Risk Factors for Deliberate Self-harm and Suicide Among Adolescents and Young Adults With First-Episode Psychosis

Aubrey M. Moe; Elyse Llamocca; Heather M. Wastler; Danielle L. Steelesmith; Guy Brock; Jeffrey A. Bridge; Cynthia A. Fontanella


Schizophr Bull. 2022;48(2):414-424. 

In This Article



The primary outcomes were the first DSH claim (e.g., overdose, cutting, falls) during follow-up (see supplementary table S1 for ICD-9-CM and ICD-10 codes)[22,35] and suicide (ICD-10 codes X60-X84, Y87.0, *U03). Secondary outcomes included unintentional deaths (ICD-10 codes V01-X59, Y85-Y86),[36] and all-cause mortality.

Sociodemographic and Clinical Characteristics

Demographic variables included age at index diagnosis (adolescents: 15–19 or young adults: 20–24), gender, race/ethnicity, area of residence, and Medicaid eligibility. Index diagnoses were categorized as either schizophrenia-spectrum or affective disorder with psychosis using the most frequent diagnosis category (supplementary table S2). Clinical comorbidities, treated as time-varying covariates, were identified based on claims during the pre-period or the follow-up period. Youth were identified as having a specific psychiatric comorbidity if they had at least one inpatient claim and/or at least two outpatient claims associated with the diagnosis (supplementary table S2). Medical comorbidities were classified as the presence of at least one diagnosis code for either a complex chronic condition (e.g., cancer, cystic fibrosis), a non-complex chronic condition (e.g., asthma, obesity), or no chronic condition. A classification system utilized in the past analysis of administrative data was used to identify complex chronic conditions,[37,38] while the HCUP chronic condition indicator system was used to identify other non-complex chronic conditions.[37,39,40] History of child abuse and neglect was defined as having any claim with an associated diagnosis during the pre-period or follow-up period. Prior history of DSH or suicidal ideation included any claim with an associated diagnosis during the pre-period or index diagnosis date for the outcome of DSH; the follow-up period was also included for these variables with the outcome of suicide. Treatment variables included any inpatient, outpatient, or emergency room mental healthcare during the pre-period.

Statistical Methods

Descriptive statistics were examined using absolute counts and relative frequencies compared between adolescents and young adults. For each sociodemographic and clinical characteristic, the number of patients, person-years of follow-up, persons with at least one DSH claim, and the rate of DSH claims per 1000 person-years were calculated. Suicide rates per 100 000 person-years were also calculated. Cumulative incidence curves for DSH and suicide were graphed for each age group with Gray's test used to assess equivalency. The associations between sociodemographic and clinical characteristics and the outcomes were modeled using Cox proportional hazards regression. We checked the proportional hazards assumption by testing whether Schoenfeld residuals were correlated with time. Certain variables violated the proportional hazards assumption. However, given the high power of our large sample size to detect violations of the proportional hazards assumption and the lack of clinical meaningfulness of the interaction of these variables with time, we chose to present the main effects only. Hazard ratios (HR) and associated 95% confidence intervals (CI) are presented for both unadjusted and adjusted analyses. DSH and suicide-adjusted models controlled for age, gender, and race/ethnicity. Due to potential developmental differences, we examined differences in risk factors for DSH between adolescents and young adults by fitting models including an interaction term (stratification variable by age group). The false discovery rate was used to adjust for multiple comparisons. Rates of suicide, unintentional death, and all-cause mortality were compared with Ohio rates by calculating corresponding rates in the general population matched by age, gender, and calendar year.[41] Statistical analyses were done using SAS version 9.4 and R version 4.0.3.[42,43] Open Source Epidemiologic Statistics for Public Health was used to calculate 95% CI for standardized mortality ratios using the mid-P exact test.[44]