Sample
Data for this study was derived from the National Survey of Drug Use and Health (NSDUH) for 2015-19. The annual weighted responses of adults who were sampled through a multistage probability sampling frame were analyzed. The full sample included 214,505 adults.
Measures
Past-year suicidal ideation: Respondents were asked, “At any time in the past 12 months did you seriously think about trying to kill yourself?” A total of 13,195 adults reported past-year suicidal ideation.
Past-year suicide attempt: If a respondent answered yes to the question above, they were then asked, “At any time in the past 12 months did you try to kill yourself?” A total of 2,009 of the 13,195 who reported suicidal ideation also reported a suicide attempt in the past year.
Sociodemographic characteristics: Respondents reported certain sociodemographic information about themselves, including sex (males, females), age (18-25, 26-34, 35-49, 50-64, 65+), race/ethnicity (White, Black, Hispanic, Asian/Pacific Islander, Native American/Alaska Native, multiracial/other), education (less than high school, high school or equivalent, associate degree, some college, college degree), employment (full-time, part-time/other, unemployed), income (<$20,000, $20,000-$49,999, $50,000-$74,999, $75,000+), insurance (private, Medicare, Medicaid, military, none/other), marital status (single, married, divorced/separated/widowed), sexual orientation (heterosexual/straight, lesbian/gay, bisexual, unknown), and urbanicity (urban/rural). Data presented in Parts I, II, and IV is adjusted for sociodemographic characteristics.
Behavioral characteristics (no past-year substance use, past-year use without a disorder, or past-year substance use disorder): Data presented in Parts I, II, and IV of the interactive is adjusted for behavioral characteristics.
Clinical characteristics (health and mental health): Data presented in parts I, II, and IV of the interactive is adjusted for clinical characteristics.
Statistical analyses
Part I
To examine changes over time in suicidal ideation and attempts between groups, a series of separate logistic regressions was completed. Logistic regression aims to reveal associations between a set of characteristics and a binary outcome. In these models, the outcomes were Yes or No for suicidal ideation within the full sample, then Yes or No for suicide attempt within those who reported suicidal ideation. Each categorical characteristic listed above was included in the models as a variable or predictor of the binary outcome. For all categorical variables, a reference group (see glossary) was chosen. The reference group is generally the most common category or the category that makes interpretation easiest. Statistical interaction terms were included in these models for sex x year, age x year, and race x year in six separate models (three for suicidal ideation and three for suicide attempt). The models controlled for all other sociodemographic, substance use, and health variables. Only the results of significant interaction terms are shown in Part I.
Part II
To understand group differences within all categories, another series of logistic regression models predicting Yes or No for suicidal ideation within the full sample then predicting Yes or No for suicide attempt among those who reported suicidal ideation was performed. Each categorical characteristic listed above was again included in the models as a variable or predictor of the binary outcome with a reference group. In this analysis, for sociodemographic characteristics, generally the category with the most respondents was chosen as a reference group, except for age and urbanicity, for which the reference group was chosen to be the group at lowest suicide risk for ease of interpretability. For substance use and health variables, “No use” and better health outcomes were the reference groups, again for ease of interpretability. Because of the inclusion of all sociodemographic, substance use, and health characteristics in these analyses, results of the models are statistically controlling for all other characteristics. Results of the analysis are presented as predicted probabilities. Note that only comparisons between sociodemographic groups are reported in this brief. To see comparisons of substance use groups and health groups, see Table 2 in the manuscript linked below.
Part III
Prevalence of both suicidal ideation and suicide attempts was calculated and survey-weighted for each year of the 2015-19 dataset. Statistical tests determined whether ideation and attempt prevalence significantly changed from 2015 to 2019. These changes were tested over time only within each group (e.g., change over time for men, change over time for females). Group differences in prevalence trends of ideation and attempts were not compared (e.g., change over time for males was not compared with females) in these analyses. Note that only prevalence trends for sociodemographic groups are reported in this brief. To see prevalence trends for substance use groups and health groups, see Supplemental Tables S1 and S2 in the manuscript linked below.
Part IV
A final series of logistic regression models was conducted to understand the probability of health care utilization in various settings in the past year. For those with past-year suicidal ideation, one series of models predicted the probability of general health care usage in four different settings: any health care, inpatient, emergency department, outpatient. The next series of models predicted the probability of past-year mental health care usage in four different settings—any mental health care, inpatient, outpatient, and prescription services—for those with past-year suicidal ideation. Each model included all control variables of sociodemographic, substance use, and health characteristics. Given the potential to address suicide risk through clinical risk assessment, Part IV visualizes differences in probability of health care utilization of those with past-year suicidal ideation for sex, age, and race—characteristics for which data is regularly collected in clinical settings.
For more detailed information on the methodology and results, see the study in the journal Psychiatric Services.
link