Latent classes of posttraumatic psychiatric comorbidity in the general population

Abstract

Some narrow patterns of posttraumatic psychiatric comorbidity are well-established (e.g., posttraumatic stress disorder and substance use). However, broad multi-diagnosis profiles of posttraumatic comorbidity are poorly characterized. The goal of the current study was to use latent class analysis (LCA) to identify profiles of posttraumatic psychopathology from 11 International Classification of Diseases (ICD-10) diagnostic categories (e.g., stress, substance, depressive, psychosis, personality). Danish national registries were used to identify 166,539 individuals (median age = 41 years, range = <1 to >100) who experienced a traumatic event between 1994–2016 and were diagnosed with one or more mental disorders within 5 years. Two through 14-class LCA solutions were evaluated. A 13-class solution (a) provided the best fit, with the Bayes and Akaike Information Criteria reaching a minimum, (b) was broadly consistent with prior LCA studies, and (c) included several novel classes reflecting differential patterns of posttraumatic psychopathology. Three classes were characterized by high comorbidity: broad high comorbidity (Mean # diagnoses = 4.3), depression with stress/substance use/personality/ neurotic disorders (Mean # diagnoses = 3.8), and substance use with personality/stress/psychotic disorders (Mean # diagnoses = 3.1). The other 10 classes were characterized by distinct patterns of mild comorbidity (Mean # diagnoses = 1.6–2.0) or negligible comorbidity (Mean # diagnoses = 1.0–1.4). Compared to the mild and negligible comorbidity classes, individuals in high comorbidity classes were younger, had lower income, and had more pre-event psychiatric disorders. Results suggest that several specific comorbidity patterns should be assessed when studying and treating posttraumatic psychopathology.

Introduction

Psychopathology is highly comorbid, with many individuals who receive one diagnosis experiencing several different types of mental disorders either concurrently or cumulatively over a lifetime (Conway et al., 2006; Kessler et al., 2005). Early studies of psychiatric comorbidity following exposure to a traumatic event, defined as an event involving threatened death or serious injury, largely focused on disorders comorbid with posttraumatic stress disorder (PTSD; Brady et al., 2000). These studies identified a limited number of common comorbidity patterns such as PTSD with depression and PTSD with substance use (Debell et al., 2014; Stander et al., 2014). As a result, much of the literature on posttraumatic comorbidity etiology, assessment, and treatment is focused on PTSD plus one other disorder. However, there also is evidence that a range of other psychiatric diagnoses and combinations of diagnoses may occur following a traumatic event (Bryant et al., 2010). Assessment and treatment strategies focused on one or two forms of psychopathology may be inadequate for patients suffering from broad posttraumatic psychopathology.

Over the past decade, latent class analysis (LCA) has been used to identify data-driven profiles that capture novel patterns of psychopathology and comorbidity following the experience of a traumatic event. Consistent with the focus on PTSD in the broader trauma literature, several studies used LCA to identify PTSD typologies, or subtypes, with distinct PTSD symptom profiles (e.g., Campbell et al., 2020; Horn et al., 2016). Other studies used LCA to identify symptom profile subtypes that capture the co-occurrence of PTSD symptoms with one additional type of psychopathology, including dissociative symptoms (Wolf et al., 2012a), grief symptoms (Nickerson et al., 2014), and borderline personality disorder symptoms (Cloitre et al., 2014). These studies typically have found that individuals in classes characterized by more severe PTSD/associated symptoms are more likely to experience a comorbid anxiety, mood, or substance use disorder.

LCA also has been applied to a range of different diagnostic and personality indicators to identify broader profiles of psychiatric comorbidity among individuals with PTSD (e.g., LCA of >3 disorders or traits). For example, three latent classes of lifetime posttraumatic psychopathology were identified in the National Comorbidity Survey – Replication (NCS-R), characterized by PTSD alone (no comorbidity), PTSD with depressive and anxiety disorders, and PTSD with depressive, anxiety, and substance use disorders (Galatzer-Levy et al., 2013). Latent classes of severe and pervasive comorbidity, characterized by many different anxiety, mood, and substance use disorders, also have been identified in samples not restricted to individuals with a PTSD diagnosis (Contractor et al., 2015; Hruska et al., 2014; Müller et al., 2014). Some LCA studies of posttraumatic personality traits and disorders have found evidence to support distinct classes of posttraumatic internalizing and externalizing psychopathology, with some individuals characterized by negative emotionality, introversion, decreased positive emotionality, and schizoid and avoidant personality; while others are characterized by negative emotionality, aggressiveness, disinhibition, and antisocial and narcissistic personality (Forbes et al., 2010; Wolf et al., 2012b). Other studies have not found distinct posttraumatic internalizing and externalizing comorbidity profiles (Jongedijk et al., 2019; Tsai et al., 2014).

The identification of data-driven diagnostic profiles, particularly those that reflect complex and pervasive patterns of comorbidity, can inform transdiagnostic models of psychopathology and transdiagnostic approaches to assessing, preventing, and treating comorbidity. The co-occurrence of disorders may suggest an underlying transdiagnostic vulnerability or mechanism of psychopathology, which could be targeted through prevention or treatment in the immediate aftermath of trauma. Moreover, knowledge of common patterns of comorbidity can assist treatment providers in knowing what disorders to assess in individuals experiencing posttraumatic psychopathology, and how interventions might need to be combined to treat them. However, research has been limited by studying a select number of disorders and potential patterns of comorbidity. For example, several LCA studies assessed internalizing and substance use disorders but not personality or psychotic disorders (Contractor et al., 2015; Forbes et al., 2015; Galatzer-Levy et al., 2013; Hruska et al., 2014). In addition, most studies have focused exclusively on comorbidity among individuals with PTSD. However, there is evidence of profiles of posttraumatic comorbidity that do not include PTSD (Cozza et al., 2019; Forbes et al., 2015). Accordingly, to fully understand the expression, prevention, and treatment of posttraumatic comorbidity, research is needed to assess the full spectrum of psychopathology in a population of individuals exposed to traumatic events.

Population-based electronic health registries provide an opportunity to identify novel patterns of psychopathology and comorbidity following a traumatic event. A recent study of the population of Denmark used medical and accident registry codes to identify approximately 1.4 million individuals exposed to a traumatic event from 1994 to 2016 and found elevated standardized rate ratios for all individual International and Classification of Diseases - 10^threvision (ICD-10) mental disorder categories under consideration (e.g., stress, substance use, psychotic, personality, depressive) compared to individuals who experienced a non-traumatic life stressor (Gradus et al., under review). The goal of the current study was to identify and characterize novel latent classes of posttraumatic psychiatric comorbidity in this cohort.

Material and Methods

Sample

Data were obtained from a cohort of approximately 1.4 million individuals who had one or more traumatic events recorded in Danish national healthcare and social registries between 1994 and 2016 (Gradus et al., under review). The base population of this cohort was the population of Denmark from 1994 to 2016 with no age restrictions (N = 7,420,888). The registries and cohort are summarized below; additional details surrounding cohort development are described elsewhere (Gradus et al., under review). We restricted the analytic sample to the 166,539 individuals who experienced any ICD-10 mental disorder in the five years following exposure to their traumatic event, regardless of PTSD status. This approach permitted identification of diagnostic profiles that did not include PTSD and omitted the group of “resilient” individuals who did not have recorded posttraumatic psychopathology.

Registries

A unique personal identifier is assigned to all residents of Denmark at birth or upon immigration (the Civil Person Register number). This identifier was used to merge healthcare and social data from the following registries (Schmidt et al., 2019).

The Danish Psychiatric Central Research Register (DPCRR) includes information on psychiatric care (Mors et al., 2011). The Danish National Patient Registry (DNPR) includes information on somatic and psychiatric care received in non-psychiatric treatment settings (Schmidt et al., 2015). The DPCRR and DNPR both record inpatient, outpatient, and emergency department treatment and admission dates. For each record, one primary ICD diagnosis and up to 19 secondary ICD diagnoses are coded. The DNPR also includes accident and injury codes based on the ICD-10 and the Nordic Classification of External Causes of Injury. Validation studies support the high quality of DPCRR and DNPR diagnoses (Mors et al., 2011; Schmidt et al., 2014).

Additional traumatic event and sociodemographic information was gathered from other registries. The Danish Medical Birth Register (MBR) records information on all live births and stillbirths by persons residing in and giving birth in Denmark, including second trimester miscarriages and spontaneous deliveries, preeclampsia, and other pregnancy-related traumas (Bliddal et al., 2018). The Cause of Death Registry (CDR) contains information about manner of death, including death by suicide (Helweg-Larsen, 2011). The Civil Registration System (CRS) stores information about sex, age, and marital status, and can be used to link information across families (Schmidt et al., 2014). The Income Statistics Register (ISR) tracks income data (Petersson et al., 2011). Numerous studies document the high validity of information recorded in these registries (Klemmensen et al., 2007; Mors et al., 2011; Sneider et al., 2015; Tøllefsen et al., 2015; Vest-Hansen et al., 2013).

Traumatic events

Incident (first-recorded) traumatic events were identified using ICD-10 and Classification of External Causes of Injury codes recorded in the DNPR, DPCRR, MBR, CRS, and CDR (Supplemental Table 1). Code definitions were reviewed to identify all codes that were consistently used over the study period and could plausibly represent a traumatic event, defined as exposure to threatened death, serious injury, or unexpected death (codes that could represent DSM Criterion A events).

Many similar and overlapping codes were identified and thus were categorized into broader traumatic event types. In total, eight traumatic event types were operationalized. Exposure to a fire or explosion was defined based on codes representing exposure to fire, or treatment for burns, as these codes were frequently recorded on the same day as the incident. Traumatic brain injuries were identified and distinguished from other injury codes because of their much higher prevalence compared to other injury types. Codes representing assault were distinguished into two groups based on whether or not the assault involved a weapon (which is rare in Denmark). Pregnancy-related traumas were identified from codes representing complications of labor and delivery (e.g., stillbirth). Accidents were identified based on several different injury codes (e.g., falls), although this group predominately included transportation accidents. Suicidal death of a parent, child, or spouse was operationalized as a form of unexpected death and identified based on codes for suicide-related causes of death. Finally, we identified codes representing exposure to a toxic substance or other medical complications due to external causes as a broad category representing other health-related trauma. The toxic substance/medical complications group was heterogeneous (see Supplemental Table 1 for the most common codes comprising this group) and included traumatic health events due to medications (e.g., allergic reactions), illicit drugs (e.g., overdosing or underdosing), and medical procedures (e.g., infections). To increase the likelihood of operationalizing true traumatic experiences (e.g., involving threat of serious injury or death), cohort entry required two or more days of hospitalization following transportation accidents and toxic substance exposure/medical complications (i.e., longer than a standard overnight observation period) and three or more days of hospitalization following pregnancy-related events (i.e., longer than the standard two days following delivery). Individuals with multiple incident events on the same day were coded as “multiple traumas.”

Latent class indicators

Posttraumatic psychopathology was identified using mental disorder diagnoses recorded in the DPCRR or DNPR between the date of the incident (first-recorded) trauma and the five years that followed, until migration or death, or the end of the study period on December 31, 2016, whichever came first. A five-year follow-up was selected because many individuals delay seeking treatment for several years following trauma exposure (Wang et al., 2005). Thus, trauma-related psychiatric diagnoses may not be recorded in the DPCRR or DNPR until years after a traumatic event is recorded in one of the other registries.

Latent class indicators were selected from the 11 broad ICD-10 psychiatric disorder groups with three modifications. First, we omitted the mental retardation (F70-F79) and unspecified mental disorders categories (F99) due to having a low prevalence (<1% of the total cohort had these diagnoses) and limited conceptual/substantive value (e.g., unclear interpretation of “unspecified mental disorder”). Second, we distinguished PTSD and other stress disorders from neurotic and somatoform disorders because PTSD is the most widely studied type of posttraumatic psychopathology. Third, we distinguished bipolar and unipolar mood disorders because unipolar depression also has been widely studied in the posttraumatic psychopathology literature (Stander et al., 2014). Accordingly, the 11 indicators of psychopathology were: organic mental disorders (e.g., dementia); mental and behavioral disorders due to psychoactive substance use; schizophrenia, schizotypal and delusional disorders; manic episode or bipolar disorders; depressive disorders; neurotic and somatoform disorders; stress disorders; behavioral syndromes associated with physiological disturbances and physical factors (e.g., eating and sleep disorders); disorders of adult personality and behavior; disorders of psychological development (e.g., learning disorders), and behavioral and emotional disorders with onset usually occurring in childhood and adolescence (e.g., conduct disorder, tic disorder).

Statistical Analysis

We examined descriptive characteristics of the cohort including the proportions in demographic, diagnostic, and traumatic event categories, as well as categories related to number of diagnoses.LCA was used to identify data-driven profiles of posttraumatic psychopathology from the 11 ICD categories based on diagnoses assigned during healthcare visits over the five-year post-event period. Competing solutions were assessed based on interpretability and model fit. A solution was determined to be interpretable if the classes reflected distinct (non-overlapping) patterns of psychopathology. In other words, conditional probabilities of the 11 diagnostic categories (i.e., the likelihood an individual in the class has a diagnosis) should be different across classes and not reflect the same patterns of psychopathology at varying probability levels (e.g., two classes differentiated by “moderate” vs. “high” probabilities of all 11 disorders). Fit statistics included the log-likelihood, Akaike Information Criteria (AIC), unadjusted and sample-size adjusted Bayes Information Criteria (BIC), and entropy. Smaller values of log-likelihood, AIC, and BIC indicate better fit to the data. Higher entropy values represent better between-class distinction, though entropy should not be used to guide model selection (Nylund-Gibson and Choi, 2018). BIC was prioritized given evidence it is one of the most robust parameters of LCA model fit (Nylund et al., 2007).

Psychopathology classes were further evaluated by assigning individuals to the class with the highest predicted likelihood of membership and examining differences in number of diagnoses, sociodemographic characteristics at the time of the traumatic event, and type of incident trauma. The mean number of posttraumatic diagnoses among individuals in each class was calculated to confirm which classes were characterized by high comorbidity (M # diagnoses, range = 1 to 11). Analyses were conducted in SAS version 9.4. This work was approved by the Institutional Review Board at Boston University and the Danish Data Protection Agency (2015–57-0002). All procedures contributing to this work complied with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.

Results

Sample characteristics

The sample was largely female (58%) and single (47%), with a median age of 41 years at 1 the time of incident trauma occurrence (range = <1 to >100; Supplemental Table 2). Exposure to toxic substance and other medical complications (50%), traumatic brain injury (26%), and pregnancy-related events (14%) were the most common traumatic experiences. Substance use (35%), depressive (26%), and stress disorders (21%) were the most common psychiatric diagnoses. One-third (35%) of individuals had multiple disorders across the 11 diagnostic categories within five years of the traumatic event (exactly one diagnostic category = 65%, exactly two = 22%, three or more = 12%).

Latent classes

Log-likelihood, BIC, and AIC reached a minimum in the 13-class solution (Table 1). The 11- and 12-class solutions also were interpreted to ensure that additional classes identified in the 13-class solution were distinct. The 11-, 12-, and 13-class solutions differed in the number of “high comorbidity” classes, characterized by having three or more categories of disorders, on average. Whereas three high comorbidity classes were identified in the 13-class solution, one high comorbidity class was identified in the 11-class solution and two were identified in the in the 12-class solution (Supplemental Table 3). We selected the 13-class solution because it was characterized by three unique profiles of high comorbidity (in addition to providing the best model fit).

Table 1.

Latent class analysis goodness of fit statistics

Model	Log-likelihood	BIC	Adjusted BIC	AIC	Entropy
2 classes	−652,223	106,365	106,292	106,134	0.93
3 classes	−646,653	95,369	95,258	95,018	0.84
4 classes	−631,638	65,482	65,333	65,011	0.84
5 classes	−626,094	54,539	54,351	53,948	0.82
6 classes	−623,067	48,629	48,404	47,918	0.79
7 classes	−619,302	41,243	40,979	40,411	0.80
8 classes	−617,373	37,530	37,228	36,578	0.84
9 classes	−609,237	21,402	21,062	20,330	0.96
10 classes	−604,239	11,551	11,172	10,358	0.97
11 classes	−602,910	9,036	8,619	7,723	0.91
12 classes	−602,542	8,445	7,990	7,012	0.91
13 classes	−602,106	7,718	7,225	6,164	0.88
14 classes	−602,121	7,891	7,360	6,217	0.89

Abbreviations: AIC= Akaike Information Criteria; BIC= Bayes Information Criteria

In the 13-class solution, a “broad high comorbidity” class was characterized predominately by stress disorders (conditional probability = 0.56) and behavioral/emotional disorders of childhood (probability = 0.48), together with six other disorder categories (e.g., personality, schizophrenia) also having conditional probabilities between 0.26 and 0.40 (1.8% of the sample, M # of posttraumatic diagnoses = 4.3, Figure 1). Thus, in this class, the prevalences of stress disorders and behavioral/emotional disorders of childhood were 56% and 48%, respectively, while the prevalence of six other disorders was between 26% and 40%. A “depression with high comorbidity” class (4.6%, M diagnoses = 3.8) was characterized predominately by depressive disorders (probability = 1.0) but also stress disorders (probability = 0.47), substance use disorders (probability = 0.45), personality disorders (probability = 0.43), and neurotic/somatoform disorders (probability = 0.28). A “substance use with high comorbidity” class (5.0%, M diagnoses = 3.1) was characterized predominately by substance use disorders (probability = 1.0) but also personality disorders (probability = 0.49), stress disorders (probability = 0.32), and schizophrenia/psychotic disorders (probability = 0.26).

Figure 1.

Conditional probabilities of diagnostic categories for the three high comorbidity classes

Three other classes were characterized by one diagnostic category with a conditional probability of 1.0 plus one or two other probabilities ~0.25, i.e., classes with 100% of individuals experiencing a particular disorder plus roughly 25% experiencing one or more other disorders (“mild comorbidity” classes, Figure 2). These classes captured personality disorders with depressive and/or stress disorders (7.2%, M diagnoses = 2.0), bipolar disorders with depressive and/or substance use disorders (3.8%, M diagnoses = 1.9), and schizophrenia/psychotic disorders with substance use disorders (5.6%, M diagnoses = 1.6).

Figure 2.

Conditional probabilities of diagnostic categories for the three mild comorbidity classes

Remaining classes were characterized by one diagnostic category with a conditional probability of 1.0 and very low probabilities of other disorders (Figure 3). These “negligible comorbidity” classes were predominately, but not exclusively, characterized by a single disorder, including “stress only”, “depression only”, “substance use only”, “organic only”, “physiological only”, and “neurotic/somatoform only” (2.2% to 19% of the sample, M diagnoses = 1.0 to 1.4). In addition, a “childhood disorders” class, characterized by disorders of psychological development and/or behavioral/emotional disorders of childhood, was identified (4.8%, M diagnoses = 1.3).

Figure 3.

Conditional probabilities of diagnostic categories for the seven negligible comorbidity classes

Class characteristics

We prioritized evaluating covariate distributions across the three high comorbidity classes and three most prevalent negligible comorbidity classes (“substance use only”, “depression only”, “stress only”) to identify differences in individuals who develop different patterns of high comorbidity, as well as to compare individuals with many disorders to individuals with very few disorders following a traumatic event. Supplemental Tables 4-7 present covariate distributions across the other seven classes.

Socio-demographic.

Individuals in the “broad high comorbidity” class were younger at the time of the traumatic event (median age = 17.0) versus those in the other classes (median ages = 29.7 to 51.6; Table 2). One third of the “substance use only” class was female (34%), compared to 47% of “substance use with high comorbidity” class, and roughly two-thirds of the “depression only”, “depression with high comorbidity”, “stress only”, and “broad high comorbidity” classes (63–69%). Over half of the “broad high comorbidity” class was in the lowest income quartile (56%), versus 23–38% in the other classes.

Table 2.

Diagnosis, sociodemographic, and trauma characteristics of selected high and low comorbidity classes¹

	Substance use with high comorbidity (personality/stress/psychosis) (N = 6,298)		Substance use only (N = 31,610)		Depression with high comorbidity (stress/substance/personality/neurotic) (N = 6,304)		Depression only (N = 22,597)		Broad high comorbidity (N = 1,279)		Stress only (N = 23,226)
	N	%	N	%	N	%	N	%	N	%	N	%
Sociodemographic
Median Age2 (Q1, Q3)	30.2 (20.8, 41.7)		46.2 (31.7, 58.3)		38.8 (26.3, 51.0)		51.6 (30.5, 71.7)		17.0 (14.6, 22.0)		29.7 (20.0, 43.3)
Gender (female)	2,982	47	10,640	34	3,979	63	15,591	69	888	69	15,776	68
Income quartile
<Q1	2,375	38	7,177	23	1,687	27	5,171	23	715	56	7,124	31
Q1–Q2	2,111	34	13,407	42	2,083	33	8,110	36	154	12	5,494	24
Q2–Q3	1,091	17	6,210	20	1,450	23	5,260	23	56	4.4	4,902	21
Q3<=	400	6.4	3,930	12	907	14	3,570	16	28	2.2	3,385	15
Child	121	1.9	410	1.3	58	0.92	237	1.0	224	18	1530	6.6
Missing	200	3.2	476	1.5	119	1.9	249	1.1	102	8.0	791	3.4
Marital Status
Married/partnership	878	14	7,613	24	1,919	30	8,443	37	58	4.5	6,979	30
Single	4,332	69	14,149	45	2,879	46	7,263	32	1,195	93	12,902	56
Divorced	986	16	7,898	25	1,199	19	3,194	14	20	1.6	2,444	11
Widowed	85	1.3	1,839	5.8	292	4.6	3,649	16	<5	--	757	3.3
Unknown	17	0.27	111	0.35	15	0.24	48	0.21	<10	--	144	0.62

Abbreviations: Q=quartile.

¹In order to evaluate the distribution of covariates across classes, individuals were assigned to the single class with the highest predicted likelihood of membership. Class sizes reported here are based on most-likely class membership. These values are slightly different than the model-based class size estimates reported in text because model-based estimates account for uncertainty in class membership (i.e., an individual can be a partial member of multiple classes).

²Median age is reported rather than mean age because age did not have a normal distribution

Traumaticeventtypes.

Several traumatic events were more common (1.5-fold or greater) in certain classes than others (Table 3). Traumatic brain injury was more common in the “substance use with high comorbidity” and “substance use only” classes (25–38%) than in the other classes (17–18%). Pregnancy-related traumatic events were more common in the “depression only” and “stress only” classes (20–23%) than the other classes (4.8–9.8%). Suicidal death of a family member was more common in the “stress only” and “broad high comorbidity” (1.1–1.2%) classes than the other classes (0.29–0.49%).

Table 3.

Trauma and diagnostic characteristics of selected high and low comorbidity classes¹

	Substance use with high comorbidity (personality/stress/psychosis) (N = 6,298)		Substance use only (N = 31,610)		Depression with high comorbidity (stress/substance/personality/neurotic) (N = 6,304)		Depression only (N = 22,597)		Broad high comorbidity (N = 1,279)		Stress only (N = 23,226)
	N	%	N	%	N	%	N	%	N	%	N	%
Traumatic event type
Fire or explosion	290	4.6	1,740	5.5	210	3.3	680	3.0	66	5.2	1,013	4.4
Transportation accident	79	1.3	621	2.0	73	1.2	378	1.7	21	1.6	308	1.3
Exposure to toxic substance	3,743	59	13,976	44	4,122	65	12,506	55	825	65	11,709	50
Traumatic brain injury	1,578	25	11,960	38	1,080	17	4,034	18	229	18	3,884	17
Physical assault	223	3.5	924	2.9	135	2.1	241	1.1	42	3.3	568	2.4
Assault with a weapon	<5	--	14	0.044	7	0.11	7	0.031	<5	--	17	0.073
Pregnancy related trauma	302	4.8	1,962	6.2	619	9.8	4,499	20	65	5.1	5,289	23
Suicidal death of family member	18	0.29	109	0.34	25	0.40	110	0.49	14	1.1	274	1.2
Multiple traumas	61	0.97	304	0.96	33	0.52	142	0.63	15	1.2	164	0.71
Pre-event psychiatric diagnoses
Mean # diagnoses (SD)	1.4 (1.5)		0.6 (0.8)		1.3 (1.5)		0.6 (0.9)		1.4 (1.6)		0.4 (0.8)
Diagnostic categories
Organic	157	2.5	542	1.7	152	2.4	590	2.6	25	2.0	207	0.89
Substance use	2,593	41	11,572	37	1,817	29	2,629	12	126	9.9	1,831	7.9
Schizophrenia/psychotic	914	15	384	1.2	336	5.3	284	1.3	125	9.8	363	1.6
Manic episode/bipolar	322	5.1	244	0.77	385	6.1	474	2.1	35	2.7	173	0.74
Depressive	768	12	1,429	4.5	1,889	30	4,763	21	210	16	1,497	6.4
Neurotic/somatoform	616	9.8	756	2.4	797	13	1,171	5.2	161	13	823	3.5
Stress	1,112	18	1,399	4.4	1,318	21	1,529	6.8	285	22	2,599	11
Physiological3	199	3.2	158	0.50	231	3.7	257	1.1	106	8.3	282	1.2
Personality	1,635	26	881	2.8	1,105	18	766	3.4	173	14	749	3.2
Psychological development	117	1.9	142	0.45	40	0.63	93	0.41	270	21	177	0.76
Behavioural/emotional of childhood	564	9.0	478	1.5	168	2.7	239	1.1	333	26	482	2.1
Post-event psychiatric diagnoses
Mean # diagnoses (SD)	3.1 (0.9)		1.0 (0.1)		3.8 (0.9)		1.4 (0.6)		4.3 (1.1)		1.4 (0.6)

Abbreviations: SD=standard deviation.

¹In order to evaluate the distribution of covariates across classes, individuals were assigned to the single class with the highest predicted likelihood of membership. Class sizes reported here are based on most-likely class membership. These values are slightly different than the model-based class size estimates reported in text because model-based estimates account for uncertainty in class membership (i.e., an individual can be a partial member of multiple classes).

Diagnoses.

Individuals in the three high comorbidity classes had more pre-event diagnoses (Ms = 1.3–1.4) than individuals in the three “negligible comorbidity” classes (Ms = 0.4–0.6). Some individuals in each class had the same diagnoses assigned prior to the traumatic event. For example, 37–41% of the “substance use only” and “substance use with comorbidity” classes had a substance use disorder prior to the traumatic event compared to 7.9–12% of the “depression only” “stress only” and “broad high comorbidity” classes. Likewise, 21–30% of individuals in the “depression only” and “depression with high comorbidity” classes had a depressive disorder prior to the traumatic event compared to only 4.5–6.4% of “substance use only” and “stress only” classes. Across all classes, individuals had more post-event diagnoses than pre-event diagnoses.

All three of the high comorbidity classes were characterized by an elevated conditional probability of personality disorders. To explore heterogeneity across classes, we evaluated the frequency of specific posttraumatic personality disorders in each class (Supplemental Table 6). Borderline (19–31%) and unspecific (18–28%) personality disorders were somewhat common in all three high comorbidity classes. Antisocial personality disorder was more common among individuals in the “substance use with high comorbidity’ class (10%) than the other two high comorbidity classes (1.7–2%). Other personality disorders were uncommon (<5%) across the high comorbidity classes.

Discussion

Our results are consistent with prior LCA studies that found some classes characterized by severe and pervasive comorbidity, and other classes characterized by the experience of a single disorder and negligible comorbidity. Specifically, the “substance use only”, “stress only”, “substance use with high comorbidity”, and “broad high comorbidity” classes are similar to previously identified classes of alcohol use disorders with low versus high comorbidity (Forbes et al., 2015) and PTSD with low versus high comorbidity (Galatzer-Levy et al., 2013; Hruska et al., 2014). Our “broad high comorbidity” and “depression with high comorbidity” classes are also generally similar to the two high comorbidity classes (of seven total classes) identified in the NCS-R, which included individuals who had not experienced a traumatic event (Kessler et al., 2005).

Consistent with the distinction between externalizing and internalizing posttraumatic comorbidity, antisocial personality disorder was much more common among individuals in the “substance use with high comorbidity” class than the “broad high comorbidity” and “depression with high comorbidity” classes. However, the “substance use with comorbidity” class also was characterized by an elevated conditional probability of stress disorders (internalizing disorders), while the “depression with high comorbidity” and “broad high comorbidity” classes were characterized by an elevated probability of substance use disorders (externalizing disorders). In other words, the highest comorbidity classes were differentially characterized by internalizing or externalizing disorders, but many individuals in these classes had both internalizing and externalizing disorders. Hierarchical models of psychopathology suggest that a transdiagnostic general psychopathology factor (vulnerability) may account for broad comorbidity across internalizing and externalizing disorders (Kotov et al., 2017). Negative emotionality is one such transdiagnostic construct associated with both internalizing and externalizing psychopathology (Tackett et al., 2013; Wolf et al., 2012b). Alternately, the internalizing-externalizing distinction may have been less pronounced in the current study because broad diagnostic indicators were used (see also Tsai et al., 2014). Most evidence of distinct classes of posttraumatic internalizing and externalizing psychopathology is from studies of symptom or trait-level measures (e.g., introversion, disinhibition) rather than diagnoses (Forbes et al., 2010; Wolf et al., 2012b).

Prior LCA studies of posttraumatic psychopathology typically have identified only three (Galatzer-Levy et al., 2013; Müller et al., 2014) or four (Forbes et al., 2015; Hruska et al., 2014) latent classes, regardless of whether the sample was restricted to individuals with PTSD. Three interrelated factors may have contributed to identifying a larger number of classes in the current study. First, our sample was over 50-fold larger than earlier LCA studies of posttraumatic diagnoses, whose sample sizes range from a few hundred to a few thousand (<3,000). Some of the classes identified here may be too uncommon to be detected by LCA in smaller studies. Second, about two-thirds of the sample had exactly one diagnosis across the 11 categories. This likely contributed to the identification of seven distinct classes characterized by negligible comorbidity. Third, we evaluated a broader scope of diagnostic categories than earlier studies, without age restrictions, and therefore had a larger number of potential unique disorder combinations. Replication of the current classes in other populations is needed, using registry data or survey/interview assessments that permit identification of the full spectrum of potential posttraumatic psychopathology. In addition, research is needed to determine the utility of distinguishing between the latent classes, particularly the three high comorbidity classes. Although these classes were somewhat rare, this does not preclude clinical utility as they may differentially predict outcomes such as treatment prognosis, role impairment, or suicidality.

LCA findings also can inform the assessment of posttraumatic psychopathology. Our results support prioritizing the assessment of substance use, depression, and PTSD when treating or studying posttraumatic psychopathology, as the three largest classes were characterized by these disorders with negligible comorbidity. Moreover, these disorders also were common, to varying degrees, in the three high comorbidity classes. At the same time, the distinct patterns of high comorbidity suggest that broader disorder combinations might be assessed rather than the typical triad of PTSD, depression, and substance use. For instance, identification of the “substance use with high comorbidity” class indicates potential utility in assessing comorbid personality disorders (borderline and antisocial), psychosis, and stress disorders among individuals with a primary substance use disorder. The “depression with high comorbidity” class suggests individuals presenting with primary depressive disorders should also be assessed for stress, substance use, personality (borderline), and neurotic disorders. Likewise, the “broad high comorbidity” class indicates the potential need to assess the gamut of psychopathology when individuals present with stress and childhood disorders. Collectively, our findings are consistent with evidence that traumatic experiences are associated with the subsequent development of a broad range of psychopathology (Gradus et al., under review; McGrath et al., 2017; Paris, 1998). Explication of the three distinct high comorbidity classes, which collectively comprised 11% of all individuals who experienced posttraumatic psychopathology, underscores a potential need to assess broader but targeted patterns of posttraumatic psychopathology. Along these lines, research on the risk, prediction, and course of posttraumatic psychopathology should consider assessing multidimensional and transdiagnostic posttraumatic outcomes in addition to, or in place of, individual disorders of interest.

Sociodemographic factors varied across the classes in ways consistent with prior posttraumatic psychopathology LCA studies (e.g., in the NCS-R; (Galatzer-Levy et al., 2013; Kessler et al., 2005). For instance, most members of the substance use classes were male, while most members of the depressive and stress disorder classes were female. In addition, the three high comorbidity classes were composed of persons who were younger and had lower incomes compared with their respective “single disorder” classes. The median age at trauma exposure for members of the “broad high comorbidity” class was only 17 years, suggesting that a subgroup of individuals exposed to trauma in childhood is diagnosed with severe/pervasive psychopathology (including childhood disorders) by late adolescence/early adulthood. Research is needed to identify risk and protective factors for “broad high comorbidity” following trauma exposure.

Traumatic event types also varied across classes, and three findings warrant additional comment. First, pregnancy-related traumas were more common in “negligible comorbidity” classes of internalizing psychopathology. These findings suggest that pregnancy-related trauma may be associated with circumscribed psychopathology such as postpartum depression, rather than diffuse or highly comorbid psychopathology. Second, traumatic brain injury was most common in the “substance use with high comorbidity” and “substance use only” classes. The relationship between traumatic brain injury and substance use is likely bidirectional (West, 2011), possibly due to similar underlying mechanisms such as impulsivity and disinhibition (Olson-Madden et al., 2012). Third, suicidal death of a family member was most common in the “broad high comorbidity” class as well as the negligible comorbidity “stress only” class. Other risk and protective factors, such as social support, likely moderate the effects of family member suicide death on subsequent psychopathology and comorbidity (Callahan, 2000).

Several limitations associated with the use of registry data should be noted. For example, it was not possible to include individuals with traumatic events or psychopathology unrecorded in the registries (e.g., individuals who did not receive treatment for trauma/injury or mental disorders). Along these lines, we did not have access to sexual assault data. Information pertaining to the remission of disorders was also unavailable. The classes thus reflect cumulative posttraumatic comorbidity rather than definitively co-occurring disorders. There also are limitations associated with studying diagnostic categories, which do not capture heterogeneity in symptom expression within a diagnosis, or symptom severity. Leading alternate models of psychopathology emphasize studying transdiagnostic dimensions rather than diagnoses (e.g., worry, negative emotionality, disinhibition, anhedonia; Kotov et al., 2017). It is also unclear how our results generalize to other countries and cultures, particularly the United States’ population, where diagnoses are assigned using the Diagnostic and Statistical Manual of Mental Disorders. Similarities with classes found in representative United States samples somewhat dampen this concern (Galatzer-Levy et al., 2013; Kessler et al., 2005). Relatedly, although our goal was to identify patterns of posttraumatic psychopathology in the full population of Denmark, studies are needed to evaluate if latent classes vary across populations or subpopulations defined by culture, race/ethnicity, age, sex, and other factors (e.g., testing measurement invariance).

Our prior study found elevated rates of all mental disorder categories in this trauma cohort compared to individuals who experienced a family member’s non-suicide death (Gradus et al., under review). In the current study, all classes were characterized by having more diagnoses in the five years following the traumatic event than in the entire period before. Although these findings suggest that our registry-defined traumatic events are associated with subsequent psychopathology, firm causal conclusions cannot be made. Some individuals had at least one similar diagnosis prior to the traumatic event. Further, the number of pre-event diagnoses likely was underestimated because information about psychiatric diagnoses before 1994 was unavailable (during the ICD-8 classification period). At the same time, having access to any longitudinal data on pre-event diagnoses is a strength; the literature has relied heavily on retrospective self-reports of both traumatic events and psychopathology. The incidence or recurrence of psychopathology following a traumatic event is influenced by a combination of factors, including aspects of the traumatic event, pre-event psychopathology and functioning, and a number of other factors occurring during and after the traumatic event, which were not measured in this study (Sayed et al., 2015). Psychopathology also may increase risk of trauma (e.g., bidirectionality between substance use and traumatic brain injury). Causal inference surrounding posttraumatic outcomes is complex, and additional research using robust methods is needed (e.g., targeted maximum likelihood estimation via super learning, van der Laan & Gruber, 2010).

Within the context of these limitations, the current study identified novel profiles of posttraumatic psychopathology. Understandably, many trauma-focused clinicians and researchers prioritize the assessment and study of PTSD, substance use, and depression as the prototypical expressions of posttraumatic psychopathology. Nevertheless, it is becoming increasingly clear the field needs to pivot towards assessing and treating more nuanced multidimensional profiles of posttraumatic outcomes, in order to better understand the full expression, risk, and prognosis of posttraumatic psychopathology and comorbidity.