Suicide is the second leading cause of death among adolescents and young adults in the United States (
1). Rates of suicide increased from 1999 to 2014 for all age groups under age 75, with young adolescent girls (ages 10–14) demonstrating the largest increase (200%) in the 15-year period (
2). Recent estimates from the Youth Risk Behavioral Surveillance Survey indicated that nearly one in five students (17.7%) seriously considered attempting suicide and that almost one in ten (8.6%) reported making a suicide attempt (
1).
Self-injurious behavior, with or without intent to die, is the most reliable predictor of future suicide attempts and death by suicide (
3,
4). Although there are important distinctions and incomplete overlap between youths who engage in nonsuicidal self-injury and those who attempt suicide, previous research found that nonsuicidal self-injury was second only to suicidal ideation as a predictor of suicidal behavior (
5). Despite generative efforts to understand, develop, and evaluate interventions targeting self-injury in outpatient psychiatric (
6–
9), emergency medical (
10), and community-based settings (
11–
13), the clinical course and outcomes among adolescents who have prior self-injury is not well characterized and remains understudied (
14). Information about the typical patterns of service use is vital to improve outcomes for youths, because it can underscore gaps in usual care (
15). This may be particularly important for addressing the secondary and tertiary prevention needs of youths who have presented to the hospital following self-injury.
There are substantial gaps in understanding the role of hospitals in providing treatment for self-injury. Recent scholarship has emphasized the importance of comprehensive approaches to suicide prevention (e.g., Zero Suicide [
16]) to increase the effectiveness of the response to individuals at risk. Risk identification across published studies is poor (
17), and elucidating patterns of repeated self-injury in the context of typical service use patterns would be valuable to direct outreach efforts. Knowledge about usual care for adolescents who have medically treated self-injuries is limited (
18–
21), and existing data sources can be leveraged to understand the trajectories associated with conventional service use patterns among adolescents who have history of self-injury.
Data from adult psychiatric inpatient populations have suggested that there may be limitations to conventional health care responses to acute suicidality and self-injury (
22–
24), in that inpatient psychiatric care may not yield clinically meaningful reductions in suicidality over time (
22). Access to and engagement in outpatient psychiatric services among adolescents recently discharged from acute psychiatric care remain limited (
25–
29). The findings illustrating difficulties with engagement in and use of psychiatric services are based on clinical trials, and evaluation of usual care patterns at a national level with a generalizable sample of youths followed over time has not been published.
Prior work has examined the impact of time and of demographic and clinical factors on encounters related to suicidal thoughts and behaviors in emergency departments and acute care hospitals by using the Pediatric Health Information System data set. This research highlighted that hospital care for youths with suicidal thoughts or behaviors doubled in the time range evaluated from 2008 to 2015 (
30). Furthermore, increases were observed in all age groups and racial and ethnic groups, with more substantial change among older adolescents, girls, and non-Hispanic white youths (
30). These authors did not differentiate between youths who reported thinking about suicide and youths who engaged in self-injury. We sought to extend this study by evaluating the demographic and service use characteristics of youths who had a prior medically treated self-injury, given that this is the most reliable predictor of a subsequent suicide (
3,
17) and that only a subset of youths who consider suicide will harm themselves (
1).
Given the limited evidence focused directly on youths who have medically treated self-injury, this study sought to compare the clinical characteristics of children and adolescents who were admitted for inpatient psychiatric care and those who were not. In addition, this study explored differences in the risk of repeated self-injury following an index episode of medically treated self-injury. Having more information about typical service use practices and behavioral patterns for high-risk youths will contribute to a better understanding of opportunities for designing suicide prevention efforts.
Methods
Data Source
Data were collected from the IBM/Watson (formerly Truven Health Analytics) MarketScan Commercial Claims and Encounters Databases (2007–2016). These databases contain inpatient, outpatient, and pharmacy claims for patients covered by employer-sponsored commercial insurance across the United States. The inpatient and outpatient claims databases include procedure- and visit-level details from medical claims, including ICD-9-CM and ICD-10-CM diagnoses and CPT medical procedure codes, dates of service, and variables describing the financial expenditures of both the patients and their insurance plans. The outpatient pharmacy claims database provides prescription dispensing details that include National Drug Code and generic identifiers of the drugs dispensed, dates dispensed, quantities, days’ supply, and payments made for each claim. A separate eligibility and demographics file provides additional information about each patient, such as age, gender, insurance plan type, geographic location, and enrollment status by month.
Inclusion and Exclusion Criteria
We identified patients with inpatient or outpatient codes for self-injury or suicidal ideation accompanied by a poisoning or a wound diagnosis code by using the algorithm identified by Simon and colleagues (
31) (a table listing
ICD-9-CM and
ICD-10-CM codes used for inclusion criteria is available in the
online supplement to this article). For patients with multiple instances of self-injury, we counted the first appearance of these codes in the data as the index self-injury. We excluded patients who were under age 6 and over age 18 at the time of the index self-injury. Because we were interested in health care utilization for 1 year before and after the index self-injury, we included patients whose index event was between January 1, 2008, and December 31, 2015, and excluded patients who did not have at least 365 days of enrollment before and after their index self-injury. Furthermore, because we were interested in associations between hospitalization in a psychiatric facility and subsequent self-injury, we excluded patients who had inpatient hospitalizations in psychiatric facilities more than 7 days before or after the index self-injury.
Variable Definitions
The main outcome variable of interest was whether patients had a subsequent medically treated self-injury in the year following the index self-injury. We used the Simon and colleagues’ (
31,
32) algorithm to identify subsequent medically treated self-injury and required at least 8 days to have elapsed between the index and the subsequent self-injury. Five categories of psychiatric disorders in the year prior to the index self-injury, including unipolar mood disorders, bipolar mood disorders, psychotic disorders, behavior disorders, and anxiety disorders were created, and a summary score ranging from 0 to 5 was used to indicate the number of disorder diagnoses recorded for each patient. Health care utilization in the year prior to and the year following the index self-injury, including inpatient admissions at any type of facility (psychiatric facilities, residential substance abuse facilities, and comprehensive inpatient rehabilitation facilities), was assessed. Similarly, we examined outpatient visits at psychiatric facilities, including community mental health centers, residential substance abuse facilities, psychiatric residential treatment centers, nonresidential substance abuse facilities, and comprehensive rehabilitation facilities.
We also assessed emergency department visits. Medications filled for any indication in the following therapeutic classes were counted: antihistamines and combinations, anticholinergic-antiparkinsonian agents, anticholinergic-antispasmodic agents, sympathomimetic agents, centrally acting alpha2-adrenoceptor agonists, alpha or beta blockers, opiate agonists, benzodiazepines, succinamides, miscellaneous anticonvulsants, antidepressants, and tranquilizers-antipsychotics. Health care use that occurred within 1 week before or after the index self-injury was considered to have been related to the index event and thus was not included as pre- or postindex use. We also examined whether patients received mental health assessments for their index self-injury (defined as having ICD-9-CM codes V70.1 or V70.2 or ICD-10-CM code Z04.6 within 1 day of the index self-injury).
Analytic Strategy
We performed descriptive statistics for demographic variables, including age at the time of the index self-injury, gender, type of health insurance plan, geographic region, the calendar year of the index self-injury, whether the patients received mental health assessments during their index self-injury event, and whether they received diagnoses for mood disorders in the year prior to the index self-injury.
For each health care utilization described above, we examined the proportion of our population that had at least one utilization in the year prior to the index self-injury, and we separately examined the proportion that had at least one utilization in the year after the index self-injury (again excluding the 7 days before and after the index self-injury). We also examined the mean and standard deviation of the number of times each utilization occurred in each of the 1-year periods before and after the index self-injury. All utilizations were stratified by whether patients had hospitalizations in a psychiatric facility within 7 days of the index self-injury. We tested the significance of the differences between those who were and were not hospitalized in a psychiatric facility by using chi-square tests for categorical variables and t tests for continuous variables. Alpha was adjusted (i.e., Bonferroni correction) for multiple comparisons (k=103), and only p values less than 0.0005 were considered statistically significant.
To assess the risk of subsequent self-injury, we used Cox proportional hazards regression to find hazard ratios (HRs) and 95% confidence intervals (95% CIs). We also performed a log-rank test to determine whether the hazard curves significantly differed depending on whether the patients were hospitalized in a psychiatric facility in the 7 days before or the seven days after the index self-injury. We adjusted for patient age (continuous), sex, region of residence, health insurance plan type, calendar year of index self-injury, and number of psychiatric diagnosis categories that the patients had in the year prior to the index self-injury. Because some patients were missing data on geographic region and health insurance plan type, we performed multiple imputation (
33) to impute these adjustment variables.
Finally, for sensitivity analyses, we performed the same analyses stratified by whether patients were hospitalized at any time in the year before and the year after the index self-injury, in the 8 to 365 days before the index self-injury, and in the 8 to 365 days after the self-injury event.
Compliance With Ethics Guidelines
This study was exempt from review by the University of Washington Institutional Review Board by self-determination under category 4 secondary data analysis.
Results
Patient Characteristics
A total of 262,034 patients had codes for at least one self-injury event from 2008 to 2015 (see patient flow diagram in online supplement). Of these patients, 28% (N=73,657) were between the ages of 6 and 18 at the time of their index self-injury. Of these, 32,281 (44%) had at least 1 year of enrollment prior to and after the index self-injury, and of those, 1,134 youths who were hospitalized at some other time within the 2-year period were excluded, leaving 31,147 (96%) who either did not have a hospitalization in a psychiatric facility at any time in the year before or after the index self-injury or had a hospitalization related to the index injury. Among the 31,147 patients, 128 (0.4%) died—most (N=100, 78%) from the index self-injury; most of those who died (N=80, 63%) were ages 15–18. Of the 31,147 patients who met our inclusion criteria, 527 (1.7%) had a subsequent medically treated self-injury in the following year. No patients had more than one subsequent medically treated self-injury in the year following the index self-injury.
Table 1 presents data on the characteristics of the 31,147 patients in the study cohort, by whether or not they were hospitalized within 7 days of the index self-injury. The proportion of females was greater among the hospitalized patients (73%), compared with those who were not hospitalized (54%). The region with the largest proportion of hospitalized patients was the West. The proportions of youths who had diagnoses of the psychiatric disorders evaluated in this study were larger among hospitalized youths (
Table 2).
Services Utilization Prior to Index Self-Injury
Table 3 compares service utilization by whether patients were hospitalized within 7 days of the index self-injury. In the 12 months prior to the index self-injury, patients who were hospitalized had higher levels of psychiatric and health services utilization, compared with those who were not hospitalized (
Table 3). In the year prior to the index self-injury, the rate of attending one or more outpatient visits for psychiatric services was 12% among hospitalized patients, compared with 2% in the nonhospitalized group. However, among patients with at least one outpatient visit, the mean number of outpatient visits in the year prior to the index did not differ significantly between the groups (
Table 3). In the year prior to the index, the hospitalized group had a higher rate of use of any outpatient health services, compared with the nonhospitalized group. A larger proportion of hospitalized patients filled prescriptions for at least one indicated medication, compared with the nonhospitalized patients (74% versus 62%) (
Table 3).
Services Utilization Subsequent to Index Self-Injury
As shown in
Table 3, in the year after the index self-injury, the rate of inpatient psychiatric hospitalizations was higher among patients hospitalized within 7 days of the index injury, compared with those who were not (31% versus 11%). Similarly, the rate of emergency department use was higher among the hospitalized patients (49% versus 35%). A larger proportion of hospitalized patients used any outpatient psychiatric services in the year after the index self-injury (25% versus 3%). Among patients with at least one outpatient visit in the year after the index self-injury, the mean number of visits was significantly higher among nonhospitalized patients, compared with the hospitalized group. A small but statistically significant difference was noted between nonhospitalized (95%) and hospitalized (98%) patients with regard to accessing any outpatient health services in the year following the index self-injury. The mean number of visits to any type of facility was higher among hospitalized patients, compared with nonhospitalized patients (26.1 versus 14.4 visits). As in the year prior to the index self-injury, the proportion of patients who filled prescriptions for at least one indicated medication was higher among the hospitalized patients (82% versus 65%).
Self-Injury in the Year Following the Index Self-Injury
In the year after the index self-injury, a larger proportion of the hospitalized patients sought medical attention for self-injury, compared with the nonhospitalized group (5.4% versus 1.4%) (
Table 3). As shown in
Figure 1 and
Table 4, youths who were hospitalized in a psychiatric facility within 7 days of the index self-injury were twice as likely as those who were not hospitalized to have another medically treated self-injury in the year after their index self-injury (adjusted HR=2.13, 95% CI=1.72–2.67), after the analysis adjusted for age, gender, geographic region, year of index self-injury, and number of categories of psychiatric diagnoses in the year prior to the index (p value for log-rank test <0.001). However, no significant difference was found between the hospitalized and nonhospitalized group with regard to the median time to onset of another medically treated self-injury (hospitalized, median=95 days; nonhospitalized, median=117 days; p=0.34).
We observed similar trends of when we compared patients who were hospitalized at any time before or after the index self-injury (see tables and figures in online supplement).
Discussion
The primary goals of this research were to understand psychiatric services use and the recurrence of medically treated self-injury for youths over 2 years of consecutive coverage in a commercial claims data set. There were 31,147 encounters for medically treated self-injury during the study period (2008–2015). Of these, 6.5% were admitted to a psychiatric inpatient facility within 7 days of the index self-injury, and 1.7% had a medically treated self-injury in the year after the index self-injury. Similar to past research (
30), this analysis of a large nationwide sample demonstrated an increase in visits specifically for medically treated self-injury over the study period. The results of this study indicate that youths who were hospitalized in psychiatric facilities within 7 days of the index self-injury were more likely to be female and to have had a hospitalization in the year prior to their index self-injury.
This study yielded important insights into the opportunities for suicide prevention from medical services. First, in the year prior to the index self-injury, an overwhelming majority of youths were seen in medical settings (94%−96%), whereas a smaller proportion were seen in outpatient psychiatric services (2%−12%). Consistent with prior data highlighting frequent health care contact in the year prior to suicide death (
34,
35), this finding suggests that the opportunity for suicide prevention may arise for medical service providers from outside psychiatric services; these professionals need to be aware of suicide risk, provide screening for suicide, and encourage treatment options (
36,
37). Estimates of the number of deaths prevented through various suicide prevention strategies suggest that training general practitioners in these activities may significantly decrease the number of lives lost to suicide (
36,
38).
Patients in our study who were engaged in outpatient psychiatric treatment received far fewer sessions than would be recommended with a suicide-specific intervention, with an average of seven or eight visits per year. Following hospitalization, use of outpatient psychiatric services increased, with 25% of youths receiving some sort of outpatient treatment, suggesting that inpatient care may be an entree into services but does not necessarily yield higher frequency of service engagement over time. The low level of engagement in psychiatric services highlights the need to make suicide-specific outpatient services more accessible, appealing, and engaging for youths and families. It may be that our current approach to intervening with youths is not “sufficiently effective and sufficiently benign” (
36) and that alternative service delivery strategies are needed.
Youths were seen on average two times in the ED each year, indicating that the ED is a common portal for entry into psychiatric services (
39). There is an important role for emergency department interventions that are therapeutic and offered to all patients with self-injury, consistent with guidelines from the National Institute for Health and Care Excellence and recommended standard care from the National Action Alliance for Suicide Prevention; such interventions include assessment, safety planning, lethal-means counseling, and discharge planning (
37,
40–
43). Interventions integrated within emergency services show promise in reducing the burden of suicide, increasing connection to ongoing treatment (
44,
45), and providing support during transition from acute to outpatient care (
46). We observed a strong positive association between psychiatric hospitalization with 7 days of the index self-injury and future medically treated self-injury, similar to prior work evaluating the impact of hospitalization on suicidal ideation (
47). It is important to explore innovations in suicide care delivered on inpatient psychiatric units that can enhance the durability of crisis stabilization and improve patient outcomes over longer intervals.
This study had limitations that need to be acknowledged to contextualize its contribution to the literature. First, the MarketScan data set is limited to commercial insurance plans and does not include public insurance information. The data set also has limited social, economic, clinical, and demographic information, and the analyses were necessarily constrained by that limitation. Administrative data may not have fully captured the spectrum of self-injuries, because many are not medically treated. Geulayov and colleagues (
48) observed that suicide deaths and medically treated self-injuries are overt indicators of broader patterns of covert self-injury, including nonsuicidal self-injury. It is important to consider the extent to which our study may have underestimated the larger ecology of self-injurious behavior among adolescents with differing patterns of psychiatric services utilization. Our data were also limited with regard to important metrics of mental illness, such as disease severity, change in symptoms, and response to treatment. In particular, the observed differences may reflect differences in psychiatric illness severity, rather than a direct effect of inpatient psychiatric care upon future risk of suicidal thoughts and behaviors. Second, our inclusion criteria required 2 consecutive years of data in order to provide person-level indicators over time, and thus there may have been a selection bias in the sample for families who have more occupational stability. Finally, although we observed a statistically significant relationship between inpatient hospitalization and subsequent self-injury, we cannot be certain that this is not the result of residual confounding.
Conclusions
Our study represents a large evaluation of service use among youths who were medically treated for a self-injury from 2008 to 2015. The findings support prior research that reported increased numbers of medical and emergency visits for medically treated self-injury among youths. Our study aimed to evaluate individual patterns of care to understand opportunities to augment suicide prevention strategies within the health care system. In this retrospective, observational study examining characteristics of repeated medically treated self-injury, psychiatric hospitalization was strongly associated with recurrence of a medically treated self-injury in the subsequent year. This suggests that hospitalization is utilized for high-risk groups and represents a potent indicator of increased risk of later self-injury.