), poisoning by gases in domestic use (ICD-9-CM code E951), charcoal burning and other gas poisoning (ICD-9-CM code E952), hanging (ICD-9-CM code E953), drowning (ICD-9-CM code E954), firearms and explosives (ICD-9-CM code E955), cutting (ICD-9-CM code E956), and jumping from a high place (ICD-9-CM code E957). International Classification of Diseases, Tenth Revision, Clinical Modification codes have been used in the Taiwan health care system after 2011, and codes for suicide methods after 2011 were converted to ICD-9-CM codes in this study.

Nested case–control design

We defined the identified 935 adolescent suicide decedents as cases. Subsequently, we performed a nested case–control study. The date of suicide death was defined as the index date. For each case, 20 controls were randomly selected from the entire population to ensure adequate statistical power and were matched on the same birth year and sex through risk set sampling. Furthermore, this study included selected controls who were alive on the date of the suicide mortality of the matched case; this date was used as the corresponding index date for controls. A control candidate who died before the corresponding index date was excluded, and the next suitable control was selected. Finally, this study included 935 case–control pairs (i.e. 935 cases v. 18 700 controls).

Figure 1 shows the flowchart for the inclusion of the national cohort, the process of identifying deceased adolescents, and the selection of suicide cases and suitable controls.

Fig. 1. Study flow diagram.

Variables

Information regarding various variables, including demographics and clinical diagnoses, was obtained by searching all claims data in the NHIRD. Data regarding demographic variables including age, sex, urbanization level, and Charlson Comorbidity Index (CCI) were obtained at the index date (death date). Employment status was recorded within 1 year before the index date. We adopted urbanization stratification specifically used in Taiwan (Liu et al., 2006), and the urbanization level was categorized into levels 1 (highly urbanized area), 2 (moderately urbanized area), 3 (newly urbanized area), 4 (township area), and 5 (rural area). The CCI (Charlson, Pompei, Ales, & MacKenzie, 1987) categorizes the severity of physical comorbidities into 0, 1, or ⩾2 according to ICD-9-CM diagnosis codes in the NHIRD.

Medical utilization was evaluated by determining the proportion and number of health care encounters in inpatient, emergency departments, and specialty outpatient settings 3 months before suicide. Psychiatric and physical comorbidities were assessed according to the ICD-9-CM coding system. To exclude encounters and physical or psychiatric disorders that were the direct result of a fatal suicide attempt, all health care encounters and physical or psychiatric diagnoses were excluded if they occurred on the death date.

Statistical analysis

According to the nested case–control study design, we performed univariable conditional logistic regression analysis to compare factors, namely demographic variables, medical utilization, and physical and psychiatric comorbidities, between cases and controls. Subsequently, we conducted multivariable analysis by using the backward variable selection method to examine two models (i.e. psychiatric disorders and physical illnesses); only strongly associated variables (p < 0.01) were retained in the final multivariate model. We plotted curves to examine medical departments visited (percentage with an encounter) within 12 months before suicide among cases and controls. Differences in the curves between cases and controls were estimated using Gehan's Wilcoxon test (Lee, 1992) through survival (life-table) analysis. All analyses were conducted using SAS (version 9.4; SAS Institute, Cary, NC, USA). p < 0.01 was considered statistically significant.

Results

Demographic and clinical characteristics of cases and controls

Table 1 lists the demographic and clinical characteristics of the cases who died of suicide and living controls. Sex and age at suicide were well matched between the cases and controls. More than half of those who died from suicide were male (62.0%). Compared with living controls, a higher proportion of cases who died from suicide had CCI scores of >1 [risk ratio (RR) = 2.86 and 6.53 for CCI = 1 and ⩾2, respectively]. Methods employed in suicide cases were stratified into two age groups e-Table 1, see online Supplementary material). The predominant suicide methods used were hanging (32.5%), jumping from height (29.1%), and charcoal burning (22.6%). Hanging (41.6%) and jumping from height (40.6%) were more common in the younger age group (those aged between 10 and 14 years).

Table 1. Demographic and clinical characteristics of adolescents who died by suicide (cases) and controls (ratio = 1:20) at the index (corresponding) date

^a Estimated using univariable conditional logistic regression.

^b The level of urbanization was categorized as levels 1 (highly urbanized area), 2 (moderately urbanized area), 3 (newly urbanized area), 4 (township area), and 5 (rural area). Urbanization values were missing for 32 cases and 115 controls.

Patterns of medical utilization before suicide

Figure 2 presents the backward cumulative proportions of health care encounters (including outpatient and emergency visits) within a given time interval before the index date for adolescent suicide victims compared with controls. The percentage of such encounters at 1 year prior to suicide was similar in suicide cases and controls; however, it was significantly higher among suicide cases than among controls 3 months prior to suicide (the vertical dotted line in Fig. 2a). A similar trend was observed in the percentage of adolescents having any non-psychiatric health-care encounter (Fig. 2b). The percentage of adolescents having any psychiatric health care encounter was significantly higher among suicide cases than among controls during the entire 1-year observation period (Fig. 2c), and a similar trend was observed in the percentage of those having any emergency department visits (Fig. 2d).

Fig. 2. Medical utilization within 12 months before suicide among adolescents who died from suicide compared with the matched controls (95% confidence intervals shown as dotted curves). (a) All outpatient department visits and emergency department visits. (b) Non-psychiatric visits. (c) Psychiatric visits. (d) Emergency visits.

Table 2 shows the patterns of medical utilization 3 months before suicide in cases with suicide mortality and controls. Cases had more outpatient and inpatient visits at non-psychiatric and psychiatric departments. They also had more emergent visits at non-psychiatric departments. Cases had a significantly higher proportion of having any outpatient visit than the living controls (68.2% v. 63.2%, RR = 1.26, p = 0.002). Among adolescents who died by suicide, only 18.6% had an outpatient psychiatric visit within the previous 3 months. The suicide cases had more subspecialty outpatient visits, including those to internal medicine, surgery, gynecology, neurosurgery, urology, neurology, and plastic surgery departments.

Table 2. Patterns of medical utilization within 3 months prior to the index date (suicide death) of cases and the corresponding control group (ratio = 1:20)

^a Estimated using univariable conditional logistic regression.

Distribution of psychiatric disorders and physical illnesses

Tables 3 and 4 present the results of the univariate analysis of psychiatric and physical comorbidities among the cases and controls, respectively. As shown in Table 3, the prevalence of having any psychiatric disorder was 23% among the suicide cases. The cases had significantly higher risks of all psychiatric disorders than did controls. As shown in Table 4, the cases had a significantly higher risk of other forms of heart disease, congestive heart failure, cerebrovascular disease, pneumonia, ulcer disease, diabetes mellitus, and moderate or severe renal disease than did controls.

Table 3. Distribution and multivariate conditional logistic regression of psychiatric disorders within 3 months prior to the index date (suicide death) of cases and the corresponding controls (ratio = 1: 20)

^a Estimated using univariable conditional logistic regression.

^b Estimated using multivariable conditional logistic regression. Variables with p < 0.01 were retained in the final regression model.

*p < 0.01.

Table 4. Distribution and multivariate conditional logistic regression of physical illnesses within 3 months prior to the index date (suicide death) of cases and the corresponding controls (ratio = 1: 20)

^a Estimated using univariable conditional logistic regression.

^b Estimated using multivariable conditional logistic regression. Variables with p < 0.01 were retained in the final regression model.

*p < 0.01.

Multivariable regression analysis

We conducted a multivariable regression analysis of comorbidities in the cases and controls. Among various psychiatric disorders (Table 3), the cases had a higher risk of drug-induced mental disorder (RR = 11.30, p < 0.001), schizophrenia (RR = 24.90, p < 0.001), bipolar disorder (RR = 18.37, p < 0.001), depressive disorder (RR = 20.30, p < 0.001), anxiety disorder (RR = 2.25, p = 0.006), adjustment disorder (RR = 7.52, p < 0.001), and sleep disorder (RR = 3.94, p < 0.001) than did controls. Regarding physical illness (Table 4), cases had a higher risk of other forms of heart disease (RR = 11.07, p < 0.001), pneumonia (RR = 3.14, p = 0.001), and ulcer disease (RR = 3.97, p < 0.001) than did controls.

Discussion

Main findings

The results of this study demonstrated that compared with controls, a higher proportion of adolescents with suicide mortality had visited the psychiatric department, but a similar proportion of adolescents with suicide mortality had visited any non-psychiatric medical department within 1 year prior to suicide. Moreover, a higher proportion of adolescents with suicide mortality had utilized non-psychiatric services within 3 months prior to suicide, particularly services from emergency, surgery, and internal medicine departments. In addition, adolescents with suicide mortality had higher risks of several psychiatric disorders and physical illnesses, including heart diseases, pneumonia, and ulcer disease, than did controls.

Patterns of medical utilization before suicide and suicide methods

The results of the current study revealed that only 18.6% of adolescent suicide victims had contacted with a psychiatric department 3 months before suicide; this finding is in agreement with those of previous studies indicating that although a substantial proportion of adolescents had a health care encounter shortly prior to suicide, only less than one-fifth of them had met a psychiatrist (Chang et al., 2012; Renaud et al., 2009), indicating a large, unmet treatment need for mental health problems prior to suicide. However, 71.8% of adolescent suicide victims had contact with any health service 3 months before suicide; this finding further extended the existing literature by indicating specialty departments that adolescents visited in the 3 months preceding suicide. As expected, the utilization of psychiatric services significantly increased in adolescent suicide victims. Moreover, substantially increased visits to the emergency department were noted; this result is consistent with that of a previous study indicating that emergency department visits due to attempted suicide and severe psychopathology were strongly associated with subsequent suicide risk (Rhodes et al., 2019). These findings highlight the critical role of active suicide screening and identification of at-risk youth in emergency departments (Ballard et al., 2017). In addition, we observed significantly higher utilization of neurosurgery, neurology, urology, plastic surgery, gynecology, surgery, and internal medicine departments, suggesting impaired physical health status prior to suicide in adolescents; this finding concurs with that of a previous study indicating that a substantial proportion of adolescent suicide victims had physical illness (Viilo et al., 2005).

The findings of the current study demonstrated that the discrepancy in health care utilization between adolescent suicide victims and controls tended to be more prominent with shorter delays to suicide during the 1-year observation period, especially for non-psychiatric health services. These findings are consistent with those of a previous study reporting a particularly increased frequency of contact with non-psychiatric hospital-based providers among adolescent suicide victims 1 month prior to suicide (Chang et al., 2012) and imply the importance of enhancing educational programs on suicide prevention for primary care providers. Taken together, these results indicate that adolescent suicide victims received care before suicide not only from mental health services but also from other medical specialties, and the increased medical utilization shortly before suicide provides an opportunity for adolescent suicide prevention.

The current study revealed that violent methods, including hanging and jumping from height, accounted for more than half of all suicide deaths among adolescents, whereas self-poisoning was rare; this finding differed from that observed in the general population in Taiwan, in which hanging, charcoal burning, and pesticides were the three most commonly used methods (Lin, Chang, & Lu, 2010). The particularly high rate of jumping from height among adolescents can be attributed to several factors. First, tall buildings are more accessible to adolescents in Taiwan, a densely populated country (Lin & Lu, 2006), compared with drugs, pesticides, and charcoal. Second, jumping is a highly lethal method that requires little preparation, possibly reflecting the increased impulsivity of adolescents (Renaud et al., 2008). Third, younger people are more exposed to social media and thus more vulnerable to media portrayals of dramatic, detailed circumstances of suicide by jumping (Hawton, Saunders, & O'Connor, 2012). These findings provide crucial information for developing measures for preventing adolescent suicide, especially jumping from height.

Psychiatric comorbidities

The proportion of psychiatric disorders 3 months prior to suicide was only 23% in this study, indicating that a substantial proportion of adolescent suicide victims had not received any psychiatric diagnoses shortly before suicide. This finding is in contrast to the high prevalence of psychiatric disorders (48–94%) observed in previous psychological autopsy studies (Fleischmann et al., 2005; Renaud et al., 2008; Zhang et al., 2010); this difference can be mainly due to the underestimation of psychiatric disorders ascribed to the low rates of seeking treatment in this population. However, adolescents who attempted suicide were more motivated by psychosocial factors, such as interpersonal difficulties, school problems, and family discord, than by illness-related problems compared with their adult counterparts (Lee et al., 2019). These characteristics might also explain the low prevalence of psychiatric disorders in the current study sample and imply that suicide prevention programs should be applied not only in the medical system but also in other treatment venues, such as school-based or social welfare systems, to identify more at-risk youths.

In the current study, adolescent suicide victims had a more than 20-fold increased risk of schizophrenia, bipolar disorder, and depressive disorder compared with healthy controls, confirming the finding of a previous study indicating that a psychiatric history is a strong risk factor for adolescent suicide (Gould et al., 2003; Pelkonen & Marttunen, 2003; Stenager & Qin, 2008).

Mood disorders are the most prevalent disorders among adolescent suicide victims, with their prevalence ranging from 49% to 76% in psychological autopsy studies (Pelkonen & Marttunen, 2003). Adolescents with depression had a sixfold higher risk of attempting suicide (Nock et al., 2013), and approximately one-fourth of adolescents with bipolar disorder had attempted suicide (Hauser, Galling, & Correll, 2013). Furthermore, the severity and clinical characteristics of mood disorders was correlated with suicide risk. For instance, several specific depressive symptoms, such as hopelessness and anhedonia, predicted incident suicide attempts and differentiated those who attempted suicide from those with suicidal ideation (Asarnow et al., 2011; Auerbach, Millner, Stewart, & Esposito, 2015). Characteristics such as earlier onset age, more severe course, and mixed features, which are particularly observed in the pediatric-onset subtype of bipolar disorder, were associated with a higher risk of suicide attempt in adolescents with bipolar disorder (Hauser et al., 2013). These findings collectively underscore the importance of adequate treatment of mood disorders among adolescents in reducing suicide risk.

In this study, schizophrenia accounted for a substantial portion (5.9%) of all youth suicides. The effect of schizophrenia on youth suicide is indispensable; the present study revealed that adolescent suicide victims had a nearly 25-fold increased risk of schizophrenia, the highest among all psychiatric disorders, compared with healthy controls. Moreover, young patients with schizophrenia with a more severe clinical course, non-adherence to treatment, and active psychotic or depressive symptoms have a higher risk of suicide (De Hert, McKenzie, & Peuskens, 2001). Mechanisms linking psychotic experiences and suicidality encompass the neurobiological and psychosocial aspects of adolescents (e.g. levels of psychological distress, traumatic experiences, poor stress coping, and deficits in processing speed resulting in vulnerability to stress) (Hielscher et al., 2021; Kelleher, Clarke, Rawdon, Murphy, & Cannon, 2013). Taken together, these findings imply the key role of evaluating psychotic disorders while assessing the risk of suicide among adolescents.

Physical illnesses

The present study demonstrated that adolescent suicide victims had an increased risk of several physical illnesses 3 months before suicide; this finding is consistent with that of a previous study (Ahmedani et al., 2017). Studies have reported the association of several physical illnesses, including epilepsy, asthma, migraine, concussion, infection, and peptic ulcer disease, with an increased risk of suicide behavior (Bahmanyar, Sparen, Rutz, & Hultman, 2009; Fralick et al., 2019; Lund-Sorensen et al., 2016; Singhal, Ross, Seminog, Hawton, & Goldacre, 2014). In this study, we determined that the risks of ulcer disease, pneumonia, and other forms of heart disease were significantly increased among adolescent suicide victims. The association between ulcer disease and suicide is consistent with previous data revealing a 60–70% excess risk of death due to suicide among adult patients hospitalized for peptic ulcers (Bahmanyar et al., 2009). Moreover, ulcer disease has been proposed to be strongly associated with stress and anxiety (Goodwin & Stein, 2002; Levenstein, Rosenstock, Jacobsen, & Jorgensen, 2015), partially explaining the high suicidality in this population. Furthermore, peptic ulcer disease was found to be associated with Helicobacter pylori infection (Lanas & Chan, 2017), and the infection was reported to be significantly associated with hospitalization for an infection (Lund-Sorensen et al., 2016). Future studies should investigate the effect of H. pylori infection on the risk of suicide in youth. Pneumonia usually indicates the presence of severe systemic infection, which was found to be positively correlated with suicide risk in a graded manner (Lund-Sorensen et al., 2016), and neuroinflammatory mechanisms might lead to high suicidality (Brundin, Erhardt, Bryleva, Achtyes, & Postolache, 2015).

Although no study has directly examined the association between heart disease and suicide in adolescents, some evidence has revealed that medically unexplained chest pain, a common form of functional somatic symptoms in adolescents, was associated with high levels of underlying psychosocial problems such as depression, anxiety, stressful life events, and suicidal ideation (Campo, 2012; Eliacik et al., 2017). In clinical practice in Taiwan, a diagnosis of other forms of heart disease is frequently established when adolescents have medically unexplained chest pain, which might partly explain the association between other forms of heart disease and suicide in the current study.

Strengths and limitations

This study has several strengths. This study investigated the differences in medical utilization and psychiatric and physical comorbidities between adolescent suicide victims and living controls by using a large population-based sample that ensured the enrollment of an adequate number of cases and age- and sex-matched controls. In addition, the indices of medical utilization and the diagnoses of physical and psychiatric diseases were fully recorded in the health insurance database, thus minimizing record bias compared with that encountered in self-report or psychological autopsy designs. Nevertheless, this study has several limitations that should be considered. First, we used claims data from a health insurance database, which provided only demographic variables and lacked detailed circumstances of suicide, such as geographic locations, settings of suicide, and antecedents of suicide. Other crucial factors affecting suicide risk, including childhood maltreatment, family characteristics, school problems, and socioeconomic status (Gould et al., 2003), also could not be obtained. Second, information on the history of suicidal attempts, which can substantially affect suicide risk, could be underreported in the diagnostic coding system. This study observed that six of the 935 cases with suicide mortality and only one of the 18 700 controls had attempted suicide previously. Therefore, we included all suicide cases without excluding the six cases of previously attempted suicide. For comparability, we applied the same selection criteria for cases and controls. Future studies should investigate the association of suicide attempts with suicide mortality in adolescents. Third, we used a national mortality database to track each suicide event. However, those who died from suicide but whose death was misclassified as mortality from other causes might have been neglected. The most prominent could be accidental death, which was used to prevent the potential stigma associated with suicide. Finally, we estimated psychiatric and physical illness among adolescent suicide victims by using diagnostic codes recorded in the health insurance database; this could have caused adolescents with subsyndromal psychopathology or those who did not enter the medical system 3 months before suicide to be overlooked, thus possibly leading to the underestimation of the prevalence of psychiatric or physical illness among adolescent suicide victims.

Implications

The present study highlights that adolescents who died from suicide had higher medical utilization within 3 months before suicide across several medical departments, particularly emergency, surgery, and internal medicine departments. However, the potential warning signs could not be detected earlier in the present health-care system. Implementing screening programs into such specific departments and strengthening referral systems to mental health subspecialties might reduce suicide risk in this patient population. Furthermore, the finding of the higher risks of several physical and psychiatric comorbidities among adolescent suicide victims is crucial for developing specific interventions for suicide prevention in this patient population.