Suicide risk stratification in adolescents with psychiatric disorders

Abstract

Suicidal ideation and attempts are more prevalent during adolescence than at any other stage of life. Suicidal thoughts are the best predictors of a subsequent suicide attempt in children and adolescents. The purpose of this study was to identify latent classes or groups of youth with similar patterns of suicidal ideation using their responses to the five items of the Columbia-Suicide Severity Rating Scale, that measures suicidal ideation, plus a sixth item measuring intent to carry out suicide. This study included 388 adolescents admitted to primary and secondary mental health care programs in the Maule Region, Chile. Person-centered latent class analysis was utilized. Five classes were detected: ‘planners with intent’, ‘unintentional planners’, ‘non-planners with specific ideation’, ‘non-planners with nonspecific ideation’, and ‘non-ideators’. The class of ‘planners with intent’ was the most frequent (29.0%) and severe in all the indicators evaluated, comprising adolescents who reported self-injury, suicidal ideation, planning and at least one suicide attempt in their lifetime. The highest percentage of this class is represented by females (76.3%), middle adolescents (69.3%), living with both parents (43%) and presenting depressive disorder (57.9%). To enhance preventive strategies within healthcare services, it is recommended to include an assessment of the severity of suicidal ideation. This approach aims to mitigate risk factors that could potentially escalate to active suicidal ideation and suicide.

REVIEWING EDITOR:

• Daryl O’Connor, University of Leeds, United Kingdom

Keywords:

• Suicidal ideation
• latent class analysis
• person-centered approach
• adolescents
• suicidal prevention 

SUBJECTS:

• Health psychology
• Psychological science
• Mental health

Introduction

Suicidal behavior is the leading cause of premature death during adolescence (Bachmann, 2018; Shain and American Academy of Pediatrics Committee on Adolescence, 2007). Indeed, suicidal ideation and attempts are more prevalent in adolescence than at any other stage of life (Miranda et al., 2014). In the age group of 13 to 18 years, the estimated lifetime prevalence of suicidal ideation stands at 12.1%, while suicide attempts are estimated at 4.1%. (Blum et al., 2014; Nock et al., 2013). Suicidal ideation is three times more prevalent in adolescent females compared to males of the same age (Nock et al., 2013). For every suicide death in young people, there are between 100 and 200 suicide attempts (Sarchiapone et al., 2015; Wasserman et al., 2010).

A comparative study of the adolescent suicide rate (10 to 19 years) among OECD member countries identified Chile as the fourth country with the highest suicide rate, among the countries analyzed (Roh et al., 2018). In recent years, an increased risk of suicide in the adolescent population has been detected in Chile, especially after the COVID-19 pandemic. According to the report developed in 2023 by the CONVIVO Foundation (Gutiérrez-Lobos, 2023), based on the analysis of national statistics on deaths and their causes, provided by the Department of Health Statistics and Information (DEIS), between the years 2021 and 2022, adolescent suicide increased by 31% overall and by 68% in women between 15 and 19 years of age.

Suicidal behavior presents unstable and fluctuating patterns (Miranda et al., 2014; Nock et al., 2013). Suicidal thoughts have been associated with subsequent suicide attempts and their repetition (Miranda et al., 2014), and they are also the main predictor of suicide in children and adolescents (Chamberlain et al., 2009; Park and Jang, 2018). One-third of adolescents with suicidal thoughts attempt suicide within the next year (Nock et al., 2013). Given that most adolescent suicides are first attempts (McKean et al., 2018), risk estimation cannot be based only on the characteristics of suicide attempters. Instead, combining the risk factors associated with suicidal thoughts and attempts may help to identify vulnerable groups.

The characterization of subgroups at high risk of suicide through clinical and psychological factors may allow the development of more effective and focused prevention strategies.

Research into the categorization of adolescent suicide risk groups has primarily focused on clinical factors, including specific mental disorders (Auerbach et al., 2023; De Neve-Enthoven et al., 2023; Lee et al., 2023), instances of self-harm (He et al., 2023), and cognitive as well as emotional challenges (Díez-Gómez et al., 2020). However, variables associated with continued risk suicide have not been specifically addressed.

The present study seeks to identify suicide risk subgroups in a clinical population of Chilean adolescents through the analysis of latent classes. We hypothesize that there are latent groups of adolescents with varying levels of vulnerability to suicidal behavior, and these groups can be separated by their demographic, psychological, and clinical features. Using latent classes could allow a broader understanding of the behavioral patterns underlying each suicide risk subgroup (Díez-Gómez et al., 2020).

Material and methods

Design

A quantitative study was carried out, with a non-experimental cross-sectional design.

Participants

The sample included 388 adolescents and young adults, 230 females (59.3%) and 158 males (40.7%), aged 10 to 21 years, with a mean age of 15.63 years (SD = 1.98). Participants were selected by non-probability purposive sampling, specifically from patients enrolled in various primary and secondary mental health care programs within the Maule Region, Chile.

Inclusion criteria: Adolescents between 10 and 19 years of age, with a diagnosis of psychiatric disorder (with or without suicidal ideation or attempts) and who are being treated in mental health units of hospitals, CESFAM or COSAM in the Maule Region.

Exclusion criteria: Adolescents unable to read or write, adolescents with a cognitive disability and adolescents with symptoms of active psychosis.

In terms of diagnosis according to the DSM-5 criteria, 29.9% of adolescents had a depressive disorder, 15.6% an anxiety disorder, 14.8% a neurodevelopmental disorder, and 22.6% other. No diagnosis was recorded in the clinical records of 17.1% of the sample. In addition, the highest percentage is in the stage of middle adolescence (55.9%) and cohabit with both parents (48.7%).

Research instruments

Following Méndez-Bustos et al. (2022), sociodemographic variables and diagnostic classification based on DSM-V were obtained from adolescents’ clinical records. The Spanish version of the Columbia-Suicide Severity Rating Scale (C-SSRS Chile/Spanish 5.1) (Gipson et al., 2015; Posner et al., 2011) was used to evaluate suicidal risk. This instrument was designed to gather information about suicidal ideation and behavior, as well as non-suicidal self-injuries. It allows quantifying the severity of both suicidal ideation intensity and behavior lethality along specific time periods. This instrument has a mixed format, with open, closed and Likert-format questions. The University of Columbia Suicide Severity Rating Scale has acceptable predictive power for future suicidal attempts for clinical and adolescent populations. The C-SSRS was specifically designed to assess suicidal ideation and behavior and is composed of four subscales. The subscale used in this study measures lifetime severity of suicidal ideation and suicide planning, and comprises five questions (e.g. have you wished you were dead?) which must be answered dichotomously (0 = No, 1 = Yes). A higher score indicates greater severity of suicidal ideation. In addition, four additional questions inquired about non-suicidal self-injurious behaviors, specific preparations for a suicidal act, the intention to carry out a suicidal act, and suicide attempts before the survey. These questions also had to be answered dichotomously (0 = No, 1 = Yes). Table 1 shows the complete wording of the nine items used.

Table 1. Response probabilities to suicidal ideation item conditional on class membership (five-class model).

	Five-class model
Latent Class	Self-harm	Desire for death	Suicidal ideation	Ideation method	Intention 1	Planning	Intention 2	Attempt preparation	Suicide attempt	Probabilities of membership
1. Planners with intent	.856	.999	.999	.999	.958	.872	.631	.985	.830	.290
2. Unintentional planners	.248	.230	.001	.969	.969	.798	.462	.797	.001	.083
3. Non-planners with specific ideation	.676	.999	.972	.864	.793	.341	.328	.110	.008	.163
4. Non-planners with nonspecific ideation	.805	.999	.794	.001	.084	.001	.001	.105	.405	.189
5. Non-ideators	.171	.375	.001	.008	.001	.001	.001	.011	.001	.272

1. Have you had a self-harming behavior -self-harm- not suicidal? (Self-harm).

2. Have you ever wished you were dead or wished you could sleep and not wake up? (Desire For Death).

3. Have you ever thought of committing suicide? (Suicidal Ideation).

4. Have you thought about how you would carry it out? (Ideation Method).

5. Have you had these ideas and, to some degree, the intention to carry them out? (Intention 1).

6. Have you started to elaborate or worked out the details of how to commit suicide? (Planning).

7. Do you intend to carry out this plan (Intention 2)?.

8. Have you done something, started to do something, or prepared to do something to end your life? (Attempt Preparation).

9. Dichotomized Suicide attempt (Suicide Attempt).

Analysis plan

A latent class analysis was conducted to identify individual differences or heterogeneity in patterns of suicidal ideation and behavior. This analysis aimed to identify latent classes or groups of youths with similar patterns of beliefs and behaviors using their responses to the five C-SSRS items measuring suicidal ideation, plus the four additional items described in the previous section. In the analysis of latent classes, a mathematical model probabilistically relates discrete manifest variables to a single latent categorical variable, where different levels of the latent variable correspond to different subgroups or types of people (Cumsille et al., 2009). The relationships between each item and the latent categorical variable are estimated as probabilities and the individual membership in a specific latent class, given its response pattern to the items, is also probabilistic. All analyses were performed with the program MPLUS version 7.4. (MPLUS (Version 7.4). [Computer Software]. Los Angeles, CA: Muthén & Muthén.

Model fitting and selection

A two-class model was fitted and then the number of classes was increased by one until convergence was lost. The AIC, BIC, and adjusted BIC indices were used to compare the models, where lower values indicate a better fit. These indices apply different penalties (i.e. by the number of parameters or according to the sample size), and their results may differ (Nylund et al., 2007). Thus, the Lo-Mendel-Rubin test (LMR) and the Bootstrap Likelihood Ratio Test (BLRT) were also used, both tests provide a p-value that determines whether there is a statistically significant improvement in model fit when an additional class is included. Finally, an interpretability criterion was applied (Lanza et al., 2007). Therefore, to select the final model, we first discarded those models that did not show a significantly better fit through LMR and BLRT. Next, we compared those models with lower AIC, BIC, and adjusted BIC and discarded those models that showed classes with a size of less than 2% of the total sample and those models with additional classes whose interpretation was essentially the same as the classes already contained in the more parsimonious models.

Ethical considerations

The research was approved by the Ethics Committee of the Maule Health Service and by the Scientific Ethics Committee of the Universidad Católica del Maule de Chile (Number 84/2017), meeting principles established by the Universal Declaration of Human Rights, international regulations of the Council for International Organizations of Medical Sciences (van Delden & van der Graaf, 2017) and Law 20.120 regarding scientific research involving human beings in Chile. The authors declare that necessary ethical safeguards have been adopted for working with people, as proposed in the Helsinki Declaration (World Medical Association, WMA, 2018) and the Belmont report (Zucker, 2014).

Results

Model selection

Table 2 reports the BIC, AIC, LMR, and BLRT of each of the models of suicidal ideation. It can be observed that the p-value of the LMR/BLRT of the eight-class model was not significant, which implies that this model does not present a significantly better fit than the seven-class model. The lowest AIC was found in the seven-class model, the lowest BIC was found in the five-class model and the lowest adjusted BIC was found in the six-class model. Since one of the groups of the six-class model and two of the groups of the seven-class model showed trivial class membership probabilities (less than 2%), the five-class model was considered to better explain the heterogeneity in adolescent suicidal ideation. The five-class model presented an entropy of. 931; where values close to one indicate a clear separation of the classes from each other (Celeux and Soromenho, 1996).

Table 2. Comparison of suicidal ideation models.

Number of Classes	AIC	BIC	Adjusted BIC	LMR p-value	BLRT p-value	Entropy
2	3400.82	3476.08	3415.79	<.001	<.001	.956
3	3204.49	3319.36	3227.34	<.001	<.001	.912
4	3029.06	3183.54	3059.80	<.001	<.001	.935
5	2940.94	3135.03	2979.56	<.001	<.001	.931
6	2932.04	3165.74	2978.54	=.005	<.001	.936
7	2928.12	3201.43	2982.50	<.001	.002	.946
8	2933.98	3246.90	2996.24	.131	.307	.936

Note: AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion; LMR = Lo-Mendell-Rubin Test; BLRT = Bootstrap Likelihood Ratio Test.

In bold the lowest AIC, BIC, and Adjusted BIC.

Five-class model: interpretation of each group

The conditional probabilities of responding to each item for the five-class model are reported in Table 1. These parameters represent the probability of saying ‘yes’ (i.e. presenting suicidal ideation) to each item, conditional on belonging to a class. The further a probability moves away from .50 the more the item is representative of a class. Response probabilities close to .50 indicate that the response to that item does not relate to that class. The observed response patterns show one class with high probabilities of positive response to all the items (all greater than .63). This class comprises ‘Planners with intent’ endorsing self-harm, suicidal ideation, planning, and at least one suicide attempt during their lifetime.

A second class could be labeled as ‘unintentional planners’. In this class, the probabilities of positive response were above .79 for the items of specific suicidal ideation and planning and below .24 for the items of nonspecific suicidal ideation, intention to carry out the plan and attempts. Adolescents in this class show a low probability of endorsing lifetime suicidal ideation, but a high probability of planning a suicide attempt.

A class labeled ‘non-planners with specific ideation’ also emerged, where the probability of positive response to the specific suicidal ideation items is higher than .79 and the probabilities of positive response to the suicide planning and attempt items were all lower than .34; therefore, adolescents classified in it show a high probability of having presented suicidal ideation with general method and intention at some point in their life, but a low probability of having presented suicide plans or attempt.

The fourth class that emerged was labeled as ‘non-planners with nonspecific ideation’, where the probability of positive response to the nonspecific suicidal ideation items was above .80 and the probabilities of positive response to the rest of the items were all below .40. Adolescents classified within this group exhibit a high likelihood of having experienced thoughts of wanting to die and having suicidal ideation at some point in their lives. However, they have a low likelihood of displaying more specific ideation, planning, and actual suicide attempts.

Finally, a class labeled as ‘non-ideators’ emerged, where the probabilities of positive response to the items were all below .37. Consequently, adolescents classified in this category exhibit a low likelihood of having engaged in self-harm, experiencing suicidal ideation, making plans, or attempting suicide at any point in their lives.

The ‘planners with intent’ and ‘non-ideators’ were the most common (29.0% and 27.2%, respectively), followed by the class ‘non-planners with nonspecific ideation’ (18.9%) and the class ‘non-planners with specific ideation’ (16.3%). The least common class, meanwhile, was the class ‘unintentional planners’ (8.3%).

Table 3 shows the average posterior probabilities, or the average probability of the class model accurately predicting class membership for individuals. These represent the average probability of a person being classified in a class due to the scores obtained in the items that serve to model the classes. Higher diagonal values (i.e. closer to 1.0) and lower values off the diagonal (i.e. closer to 0) are desirable (Weller et al., 2020). Thus, it can be observed that the five-class model shows a high probability of accurately predicting the membership in each class of the individuals in the sample.

Table 3. Average latent class probabilities.

	Class 1	Class 2	Class 3	Class 4	Class 5
Class 1	.973	.000	.027	.000	.000
Class 2	.000	.997	.010	.007	.000
Class 3	.033	.003	.951	.074	.000
Class 4	.000	.000	.058	.939	.003
Class 5	.000	.000	.001	.000	.925

Table 4 shows the sociodemographic and clinical characteristics for each class. It is observed a higher percentage of women in the ‘planners with intent’, ‘unintentional planners’, and ‘non-planners with specific ideation’ classes and a higher percentage of men in the ‘non-planners with nonspecific ideation’ and ‘non-ideators’ classes. In all classes there was a higher percentage of middle adolescents (14-16 years old) and a higher percentage of youth living with both parents. Finally, most of the subjects belonging to the ‘planners with intent’ class showed a diagnosis of depressive disorder.

Table 4. Sociodemographic and clinical characteristics for each class.

	Non-planners with specific ideation		Planners with intent		Non-ideators		Non-planners with nonspecific ideation		Unintentional planners
	N	%	N	%	N	%	N	%	N	%
Sex
Female	38	64,4	87	76,3	51	44,7	29	42.0	25	78.1
Male	21	35,6	27	23,7	63	55,3	40	58.0	7	21.9
Age
Early (10–13)	5	8.5	5	4.4	28	24.6	8	11.6	5	15.6
Middle (14–16)	43	72.9	79	69.3	73	64.0	52	75.4	27	84.4
Late (17–21)	11	18.6	30	26.3	13	11.4	9	13.0	0	0.0
Do you live with?
Both parents	27	45.8	49	43.0	57	50.0	39	56.5	17	53.1
Mother only	25	42.4	43	37.7	45	30.5	27	39.1	12	37.5
Father only	4	6.8	7	6.1	6	5.3	1	1.4	1	3.1
Neither	3	5.1	15	13.2	6	5.3	2	2.9	2	6.3
Siblings
No siblings	33	55.9	69	60.5	64	56.1	37	53.6	23	71.9
One or two	5	8.5	12	10.5	7	6.1	5	7.2	1	3.1
Three or more	21	35.6	33	28.9	43	37.7	27	39.1	8	25.0
Depressive disorder
Yes	22	37.3	66	57.9	13	11.4	18	26.1	2	6.3
No	37	62.7	48	42.1	101	88.6	51	73.9	30	93.7
Anxiety disorder
Yes	7	11.9	17	14.9	22	19.3	13	18.8	3	9.4
No	52	88.1	97	85.1	92	80.7	56	81.2	29	90.6
Neuro disorder
Yes	2	3.4	4	3.5	30	26.3	11	15.9	12	37.5
No	57	96.6	110	96.5	84	73.7	58	84.1	20	62.5

Finally, through the three-step method (Asparouhov & Muthen, 2013), sex, age and living with parents were incorporated into the model as covariates to establish whether these characteristics predict membership in each of the classes of suicidal ideation. The analysis was performed as a multinomial logistic regression (see Table 5). In this way, the estimated parameters represent the effect of the covariates on the odds of being classified in one of the classes of suicidal ideation compared to a reference class (i.e. the No ideator class).

Table 5. Multinomial regression. Relationship between sex, age, living with parents, and membership in the suicidal ideation classes.

	Unintentional planner vs. non-ideator			Non-planner with specific ideation vs. non-ideator			Planner with intent vs. non-ideator			Non-planner with nonspecific ideation vs. non-ideator
	B	OR	p	B	OR	p	B	OR	p	B	OR	P
Sexo	−1.35	0.25	.005	−0.85	0.42	.020	−1.35	0.25	<.001	0.20	1.22	.621
Edad	0.47	1.60	.361	1.41	4.11	.010	2.06	7.87	<.001	1.22	3.40	.022
Padres	−0.37	0.69	.664	0.96	2.62	.415	−0.92	0.39	.147	0.66	1.94	.454

Note: OR = odds ratio.

Thus, it was observed that females show higher odds than males of being classified in the Unintentional planner, Non-planner with specific ideation, and Planner with intent classes compared to the reference Non-ideator class. Additionally, the older adolescent age group (14–21 years) showed higher probability than their younger counterpart (10–13) of being classified in the Non-planner with nonspecific ideation, Non-Planner with specific ideation and Planner with intent classes compared to the Non-ideator reference class. Finally, living with parents was not significantly related to class membership.

Discussion

There is limited evidence on the patterns or time course of suicidal ideation and attempts in adolescents associated with changes in the severity and intensity of suicidal thoughts (Czyz & King, 2015; Miranda et al., 2014, Nock et al., 2013). Increasing severity of suicidal ideation over the course of hours and days often precedes a suicide attempt or non-suicidal self-injury (Millner et al., 2016). There is a need for risk profiling to identify early the risk of self-harm in adolescents.

Latent class analysis (LCA) has already been used in studies on adolescent suicidal behavior (De Luca et al., 2014; Díez et al., 2020; Stoep et al., 2009) allowing to elaborate behavioral profiles useful in the detection of suicidal risk in adolescents (Díez et al., 2020).

The results obtained allow us to classify the adolescents into clearly differentiated groups according to their latent risk of suicide and independently of their psychiatric diagnosis. The first two groups ‘planners with intent’ and ‘unintentional planner’ have a higher probability of presenting suicidal ideation and intent to carry out the suicidal act. In contrast, the groups ‘non-planners with specific ideation’ and ‘non-planners with nonspecific ideation’ present nonspecific thoughts with a low probability of planning and attempt. The last group ‘non-ideators’ presents a lower level of risk.

This classification, based on the assessment of the intensity of suicidal thoughts, may allow healthcare professionals to identify and classify adolescents into clinical risk subgroups facilitating decision making in the choice of specific and differentiated intervention strategies. Latent class analysis (LCA) has potential benefits in the early identification of behavioral patterns associated with the severity of suicidal ideation and challenges the prevailing misconception that suicidal behavior corresponds to a single type of behavior and that suicidal thoughts are less severe than suicide attempts (Díez-Gómez et al., 2020).

The study has some limitations. First, the data were provided by the adolescents themselves, which could generate a problem with the variance of the shared method. Second, given the characteristics of the population studied and the type of design used, the results lack generalizability. It is necessary to generate more evidence regarding the clinical usefulness of risk profiles in adolescents with suicidal thoughts by incorporating other clinical and sociodemographic variables that generate greater predictive capacity. At the local level, the results obtained could be considered in the identification of suicidal risk and as a guide in the decision-making process of clinical teams that allow timely referral to mental health teams.

Despite these limitations, the results obtained warn about the importance of promoting the development of strategies for the prevention and early detection of suicidal behavior through the characterization of the severity of suicidal thoughts in adolescents.

In conclusion, our findings support the importance and usefulness of this psychometric approach in the identification of risk groups based on the assessment of the severity of suicidal ideation, decreasing the probability of progression to more critical behaviors, such as planning or attempting suicide. Risk profiles based on the severity of suicidal ideation can help health care teams to make timely screening and referral decisions to more specialized clinical intervention programs.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are available from the corresponding author, [PMB], upon reasonable request.

Additional information

Funding

This study was supported by the National Fund for Scientific and Technological Development (Fondo Nacional de Desarrollo Científico y Tecnológico-FONDECYT No.11170342) of the Government of Chile. The fourth author thanks ANID, Becas Chile/Doctorado Nacional N° 21221222.

Notes on contributors

Pablo Mendez-Bustos

Pablo Méndez Bustos (PhD), Psychologist. University of Concepción - Family Therapist. Accredited as Specialist in Psychotherapy by the National Commission of Accreditation of Clinical Psychologists - Master in Social Research and Development - Master in Research Designs and Applications in Psychology and Health - Ph.D in Clinical and Health Psychology. Universidad Autónoma de Madrid- Affiliated Researcher of the Center of Excellence for Cultural Competence of the Institute of Psychiatry at Columbia University, New York. He currently works at the Catholic University of Maule, Chile.

Carlos Mellado

Carlos Mellado (PhD) is a Psychologist with extensive experience in research methodology and statistical analysis. His research focus includes Human Development in adolescence and emerging adulthood and its relationship with parenting, peers and temperament. In parallel, he studies the effect of disasters such as forest fires on the mental health of affected children and adolescents. He currently works at the Universidad Santo Tomás, Chile.

Jorge Lopez-Castroman

Jorge López Castromán, Full Professor of Psychiatry at the University of Montpellier and currently Visiting Professor at the University Carlos III of Madrid, has specialized in Psychiatry at the Fundación Jiménez Díaz and completed his PhD in 2008 at the Universidad Autónoma de Madrid. His work focuses on improving the assessment, prevention and treatment of suicidal behaviors and affective disorders. With more than 140 articles published in international peer-reviewed journals and more than 10 book chapters, he has held prominent roles as head of the departments of emergency and liaison psychiatry, as well as child and adolescent psychiatry at the University Hospital of Nimes (France). In addition, he has served as co-chair of the European Psychiatric Association’s section on suicidology and is currently a member of its Education Committee.

Oriana Arellano-Faúndez

Oriana Marisol Arellano-Faúndez, Social Psychologist. Master in Social Psychology, Universidad de Talca. Diploma in Innovation and Collaborative Management for Teaching in Higher Education, Universidad Autónoma de Chile. Certified in Local Development Strategies. International Training Center Turin, Italy International Labor Organization (ILO). Doctoral candidate in Psychology, Universidad Católica del Maule. Doctoral Fellow of the National Agency for Research and Development, Advanced Human Capital. Her lines of research are gender, life trajectory and mental health.