A Latent Profile Analysis of At-Risk First-Year Undergraduate Students

Abstract

Objective:

We examined whether meaningful subgroups of self-injurious behaviors (SIBs) would emerge within a pool of first-year college students already deemed at elevated risk.

Participants:

First-year undergraduates (N=1068) recruited in 2015-2018 Fall terms.

Methods:

Past-year nonsuicidal self-injury (NSSI) frequency, past-year number of NSSI methods used, lifetime suicide attempt (SA) history, and recency of SA were included in a latent profile analysis.

Results:

Four subgroups emerged: low SIB (n=558, 52%), high NSSI only (n=182, 17%), high SIB (n=141, 13%), and high SA only (n=187, 18%). Students in the high SIB group reported higher levels of suicidal ideation at baseline and follow-up in comparison to all groups. Those in the high NSSI only or high SIB groups had relatively higher levels of NSSI at baseline and follow-up.

Conclusions:

Findings highlight the amount of heterogeneity within a high-risk group, along with the importance of considering distal and proximal SIBs in university screening efforts.

Introduction

The transition from high school to university presents an important developmental stage, albeit a stressful one. First-year undergraduate students are in the process of separating from their families and experiencing greater levels of academic, financial, and social stress¹. Adolescents and young adults in this age group are also at increased risk for depression onset, increased use of alcohol and other drugs, suicidal behavior, and suicide². In the United States, suicide is the second-leading cause of death in individuals 15-24 years old³. Identifying which students are vulnerable to self-injurious thoughts and behaviors upon entry to university is imperative for early intervention efforts.

Nonsuicidal self-injury (NSSI), defined as self-injury without the intent to die⁴, and lifetime suicide attempt (SA) history have both been identified as important predictors of future suicidal behavior^5,6. One reason for this could be related to the construct of acquired capability, first proposed in the interpersonal psychological theory of suicide⁷. According to this theory, repeated exposure to painful and provocative events increases one’s acquired capability to follow through with suicidal thoughts, thereby increasing the likelihood that they would engage in suicidal behavior^7,8. Interestingly, one study has shown that past-year NSSI engagement was linked to increased capability for suicide one year later in a sample of first-year undergraduate students ⁹. These findings are particularly important to consider, given that both NSSI and SA history are prevalent at alarming rates in first-year college students. Nearly 18% of first-year college students reported lifetime engagement in NSSI, and approximately 8% reported engaging in NSSI in the past 12-months¹⁰. Moreover, 13.6% of first-year students reported ever planning a SA, with 4.3% reporting at least one lifetime SA¹¹. This underscores the importance of assessing past-year NSSI engagement and lifetime SA history in incoming undergraduate students, in order to better identify which students are at heightened risk of acting on suicidal desires.

NSSI and SA also tend to co-occur, with 33% of students with a history of NSSI also reporting suicidal behaviors, including SA¹². Given this overlap, previous investigations have used latent class analysis to examine if college students cluster into distinct subgroups based on self-injurious thoughts and behaviors. Latent class analysis is a person-centered approach to identifying meaningful subgroups in a population¹³. Across different studies, three distinct groups have been identified, including a low-risk group, a group with some self-injurious thoughts/behaviors, and a high-risk group characterized by all self-injurious thoughts/behaviors^14-16. These investigations differed based on population, with one including students with NSSI history¹⁵ and others including all students irrespective of risk status^14,16. We are not aware of previous studies that have examined self-injurious behavior subgroups among undergraduate students already identified at risk for suicide based on a combination of risk factors¹⁷, including alcohol abuse, depression, suicidal ideation, and suicide attempt history. Prior work has shown that a combination of factors confers greater risk for suicide than singular factors (e.g., alcohol abuse and depression^18,19), indicating that students with more than one of these factors is at greater suicide risk in comparison to those endorsing only one factor. Importantly, if subgroups of students with a combination of these factors do exist, this can aid in prioritizing the limited resources on college campuses given the growing mental health needs of college students²⁰. In other words, using a tiered approach that includes subdividing an at-risk population into further subgroups could result in low vs. high-touch intervention strategies and therefore help alleviate the stress on college health and counseling centers. Further, we are unaware of previous studies that have identified latent risk profiles of undergraduate students and that have also investigated if subgroups differ with regards to risk level at follow-up assessments. Identifying if subgroups can be differentiated based on risk-level at a later time period, including if they differ on engagement in self-injurious behaviors (SIBs) and endorsement of suicidal ideation, provides additional information regarding long-term risk in first-year college students deemed at-risk prior to college entry and can help identify where to focus limited mental health resources. The identification of meaningful subgroups of students at heightened suicide risk, and investigation of their engagement in SIBs and experiences of suicidal ideation longitudinally, could therefore aid in early detection, suicide risk formulation, and targeted intervention.

In this study, we investigated if distinct classes of first-year college students at elevated suicide risk would emerge based on indices of lifetime SA (i.e., number of attempts, recency of last attempt) and 12-month NSSI engagement (i.e., frequency, number of different methods used). We then explored if differences between classes would emerge on demographic and risk variables at the beginning of the term. Finally, and importantly, we examined if class membership would also predict higher likelihood of suicidal ideation and NSSI at 6-month follow-up, thereby adding to the limited literature on follow-up outcomes.

Materials and Methods

Participants

Data comes from a larger intervention study of students enrolled during the fall semesters of 2015-2018 from four universities in Midwest and Western regions of the United States²¹. For a detailed description of primary sample characteristics and procedures, see previous publication²¹. The final analytic sample for this study consisted of 1068 first-year university students who screened positive for suicide risk via any two of the following: lifetime SA history, past-year suicidal ideation, past 2-week positive screen for depression, and past 2-month positive screen for alcohol misuse. Previous literature has shown that endorsement of two risk factors, including depression and alcohol abuse¹⁹ and ideation and past attempt history¹⁸ was related to heightened suicide risk than any one of these factors alone. Because the focus of the main intervention study was on facilitating access to mental health care, students were only included in the main study if they reported no current use of medications for mental health reasons or counseling/therapy. Most participants identified as female (70.4%) and were primarily White (76.5%). See supplemental table 1 for additional demographic information. IRB approval was obtained for this study (protocol #: HUM00091681) from the University of Michigan. All participants provided informed consent prior to participation.

At 6-month follow-up, 801 (75%) individuals completed the assessment, of which 778 provided data for at least one outcome of interest. Retention did not vary based on the following: gay/lesbian, gender minority, bisexual/pansexual, sexual orientation marked as other, all race categories, NSSI frequency or severity, number of lifetime SA or recency of attempt, suicidal thoughts, depression scores and alcohol use. They did differ on: male gender (39% vs. 24% at follow-up; χ²(1) = 25.28, p < .000), female gender (59% vs. 75% at follow-up; χ²(1) = 25.04, p < .000), straight status (67% vs. 60% at follow-up; χ²(1) = 4.12, p = .043), mostly straight status (11% vs. 17% at follow-up; χ²(1) = 5.08, p = .024), cannabis use (M_baseline = 1.19 vs. M_follow-up = 0.86, t(456.59) = 3.38, p = .001), and impairment scores (M_baseline = 1.42 vs. M_follow-up = 1.27, t(429.03) = 2.57, p = .011).

Measures

Nonsuicidal self-injury.

Using items from two measures^22,23, participants were asked to report on frequency of NSSI (0, 1, 2 or 3, 4 or 5, and 6 or more times) over the past 12-months at baseline. If they endorsed engagement in NSSI, they were asked to specify NSSI method from the following list (could check all that apply): cutting or carving on skin, picking at wound, hitting self, scraping skin to draw blood, biting self, picking areas of the body to the point of drawing blood, inserting objects under the skin or nails, tattooing self/burning skin, pulling out one’s own hair, erasing skin to draw blood, other (provide details). An NSSI severity score was computed based on number of different methods endorsed (0, 1, 2, 3, and 4 or more). At 6-month follow-up, participants were asked about frequency of NSSI for the past 3-months, a score that was converted to yes/no response due to limited frequency.

Suicidal ideation and attempt history.

The National Comorbidity Survey²⁴ was used to assess suicidal thoughts and attempt history at baseline, and suicidal ideation at 6-month follow-up. At baseline, items assessed thoughts of wanting to be dead and thoughts of committing suicide in the past 12-months and then in the past month (yes/no), lifetime SA (yes/no), number of lifetime SA (converted to 0, 1, 2, 3 or more), SA in the past 12-months (yes/no), and SA in the past month (yes/no). A SA recency score was computed with no SA = 0, SA beyond the last 12-months = 1, SA within the last 12-months = 2¹. At 6-month follow-up, we used a single-item to assess thoughts of wanting to commit suicide in the past-month (yes/no).

Drug use.

Items adjusted from the Youth Behavior Risk Survey²⁵ were used to assess recreational substance use at baseline. In this study, we focused on cannabis use. Participants reported how often they used cannabis in the past three months: 0=Never, 1=Once or twice, 2=Monthly, 3=Weekly, and 4=Daily or almost daily.

Alcohol use.

The 10-item Alcohol Use Disorders Identification Test (AUDIT)²⁶, was used to assess frequency, dependency, and alcohol-related consequences. Individuals were asked to respond based on their pattern of alcohol use over the past 2-months. Internal consistency of the AUDIT in our sample was 0.84.

Depression and level of impairment.

The Patient Health Questionnaire-9 (PHQ-9)²⁷ was used to assess symptoms of depression in the past two weeks. Individuals were asked to respond using a four-point Likert scale, ranging from 0=not at all to 3=nearly every day. Internal consistency of the PHQ-9 in our sample was 0.84. Additionally, participants were asked to respond to a single item (i.e., PHQ-10) assessing how difficult it was to function across domains (school, home, interpersonal relationships) due to psychological/behavioral difficulties, using a four-point scale (range: 0=not at all difficult to 3=extremely difficult).

Data analytic Plan

Latent profile analysis (LPA) was conducted using R coding software²⁸ and tidyLPA²⁹. LPA variables included: past 12-month NSSI frequency, number of NSSI methods, lifetime number of SA and SA recency². We estimated a one class solution and added classes until identifying the solution that best fit the data and was theoretically meaningful³⁰. To determine statistical fit, we used: (a) Akaike information criterion (AIC; lower = better fit), (b) sample adjusted Bayesian information criterion (saBIC; lower = better fit), (c) entropy (higher = better fit), (d) bootstrap likelihood ratio test (BLRT), and (e) Vuong–Lo–Mendell–Rubin adjusted likelihood ratio test (VLMR-LRT). We aimed for average latent class posterior probabilities that were no less than 0.80³⁰ and a class solution with no less than 5% of the sample in any single class³⁰. Following identification of the best-fitting solution, we examined bivariate residuals within each profile to examine if the local independence assumption was violated, using a cut-off of 3.84 (i.e., residuals larger than 3.84 indicate that correlations between the indicator pairs are not adequately explained by the latent-class model and suggest violation of local independence³¹).

Independent samples t-tests were used to examine differences between LPA classes. Then, separate multinomial logistic regressions, using mlogit³², were conducted to examine differences between classes on: (a) gender identity, (b) sexual orientation, (c) racial identity, (d) depression³ and suicidal ideation, and (e) alcohol and cannabis misuse. Class membership was the dependent variable and variables of interest were the independent variables. Reference groups were adjusted to examine differences between all classes. The Benjamini-Hochberg³³ correction was used across all comparisons (n=118), with a 5% false discovery rate.

To examine differences at 6-month follow-up, a series of logistic regression models, using gmodels³⁴, were conducted. Class membership was the independent variable and either presence of suicidal ideation (model 1) or NSSI (model 2) were the dependent variables. An intervention indicator was included as a covariate in all models and baseline suicidal ideation was included as a covariate in model 1. The same correction was applied across all comparisons (n=12).

Results

Latent Profile Descriptions

A four-class solution was identified as most parsimonious and theoretically meaningful (supplemental table 3). While AIC and saBIC values kept decreasing with increasing class solutions, and BLRT and VLMR-LRT values indicated significant change in model fit when adding an additional class, entropy was highest in the four-class solution. Additionally, when further investigating the five-class solution, it was identified that one of the groups split into two based on degree of severity of latent profile indicators rather than on categorical distinctions, suggesting that these two groups are theoretically very similar (see Figure 1 in supplemental material comparing class 4 and class 5 solutions). Average class posterior probabilities for the four-class solution were ≥ 0.97 and no class had fewer than 5% of the sample, indicating that a four-class solution was acceptable. Bivariate residuals between pairs of indicators within each class indicated that the local independence assumption was not violated (Median = −0.57 – 0.15, Min = −1.90 – −0.14, Max = 0.03 – 2.28).

Figure 1 and table 1 illustrate differences across the classes on LPA variables. Class 1 (52%), the low SIB group, was characterized by low levels across all LPA indices. A series of independent samples t-tests were conducted to examine differences between the other groups. Class 3 (13%) had significantly higher NSSI frequency (t(306.19) = −5.43, p = .000) and severity (t(306.59) = −3.61, p = .000) in comparison to class 2 (17%). Additionally, class 3 had significantly higher lifetime SA (t(270.11) = 4.51, p = .000) and SA recency scores (t(247.35) = 3.71, p = .000) than class 4 (18%). As such, class 2 was termed the high NSSI only group, class 4 the high SA only group, and class 3 the high SIB group. See supplemental tables 1-2 for demographic and mental health characteristics across classes.

Figure 1.

NSSI = nonsuicidal self-injury. SA = suicide attempt. SIB = self-injurious behavior. NSSI frequency range = 0 – 6. NSSI severity range = 0 – 4. Lifetime SA number range = 0 – 3. SA recency range = 0 – 2.

Table 1.

Means and Standard Deviations of Latent Profile Analysis Variables in the Full Sample and Across Groups.

Latent Profile Analysis Variables	Total (n=1068)	Low SIB (n = 558; 52%)	High NSSI only (n=182; 17%)	High SIB (n=141; 13%)	High SA only (n= 187; 18%)
	M (SD)	M (SD)	M (SD)	M (SD)	M (SD)
NSSI frequency a	0.93 (1.39)	0.07 (0.25)	2.67 (0.96)	3.05 (0.92)	0.21 (0.52)
NSSI severity b	0.80 (1.25)	0.07 (0.25)	2.14 (1.07)	2.78 (1.03)	0.19 (0.44)
Lifetime SA number c	0.54 (0.93)	0 (0)	0 (0)	1.99 (0.87)	1.58 (0.73)
Recency of last SA d	0.37 (0.61)	0 (0)	0 (0)	1.30 (0.46)	1.13 (0.34)

Notes. M = mean. SD = standard deviation. NSSI = nonsuicidal self-injury. SA = suicide attempt. SIB = self-injurious behavior.

^aNSSI frequency responses: 0, 1, 2 or 3, 4 or 5, and 6 or more times

^bNSSI severity: sum of all different methods endorsed. Possible range = 0 – 4

^cLifetime SA number: number of lifetime suicide attempts reported. Possible range = 0 – 3

^dRecency of last SA: greater numbers are indicative of more recent attempts. 0 = no suicide attempt. 1 = most recent attempt was before past-year. 2 = most recent attempt fell within the past-year.

Differences between Groups at Baseline

A series of multinomial logistic regressions were conducted to examine differences between classes on demographic variables (supplemental table 4) and clinical characteristics (table 2). In comparison to the low SIB group, students in the high SIB group were more likely to self-identify as bisexual/pansexual (Odds Ratio [OR] = 8.20), endorse wish to be dead and suicidal thoughts (OR range = 2.45 – 5.20), and report greater levels of impairment (OR = 1.49). Additionally, individuals in the high NSSI only group were more likely to endorse suicidal thoughts (OR = 2.27) vs. the low SIB group. Finally, the low SIB group was characterized by greater depression severity when compared to the high SA only group (OR = 1.06). No differences in alcohol or cannabis use were found.

Table 2.

Multinomial Logistic Regression with Clinical Characteristics predicting Group Membership.

Baseline Characteristics	Class 1 vs. 2	Class 1 vs. 3	Class 1 vs. 4	Class 2 vs. 3	Class 2 vs. 4	Class 3 vs. 4
	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)
Model 1
PHQ-9 Total Score	1.05 (1.01, 1.08)*	1.01 (0.97, 1.05)	0.94 (0.91, 0.97) **	0.97 (0.92, 1.01)	0.90 (0.86, 0.94)***	0.93 (0.89, 0.97) **
PHQ-10 single item	1.22 (0.97, 1.55)	1.49 (1.14, 1.95) **	1.07 (0.83, 1.37)	1.22 (0.90, 1.65)	0.87 (0.65, 1.17)	0.72 (0.52, 0.98)*
Wish to be dead	1.15 (0.79, 1.68)	2.45 (1.47, 4.10)***	1.09 (0.76, 1.55)	2.13 (1.19, 3.80)*	0.95 (0.60, 1.50)	0.45 (0.25, 0.79) **
Thoughts of killing oneself	2.27 (1.58, 3.27)***	5.20 (3.17, 8.53)***	1.57 (1.12, 2.21)**	2.29 (1.31, 4.03) **	0.69, 0.45, 1.08)	0.30 (0.17, 0.52)***
Model 2
AUDIT	0.97 (0.93, 1.01)	0.96 (0.91, 1.00)*	0.98 (0.94, 1.02)	0.99 (0.94, 1.04)	1.01 (0.97, 1.06)	1.02 (0.97, 1.08)
Cannabis Use	1.08 (0.93, 1.25)	1.23 (1.05, 1.43)**	1.12 (0.97, 1.29)	1.14 (0.95, 1.38)	1.04 (0.87, 1.24)	0.91 (0.76, 1.09)

Notes. * p < .05. ** p < .01. Bolded findings maintained significance following correction for multiple comparisons. CI = confidence interval. OR = odds ratio. AUDIT = Alcohol Use Disorders Identification Test. PHQ-9 = Patient Health Questionnaire-9. PHQ-10 = Patient Health Questionnaire single impairment item. Class 1 = Low SIB. Class 2 = High NSSI only. Class 3 = High SIB. Class 4 = High SA only. SIB = self-injurious behavior.

In comparison to the high SA only group, students in the high SIB group reported greater depression severity (as did those in the high NSSI only group, OR range = 1.07 – 1.11) and were more likely to endorse a wish to be dead and suicidal thoughts (OR = 2.22 – 3.33). Students in the high SIB group were also more likely to endorse suicidal thoughts when compared to the high NSSI only group (OR = 2.29).

Differences between the high NSSI only and high SIB groups on type of NSSI method endorsed at baseline were also investigated using logistic regression (supplemental table 5). Cutting (OR = 3.62) and tattooing/burning self (OR = 6.57) were endorsed at higher rates in the high SIB group in comparison to the high NSSI only group.

Differences between Groups at 6-month Follow-up

Differences between groups on NSSI and suicidal ideation were examined at 6-month follow-up (table 3). Students in the high SIB group were characterized by greater likelihood to endorse suicidal ideation in comparison to all groups (OR range = 2.13 – 2.19). Additionally, students in the high SIB group or those in the high NSSI only group were more likely to endorse engagement in NSSI at follow-up in comparison to the other two groups (OR range = 6.23 – 8.83). No significant differences at follow-up were found between the low SIB and high SA only groups.

Table 3.

Series of Logistic Regression Analyses Examining Class Membership as Predictors of Suicidal Ideation or Nonsuicidal Self-Injury (NSSI) at 6-month Follow-up.

Variables at 6- month follow-up	Class 1 vs. 2	Class 1 vs. 3	Class 1 vs. 4	Class 2 vs. 3	Class 2 vs. 4	Class 3 vs. 4
	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)
Thoughts of killing oneself a, b	0.99 (0.63, 1.54)	2.16 (1.35, 3.47) **	1.01 (0.64, 1.59)	2.19 (1.27, 3.80) **	1.03 (0.60, 1.76)	0.47 (0.27, 0.82) **
NSSI b	8.83 (5.53, 14.34) **	8.00 (4.81, 13.46) **	1.28 (0.67, 2.35)	0.91 (0.54, 1.51)	0.15 (0.08, 0.26) **	0.16 (0.08, 0.30) **

Notes. * p < .05. ** p < .01. Bolded findings maintained significance following correction for multiple comparisons. CI = confidence interval. NSSI = nonsuicidal self-injury. OR = Odds Ratio. Class 1 = Low SIB. Class 2 = High NSSI only. Class 3 = High SIB. Class 4 = High SA only. SIB = self-injurious behavior.

^aCovariate: baseline thoughts of killing oneself

^bCovariate: intervention indicator

Discussion

In a large sample of first-year college students who initially screened positive for suicide risk, we examined distinct subgroups based on endorsement of suicidal and nonsuicidal forms of SIBs, and then explored differences between these subgroups on demographic and psychological variables at baseline and follow-up. Four unique subgroups emerged, with slightly over half of students falling in the low SIB group. Among the remaining half, students were almost evenly distributed across high NSSI only (high NSSI and no lifetime SA), high SIB (high NSSI and high lifetime SA), and high SA only (high lifetime SA and low rates of NSSI) groups. Differences between these groups are discussed below.

The latent groups identified in this study mostly converge with previous investigations using similar risk indices of SIBs^14-16. Specifically, prior studies have found three latent classes, characterized by a low-risk group, a group with some SIBs, and a group endorsing all SIBs measured. A key exception was that, in the current study, individuals in the high SA only group were characterized by high lifetime suicide attempt indices (higher number and recency) but low 12-month NSSI rates. Given that only recent (and not lifetime) NSSI engagement was measured in this study, it is possible that the high SA only group includes individuals with a history of NSSI who have not engaged in SIBs in the past 12-months, those with no NSSI history, or both. Importantly, measuring both lifetime SA and recent NSSI engagement appears to be important in highlighting which students are exhibiting greater current risk.

With regards to group differences, we found that the two groups characterized by recent (i.e., 12-month) NSSI engagement (i.e., high NSSI only and high SIB groups) were also the students who endorsed higher rates of suicidal thoughts, wish to be dead, and had greater depression severity in comparison to the remaining groups at baseline. Additionally, these two groups were characterized by greater self-injurious thoughts and behaviors at follow-up, such that the high SIB group was more likely to endorse suicidal ideation at 6 months, and both groups were more likely to endorse engagement in NSSI at follow-up. This pattern of findings suggests that university screening efforts should consider engagement in SIBs in the 12-months prior to college entry to identify students most vulnerable to subsequent self-injurious thoughts and behaviors, instead of only relying on lifetime indices. Nevertheless, the high SIB group (characterized by both recent NSSI and lifetime SA) was differentiated from other groups by greater initial and follow-up suicidal ideation. Therefore, our findings highlight that screening should focus on both recent SIB and lifetime SA, in line with a recent conceptual framework for student risk formulation³⁵. This is further in line with prominent theories of suicide, such that acquired capability for suicide resulting from repeated exposure to painful and provocative events, like NSSI/SA, only confers greater risk for acting on suicidal urges in the presence of current suicidal desires, highlighting the importance of considering both historical and current functioning. These findings also align with a previous investigation, which found stronger relationships between recent NSSI engagement and indices of psychopathology in comparison to frequency of NSSI engagement³⁶.

While both the high NSSI only and the high SIB groups showed greater engagement in NSSI at baseline, it is important to highlight that the high SIB group reported significantly greater NSSI frequency and severity when compared to the high NSSI only group. Previous literature has shown that individuals with more severe and frequent NSSI were less likely to cease self-injury than others³⁷. Despite reporting similar rates of NSSI at follow-up, it is less clear if these groups differed in terms of follow-up NSSI frequency and severity, particularly as the high SIB group was more likely to initially endorse more severe forms of NSSI compared to the high NSSI only group. Given that the high SIB group reported more frequent, more versatile (i.e., number of methods), and more severe types of NSSI (cutting/tattooing or burning) suggests that they are already at greater risk of more severe SIBs, including suicidal behaviors. It is unclear at this time if certain forms of NSSI confer greater risk for severe suicidal behaviors, a question that should be explored in future research.

Additionally, findings from this study illustrated that NSSI indices (frequency and severity) seem to “hang together,” such that groups were either high or low on both sets of indices. Although previous literature has highlighted the role of NSSI severity as having a stronger relationship to suicidal thoughts/behaviors^38,39, our findings suggest that those who more frequently engage in NSSI tend to be the same individuals who engage in different NSSI forms. This is in line with other investigations showing that both NSSI severity and frequency are important when considering level of risk³⁷. These results suggest that university screening measures could rely on only one index of NSSI engagement (frequency or severity), particularly if resources or time are limited. However, it is possible that, due to inherent dependencies in the data (i.e., a score of 0 on NSSI frequency means that NSSI severity will also have a score of 0), NSSI indicators “hang together” as an artifact of the data itself. While the assumption of local independence was not violated based on bivariate residual estimates, we recommend that future work examine these constructs in individuals with some level of engagement in self-injury (thereby eliminating inherent dependencies in the data based on zero engagement in self-injurious behaviors). As such, this finding is in need of further replication given conflicting evidence and given the way in which our data was coded. Similarly, we found that SA indices (lifetime number and recency of last SA) also co-occur. However, with only 6.3% of students endorsing a past-year SA, and similar issues with dependencies, it is likely that we did not have sufficient variability in recency of last SA to capture different combinations and we recommend future research examine if these indicators truly co-occur.

Interestingly, the high SA only group, characterized by high lifetime SA, was similar in its clinical profile to the low SIB group. Unexpectedly, the low SIB group displayed marginally higher depression severity in comparison to the high SA only group. It is possible that the high SA only group was composed of individuals with historical mental health crises that had partially resolved prior to college entry. These findings again highlight the significance of assessing more recent (before college entry) SIBs to determine which students are currently at higher risk.

Finally, in this first-year undergraduate sample of individuals who already screened positive for higher suicide risk based on depression, alcohol misuse, and/or suicidal thoughts/attempts, slightly over half of students fell into the low SIB group based on engagement in different forms of self-injury more specifically. Beyond differences in clinical characteristics, individuals in this group were less likely to self-identify as bisexual/pansexual than the high SIB group, a finding that is in line with literature illustrating heightened risk in the LGBTQ+ community⁴⁰ and further highlighting a need for specialized services to reduce risk among sexual minority students.

In light of the differences across all four subgroups, our results show significant heterogeneity within a subgroup of students already deemed at high risk for suicide. This suggests that university counseling efforts aimed at reducing risk for suicide could be tailored even within a group of students deemed at high-risk. Given that the majority of the high-risk students (n = 745 [69.8%], those in the low SIB and high SA only groups) in our sample were comparatively lower risk than the other subgroups (high SIB and high NSSI only groups), these students might benefit from continued monitoring and low-touch interventions like online modules that focus on teaching coping strategies and providing self-care tools. However, for students classified in the high SIB and high NSSI only groups (n = 323 [30.2%]), high-touch interventions that include continued monitoring and in-person counseling might be most beneficial. Using a tiered approach that considers multiple categories within a high-risk group could help alleviate stress on under-resourced health and counseling centers across college campuses. Future research should examine if such a tiered program is feasible and effective.

Findings from this study should be considered in light of important limitations. Follow-up assessments focused on functioning at the 6-month timepoint and did not assess experiences of suicidal thoughts and NSSI across the 6-month period. The focus on functioning at the 6-month timepoint also limited our ability to examine associations between latent classes and 6-month suicide attempts (due to a limited number of attempts endorsed at the follow-up period). The sample was predominantly female and White, making it unclear if our findings would replicate in a more diverse student body. The homogenous ethnoracial / gender breakdown of our sample is important to consider when implementing university screening efforts. College health centers should consider ways to ensure that their screening efforts are truly representative of their college population, thereby reducing resource and health disparities in individuals with minoritized identities. One consideration might be to implement incentivized peer referral to increase likelihood of engaging with university screening efforts, a strategy that has been successfully implemented in health screening and treatment research focused on minoritized identities in community settings^41,42. To reduce participant burden, questionnaires with fewer items were chosen. Relatedly, all data included in this investigation relied on self-report responses, which could produce effects representing method variance like social desirability reporting. Future work should consider multi-method approaches to identify latent groups of at-risk first-year students. Given this was a secondary data-analysis using data from an intervention study focused on elevating mental health service utilization in at-risk college students, students who were at-risk but engaged in mental health services (e.g., medication use) were not included in our analyses. It is possible that students currently engaged in services are distinct in several ways (e.g., socio-demographic variables, risk-level) than students included in our analyses. Future research should include all students at elevated suicide risk based on a combination of factors, regardless of current engagement in mental health resources, to identify: (1) if distinct subgroups emerge and (2) if risk-level and/or other variables differentiate students engaged in resources than those who are not currently engaged in resources.

Conclusion

In a large sample of students deemed at elevated risk for suicide based on several well-established risk factors, we further identified four meaningful subgroups based on engagement in SIBs. Our findings highlighted the importance of considering both lifetime and more recent engagement in suicidal and non-suicidal forms of self-injury. Distinct profiles of self-injury differentiated severity of student risk upon entry to university, with more severe profiles predicting continued engagement in NSSI and endorsement of suicidal thoughts 6-months later. Overall, we recommend that colleges include assessment of past-year and lifetime history of self-injury as part of their screening efforts to identify vulnerable students and to explore if a tiered approach with a mix of low vs. high-touch intervention programs could be both effective at reducing risk for suicide and alleviating stress on an under-resourced system.

Data Availability Statement

Data available upon reasonable request from author Cheryl A. King.