INTRODUCTION
Suicide remains one of the leading causes of premature mortality worldwide, accounting for 1.3% of all deaths in 2019 [1]. South Korea, in particular, faces a severe challenge, reporting the highest suicide rate among the OECD countries in 2019, with 24.6 suicides per 100,000 population [2]. Between 2011 and 2016, suicide accounted for 5.26% of all deaths in South Korea, with 84,934 suicides out of a total of 1,615,288 deaths [3].
Psychiatric disorders are well-known risk factors for suicidal ideation, self-harm, and suicide attempts [4], and suicide prevention is therefore a critical issue in psychiatric populations [5]. Recent systematic reviews and meta-analyses have confirmed that individuals with psychiatric disorders have a markedly higher risk of suicide than those without such conditions, and that this increased risk has remained consistent across different regions and over time [6,7]. Importantly, the magnitude of suicide risk differs according to the type of psychiatric disorder, and the presence of multiple psychiatric conditions further amplifies the risk [8]. In addition, psychiatric disorders are associated not only with suicide but also with an increased risk of non-suicidal deaths, including premature mortality from cardiovascular and metabolic diseases [9,10].
However, few studies have simultaneously compared the risks of suicide and non-suicidal deaths in individuals with psychiatric disorders relative to those without such disorders. Moreover, limited research has evaluated the cumulative effects of multiple psychiatric diagnoses or examined less frequently studied conditions such as anorexia nervosa or attention-deficit/hyperactivity disorder.
To address these gaps, the present study aimed to investigate whether the risks of suicide and non-suicidal death vary according to psychiatric disorder status. We hypothesized that psychiatric disorders would confer a disproportionately higher risk of suicide mortality compared with non-suicidal mortality, and that the risk would further escalate with increasing numbers of psychiatric disorders.
METHODS
Design and ethical considerations
This population-based cohort study adhered to the standards set by the Strengthening the Reporting of Observational Studies in Epidemiology [11]. The Institutional Review Board of Seoul National University Bundang Hospital approved this retrospective cohort study (license number X-2303-819-902). The National Health Insurance Service (NHIS)’s Big Data Center (NHIS-2023-1-526) granted permission to access and share the data. Since the data were anonymized and the study was conducted retrospectively, informed consent was not required for analysis.
NHIS database
All data for this study were provided by the NHIS, the sole public insurance program in South Korea. The NHIS database, as mandated by law, maintains records of all disease diagnoses and prescription information for treatments, medications, or both. Enrollment in the NHIS is required for individuals to qualify for government-funded health insurance programs, and all diagnoses are coded according to the International Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10). Residents who have lived in South Korea for over 6 months are required to register with the NHIS. Additionally, the NHIS database includes comprehensive information on individuals’ socioeconomic status and date of death [12].
Study population
We initially requested data from the NHIS for all adult patients (≥18 years old) diagnosed with psychiatric disorders who were receiving treatment between January 1, 2017 and December 31, 2017. The ICD-10 codes for each psychiatric disorder included in this study are listed in Supplementary Table 1. After receiving information indicating a population size exceeding 5 million, we proceeded to request a stratified 10% sample based on age and sex. Thus, adults diagnosed with psychiatric disorders in 2017 were defined as the PY group in this study. After extracting all members of the PY group from 2017 in South Korea, we requested data for individuals in the non-PY group, who had never been diagnosed with psychiatric disorders, using a 1:1 stratified random sampling technique, also considering age and sex. Because we tracked deaths by suicide from 2018 to 2022, individuals who died in 2017 were excluded from the analysis.
Study endpoint
The study aimed to track the occurrence of suicide-related deaths over a specific period, from January 1, 2018 to December 31, 2022. Cases of suicide death were identified using ICD-10 codes X60-X84, as outlined in a previous study [13]. Physicians in South Korea accurately record all causes of death, which are then stored in the Statistics Korea Database. As a central government body, Statistics Korea is responsible for planning and coordinating national statistics. All other deaths, excluding those caused by suicide, were classified as non-suicidal deaths.
Analyzed covariates
Age and sex of the patients were collected as demographic variables. Socioeconomic status was assessed using household income, residence, and employment status (including self-employment). Residence was categorized as urban (in Seoul or other large cities) or rural (all other locations). Household income was divided into five groups, of which one included individuals enrolled in a medical aid program, with the remaining groups based on a four-quartile distribution. Individuals who are unable to afford insurance are categorized by the government into groups eligible for medical aid programs.
To assess the presence of multiple comorbid conditions, the Charlson Comorbidity Index score was calculated using the most recent ICD-10 codes from the NHIS database (Supplementary Table 2). To qualify for social welfare benefits in South Korea, individuals must report any disabilities to the NHIS database. Consequently, data on underlying disabilities were also collected. Disability evaluations are conducted by specialists in relevant fields in South Korea. Once a qualified physician assesses the impact of the condition on daily life, disabilities are classified as either mild-to-moderate or severe. The classification of disabilities is detailed in Supplementary Table 3.
Statistical analysis
Continuous variables were analyzed by presenting mean values and standard deviations. For categorical variables, percentages and absolute numbers were used. To minimize differences in covariates between the PY group and the non-PY group, we conducted propensity score (PS) matching, a method recognized for its ability to reduce bias in observational studies [14]. In this study, PS matching was performed using the nearest neighbor method, with a caliper width of 0.25 and a 1:1 ratio without replacement. After PS matching, the balance between the two groups was assessed using the absolute standardized mean difference (ASD). An ASD of less than 0.1 was considered indicative of appropriate PS matching.
In the PS-matched cohort, we conducted cause-specific Cox regression analyses to estimate the risk of suicide and non-suicidal deaths associated with psychiatric disorders (PY vs. non-PY). For this analysis, an event was defined as mortality due to suicide or non-suicidal causes occurring on or after January 1, 2018, and time was measured from January 1, 2018, until the occurrence of the event. To formally compare whether the association with mortality differed by cause, we additionally applied a Lunn-McNeil cause-stratified Cox model on a person-cause stacked dataset, including an interaction term between PY status and cause of death (suicide vs. non-suicidal death). A statistically significant interaction term indicated heterogeneity of the PY effect across causes of death [15]. As a sensitivity analysis, we repeated multivariable Cox regression in the entire cohort, adjusting for all covariates to confirm the consistency of results with those from the PS-matched cohort. To further explore disorder burden, we conducted multivariable Cox regression analyses comparing individuals with two, three, or more psychiatric disorders to those with only one psychiatric disorder. The number of disorders was determined based on distinct ICD-10 diagnostic categories recorded in 2017. If a patient had diagnostic codes from more than one category (e.g., schizophrenia and depression), each was counted as a separate disorder. We acknowledge that differential diagnoses may occasionally be coded alongside primary diagnoses, which could lead to overestimation.
Finally, we performed disorder-specific multivariable Cox regression analyses to examine whether the risks of suicide and non-suicidal death varied according to individual psychiatric disorders (e.g., major depressive disorder, bipolar disorder, schizophrenia, alcohol-related disorder, anxiety disorder, anorexia nervosa, attention-deficit/hyperactivity disorder). Disorder-specific heterogeneity of effects between suicide and non-suicidal mortality was also assessed using PY×cause interaction terms.
Lastly, subgroup analyses were conducted based on sex and age. Log-log plots were used to verify that the underlying assumptions of the Cox proportional hazards models were met. The results were presented as hazard ratios (HRs) with 95% confidence intervals (CIs). The multivariable model showed no evidence of multicollinearity among the variables, as indicated by a variance inflation factor of 2. All statistical analyses were conducted using R software (version 4.0.3, R Foundation for Statistical Computing), with a significance threshold set at p<0.05.
RESULTS
Study population
Figure 1 depicts the flowchart for the patient selection process. In 2017, 497,534 adults in the PY group were initially screened, compared to 493,150 adults in the non-PY group. After excluding 16,221 individuals who died in 2017, the study included 974,463 adult participants. Following 1:1 PS matching, 761,384 adult participants (380,692 in each group) were included in the final analysis. Table 1 presents the clinicopathological characteristics of the PY and non-PY groups before and after PS matching. All ASDs in the PS-matched cohort were less than 0.1, indicating that the two groups were well balanced through PS matching. Supplementary Figure 1 illustrates the distribution of PSs before and after PS matching, showing that the distributions became similar following PS matching.
Analyses in PS-matched cohort
Table 2 shows the results of survival analyses before and after PS matching. In the PS-matched cohort, 0.2% (830/380,692) of individuals in the PY group died by suicide, compared to 0.0% (13/380,692) in the non-PY group. Cox regression analysis revealed that the PY group had a 64.43-fold higher risk of suicide death (HR: 64.43, 95% CI: 37.25-111.43, p<0.001) compared to the non-PY group. Additionally, 8.6% (32,746/380,692) of the PY group died from non-suicidal causes, while 7.1% (27,043/380,692) of the non-PY group died from nonsuicidal causes. Cox regression indicated that the PY group had a 1.22-fold higher risk of non-suicidal death (HR: 1.22, 95% CI: 1.20-1.24, p<0.001) compared to the non-PY group. To formally test whether the strength of association differed by cause, we fitted a cause-stratified Cox model including an interaction between PY status and cause of death. The PY×cause interaction was statistically significant (p<0.001), indicating that the excess mortality associated with psychiatric disorders was substantially greater for suicide than for non-suicidal causes.
Analyses in entire cohort
Table 3 presents the results of the multivariable Cox regression model for suicide mortality in the entire cohort. In Model 1, the PY group had a 68.87-fold higher risk of suicide death (HR: 68.87, 95% CI: 40.71-113.13, p<0.001) compared to the non-PY group. In Models 2 and 3, individuals in the ≥2 PY group and ≥3 PY group had a 144.06-fold (HR: 144.06, 95% CI: 86.15-240.90; p<0.001) and a 266.47-fold (HR: 266.47, 95% CI: 157.93-449.63, p<0.001) higher risk of suicide death, respectively, compared to the non-PY group. In Model 4, significant associations between increased suicide mortality and specific psychiatric disorders were observed in patients with anorexia nervosa (HR: 3.10, 95% CI: 1.94-4.96, p<0.001), alcohol-related disorder (HR: 2.55, 95% CI: 2.12-3.08, p<0.001), anxiety disorder (HR: 2.34, 95% CI: 2.06-2.65, p<0.001), bipolar disorder (HR: 2.40, 95% CI: 2.06-2.80, p<0.001), major depressive disorder (HR: 3.90, 95% CI: 3.42-4.46, p<0.001), and schizophrenia (HR: 2.67, 95% CI: 2.24-3.18, p<0.001). All HRs with 95% CIs in multivariable Model 1 are presented in Supplementary Table 4.
To complement these findings, Supplementary Table 5 presents the corresponding multivariable Cox models for nonsuicidal mortality, stratified by psychiatric disorder categories. These analyses demonstrated that while non-suicidal mortality risks were modestly increased across most disorders, the relative increase in suicide mortality was substantially greater, consistent with the significant PY×cause interaction results.
Subgroup analyses
Table 4 shows the results of subgroup analyses. The PY group had a higher risk of suicide death compared to the non-PY group in both the male (HR: 68.67, 95% CI: 34.11-138.26, p<0.001) and female (HR: 67.12, 95% CI: 31.77-141.81, p<0.001) subgroups. Additionally, the PY group showed an increased risk of suicide death compared to the non-PY group in the 18-40 years age group (HR: 220.52, 95% CI: 30.89-1,574.29, p<0.001) and the ≥61 years age group (HR: 30.72, 95% CI: 17.98-52.47, p<0.001).
DISCUSSION
This retrospective population-based cohort study demonstrated that psychiatric disorders are associated with a significantly elevated risk of death by suicide in South Korea. Our investigation revealed that this increase is substantially greater than the risk of non-suicidal deaths. Additionally, the risk of death by suicide rose sharply as the number of underlying psychiatric disorders increased. When considering individual psychiatric disorders, significant associations with increased suicide mortality were observed in patients with anorexia nervosa, alcohol-related disorder, anxiety disorder, bipolar disorder, major depressive disorder, and schizophrenia.
The most important finding of this study is that psychiatric disorders were associated with a disproportionately greater increase in suicide mortality compared with non-suicidal mortality. Importantly, this difference was confirmed by a formal interaction analysis in a cause-stratified Cox model, which demonstrated significant heterogeneity of effect by cause of death. These results highlight that the excess mortality associated with psychiatric disorders is driven primarily by suicide, rather than by non-suicidal causes. Notably, the overall HR for suicide mortality in the PY group was extremely high (HR≈68). This should be interpreted in the context of the very low baseline suicide mortality observed in the non-PY group (<0.01%, only 13 deaths), which inflates relative HRs. Furthermore, the PY group encompassed a wide range of psychiatric conditions, including major depressive disorder, bipolar disorder, schizophrenia, and alcohol-related disorders, many of which are independently associated with markedly increased suicide risk. Thus, the pooled estimate reflects the aggregate impact of multiple high-risk conditions. Previous meta-analyses have similarly reported that individuals with psychiatric disorders face dramatically increased suicide risk relative to the general population, with effect sizes ranging from 20-fold to over 60-fold higher, supporting the plausibility of our findings [7,8].
Although various studies have reported an increased risk of suicide mortality and all-cause mortality among individuals with psychiatric disorders [8], risks of suicide mortality and non-suicidal mortality have not been previously compared within this population relative to individuals without psychiatric disorders. A recent cohort study reported that death due to suicide in psychiatric patients is distinct from other causes of death in South Korea [9]. However, the risk of suicide mortality and all-cause mortality was not compared relative to individuals without psychiatric disorders, and it included only four psychiatric disorders: depressive disorder, bipolar affective disorder, schizophrenia, and alcohol use disorder. Therefore, our findings provide a broader and more informative perspective than previous studies.
Our study also demonstrated that the risk of death by suicide increases dramatically with the presence of multiple psychiatric disorders. Although mental illness is reportedly associated with an increased risk of suicide [7], the growing risk of suicide in people with multiple comorbid psychiatric disorders has been poorly understood. Recent research suggests that many psychiatric disorders share genetically overlapping features, which is crucial information for clinicians [16]. This knowledge enables clinicians to identify patients with one psychiatric disorder who may also have other mental illnesses, thereby improving the screening process for those at higher risk of suicide.
Significant associations between increased suicide mortality and specific psychiatric disorders were observed in individuals with anorexia nervosa, alcohol-related disorder, anxiety disorder, bipolar disorder, major depressive disorder, and schizophrenia. Depressive disorder and alcohol-related disorder are well-known risk factors for death by suicide [17]. Anxiety disorder and bipolar disorder have also been reported as significant risk factors for suicide [18,19]. In patients with schizophrenia, suicide is reportedly not a major cause of death [20]. The second most common cause of mortality in individuals with anorexia nervosa is suicide [21]. Therefore, our study further characterizes psychiatric disorders that are already recognized as increasing suicide risk.
An interesting finding regarding age emerged from the subgroup analysis. Individuals with psychiatric disorders showed an increased risk of suicide in the 18-40 years age group and the ≥61 years age group, whereas no significant impact of psychiatric disorders on suicide risk was observed in the 41-60 years age group. The absence of a significant association in the 41-60 years subgroup may reflect differential treatment-seeking behavior in this age group. Since our data were derived from individuals who accessed psychiatric care, it is possible that a substantial proportion of middle-aged adults with psychiatric disorders remained untreated and were therefore misclassified as non-PY, attenuating the observed association. This indicates that comorbid psychiatric disorders contribute to an increased risk of suicide in both younger and elderly adults. Previous research has identified psychiatric disorders as significant risk factors for suicidal behavior among young people [22]. The prevalence of suicidal behavior among young individuals is a major public health concern [23]. Notably, approximately 90% of young people who have made suicide attempts have also been diagnosed with psychiatric disorders [23]. Additionally, the high rate of suicide among the elderly is another critical public health issue [24]. These findings can inform the development of targeted suicide prevention policies.
Our study has some limitations. First, the study was constrained by the NHIS database, which does not include information on smoking history, an important risk factor for suicide. Second, although this is a large cohort study, the data were collected retrospectively, which may have introduced unmeasured confounders and potential biases. Third, our analyses did not include certain psychiatric conditions such as posttraumatic stress disorder or substance use disorders beyond alcohol-related disorder. These conditions are either uncommon in the claims database or difficult to reliably capture using administrative ICD-10 codes, and their exclusion may have led to underestimation of psychiatric morbidity. Fourth, the number of psychiatric disorders was defined by distinct ICD- 10 codes, which may have overestimated the true prevalence of multiple disorders if differential diagnoses were coded in addition to primary diagnoses. In addition, because our study relied on insurance claims data, misclassification of exposure is possible. Individuals with psychiatric disorders who did not seek or receive treatment may have been included in the non-PY group. Such nondifferential misclassification would most likely bias the observed associations toward the null, suggesting that the true effect of psychiatric disorders on suicide risk may be even greater than our estimates. Finally, the results of this study are based on data from South Korea and may not be generalizable to other countries.
In conclusion, our study revealed that psychiatric disorders significantly increase the risk of suicide in South Korea. This increase is substantially greater than the risk of non-suicidal deaths. The risk of death by suicide rose sharply as the number of underlying psychiatric disorders increased. Specifically, anorexia nervosa, alcohol-related disorder, anxiety disorder, bipolar disorder, major depressive disorder, and schizophrenia were associated with increased suicide mortality.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Supplement1
Print
