Analyzing transdiagnostic internalizing symptoms in a global sample of trauma‐exposed children using pooled individual participant data: a latent transition analysis

Abstract

Background

Children often experience both posttraumatic stress (PTS) and depression after potentially traumatic events (PTEs). Latent class analyses (LCAs) identify subgroups with different co‐occurring symptoms, but little is known about what predicts transitions between these symptom classes over time. Analyzing these transitions could reveal factors that influence the shift to classes with fewer internalizing symptoms.

Methods

Using harmonized individual participant data (n = 787) from nine studies (United States, United Kingdom, Australia, and Switzerland) included in the Prospective studies of Acute Child Trauma and Recovery Data Archive (PACT/R), internalizing symptom classes (using PTS and depression items) were identified using LCAs for T1 (0–2 months posttrauma) and T2 (3–15 months posttrauma) separately. Latent transition analysis was used to examine predictors of transitions across symptom classes over time. Included predictors were age, gender, minority status, and trauma type.

Results

Five classes were identified at both time points: ‘low internalizing’ class (T1: 20%; T2: 40%), ‘low PTS/moderate‐high depression’ class (T1: 20%; T2: 27%), ‘moderate internalizing’ class (T1: 25%; T2: 12%), ‘moderate PTS/high depression’ class (T1: 17%, T2: 14%), and ‘high internalizing’ class (T1: 16%, T2: 6%). ‘Low internalizing’ was the most common and stable response (87% stayed in this class). At both time points, 40% of all children were classified into either a ‘low PTS/moderate‐high depression’ class or a ‘moderate PTS/high depression’ class. Children belonging to minority groups in their country of residence and children exposed to interpersonal PTEs were less likely to transition from the ‘high internalizing’ class to lower symptom classes.

Conclusions

Most children show low PTS and depression symptoms after PTEs. When symptoms do develop, they usually involve both high depression and PTS, rather than high PTS with low depression. This underscores the importance of assessing both conditions and developing interventions that target multiple disorders. Minority children exposed to interpersonal trauma require special focus in research and treatment.

Introduction

Every year, millions of children are exposed to potentially traumatic events (PTEs) worldwide (Smith, Dalgleish, & Meiser‐Stedman, 2019). Exposure to PTEs may adversely affect a child's long‐term development and well‐being on a social, emotional, and physical level (Fairbank & Fairbank, 2009; Pynoos et al., 2009). To date, research in trauma‐exposed children has mainly focused on the occurrence, predictors, and treatment of posttraumatic stress (PTS) disorder. Yet, PTS and depression symptoms frequently co‐occur, and a smaller body of research has investigated the associations among PTS and depression symptoms after exposure to a variety of traumas, for instance, natural disasters (Haws et al., 2023), family and community violence (Nöthling, Suliman, Martin, Simmons, & Seedat, 2019; Sargent, Zahniser, Gaylord‐Harden, Morency, & Jenkins, 2020), and single‐incident trauma (i.e., assault, road traffic accident (RTA) and accidental injury) (Kassam‐adams, Bakker, Marsac, Fein, & Winston, 2015). Strong correlations exist between PTS and depression symptoms in trauma‐exposed populations (Contractor, Greene, Dolan, & Elhai, 2018; Rytwinski, Scur, Feeny, & Youngstrom, 2013). Specifically, a meta‐analysis found correlations between 0.52 and 0.70 for PTSD and depression (Rytwinski et al., 2013). Individuals with co‐occurring PTSD and depression report greater functional impairment and psychological distress compared to those with only one disorder (Morina et al., 2013; Post, Zoellner, Youngstrom, & Feeny, 2012). Thus, focusing on the mechanisms underlying this highly prevalent PTSD‐depression comorbidity pattern is a critical first step in understanding posttrauma distress and the cumulative burden these internalizing disorders place on individuals (Rytwinski et al., 2013).

Yet, theoretical and empirical considerations justify their conceptual separability. Theoretically, PTS symptoms are characterized by trauma specific features including intrusive reexperiencing, avoidance of trauma reminders, and trauma‐related changes in cognitions and mood (American Psychiatric Association, 2013). In contrast, depression encompasses a broader pattern of persistent negative mood, anhedonia, and cognitive‐behavioral disturbance patterns that is not trauma‐related per se (Beck & Alford, 2009). These conditions also have different temporal patterns and etiologic requirements. For DSM definition, PTSD requires an immediate casual connection with a particular identifiable qualifying traumatic stressor, and the symptoms are closely connected with the memory of the same particular stressor (American Psychiatric Association, 2013). In contrast, major depression may emerge without trauma exposure through several mechanisms including preexisting vulnerability, genes, or secondary consequences of trauma (Breslau, 2009). Empirically, factor analytic studies have demonstrated that PTS and depressive symptoms load on separate but correlated factors, even when accounting for symptom overlap, supporting their discriminant validity, while acknowledging their shared variance (Elhai, Contractor, Palmieri, Forbes, & Richardson, 2011; Friedman, Resick, Bryant, & Brewin, 2011; Gros, Simms, & Acierno, 2010; Grubaugh, Long, Elhai, Frueh, & Magruder, 2010).

Although PTS and depression symptoms have been extensively explored theoretically and empirically (Breslau, 2009; Contractor et al., 2018; Stander, Thomsen, & Highfill‐McRoy, 2014), the mechanisms behind their co‐occurrence remain unclear. The majority of research has focused on adults exposed to trauma (Haws et al., 2023). Some perspectives have explored whether one disorder leads to the other (Breslau, 2009) or whether a subset of shared factors may predispose an individual to both PTS and depression (Stander et al., 2014). It has also been proposed by some scholars (Elhai et al., 2011) that the similarity in symptoms between these disorders (e.g., anhedonia, concentration difficulties) contributes to their high comorbidity rate. Another plausible reason for the co‐occurrence between PTS and depression, is the shared underlying pattern of internalizing symptoms (e.g., disruptions in negative affect) (Contractor et al., 2018; Elhai et al., 2015; Simms, Watson, & Doebbeling, 2002).

This latest mechanism aligns with the Hierarchical Taxonomy of Psychopathology (HiTOP) framework (Conway et al., 2019; Kotov et al., 2017), which conceptualizes psychopathology in a hierarchical framework. Higher levels in the model are characterized by broad heterogeneous constructs (e.g., internalizing, externalizing) and lower levels by homogeneous symptoms (e.g., traumatic intrusion). Within this framework, PTS and depression can be understood as related but distinct disorders that share variance through a common internalizing factor while retaining unique symptom patterns and etiological pathways (Watson et al., 2022). Some researchers propose an even stronger position – that PTS may simply reflect a broader latent factor that fully encompasses depressive symptoms rather than maintaining any distinctiveness (Elhai et al., 2011). Yet factor analytic studies support the HiTOP hierarchical view, consistently identifying separate but correlated factors for PTS and depression (Contractor et al., 2018).

Further research focusing on the relationship between symptom dimensions indicates that PTSD clusters representing dysphoria and alterations in arousal are primarily responsible for the shared variance with depression. For instance, certain PTSD symptom clusters (such as dysphoric arousal) are strongly related to somatic depression, while others (such as negative affect) mainly relate to nonsomatic depression (Contractor et al., 2018). This overlap suggests that analyzing the heterogeneity within these two primary internalizing disorders is a scientifically strategic focus because their shared underlying dimensions inherently capture the core distress factors that bridge internalizing disorders to broader psychopathology dimensions, including those that manifest as somatic complaints or emotional dysregulation (Contractor et al., 2018).

A complementary approach for understanding the co‐occurrence between PTS and depression is to view comorbidity as the result of similar symptom presentations that occur in only a subgroup of people. Person‐centered models such as latent class analysis (LCA) allow us to identify groups of people (‘classes’) based on distinct subsets of symptoms. Applying LCA makes it possible to identify whether distinct PTS or depression symptom classes exist (i.e., at least one class has only elevated PTS/depression symptoms), or whether co‐occurring symptoms reflect dysfunction across classifications (Contractor, Roley‐Roberts, Lagdon, & Armour, 2017; Haws et al., 2023). Importantly, LCA distinguishes between two competing interpretations: whether identified classes represent truly distinct symptom profiles (e.g., high PTS/low depression vs. low PTS/high depression) or simply severity gradations of a unified construct where all symptoms increase or decrease together. Evidence for distinct profiles would support the conceptual separability of PTS and depression, whereas classes showing only quantitative differences across all symptoms simultaneously would suggest a single underlying factor.

Several studies used LCA to examine co‐occurring classes of PTS and depression among adult refugees (Lenferink et al., 2022), primary care patients (Rahman et al., 2018), and parents of trauma‐exposed children (Sadeh et al., 2024). Across these LCA studies, there were three or four classes identified that differed in severity and type, which signified that the severity of PTS and depression symptoms was not consistent with each other (Lenferink et al., 2022; Rahman et al., 2018; Sadeh et al., 2024). For example, LCA conducted with primary care patients identified four classes with qualitatively different symptom classes of PTS and depression symptoms (i.e., high severity and comorbidity subclass, moderate severity and comorbidity, low PTS and high depression, and resilient class; Rahman et al., 2018). To the best of our knowledge, identifying PTS and depression symptom classes in children after PTEs exposure using LCA has not been examined yet [but see latent profile analyses (LPAs) examining PTS, depression, and additional outcomes in children (Lai, Kelley, Harrison, Thompson, & Self‐Brown, 2015; Zhen & Zhou, 2022)].

Although LCA can be used to identify latent classes of people with distinct symptom classes, it remains unclear how stable these classes are over time. A better understanding of transitions among latent classes might assist early intervention efforts by identifying factors associated with transitions, for example, classes that show recovery via transition from a high symptom class to a low symptom class. A Latent Transition Analysis (LTA) is a longitudinal extension of LCA intended for the analysis of changes in class membership (and predictors of these changes) across two or more time points (Collins & Lanza, 2010; Jung, Lee, & Lee, 2021). This approach has been used to identify PTS and/or depression symptom classes and transitions between classes over time, mainly among adults following an injury (Forbes et al., 2016), across the military deployment cycle (Boasso, Steenkamp, Larson, & Litz, 2016), as well as transitions in PTS and depression symptom classes among adult civilians exposed to conflict/war (Tay, Jayasuriya, Jayasuriya, & Silove, 2017) and among people with a refugee background (Lenferink et al., 2022).

To the best of our knowledge, only two LTAs have been conducted to examine PTS and depression symptom‐profiles using continuous outcomes (i.e., LPA) among adolescent victims of bullying (Yang et al., 2023) and following the Wenchuan earthquake (Zhou & Wu, 2019). Both studies revealed a ‘low/no symptom comorbidity’ group (30%; 55%) and a ‘high comorbidity’ group (4%; 8%) at the first time point. A subgroup of adolescent victims of bullying showed ‘predominantly PTSD symptoms’ (36%), whereas adolescents following the Wenchuan earthquake showed a ‘depression symptoms’ group (38%). Additionally, while most children following the earthquake showed stable (76%) patterns of PTSD and depression across 1 year, following bullying events only 10% showed a stable pattern of PTSD across 3 years. While different measurement time points as well as different types of traumatic events may play a central role in understanding these differences, additional studies in the field are warranted. The current study addresses this gap by using LTA to assess the transition of patterns of PTS and depression symptoms across time in children following various PTEs.

Our second aim was to examine which risk factors related to transition from one class to another over time. Based on prior research in children after PTEs exposure (Alisic et al., 2014; Memarzia, Walker, & Meiser‐Stedman, 2021; Pumariega, Jo, Beck, & Rahmani, 2022; Trickey, Siddaway, Meiser‐Stedman, Serpell, & Field, 2012; Vibhakar, Allen, Gee, & Meiser‐Stedman, 2019), we expected that female gender, younger age, belonging to an ethnic minority, and interpersonal (vs. other trauma types) would be related to classes that were characterized by more pervasive PTS and depression symptoms patterns over time.

In sum, while examining (predictors of) PTS and depression symptom classes is urgently needed to gain more insights in children following PTEs, LTA research in children is limited likely due to the relatively large samples needed to conduct this type of analysis. Consequently, we reused existing pooled participant‐level data from nine studies, providing a larger sample of this vulnerable group and allowing the use of LTA and consideration of sociodemographic predictors that would not otherwise be possible (Kassam‐Adams & Olff, 2020). We used LCA to identify classes of internalizing symptoms (using PTS and depression items) at an acute posttrauma time point (T1; 0–2 months following PTEs) and a later time point (T2; 3–15 months after the PTEs). LTA was used to examine (predictors of) transitions in PTS and depression symptom classes over time. More specifically, we examined to what extent changes in symptom classes over time were predicted by child gender and age, child's ethnic minority status, and type of PTEs.

Methods

This study used data from the Prospective studies of Acute Child Trauma and Recovery (PACT/R) Data Archive (Kassam‐Adams et al., 2020; Lenferink et al., 2020; Sadeh et al., 2024; Silverstein et al., 2022), part of the Child Trauma Data Archives (CTDA, www.childtraumadata.org). PACT/R is an international data repository of prospective studies of child traumatic stress. The archive includes deidentified, anonymized data from studies conducted with ethics approval. The Institutional Review Board (IRB) of The Children's Hospital of Philadelphia determined that reuse of data from the archive does not constitute human subjects research because all data are deidentified. These projects are associated with the Global Collaboration on Traumatic Stress (Olff et al., 2020) FAIR Data Theme, which supports a growing collection of integrative individual patient data projects (https://www.global‐psychotrauma.net/fair). We included data from the nine studies (out of 30 studies currently in the PACT/R Archive) that included at least two child measurements on PTS and depression (0–2 months and 3–15 months after the PTEs). In total, 787 children were included from nine studies conducted in the USA, United Kingdom, Australia, and Switzerland, with children ages 7–18 (de Haan, Tutus, Goldbeck, Rosner, & Landolt, 2019; Delahanty, Nugent, Christopher, & Walsh, 2005; Kassam‐Adams, Marsac, & Cirilli, 2010; Meiser‐Stedman et al., 2017; Meiser‐Stedman, Smith, Glucksman, Yule, & Dalgleish, 2008; Meiser‐Stedman, Yule, Smith, Glucksman, & Dalgleish, 2005; Nixon et al., 2010; Winston et al., 2002; Zehnder, Meuli, & Landolt, 2010). Table 1 presents demographic data for the final sample.

Table 1.

Sample demographic characteristics

Variables		N (%)
Sex	Boys	505 (64.2%)
	Girls	282 (35.8%)
Age of child (years)	M = 12.15 (SD = 2.77; 7–18)
Country of residence	USA	266 (34%)
	United Kingdom	267 (34%)
	Australia	79 (10%)
	Switzerland	163 (21%)
Ethnic minority in country of residence	No	517 (65.7%)
	Yes	270 (32.8%)
	Missing	12 (1.5%)
Trauma type	Unintentional injury (fall, sports, animal bite etc.)	439 (55.8%)
	Acute medical event (noninjury)	4 (0.5%)
	Road traffic accident requiring medical attention	284 (36.1%)
	Interpersonal violence	48 (6.1%)
	Other	12 (1.5%)

Measures

The nine studies used in this analysis had utilized 13 measures of PTS (five out of the nine studies used multiple measures for PTS symptoms) and eight measures of depression. The measures used in each study can be found in Table 2.

Table 2.

Measures of PTS and depression symptoms used in each study

Child trauma data archives dataset	Winston et al. (2002)	Kassam‐Adams et al. (2011)	Delahanty et al. (2005)	Meiser‐Stedman et al. (2005)	Meiser‐Stedman et al. (2008)	Nixon et al. (2010)	Kassam‐Adams et al. (2010)	de Haan et al. (2019)	Meiser‐Stedman et al. (2017)
	#1001	#1007	#1018	#1020	#1021	#1022	#1023	#1032	#1037
Posttraumatic stress measures
Anxiety Disorders Interview Schedule (PTSD module)				x
ASD Dissociation Criteria Interview				x	x
CAPS‐CA	x		x		x	x
Children's PTSD Inventory									x
IBS‐KJ (CAPS‐CA German)							x
Acute Stress Checklist						x
Child Acute Stress Questionnaires	x
Child and Adolescent Trauma Survey	x
Child Impact of Event Scale – Revised				x	x
Child PTSD Symptom Scale		x				x			x
Immediate Stress Reaction Checklist	x
UCLA PTSD Reaction Index for DSM IV – German								x
Depression measures
Birleson Depression Self‐Rating Scale				x	x
Child Depression Invetory – English	x					x
Child Depression Inventory 2 – German							x
Child Depressoin Inventory 3 – German								x
Child Center for Epidemological Studies Depression		x
Reynolds Adolescent Depression Scale			x
Reynolds Child Depression Scale			x
Short Mood and Feelings Questionnaire									x

ASD, acute stress disorder; CAPS‐CA, Clinician‐Administered Posttraumatic Stress Disorder Scale for Children and Adolescents; IBS‐KJ, German version of the Clinician‐Administered PTSD Scale for Children and Adolescents (CAPS‐CA); PTS, posttraumatic stress.

Given the use of pooled individual participant data (IPD) from nine studies, the measures utilized varied reporter types, including clinician‐administered scales (e.g., CAPS‐CA), parent reports, and self‐reports (Table 2). To ensure comparability across these diverse reporters and measures (which used different item response scales, e.g., Likert 1–4 vs. 1–5), the core harmonization step (described in the Data Harmonization section and in Appendix S1, Step 3) involved dichotomization of item‐level ratings into symptom presence or absence based on each measure's standard scoring rules or expert consensus where required. This provided comparable binary metrics for the LCA analysis, meaning that mathematical transformations such as rescaling or z‐scoring were not applied.

Data harmonization

The constructs being harmonized were PTS symptoms and depression symptoms. Theoretically, these were distinguished by their unique temporal requirements: PTS symptoms are trauma‐anchored and linked to a specific stressor, whereas depression encompasses broader mood disturbance not causally connected to a traumatic event.

In order to compare IPD across nine discrete studies, we created standardized harmonization processes to examine symptom‐level constructs across measures of PTS and depression. The multistep process example outlined in Figure 1 includes identifying symptom concordance across major diagnostic manuals (DSM and ICD) followed by item‐level mapping of measures onto symptom constructs and then review by the Child Trauma Data Archives Expert Advisory group to reach consensus on item‐level concordance. A table outlining item‐level harmonization across these measures and more can be found in https://childtraumadata.org/use‐data/harmonization. Additional information on the harmonization process can be found in Appendix S1.

Figure 1.

Example of multistep harmonization process: intrusive thoughts

As described in the Measures section, to address measurement comparability across these varied reporters and scales (which used different Likert formats), the final step in the harmonization process involved dichotomization of item‐level ratings into symptom presence or absence based on standard clinical cutoffs. This yielded the comparable binary metrics required as indicators for the LCA.

To ensure the analytical independence of the harmonized constructs, we specifically examined content overlap and made a methodological decision regarding the final symptom indicators; in our data harmonization process across nine studies, we identified three overlapping symptoms between PTSS and depression measures: sleep problems, difficulty concentrating, and diminished interest. This limited overlap was partly due to most measures being based on DSM‐IV criteria, which for PTSS did not include the negative cognitions and mood cluster introduced in DSM‐5, resulting in fewer potentially overlapping symptoms between PTSS and depression constructs. Despite the content similarity of these few overlapping items, we maintained all PTS and depression items, acknowledging their different temporal frameworks (trauma‐anchored vs. general time frame) and theoretical underpinnings. This approach is consistent with research demonstrating that temporal anchoring and etiological context maintain discriminant validity between constructs even when symptom content overlaps (Friedman et al., 2011; Gros et al., 2010; Grubaugh et al., 2010).

Background characteristics

Correlates of class membership were child age in years, child gender, child ethnic minority status, and type of PTE (1 = unintentional injury, 2 = acute medical event (noninjury), 3 = motor vehicle accident/road traffic accident, 4 = interpersonal violence, and 5 = other). According to data from each study about the child's race/ethnicity, ethnic minority status in the child's country of residence was represented by a dichotomized harmonized variable. Additional information on the harmonization process can be found in Appendix S1.

Analytic procedure

We conducted LTA using Mplus 8.8 (Muthén & Muthén, 2012). LTA consists of three consecutive steps. First, LCAs were conducted at T1 (0–2 months after PTEs) and separate LCAs were conducted at T2 (3–15 months after PTEs). Models with one‐ to six‐class solutions were examined. Selection of optimal class solution was based on the following comparative fit statistics: (a) lower (sample‐size adjusted) Bayesian Information Criterion (SA‐BIC and BIC) and Akaike Information Criterion (AIC), (b) higher entropy R ² value, and (c) significant p‐value (<.05) of the Vuong–Lo–Mendell–Rubin test (VLMRt), Lo–Mendell–Rubin‐Likelihood Ratio Test (LMR‐LRt), and the Bootstrap‐Likelihood Ration Test (BLRt). Parsimony and interpretability were also considered when comparing model fit. Probability estimates (i.e., indicating the rate of how many people in a certain class endorsed a symptom) >.60 were considered high, 0.16–0.59 moderate, and ≤0.15 low. Full information maximum likelihood estimation was used to account for missing data on LCA indicators.

Second, measurement invariance was tested between the classes at the T1 and T2 for the final models resulting from step 1, by comparing models in which the parameters (i.e., thresholds) were held equal across T1 and T2 (i.e., full measurement invariance) with models in which the parameters were allowed to vary freely across T1 and T2 (i.e., full measurement variance). By ensuring measurement invariance, transition probabilities can be calculated based on the symptom classes across time points (Nylund, Muth'en, Nishina, Bellmore, & Graham, 2006). A nonsignificant log‐likelihood difference test was considered to demonstrate full measurement invariance. Third, an unconditional LTA was performed with regressing T2 class‐membership on T1 class‐membership. This allows evaluation of the probability of participants transitioning from one class at T1 to another class at T2 or the probability of staying in the same class over time (i.e., transition probabilities).

Lastly, we examined to what extent age (in years), gender (0 = male, 1 = female), minority status (0 = nonminority in country of residence, 1 = minority in country of residence), and type of PTEs (specifically comparing interpersonal events and noninterpersonal events; 0 = unintentional injury/acute medical event (noninjury)/ motor vehicle accident/ road traffic accident/other, 1 = interpersonal violence) predicted latent transition probabilities from T1 to T2 classes. Associations between predictors indicated no concern for multicollinearity [tolerance range = 0.994–0.996; variance inflation factor (VIF) range = 1.004–1.006]. Missing data on predictors were imputed using multiple imputation on all demographic predictors across the two time points by generating 20 datasets. Following prior research (Coakley et al., 2010; Forbes et al., 2016; Lenferink et al., 2022), we only focused on transition probabilities for participants moving out of the ‘high symptoms’ class to one of the four lower symptom classes.

For categorical predictors, a new feature (i.e., output features and a new Calculator for LTA that make it more convenient to interpret analysis results when assessing effects of covariates on transition probabilities) in the Mplus 8.8 provides the effects of categorical predictors on transitioning probability odds into another class, as compared to staying in the same class (e.g., a probability of transition if the participant is male and a comparative conditional probability if the participant is female) (Asparouhov & Muthén, 2021).

As a final step, we conducted multinomial logistic regression analyses for age as a continuous covariate, which yielded estimates reflecting the age covariate's ability to predict moving out of high symptoms (using the trajectory of staying in the high symptom class as the reference group; Muthén & Muthén, 1998–2017c). The maximum likelihood with robust standard errors estimator was used to estimate conditional transition probabilities.

Additionally, to ensure that observed transition patterns were not disproportionately driven by particular cohorts, we included cohort membership as a covariate in the transition models. This approach allowed us to test whether transition probabilities varied significantly across the nine contributing studies, thereby assessing the robustness of our harmonization approach and the generalizability of findings across diverse international samples. This analysis was particularly important given the substantial variance in T2 assessment timing (3–15 months posttrauma) across cohorts, allowing us to evaluate whether differences in follow‐up periods affected transition patterns.

Results

Background characteristics

Children were 12.15 years old (SD = 2.77) on average and 505 were boys (64%). Most children were not an ethnic minority in their country of residence (n = 517; 67%). All studies enrolled participants after PTEs that met Criterion A for PTSD under DSM‐IV or DSM‐5. The most common type of index event was unintentional injury (56%; e.g., fall and sports). See Table 1 for more details.

PTS and depression symptom classes at T1 and T2

Table 3a,b presents the fit indices for the one‐ to six‐class models at T1 and T2. The five‐class solution had the lower AIC and SABIC compared to other classes. According to the LMR, the two‐, three‐, four‐, and five‐class solutions had a significant LMRt, which indicated that each model had a significantly better fit than the model with one less class, except for the six‐class model. All entropy values were acceptable. Between the two‐, three‐, four‐, and five‐class models, the five‐class solution was selected as the best fitting model at T1 and T2 as it had the lowest AIC and SABIC.

Table 3.

(a) T1 LCA model fit summary. (b) T2 LCA model fit summary

Model	Fit indices						Class 1		Class 2		Class 3		Class 4		Class 5		Class 6
	AIC	BIC	SABIC	Entropy	LMR adjusted p‐value	BLRt p‐value	N	%	n	%	n	%	n	%	n	%	n	%
(a)
1	20,326.97	20,448.35	20,365.78	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
2	17,739.03	17,986.45	17,818.15	0.88	.00***	.00***	508	64.54	279	35.45	–	–	–	–	–	–	–	–
3	17,202.48	17,575.94	17,321.90	0.84	.00***	.00***	184	23.38	256	32.52	347	44.09	–	–	–	–	–	–
4	16,948.08	17,447.58	17,107.80	0.84	.00***	.00***	168	21.34	243	30.87	169	21.47	207	26.30	–	–	–	–
5	16,845.70	17,471.25	17,045.73	0.82	.04 *	.00 ***	136	17.28	180	22.87	147	18.67	167	21.22	157	19.94	–	–
6	16,750.14	17,501.73	16,990.47	0.82	.47	.00***	88	11.11	81	10.29	140	17.78	175	22.23	162	20.58	141	17.91
(b)
1	17,066.19	17,187.57	17,105.01	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
2	14,501.38	14,748.80	14,580.49	0.89	.00***	.00***	–	–	–	–	–	–	–	–	–	–	–	–
3	13,963.49	14,364.96	14,091.86	0.87	.00***	.00***	312	39.64	144	18.29	331	42.05	–	–	–	–	–	–
4	13,777.73	14,314.58	13,949.39	0.85	.02*	.00***	328	41.67	116	14.74	141	17.916	202	25.667	–	–	–	–
5	13,655.03	14,327.26	13,869.98	0.85	.02 *	.00 ***	123	15.6	43	5.46	210	26.68	314	39.89	97	12.32	–	–
6	13,596.74	14,404.34	13,854.98	0.87	.45	.00***	42	5.33	53	6.73	90	11.4	123	15.629	221	28.08	258	32.78

AIC, Akaike Information Criterion; BIC, Bayesian Information Criterion; BLRt, Bootstrap Likelihood Ratio Difference test; LMR, Lo–Mendell–Rubin Adjusted Likelihood Ratio test; SABIC, Sample‐size Adjusted Bayesian Information Criterion. Bold row was chosen model.

*p < .05, **p < .01, ***p < .001.

For T1 and T2, the first class, labeled ‘low internalizing’ (T1, n = 160, 20%; T2, n = 316, 40%), was characterized by low probability estimates of all PTS and depression symptoms, except for the symptom ‘worthlessness’ (b = 0.196), and the trauma‐related amnesia item (b = 0.180). The second class, labeled ‘low PTS/moderate‐high depression’ (T1, n = 157, 20%; T2, n = 214, 27%), was characterized by low probability of 16 out of 17 PTS symptoms, with a moderate probability of seven out of nine depression symptoms. The third class, labeled ‘moderate internalizing’ (T1, n = 202, 25%; T2, n = 96, 12%) was characterized by moderate endorsement of 14 out of 17 PTS symptoms, and six out of nine of the depression symptoms. The fourth class labeled ‘moderate PTS/high depression’ class (T1, n = 139,17%; T2, n = 109, 14%) was characterized by moderate levels of 12 out of 17 PTS symptoms, and high probability of endorsing all of the depression items. The fifth and smallest class named ‘high internalizing’ class (T1, n = 128, 16%; T2, n = 52, 6%) consisted of elevated levels of 11 out of the 15 PTS symptoms, and 8 out of the 9 depression symptoms. Figure 2A,B show the probability estimates for the five‐class model at T1 and at T2.

Figure 2.

Probability estimates for the five‐class model at (A) T1 (N = 787). T1 = 0–2 months posttrauma. (B) T2 (N = 787). T2 = 3–15 months posttrauma. PTS, posttraumatic stress

Unconditional latent transition model

The loglikelihood difference test between the unconstrained (i.e., allowing thresholds to vary freely) and constrained latent transition model (where thresholds were held equal) was not significant (ΔLoglikelihood = 156.232 (Δdf = 130), p > .10). This provided support for full measurement invariance.

The probability (i.e., transition estimates) of participants remaining in the same class (diagonal in the table) (i.e., a stable trajectory) or moving to another class (off‐diagonal in the table) from T1 to T2 is shown in Table 4. The most stable trajectories (greater than 80% likelihood of remaining in the same class at T2) were for children in the ‘low Internalizing’ class at T1, or those in the ‘low PTS/moderate‐high depression’ class at T1. Children in the ‘moderate PTS/high depression’ class at T1 had about a 50% likelihood of remaining in that class and were next most likely (36% likelihood) to move to the ‘low PTS /moderate‐high depression’ class. The least stable trajectories (less than 33% likelihood of the same class at T2) were for those in the ‘moderate internalizing’ class at T1 (most likely to move to the ‘low internalizing’ class at T2) and those in the ‘high internalizing’ class. Figure 3 shows the latent transition model.

Table 4.

Transition probabilities

	Low internalizing symptoms at T2	Low traumatic stress/moderate ‐high depression symptoms at T2	Moderate internalizing symptoms T2	Moderate traumatic stress/high depression T2	High internalizing symptom at T2
Low internalizing symptoms at T1	0.87	0.05	0.05	0.01	0.00
Low PTS/moderate – high depression symptoms at T1	0.10	0.82	0.00	0.07	0.00
Moderate internalizing symptoms at T1	0.62	0.05	0.29	0.02	0.00
Moderate PTS/high depression at T1	0.05	0.36	0.02	0.50	0.05
High internalizing symptom at T1	0.19	0.10	0.19	0.17	0.32

The bold formatting on the diagonal as it emphasizes class stability. These bold values show the probability of remaining in the same symptom class across time points, which is central to understanding trajectory patterns and allows readers to quickly distinguish stability from transition probabilities.

Diagonal estimated indicates the probability of participants remaining in the same class from T1 through T2. The off‐diagonal estimates represent the probability of participants transitioning to another class from T1 to T2; T1 = 0–2 months posttrauma; T2 = 3–15 months posttrauma. PTS, post traumatic stress.

Figure 3.

Unconditional latent transition model (N = 787). PTS, posttraumatic stress. Thickness of the lines represents the magnitude of transition probabilities

Conditional latent transition model

The conditional latent transition model included age, gender, ethnic minority group, trauma type, and cohort membership. Odds ratios (ORs) for all predictors associated with transitions between the participants who stayed in the ‘high internalizing’ class compared to those who moved to other classes are shown in Table 5.

Table 5.

Odds ratio of individual covariates associated with transition of latent internalizing classes

Covariates	OR (95% CI)
High symptoms class to moderate PTS/high depression class versus staying in the high symptoms class
Gender (ref. female = 1)	5.26 (0.03, 1.16)
Ethnic minority (ref. minority =1)	7.35 (0.02, 0.83)
Trauma type (ref. interpersonal trauma = 1)	0.00 (0.00, 0.00)
Age (continuous variable 7–18 years)	1.19 (0.66, 1.05)
High symptoms class to moderate symptoms class versus staying in the high symptoms class
Gender (ref. female = 1)	1.02 (0.41, 2.36)
Ethnic minority (ref. minority =1)	1.49 (0.32, 1.40)
Trauma type (ref. interpersonal trauma = 1)	1.89 (0.19, 1.47)
Age (continuous variable 7–18 years)	1.06 (0.79, 1.11)
High symptoms class to low PTS/moderate‐high symptoms class versus staying in the high symptoms class
Gender (ref. female = 1)	9.09 (0.00, 3.10)
Ethnic minority (ref. minority =1)	7.69 (0.02, 0.83)
Trauma type (ref. interpersonal trauma = 1)	0.00 (0.00, 0.08)
Age (continuous variable 7–18 years)	1.08 (0.68, 1.26)
High symptoms class to low symptoms class versus staying in the high symptoms class
Gender (ref. female)	1.30 (0.39, 1.52)
Ethnic minority (ref. minority =1)	7.81 (0.05, 0.29)
Trauma type (ref. interpersonal trauma = 1)	3.57 (0.09, 0.82)
Age (continuous variable 7–18 years)	0.93 (0.86, 1.34)

The bold formatting as it highlights statistically significant associations (p < 0.05), identified by 95% confidence intervals that exclude 1.0. This bold denotes emphasis which covariates significantly predict class transitions. Confidence intervals provide both statistical significance and precision of effect estimates, which is more informative than p‐values alone and aligns with current best practices in statistical reporting.

Two predictors showed significant positive effects on transitioning out of the high internalizing class to classes with less symptomatology. The odds of transitioning to the ‘low PTS/moderate‐high depression’ class, the ‘moderate PTS/High depression class’, or to the ‘low internalizing’ relative to staying in the ‘high internalizing’ class is 7.35, 7.69, and 7.81 times lower for children belonging to a minority group than for children from nonminority groups and significantly so. The odds of transitioning to the ‘low internalizing’ class relative to staying in the ‘high internalizing’ class is 3.57 times lower for children following interpersonal PTEs than for children following other types of PTEs which are not interpersonal and significantly so. Age and gender were not associated with transiting out of the ‘high internalizing’ class compared with staying in this class.

Cohort membership was not significantly associated with any transition probabilities, with all confidence intervals for cohort effects including 1.0. This indicates that transition patterns were consistent across the nine contributing studies, which supports the robustness of our data harmonization approach. Importantly, this suggests that the substantial variance in T2 assessment timing (3–15 months posttrauma) across cohorts did not systematically affect transition probabilities and that our identified patterns are robust across diverse temporal frameworks.

Discussion

The two aims of this study were to identify (a) transitions in classes of PTS and depression symptom classes in children from 0–2 months to 3–15 months after PTEs exposure, and (b) factors that predict transition between classes over time. Our first main finding was that five symptom classes emerged from the data that were similar across time: (a) low internalizing, (b) moderate internalizing symptoms, (c) low PTS/moderate‐high depression, (d) moderate PTS/high depression, and (e) high internalizing classes. These classes differed both by shape/type (qualitative differences) and by symptom severity (quantitative differences).

Our first main finding shows that over time, a significant percentage of the children (40%) had low levels of PTS and depression symptoms. In line with these findings, prior studies revealed that most children following a single PTE are displaying low symptoms or are able to recover within the following months up to a year after the event (La Greca et al., 2013; Lai et al., 2021; Punamäki, Palosaari, Diab, Peltonen, & Qouta, 2014; Zhang et al., 2022).

Our second main finding concerns the co‐occurrence and relationships between PTS and depression symptoms. Importantly, when person‐centered approaches identify multiple symptom classes, the interpretation of whether these represent truly distinct symptom profiles versus gradations in overall severity requires careful consideration. Evidence for distinct profiles would include classes showing differential patterns of PTS and depression symptoms (e.g., high PTS/low depression vs. low PTS/high depression), whereas severity gradations would show primarily quantitative differences across all symptoms simultaneously. Given that 40% of all children at both time points were classified into a ‘low PTS/moderate‐high depression’ class or a ‘moderate PTS/high depression’ class, it appears that when these symptoms are seen together after children's trauma exposure, depression symptoms may be more prominent than PTS symptoms. Thus, our findings suggest a qualitative trend where PTS symptoms often occur alongside higher depression symptoms, rather than vice versa. A longitudinal study on children after the Wenchuan earthquake suggested an interactive model of PTS and depression comorbidity. Initially, PTS and depression symptoms influenced each other, while at subsequent time points, depression symptoms were predicting later PTS symptoms (and not vice versa) (Cheng, Liang, Fu, & Liu, 2018). While our analysis does not provide assumptions regarding the direction of the associations between PTS and depression symptoms over time, considering our results and those of Cheng et al. (2018) as well as three previous studies in adults (Cao et al., 2015; Contractor et al., 2017; Lenferink et al., 2022), it is likely that depression is the dominant subgroup resulting from exposure to PTEs compared to PTS.

When examining class stability over time, we found that the ‘low internalizing’ class was the most stable, with 87% of participants remaining in this class at T2 (n = 140). The ‘low PTS/moderate‐high depression’ class was the second most stable, with 82% of participants (n = 129) staying in this class. These findings suggest that PTS symptoms are more likely to be low and stable or to improve over time, while moderate‐high depression symptoms tend to be stable and persistent after PTEs. This aligns with similar recent studies which examined children following single‐incident trauma (Cheng et al., 2018; Zhang et al., 2022; Zhou & Wu, 2019). One recent explanation for the persistency of depression symptoms across time is also referred to as the internalizing paradox. To elaborate on this, children's depression has a wide variety of comorbidities and symptom combinations compared to other internalizing disorders. However, psychotherapy outcomes were found to be weak for depression, while strong for other internalizing disorders (Eckshtain et al., 2020; Weisz et al., 2019). It means that compared to other internalizing disorders, early‐onset depression might be particularly chronic and potentially genetically based and therefore resistant to change (Weisz et al., 2023).

Regarding the second aim of our study, we found that ethnic minority status was a predictor of class transitions. The odds of transitioning to the ‘low PTS/moderate‐high depression’ class, the ‘moderate PTS/High depression’ or to the ‘low internalizing’ class relative to staying in the ‘high internalizing’ class were 7.35, 7.69, and 7.81 times lower for children belonging to a minority group than for children from nonminority groups. This aligns with prior research describing the close relationship between structural racism, marginalization, and cumulative psychosocial disadvantages which may result in increasing PTS and depression in children of minority/Black, Indigenous, and People of Color (BIPOC) backgrounds (Andrews et al., 2015; Lenferink et al., 2022; Pumariega et al., 2022). The impact of cumulative PTS and historical trauma on minority/BIPOC children and families' needs to be highlighted as part of future research on interventions and preventions for PTS and depression in these populations. PTS and depression symptoms must be addressed with the understanding that these symptoms may be part of a larger, more complex network of risk factors for depression and PTS among children from minority groups (Oldehinkel & Ormel, 2023; Pumariega et al., 2022).

The second predictor related to class transitions was type of PTE. The odds of transitioning to the ‘low internalizing’ class relative to staying in the ‘high internalizing’ class were 3.57 times lower for children following interpersonal PTEs than for children following other types of PTEs. The experience of interpersonal trauma, particularly in this study, assault or violence by nonfamily members, can disrupt social relationships and potentially impact children's cognitive, psychological, and social development. Affect regulation, attention, consciousness, self‐identity, and interpersonal relationships have all been found to be adversely affected by interpersonal trauma (Peters et al., 2022), which might explain the high stability of symptoms in this group.

The main strength of this study is the use of data from the PACT/R Data Archive, combining data from many smaller prospective research studies to enable new analyses and insights. This study is one of the first to examine transitions in internalizing symptom classes beginning in the acute posttrauma phase. This was made possible via pooling individual participant data across nine studies in multiple countries. The PACT/R data resources exemplify an international collaborative joint effort, aiming to preserve and share deidentified child trauma datasets and facilitate cross‐study analyses of harmonized participant‐level data, which are available to any qualified researcher upon request (Kassam‐Adams et al., 2020). As noted in a recent research review (Oldehinkel & Ormel, 2023; Sadeh, Denejkina, Karyotaki, Lenferink, & Kassam‐Adams, 2023), advancing both child mental health research and global mental health research requires expanding and intensifying collaboration among different research groups, facilitating the harmonization of measures, and enabling data sharing and harmonization to allow investigation of developmental processes and cultural variation in mental health across studies.

A limitation of this study is that combining cross‐study data with different measures reduces specificity and granularity regarding PTS and depression severity. Given the use of pooled IPD from nine studies, the measures utilized varied by reporter types, including clinician reports, parent reports, and self‐reports. Research shows that parents and children often provide divergent ratings of the same constructs, with correlations typically in the low‐to‐moderate range, which may reflect genuine differences in perspectives across contexts or variations in symptom observability (De Los Reyes et al., 2022). The varied reporter types across studies may therefore introduce measurement heterogeneity, potentially affecting the comparability of findings. Additionally, our harmonization approach maintained overlapping PTSS and depression items (i.e., sleep problems, difficulty concentrating, and diminished interest) based on their temporal frameworks rather than their content. While this decision was theoretically informed and supported by prior research on discriminant validity (Friedman et al., 2011; Gros et al., 2010; Grubaugh et al., 2010), it may have contributed to shared variance between PTS and depression constructs. Future research could benefit from more detailed examinations of these overlapping symptoms, potentially using advanced psychometric approaches such as bifactor modeling or item response theory to better differentiate shared versus unique variance. Additionally, studies employing experimental designs that manipulate temporal anchoring instructions could provide valuable insights into the theoretical distinction between trauma‐specific versus general manifestations of these symptoms.

Additionally, cross‐study (and cross‐national) variation in assessment of demographic variables limited the availability of sociodemographic variables to be included as predictors. This may have excluded other important predictors of symptom classes in children, such as family relationships (i.e., parent–child and parental relationships) (Cheng et al., 2018; Yap, Pilkington, Ryan, & Jorm, 2014), as well as prior traumatic experiences (Liang, Zhou, & Liu, 2021; Osofsky, Osofsky, Weems, King, & Hansel, 2015). Another limitation is the substantial variance in timing of the T2 measure (3–15 months posttrauma). While our cohort covariate analysis showed that transition probabilities did not differ systematically across studies with different follow‐up periods, providing evidence for the robustness of our findings across this temporal range. The broad time window, while reflecting the practical realities of multisite data harmonization, represents both a limitation and a strength – demonstrating that our identified transition patterns are consistent across diverse temporal frameworks. Future research could benefit from more standardized follow‐up intervals and provide additional insights into how specific timing intervals affect transition probabilities and class membership stability.

An additional important limitation is our sole concentration on internalizing symptoms (PTS and depression). Although 40% of children fell into the ‘low internalizing’ class at T2, this should not be taken as proof of resilience or absence of trauma‐related difficulties. Children and adolescents may be experiencing externalizing behaviors (e.g., aggression, conduct problems, substance use), somatic complaints, academic difficulties, or impairments in social functioning that are not measured by internalizing symptom measures (Alisic et al., 2014). Additionally, some youth may exhibit delayed onset of symptoms or exhibit distress in other ways that are not assessed in our study, such as through heightened risk‐taking behaviors or increased challenges in maintaining healthy peer relations or academic standards (McLaughlin et al., 2013; Trickey et al., 2012). Our focus was therefore limited in scope, omitting children who may exhibit these other symptoms after trauma exposure. If children develop complaints after trauma exposure, then PTSD and depression are (one) the most distressing/disabling symptoms. It is important to focus on these two conditions given that they frequently co‐occur and strong correlations exist between them (e.g., .52–.70 in Rytwinski et al., 2013) and individuals with co‐occurring symptoms report greater functional impairment and psychological distress (Morina et al., 2013; Post et al., 2012). While these other trauma‐related difficulties warrant future investigation, our current study focused on PTS and depression due to their structural centrality and high cumulative functional burden, as elaborated in the Introduction. This specific focus was also constrained by methodological requirements; LTA is a highly demanding statistical method, only feasible with a relatively large sample size (n = 787 in our case, which is unusually large for child trauma research). Future investigations should include multifaceted assessments, encompassing externalizing symptoms, academic performance, peer relationships, and family functioning to yield a more complete picture of mental health posttrauma in children and adolescents.

Lastly, the studies included in the analyses were all conducted in western industrialized countries, limiting the generalizability of the results globally due to cultural differences in expression of PTS and depression symptoms, including language and terminology used to describe symptoms.

Notwithstanding these limitations, using pooled harmonized participant‐level data from nine different studies allowed us to employ an advanced longitudinal statistical method (i.e., LTA) that has been rarely used in trauma samples before, and in children specifically. In order to better understand the (co‐) occurrence of PTS and depression symptoms and key factors to address this issue, this study provides a unique perspective on (predictors of) symptom classes over time using a unique data resource which is a result of a growing collaborative effort that can inspire work in other topic areas and research communities in the mental health field. Our findings suggest that following a child's PTEs, internalizing reactions have individual heterogeneity with five distinct classes of individuals – including 40% of the children in the ‘low internalizing’ class and 37% of the children transitioning to one of the other lower level classes of symptoms across time. Belonging to a minority group in one's country, as well as interpersonal trauma exposure, were the most important predictors for staying in the ‘high internalizing’ class rather than transitioning to a class consisting of fewer internalizing symptoms across time.

Our five identified classes imply that different treatment approaches may be needed.

The large proportion of children classified in the stable ‘low internalizing’ class (40% at T2), alongside the high stability of this class (87% remained low), suggests that the primary clinical implication is not that all trauma‐exposed children require mental health intervention, but that effective triage is paramount. Children in the ‘high internalizing’ group could be best served by integrated, transdiagnostic, intensive treatments targeting PTS and depression simultaneously, whereas children in classes with mainly PTS symptoms may benefit from standard trauma‐focused interventions, such as TF‐CBT or EMDR.

For clinicians, recognizing and treating internalizing symptoms in trauma‐exposed children may be aided by the use of interventions designed to be transdiagnostic (i.e., target multiple disorders), consisting of a flexible modular treatment, with menus of treatment procedures from which therapists select the combination needed for each individual client (Weisz et al., 2023). Some of the challenges of the internalizing paradox described above may be addressed via transdiagnostic modular therapy which has shown good outcomes in comparison with usual clinical care (Chorpita et al., 2017; Chorpita & Weisz, 2009). Additionally, these classes can inform clinical triage, with ‘high internalizing’ children warranting immediate intensive intervention while moderate symptom classes receive standard care with monitoring.

Given the finding that exposure to interpersonal PTEs was associated with 3.57 times lower odds of transitioning out of the ‘high internalizing’ class to the ‘low internalizing’ class, compared to staying in the ‘high internalizing’ class, and that such trauma disrupts self‐identity and interpersonal relationships, clinicians should prioritize interventions that specifically address the relational and attachment disruptions caused by such events. Approaches incorporating relational, attachment‐based, or complex trauma frameworks may be particularly relevant for these children, focusing not only on symptom reduction but also on core deficits in affect regulation and trust (Steele & Steele, 2019).

In addition to our findings and inspired by the HiTOP model, future research should aim to produce transdiagnostic profiles that reflect individual differences in specific domains. This approach underscores that diagnostic comorbidity is a meaningful indicator rather than mere background noise, making it a valuable target for research and clinical efforts (Levin‐Aspenson & Greene, 2024).

Ethical considerations

This study used data from the Prospective studies of Acute Child Trauma and Recovery (PACT/R) Data Archive (Kassam‐Adams et al., 2020; Lenferink et al., 2020; Sadeh et al., 2024; Silverstein et al., 2022), part of the Child Trauma Data Archives (CTDA, www.childtraumadata.org). PACT/R is an international data repository of prospective studies of child traumatic stress. The archive includes deidentified, anonymized data from studies conducted with ethics approval. The Institutional Review Board (IRB) of The Children's Hospital of Philadelphia determined that reuse of data from the archive does not constitute human subjects research because all data are deidentified (RB# 09‐007361, 2/6/2018).

Key points.

What's known?

• Internalizing symptoms like posttraumatic stress (PTS) and depression often co‐occur in children after potential traumatic events (PTEs).

What's new?

• The study leveraged pooled harmonized participant‐level data from nine different studies, providing a robust and unique dataset that enhances the understanding of symptom trajectories over time.

• Latent Transition Analysis (LTA), rarely applied in trauma research with children, offers new insights into the long‐term co‐occurrence of PTS and depression symptoms.

What's relevant?

• Findings revealed significant heterogeneity in children's internalizing reactions, identifying five distinct symptom classes, with 40% of children in the ‘low internalizing’ class.

• Minority status and interpersonal trauma were key factors keeping children in the ‘high internalizing’ class, underscoring the need for targeted and transdiagnostic interventions.

Supporting information

Appendix S1. Background characteristics and data harmonization procedures.

Data availability statement

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