What can we learn about polytrauma typologies by comparing population-representative to trauma-exposed samples: A Nepali example

Abstract

Background:

Potentially traumatic events (PTEs) are common and associated with detrimental outcomes over the life-course. Previous studies exploring the causes and consequences of PTE-exposure profiles are often from high-income settings and fail to explore the implications of sample selection (i.e., population-representative versus PTE-restricted).

Methods:

Among individuals in the Nepal Chitwan Valley Family Study, latent class analyses (LCA) were performed on 11 self-reported PTEs collected by the Nepali version of the World Mental Health Consortium’s Composite International Diagnostic Interview 3.0 from 2016 to 2018, in a population-representative sample (N = 10,714), including a PTE-restricted subsample (N = 9183). Multinomial logistic regressions explored relationships between sociodemographic factors and class membership. Logistic regressions assessed relationships between class membership and psychiatric outcomes.

Results:

On average, individuals were exposed to 2 PTEs in their lifetime. A five-class solution showed optimal fit for both samples; however, specific classes were distinct. No single sociodemographic factor was universally associated with PTE class membership in the population-representative sample; while several factors (e.g., age, age at incident PTE, education, marital status, and migration) were consistently associated with class membership in the PTE-subsample. PTE class membership differentiated psychiatric outcomes in the population-representative sample more than the PTE-subsample.

Limitations:

Primary limitations are related to the generalizability to high-income settings, debate on LCA model fit statistic usage for final class selection, and cross-sectional nature of data collection.

Conclusions:

Although population-representative samples provide information applicable to large-scale, population-based programming and policy, PTE-subsample analyses may provide additional nuance in PTE profiles and their consequences, important for specialized prevention efforts.

1. Introduction

Potentially traumatic events (PTEs) – defined as exposure to death (actual or threatened), serious injury or harm, including sexual violence (American Psychological Association, 2013) – represent a nearly ubiquitous experience (Benjet et al., 2016). Seventy percent of World Mental Health Survey respondents from 24 countries reported exposure to at least one PTE in their lifetime (Kessler et al., 2017). PTE exposure increases the risk of developing deleterious health outcomes, such as posttraumatic stress disorder (PTSD) (Dursa et al., 2014; Lowe et al., 2014; McCutcheon et al., 2010; Wolf et al., 2015), anxiety (Benjet et al., 2020; Gill et al., 2014; McCutcheon et al., 2010; Schweininger et al., 2015), major depressive disorder (MDD) (Giordano et al., 2014; McCutcheon et al., 2010), binge drinking and substance use disorders (Giordano et al., 2014; Ziobrowski et al., 2020), psychosis (Duhig et al., 2015), suicidal ideation (Jankovic et al., 2013), borderline personality disorder (Westphal et al., 2013), sexually transmitted infections (Walsh et al., 2012), and other chronic diseases (Schweininger et al., 2015; Sumner et al., 2015).

Prevalence estimates of PTE exposure and its deleterious consequences vary substantially across sociodemographic and contextual factors. The literature has predominantly focused on high-income settings (O’Donnell et al., 2017), although PTE exposure burden is generally higher in low- and middle-income countries (LMICs) (Atwoli et al., 2017; Kessler et al., 2017). For instance, women in LMICs face a greater burden of exposure to certain PTEs (e.g., intimate partner violence) than women residing in high-income countries (Esie et al., 2019). Experiencing a natural disaster in a LMIC further increases the risk of physical injury, loss of shelter, and death, due to factors such as delayed emergency response, poor infrastructure, and displacement (Gill et al., 2014; Pollack et al., 2016). Additional research in LMIC contexts is essential to both better understand the universality of previously identified PTE profiles and to better understand the unique experiences of those in LMIC settings.

Beyond a simple focus on individual PTEs in high-burden settings, examining the heterogeneity of PTE exposures profiles can yield a deeper understanding of overlapping traumas (O’Donnell et al., 2017), providing insight into patterns of specific health outcomes in different subgroups, that can be leveraged to improve prevention and treatment efforts. PTE profiles may reveal how exposures may cluster in ways that have particular mental health consequences above and beyond individual PTEs, and tell us which PTEs require the greatest attention. Exposure to multiple PTEs (i.e., polytraumatization) can increase the risk of poor outcomes more than exposure to any single event (Sullivan et al., 2017). The potentially compounding short- and long-term effects of polytraumatization also differ by type of PTE, duration of exposure, age at exposure, and concurrent and historic trauma exposures (Esie et al., 2019; O’Donnell et al., 2017; Tang et al., 2018; Ziobrowski et al., 2020). Given higher burden of polytraumatization in LMIC, additional research into understanding PTE profiles is an important public health question.

Latent class analysis (LCA) is well suited to explore polytraumatization patterns, and has been used extensively to do so (Golder et al., 2012; O’Donnell et al., 2017; Sullivan et al., 2017; Young-Wolff et al., 2013; Ziobrowski et al., 2020). LCA-based PTE models are able to organize heterogeneous data into more homogenous groups by identifying similar clusters of PTE endorsements and assigning class membership accordingly. Existing LCA literature on PTE models, predominantly in high-income settings, has typically identified three to five classes of trauma exposures, across a variety of populations (Contractor et al., 2018a; O’Donnell et al., 2017), most often within trauma-exposed groups (e.g., war, intimate partner violence, natural disaster survivors, community violence, etc.) (Kelber et al., 2019; Pollack et al., 2016; Sullivan et al., 2017). Identified classes in these groups have been related to specific PTE categories such as grief, child maltreatment, or witnessing physical violence, in addition to including PTE severity dimensions ranging from no/low reported PTE to high PTE burden (O’Donnell et al., 2017; Young-Wolff et al., 2013; Ziobrowski et al., 2020).

As a data-driven analytic method, empirically identified classes of PTE exposure are sensitive to design features of the primary datasets, such as restrictions to a specific PTE-exposed population (e.g. car accident). Identified classes may differ in PTE-exposed sub-populations compared with a broader population not selected for these exposures. It could be that in a general population-based sample, the difference between trauma and no-trauma individuals is the major distinction (i.e., source of variation), with the resulting profiles mostly capturing this difference. With the trauma-only sample, it is possible we might uncover more nuanced profiles. However, this comparison has not been directly made, though would inform downstream research and knowledge. LCA studies that critically examine PTE patterns in these samples can provide nuanced insights into individuals who meet these conditions.

Leveraging a large-scale, multiple cohort panel study in Nepal, a low-income country with high PTE occurrence (Benjet et al., 2020; Williams et al., 2012), this study compares PTE profiles from a population-representative sample to those in the PTE-restricted subsample. To highlight similarities and differences in these approaches in the minimal extant literature, for the first time we explore the differential associations of resulting PTE classes with sociodemographic factors and psychiatric sequelae, thus both contributing to the population-based and PTE-restricted literature simultaneously.

2. Methods

2.1. Sample

The Chitwan Valley Family Study (CVFS) (University of Michigan, 2019) is a panel study originally designed to replicate demographic U.S.-based studies, includes more than two decades of data, beginning in 1995, with a cohort from 151 neighborhoods that varied by access to roads, schools, health services, employers, and other social services in Western Chitwan, Nepal (Barber et al., 1997). The CVFS has high survey response rates with minimal missing data and reliable measures (Axinn et al., 2012). In 2016–2018, the Nepali version of the World Mental Health consortium’s Composite International Diagnostic Interview (WMH-CIDI) 3.0 was administered to all study participants ages 15–59, with a response rate of 93 % (N = 10,714, detailed methods previously published) (Scott et al., 2020). This analysis includes all respondents from this WMH-CIDI sample (population-representative sample) and a subsample of those exposed to any PTE below (N = 9183; PTE-restricted subsample). Interviews were conducted in person and all study participants provided written or verbal informed consent. The study was approved by both the University of Michigan Institutional Review Board and the Nepal Health Research Council.

2.2. Potentially traumatic events

The WMH-CIDI included 29 items related to PTEs. This analysis examined the 11 lifetime PTEs that were endorsed by at least 2 % of all participants: life-threatening car accident, any other life-threatening accident other than automobile accidents, natural disaster, life-threatening illness, beaten by parents/caregivers as a child, beaten by spouse/partner, beaten by anyone else, someone close to you died unexpectedly, son or daughter had life threatening illness, witness serious physical fights as a child, or witness someone badly injured, killed, or a dead body. Among the PTE-restricted subsample, average PTE counts and age at incident PTE were drawn from these items.

2.3. Mental health outcomes

The WMH-CIDI survey assessed diagnostic criteria for lifetime and 12-month mental health disorders at the time of interview, with documented strong validity with the Structured Clinical Interview for Diagnostic and Statistical Manual of Mental Health Disorders (DSM-IV) Disorders (Axinn et al., 2020a; Ghimire et al., 2013). Life history calendars were used to enhance lifetime recall of psychiatric disorders, a method shown to improve psychiatric disorder measurement accuracy (Axinn et al., 2020a). Any mental health disorder, whether lifetime or 12-month, was defined by meeting diagnostic criteria for any of the following psychiatric disorders: major depressive disorder (MDD), bipolar disorder broad (bipolar I, II and subthreshold), panic disorder, generalized anxiety disorder (GAD), alcohol abuse or alcohol dependence indicating alcohol use disorder (AUD), post-traumatic stress disorder (PTSD), minor depression, recurrent brief depression, and dysthymia disorders. Of these, MDD and AUD, as the outcomes with sufficient sample size (e.g. most prevalent) and most studied outcomes in this dataset(Axinn et al., 2022, 2020b; Benjet et al., 2020; Cole et al., 2021; Scott et al., 2020), were explored further than the other measured mental health outcomes for this study.

2.4. Sociodemographic variables

Age (continuous in years), gender (male/female), education (passed school leaving certificate [SLC] versus less than SLC), ethnicity (Brahmin/Chhetri, as compared to other ethnicities [Hill Janjati, Dalits, Newar, and Terai Janjati]), marital status (married, separated/divorced/widowed, never married), self-rated health (excellent/very good/good, fair/poor) and ever having migrated domestically or internationally (yes/no) were self-reported by participants. Historically Brahmin/Chhetri communities have had increased access to education, wealth, and income opportunities, as compared to other ethnic groups within the Nepal.

2.5. Statistical analyses

First, descriptive analyses were performed to examine sociodemographic characteristics and prevalence of psychiatric disorders among both the population-representative sample and PTE-restricted subsample. Because the PTE-only sample is a subset of the population-based sample, the samples are not independent nor identical, so any descriptive comparisons are qualitative in nature. Second, LCA were conducted using Mplus 8 software to empirically identify latent profiles of PTEs in each of the two analytic samples, ranging from two- to seven-class solutions (Muthen and Muthen, 2017). As there is limited consensus in the literature regarding the optimal fit statistic for selecting final class solutions (Nylund et al., 2007), we compared LCA results across multiple fit statistics and incorporating conceptual plausibility to select final class solutions (Nylund-Gibson and Hart, 2014a). The following model statistics were used to inform model selection: log likelihood (non-statistically significant values indicate that the additional class does not improve statistical fit over current model) (Lo et al., 2001), Akaike information criterion (AIC, lower value indicates better model fit) (Akaike, 1987), Bayesian information criterion (BIC, lower value indicates better model fit) (Schwarz, 1978), sample-size adjusted BIC (ssaBIC, lower value indicates better model fit) (Schwarz, 1978), entropy (higher value suggests better class differentiation fit) (Ramaswamy et al., 1993), G² (statistic lower than degrees of freedom indicate a good fit), and chi-squared test and p-value (<0.05 indicates good fit). For each solution, the probabilities of endorsing each PTE were visualized by each identified class to conceptually evaluate resulting classes in an interpretable fashion (Collins and Lanza, 2009). Then, once the final class solutions were selected for both the population-representative sample and PTE-restricted subsample, the sociodemographic and psychiatric disorder prevalence across resulting classes in each sample were calculated. Class names were developed by a) reviewing probability of PTEs in each class; b) highlighting most prevalent PTE(s) per class; c) identifying PTE(s) that differentiated between classes; and d) comparing to existing classes in the literature and potential overlap of causes of PTE exposures to develop substantive class names. Supplemental Table 1 includes estimated probabilities of PTEs by class for the final class solution.

Third, associations between sociodemographic variables and PTE class membership were tested using multinomial logistic regression models, adjusting for survey design factors (i.e., survey sample wave) and using the lowest PTE burden class (i.e., class 5) as the outcome reference category. Separate multinomial logistic regression models were conducted for both the population-representative sample and the PTE-restricted subsample. Fourth, logistic regression models estimated the extent to which class membership was associated with mental health outcomes (i.e., any mental health disorder, any major depressive disorder, and any alcohol use disorder), adjusting for all sociodemographic variables in addition to survey design factors. The odds of lifetime disorders and 12-month disorders associated with class membership were examined using separate models conducted in the population-representative sample and PTE-restricted subsample. Sensitivity analyses exploring distribution of PTE before disorder onset were conducted to establish the temporal order of associations.

3. Results

There were 10,714 individuals in the population-representative sample; of these, 9183 individuals were included in the PTE-subsample. As expected, the mean number of reported lifetime PTEs was higher in the PTE-subsample (2.30 versus 2 in population-representative sample). The mean age was comparable across the two samples (33.10 and 33.20 years old in the population-representative versus PTE-subsample, respectively; Table 1), and other sociodemographic characteristics were distributed similarly across the two samples. Over half of individuals passed the school leaving certificate (SLC) in both samples (61.70 % in population-representative sample and 62.50 % in the PTE-subsample). The prevalence of any mental health disorder was qualitatively slightly higher in the PTE-subsample (5.30 %) compared with the population-representative sample (5.00 %).

Table 1.

Descriptive characteristics of the population-representative sample compared with the PTE-sub sample in the CVFS, Nepal, 2016–2018.

Characteristic	Population-based sample	PTE-subsample
	N = 10,714	N = 9183
	Mean (sd)	Mean (sd)
Age	33.12 (12.12)	33.18 (12.24)
Average PTE countsa	2 (1.39)	2.33 (1.22)
Mean age at incident PTEa (years)	24.97 (10.36)	24.97 (10.37)
	N (%)	N (%)
Female	5790 (54.04)	4874 (53.07)
Ethnicity (Brahmin/Chhetri)b	4634 (43.25)	3932 (42.82)
SLC pass	6615 (61.74)	5736 (62.46)
Any MHDc
Lifetime	2431 (22.69)	2208 (24.04)
12-month	533 (4.97)	489 (5.33)
PTSD
Lifetime	283 (2.64)	283 (3.08)
12-month	56 (0.52)	56 (0.61)
MDD
Lifetime	1622 (15.14)	1455 (15.84)
12-month	292 (2.73)	260 (2.83)
AUDd
Lifetime	629 (8.64)	595 (9.40)
12-month	119 (1.64)	114 (1.80)
GAD
Lifetime	603 (5.63)	543 (5.91)
12-month	148 (1.38)	138 (1.50)
PTE exposures
Any PTEa	9218 (86.04)	9183 (100.00)
Car accident	1097 (10.24)	1097 (11.95)
Other accident	664 (6.20)	664 (7.23)
Natural disaster	7149 (66.73)	7149 (77.85)
Illness	1954 (18.24)	1954 (21.28)
Beaten by caregiverse	768 (7.17)	768 (8.36)
Beaten by spouse	350 (3.27)	350 (3.81)
Beaten by other	246 (2.30)	246 (2.68)
Sudden death	2577 (24.05)	2577 (28.06)
Sick child	864 (8.06)	864 (9.41)
Witness physical fightse	882 (8.23)	882 (9.60)
Witness injury/dead	3306 (30.86)	3306 (36.00)

Notes: CVFS = Chitwan Valley Family Study; PTE = potentially-traumatic event; MHD = mental health disorder; SLC = school leaving certificate; PTSD = posttraumatic stress disorder; MDD = major depressive disorder; AUD = alcohol use disorder; GAD = generalized anxiety disorder.

^aIndicates estimates that included all measured PTEs (n = 29) and not just the 11 in the final LCA model.

^bEthnicity categorized as Brahmin/Chhetri, as compared to other ethnicities Hill Janjati, Dalits, Newar, and Terai Janjati.

^cAny mental health disorder includes: major depressive disorder, bipolar disorder broad (bipolar I, II and subthreshold), panic disorder, generalized anxiety disorder, alcohol abuse or alcohol dependence indicating alcohol use disorder, post-traumatic stress disorder, minor depression, recurrent brief depression, and dysthymia disorders.

^dAmong those with an opportunity to drink (n = 7276 in Population-representative sample, n = 6333 in PTE-subsample).

^ePTE was experienced during childhood.

The five-class solutions provided the best fit statistics, with the non-significant χ² p-value at k class solution supporting the k-1 class solution (Table 2) (Nylund et al., 2007; Nylund-Gibson and Hart, 2014b). Selected fit statistics for the population-representative sample were: Log likelihood p-value ≤ 0.001; AIC = 82,591.99; BIC = 83,021.47; Entropy = 0.54; and χ2 p-value = 0.02. For the PTE-subsample, selected fit statistics were: Log likelihood p-value ≤ 0.0001; AIC = 74,097.83; BIC = 74,518.22; Entropy = 0.58; and χ² p-value = 0.001. Fig. 1 displays participant latent profiles of PTE endorsement based on the respective five-class solutions. Table 3 descriptively summarizes the distribution of sociodemographic and mental health characteristics across these classes in the population-representative and PTE-subsample, respectively. The population-representative sample identified classes predominantly related to: interpersonal violence (4.93 %), accident (21.54 %), illness (6.88 %), natural disaster (52.19 %), low-burden (14.46 %). The following classes were identified in the PTE-subsample: interpersonal violence (10.85 %), bereavement (4.31 %), intimate partner violence/illness (7.69 %), accident/injury (14.82 %), and natural disaster (62.33 %). Supplemental Table 1 illustrates prevalence estimates of PTE items by class.

Table 2.

Model fit statistics from latent class model solutions comparing the population-representative sample and the PTE-subsample in the CVFS, Nepal, 2016–2018.

Model	Log likelihood	Log likelihood (p-value)	AIC	BIC	ssaBIC	Entropy	G2 (df)	χ2 (p-value)
Population-representative sample
2 classes	−41,650.13	<0.0001	83,346.26	83,513.68	83,440.59	0.30	1855.10 (2020)	0.00
3 classes	−41,461.49	<0.0001	82,992.98	83,247.75	83,136.53	0.55	1499.23 (2010)	<0.0001
4 classes	−41,327.79	<0.0001	82,749.58	83,091.71	82,942.35	0.46	1180.21 (1993)	0.01
5 classes	−41,236.10	<0.0001	82,591.99	83,021.47	82,833.97	0.54	1053.36 (1986)	0.00
6 classes	−41,202.01	<0.0001	82,546.02	83,062.84	82,837.21	0.54	942.34 (1970)	1.00
7 classes	−41,179.96	<0.0001	82,525.91	83,130.09	82,866.33	0.55	908.69 (1959)	1.00
PTE-subsample
2 classes	−37,420.60	<0.0001	74,887.20	75,051.08	74,977.99	0.43	2118.66 (2014)	0.00
3 classes	−37,215.13	<0.0001	74,500.27	74,749.65	74,638.42	0.45	1772.10 (2008)	<0.0001
4 classes	−37,071.60	<0.0001	74,237.19	74,572.07	74,422.72	0.57	1516.77 (1999)	<0.0001
5 classes	−36,989.92	<0.0001	74,097.83	74,518.22	74,330.72	0.58	1352.64 (1987)	0.00
6 classes	−36,924.17	<0.0001	73,990.33	74,496.21	74,270.59	0.69	1190.82 (1972)	0.96
7 classes	−36,888.94	<0.0001	73,943.89	74,535.27	74,271.51	0.73	1120.51 (1960)	0.48

Notes: PTE-subsample was restricted to the 11 PTEs included in the latent class models. Bold values indicate best-fitting for that fit statistic. PTE = potentially-traumatic event; CVFS = Chitwan Valley Family Study; AIC = Akaike information criterion; BIC = Bayesian information criterion; ssaBIC = sample-size adjusted BIC.

Fig. 1.

Scatter plot of potentially-traumatic event items by class, CVFS, Nepal 2016–2018.

Notes: N = 10,713 for population-representative sample (interpersonal violence class 4.9 %, accident class 21.5 %, illness class 6.9 %, natural disaster class 52.2 %, low class 14 %) and N = 9183 for PTE-subsample population (interpersonal violence class 10.9 %, bereavement class 4.3 %,interpersonal violence/illness class 7.7 %, accident/injury class 14.8 %, natural disaster class 62.3 %).

Table 3.

Descriptive characteristics of classes in the population-representative sample compared with the PTE-subsample in the CVFS, Nepal, 2016–2018.

	Population-representative sample					PTE subsample
	Interpersonal violence	Accident	Illness	Natural disaster	Low-burden	Interpersonal violence	Accident/injury	Intimate partner violence/illness	Natural disaster	Bereavement
N (%)	528 (4.93)	2308 (21.54)	737 (6.88)	5591 (52.19)	1549 (14.46)	996 (10.85)	1361 (14.82)	706 (7.69)	5724 (62.33)	396 (4.31)
Mean (sd)
Age	30.02 (11.54)	33.56 (11.63)	41.01 (11.03)	32.26 (12.30)	32.91 (11.40)	29.31 (11.34)	34.36 (11.56)	40.27 (11.41)	32.50 (12.33)	36.03 (10.88)
Average PTE countsa	3.93 (1.30)	2.77 (1.42)	2.91 (1.32)	1.92 (9.84)	0.04 (0.26)	3.59 (1.41)	2.79 (1.38)	2.81 (1.47)	1.97 (0.88)	1.94 (0.92)
Mean age at incident PTE (years)a	17.60 (6.30)	22.27 (8.91)	28.57 (9.27)	26.30 (10.81)	25.07 (8.51)	17.15 (6.53)	23.02 (8.45)	27.71 (9.55)	26.45 (10.77)	25.07 (9.49)
N (%)
Female	308 (58.33)	629 (27.25)	500 (67.84)	3418 (61.13)	935 (60.36)	482 (48.39)	263 (19.32)	461 (65.30)	3484 (60.87)	184 (46.46)
Other Ethnicityb	162 (30.68)	1063 (46.06)	327 (44.37)	2371 (42.41)	711 (45.90)	313 (31.43)	663 (48.71)	316 (44.76)	2442 (42.66)	198 (50.00)
SLC pass	389 (73.67)	1345 (58.28)	603 (81.82)	3382 (60.49)	896 (57.84)	679 (68.17)	773 (56.80)	561 (79.46)	3493 (61.02)	230 (58.08)
Any PTEa	528 (100.00)	2308 (100.00)	737 (100.00)	5591 (100.00)	54 (3.49)	996 (100.00)	1361 (100)	706 (100.00)	5724 (100)	396 (100.00)
Any MHDc
Lifetime	202 (38.26)	575 (24.91)	307 (41.66)	1112 (19.89)	235 (15.17)	326 (32.73)	303 (22.26)	303 (42.92)	1161 (20.28)	115 (29.04)
12-month	59 (11.17)	115 (4.98)	77 (10.45)	234 (4.19)	48 (3.10)	81 (8.13)	59 (4.34)	78 (11.05)	250 (4.37)	21 (5.30)
PTSD
Lifetime	32 (6.06)	69 (2.99)	61 (8.28)	121 (2.16)	0 (0.00)	39 (3.92)	23 (1.69)	58 (8.22)	127 (2.22)	36 (9.09)
12-month	11 (2.08)	10 (0.43)	10 (1.36)	25 (0.45)	0 (0.00)	10 (1.00)	6 (0.44)	9 (1.27)	28 (0.49)	3 (0.76)
MDD
Lifetime	139 (26.33)	289 (12.52)	247 (33.51)	769 (13.75)	178 (11.49)	200 (20.08)	142 (10.43)	229 (32.44)	798 (13.94)	86 (21.72)
12-month	33 (6.25)	53 (2.30)	49 (6.65)	122 (2.18)	35 (2.26)	45 (4.52)	25 (1.84)	46 (6.52)	130 (2.27)	14 (3.54)
AUDd
Lifetime	57 (14.21)	256 (13.14)	43 (9.21)	238 (6.78)	35 (3.68)	120 (15.52)	147 (12.27)	52 (11.38)	255 (7.07)	21 (7.02)
12-month	14 (3.49)	43 (2.21)	6 (1.28)	51 (1.45)	5 (0.53)	21 (2.72)	25 (2.09)	6 (1.31)	59 (1.64)	3 (1.00)
GAD
Lifetime	58 (10.98)	66 (2.86)	122 (16.55)	291 (5.20)	66 (4.26)	68 (6.83)	28 (2.06)	117 (16.57)	307 (5.36)	23 (5.81)
12-month	16 (3.03)	16 (0.69)	34 (4.61)	69 (1.23)	13 (0.84)	23 (2.31)	6 (0.44)	36 (5.10)	72 (1.26)	1 (0.25)

Notes: CVFS = Chitwan Valley Family Study. PTE = potentially traumatic event; MHD = mental health disorder; SLC = school leaving certificate; PTSD = posttraumatic stress disorder; MDD = major depressive disorder; AUD = alcohol use disorder; GAD = generalized anxiety disorder.

^aIndicates estimates that included all measured PTEs (n = 29) and not just the 11 in the final LCA model.

^bEthnicity categorized as Brahmin/Chhetri, as compared to other ethnicities (Hill Janajati, Dalit, Newar, and Terai Janajati).

^cAny mental health disorder includes: major depressive disorder, bipolar disorder broad (bipolar I, II and subthreshold), panic disorder, generalized anxiety disorder, alcohol abuse or alcohol dependence indicating alcohol use disorder, post-traumatic stress disorder, minor depression, recurrent brief depression, and dysthymia disorders.

^dAmong those with an opportunity to drink.

In the population-representative sample, gender (female) was associated with membership in the interpersonal violence, accident, or natural disaster classes versus the low-burden class (e.g., adjusted relative risk ratio [aRRR] = 0.24, 95 % confidence interval [CI] = 0.20, 0.28 for accident versus low-burden class), but not with the illness class (Table 4). Separated, divorced, or widowed marital status was associated with membership in the interpersonal violence class versus the low-burden class (aRRR = 2.02, 95 % CI = 1.27, 3.20), but not with the other PTE classes. Poor self-reported health showed a similar pattern of association to all other PTE classes except accident (e.g., aRRR = 1.47, 95 % CI = 1.18, 1.82) versus the low-burden class.

Table 4.

Multinomial logistic regression results of characteristics predicting class membership in both the population-representative sample and the PTE-subsample in the CVFS, Nepal, 2016–2018.

	Population-representative sample				PTE-subsample
	Ref: Low-burden				Ref: Natural disaster
	N = 10,713				N = 9183
	Interpersonal violence aRRR (95 % CI)	Accident aRRR (95 % CI)	Illness aRRR (95 % CI)	Natural disaster aRRR (95 % CI)	Interpersonal violence aRRR (95 % CI)	Accident/injury aRRR (95 % CI)	Intimate partner violence/illness aRRR (95 % CI)	Bereavement aRRR (95 % CI)
Age (years)	0.97 (0.95, 0.98)	1.00 (0.99, 1.00)	1.04 (1.03, 1.05)	1.00 (1.00, 1.01)	1.12 (1.10, 1.13)	1.09 (1.07, 1.10)	1.12 (1.10, 1.13)	1.09 (1.07, 1.11)
Mean age at incident PTE (years)a	–				0.75 (0.74, 0.76)	0.87 (0.86, 0.89)	0.89 (0.88, 0.90)	0.89 (0.88, 0.91)
Femaleb	0.65 (0.51, 0.81)	0.24 (0.20, 0.28)	1.11 (0.90, 1.37)	0.76 (0.66, 0.86)	0.92 (0.77, 1.10)	0.26 (0.22, 0.31)	1.57 (1.28, 1.92)	0.79 (0.62, 1.02)
Ethnicity: Brahmin/Chhetric	1.43 (1.12, 1.83)	0.95 (0.82, 1.01)	0.99 (0.81, 1.20)	1.07 (0.94, 1.22)	1.55 (1.31, 1.86)	0.90 (0.78, 1.04)	1.00 (0.83, 1.20)	0.90 (0.71, 1.14)
SLC passd	0.46 (0.36, 0.59)	0.96 (0.82, 1.13)	0.53 (0.41, 0.67)	0.89 (0.77, 1.01)	0.65 (0.54, 0.77)	1.15 (0.99, 1.34)	0.62 (0.50, 0.77)	1.36 (1.07, 1.73)
Marital status: separated/divorced/widowede	2.02 (1.27, 3.20)	0.94 (0.63, 1.40)	1.20 (0.80, 1.81)	0.90 (0.65, 1.25)	2.13 (1.43, 3.17)	0.56 (0.32, 1.00)	1.60 (1.12, 2.27)	2.15 (1.33, 3.50)
Marital status: never marriede	0.78 (0.59, 1.05)	0.87 (0.70, 1.07)	0.32 (0.21, 0.49)	1.00 (0.84, 1.20)	0.82 (0.67, 1.02)	0.82 (0.67, 1.00)	0.60 (0.44, 0.83)	0.53 (0.37, 0.76)
International migration (ever)	0.62 (0.49, 0.80)	1.05 (0.90, 1.22)	0.75 (0.62, 0.92)	0.44 (0.38, 0.50)	1.37 (1.13, 1.65)	2.44 (2.01, 2.84)	1.65 (1.36, 1.99)	2.93 (2.31, 3.72)
Self-reported general health status: poorf	1.47 (1.18, 1.82)	1.12 (0.97, 1.31)	1.42 (1.17, 1.73)	1.15 (1.01, 1.31)	1.01 (0.86, 1.20)	0.95 (0.82, 1.11)	1.23 (1.04, 1.46)	1.02 (0.81, 1.29)
PTE count: 2+a,g	–				7.23 (5.48, 9.55)	1.22 (1.03, 1.45)	0.82 (0.67, 1.00)	0.46 (0.36, 0.57)

Notes: CVFS = Chitwan Valley Family Study. PTE = potentially-traumatic event; aRRR = adjusted relative risk ratio. CI = confidence interval. SLC = school leaving certificate. Mean age at incident PTE and PTE count were only included in the PTE-subsample population model. Bold values indicate p < 0.05. Models additionally adjusted for design features (i.e., survey sample wave) and accounted for household clustering.

^aIndicates estimates that included all measured PTEs (n = 29) and not just the 11 in the final LCA model.

^bReference: male.

^cEthnicity categorized as Brahmin/Chhetri, as compared to other ethnicities (Hill Janjati, Dalit, Newar, and Terai Janjati).

^dReference: did not pass SLC.

^eReference: married.

^fReference: good health.

^gReference: 1 PTE.

In the PTE-subsample, age and age at incident PTE were associated with membership in all other PTE classes (e.g., interpersonal violence) versus the more common natural disaster class (Table 4). Belonging to the interpersonal violence class was more likely (aRRR = 7.23, 95 % CI = 5.48, 9.55) if an individual had more than two PTEs versus only one PTE as compared with the natural disaster class. The likelihood of belonging to the bereavement class was lower in those with two or more PTEs (aRRR = 0.46, 95 % CI = 0.36, 0.57) as compared with the natural disaster class.

PTE class membership was more consistently associated with psychiatric outcomes in the population-representative sample (which had a low-burden reference group) than the PTE-subsample (which had a group-based-exposure, natural disaster, reference group, Table 5). Membership in the interpersonal violence, accident, illness, or natural disaster classes, compared with the low-burden class, was associated with increased odds of any mental health disorder (adjusted odds ratio [aOR] range = 1.42 to 3.57). Odds of any 12-month mental health disorder (aOR = 3.57, 95 % CI = 2.42, 5.27), such as MDD (e.g., lifetime aOR = 3.00, 95 % CI = 2.28, 3.95) or AUD among those with the opportunity to drink (e.g., 12-month aOR = 6.65, 95 % CI = 2.37, 18.70), were particularly high in the interpersonal violence class relative to the low-burden class. In the PTE-subsample, class membership was also associated with any mental health disorder and MDD, but the observed strength of relationships was attenuated in this sample (which used the natural disaster class as the reference group) compared with the population-representative sample, particularly for AUD outcomes (e.g., lifetime AUD aOR = 2.30, 95 % CI = 1.78, 2.97 in the PTE-subsample versus lifetime AUD aOR = 5.23, 95 % CI = 3.27, 8.37 in the population-representative sample, Table 5). Sensitivity analyses revealed that the overall associations hold in logistic regressions adjusted for only birth-ascribed attributes (age, gender, ethnicity) and study design features (sample wave and household clustering) and that the majority (66.31 %) of those with lifetime exposure to any psychiatric disorder experienced their incident PTE before the age at their psychiatric disorder onset (data available upon request). These data do not capture the temporal specificity of every lifetime PTE, but rather the age at incident PTE exposure, thus complete temporal ordering cannot be accomplished.

Table 5.

Logistic regression results of the relationships between class membership and psychiatric disorders in the CVFS, Nepal, 2016–2018.

	Any MH lifetime aOR (95 % CI)	Any MH 12-month aOR (95 % CI)	MDD lifetime aOR (95 % CI)	MDD 12-month aOR (95 % CI)	AUD lifetimea aOR (95 % CI)	AUD 12-montha aOR (95 % CI)
Population-representative sample (N = 10,713)
Class
Interpersonal violence	3.68 (2.90, 4.67)	3.57 (2.42, 5.27)	3.00 (2.28, 3.95)	2.59 (1.58, 4.25)	5.23 (3.27, 8.37)	6.65 (2.37, 18.70)
Accident	2.26 (1.89, 2.70)	1.99 (1.42, 2.79)	1.87 (1.50, 2.31)	1.57 (1.00, 2.42)	2.85 (1.97, 4.13)	3.09 (1.21, 7.85)
Illness	2.90 (2.34, 3.59)	2.76 (1.88, 4.06)	2.76 (2.18, 3.50)	2.34 (1.48, 3.70)	2.82 (1.74, 4.57)	2.35 (0.70, 7.96)
Natural disaster	1.42 (1.21, 1.66)	1.32 (0.96, 1.81)	1.23 (1.02, 1.47)	0.92 (0.63, 1.35)	2.01 (1.39, 2.92)	2.63 (1.03, 6.73)
Low-burden	Ref	Ref	Ref	Ref	Ref	Ref
PTE-subsample (N = 9183)
Class
Interpersonal violence	2.16 (1.84, 2.53)	1.96 (1.51, 2.56)	2.01 (1.66, 2.44)	2.21 (1.55, 3.17)	2.30 (1.78, 2.97)	1.44 (0.85, 2.42)
Accident/injury	1.36 (1.16, 1.60)	1.30 (0.94, 1.79)	1.41 (1.13, 1.75)	1.47 (0.89, 2.42)	1.18 (0.94, 1.48)	0.97 (0.58, 1.61)
Intimate partner violence/illness	2.21 (1.86, 2.63)	2.18 (1.63, 2.90)	2.21 (1.82, 2.68)	2.41 (1.67, 3.47)	1.70 (1.22, 2.39)	0.82 (0.35, 1.88)
Bereavement	1.54 (1.22, 1.94)	1.25 (0.79, 1.97)	1.96 (1.49, 2.57)	1.70 (0.97, 3.17)	0.78 (0.48, 1.26)	0.56 (0.17, 1.80)
Natural disaster	Ref	Ref	Ref	Ref	Ref	Ref

Notes: CVFS = Chitwan Valley Family Study. aOR = adjusted odds ratio. Models adjusted for age (continuous variable), gender (male/female), binary ethnicity (Brahmin or Chhetri/Other), education (school leaving certificate [SLC] pass/less than SLC), marital status (married/separated, widowed, divorced/never married), ever internationally migrated, design features (i.e., survey sample wave), and accounted for household clustering. Bold values indicate p < 0.05. MH = mental health disorder including any of the following disorders: major depressive disorder, bipolar disorder (bipolar I or II), panic disorder, generalized anxiety disorder, alcohol abuse, alcohol dependence, post-traumatic stress disorder, minor depression, recurrent brief depression, or dysthymia disorders; MDD = major depressive disorder; AUD = alcohol use disorder.

^aSample size for AUD models restricted to those with an opportunity to drink, population-representative sample N = 7276 and PTE-subsample N = 6333.

4. Discussion

Using data from Nepal, this study documented the heterogeneous profiles of lifetime PTE exposure in a population-representative cohort in a high-stress LMIC setting. For the first time, this study compares profiles of a population-representative sample to those from a PTE-subsample. Our LCAs of 11 different relatively common PTEs in over 10,000 individuals in Nepal revealed two distinct five-class profiles of trauma exposure in this population that were distinguishable within this sociodemographic as well as mental health correlates.

Overall, class number and domains were similar across the population-representative and PTE-subsample. The five-class solutions identified in this study are consistent with few published LCAs (Adams et al., 2016; Ziobrowski et al., 2020) as the majority of PTE-based profiles published have identified four-class solutions (Contractor et al., 2018b; O’Donnell et al., 2017). Both published LCAs that identified five-class solutions as the best-fitting for their data included a component of age at PTE exposure in their models, with one specifically disentangling child from adolescent exposure windows (Ziobrowski et al., 2020) and the other documenting temporal exposure period heterogeneity (i.e. exposures in multiple lifetime age periods) (Adams et al., 2016). Similar to our models, both five-class publications had a low-burden class, consistent with most published LCAs to date (Adams et al., 2016; Ziobrowski et al., 2020), including the predominant four-class models (Contractor et al., 2018b; O’Donnell et al., 2017). Inconsistent with other studies that found the inclusion of non-interpersonal PTE exposures led to the identification of a separate high-burden class (O’Donnell et al., 2017), in both analyses the interpersonal violence category had the highest average PTE count per class member (3.93 in population-representative and 3.59 in PTE-subsample).

We found that specific class composition differed based on the sample (population-representative vs PTE-subsample). Given that we could only identify six population-representative LCA studies, and none have conducted a side-by-side sub-analysis of a PTE-subsample, it is not possible to directly situate these findings in the larger literature. In this study, within the PTE-subsample, we failed to observe a generally low-burden class, and rather identified a class that was distinguished by loss and bereavement. On a PTE-endorsement level, our findings are consistent with prior studies noting that the unexpected death of a loved one is one of the most commonly reported potentially traumatic event (Kessler et al., 1995). Given the strength of association between unexpected loss and psychopathology seen in this study, and in other high-income country population-representative samples (Keyes et al., 2014), along with prior work noting that the unexpected loss of a loved one is among the more common precipitant of psychiatric disorder (Breslau et al., 1998), it is fitting that a class defined by bereavement was identified. Additionally, this suggests that selecting samples based on their trauma exposure may reveal subclasses. That is, in the population, exposure to bereavement is represented by “expected” losses as well as “unexpected” losses. However, in a PTE-exposed selected sample, the prevalence of exposure to unexpected loss is likely higher and could enable clearer ascertainment of associations with psychopathology. This suggests that LCA results are dependent on input data, and selecting samples based on overall trauma exposure may reveal additional subclasses obscured in a more general population.

Identified classes in the population-representative and PTE-selected subsamples varied along meaningful sociodemographic dimensions. For example, in the population-representative sample, female participants were more highly represented within the various PTE classes except for the accident class, which consisted mostly of male participants. Female gender was negatively associated with membership in most of the various PTE classes (including interpersonal violence, accident, and natural disaster) relative to the low-burden class, except for the illness class, which showed no difference across females versus males. Within the PTE-subsample, gender was associated with intimate partner violence class (more female) and accident (more male), as compared to the natural disaster class. Literature from other settings suggests that females are more likely to experience interpersonal violence, particularly sexual assault and partner violence (Iverson et al., 2013); however, the violence-related PTEs we examined included being beaten by others or witnessing serious physical fights, rather than sexual forms of violence which were likely underreported in this setting. Data on the gendered distribution of accident exposure is inconsistent (Boughton et al., n.d.; Chandran et al., 2010; Wegman, 2017). Higher educational attainment showed protective associations with belonging to the various PTE classes, as did a history of international migration. This is consistent with literature linking higher educational attainment and economic means to reduced risk of psychopathology (potentially mediated through PTE exposure) (Araya et al., 2003; Bjelland et al., 2008; Butler et al., 2015).

While comparing sociodemographic associations across class membership is interesting and potentially useful if they identify characteristics that can be leveraged to personalize and improve prevention and treatment related to PTE exposures, the ability to compare between these classes and choose an appropriate reference group is more straightforward within the population-representative sample than the PTE-subsample. Some have suggested that isolating a high exposure category (i.e. set as referent group) would allow for more class differentiation in sociodemographic factors and predicting psychiatric disorders (O’Donnell et al., 2017); however, in this study, models were identified along PTE qualitative composition rather than burden counts. Thus, for ease of interpretation to guide clinicians and policy makers, the population-representative sample, with a low-burden group identified, has great utility over the PTE-subsample.

We also observed different patterns in the population-representative versus PTE-subsample when specifically looking at associated mental health morbidity. In the population-representative sample, membership in almost all PTE classes relative to the low-burden class was linked to increased odds of lifetime and 12-month mental health disorders, as well as MDD or AUD specifically. The interpersonal violence class showed the highest odds of lifetime and 12-month mental health disorders, with almost three-fold odds of lifetime MDD and five-fold odds of lifetime AUD compared to the low-burden class. This emphasizes interpersonal violence as a particularly strong risk factor for poor mental health, even when compared to other PTEs (Devries et al., 2011; Mercy et al., 2017). On the other hand, in the PTE-subsample, membership in other PTE classes relative to the natural disaster class was linked to increased odds of mental health disorders, but more so for MDD than AUD, with no significant associations between class membership and 12-month AUD. As natural disaster-related PTEs are more likely to be experienced collectively, they are thus often accompanied by increased community need and resultant support, as compared to individually-experienced PTEs. It is important to note that not all LMICs have had an earthquake/natural disaster. Conversely, some settings have even higher interpersonal violence (e.g., Liberia). While it may reveal some overarching patterns, this LCA approach would likely need to be applied in a variety of LMIC settings to identify tailored patterns.

This study offers a number of strengths. Not only does it expand our understanding of the implications of sample selection (population-representative sample vs PTE-subsample) in polytraumatization profiles, but it draws upon a dataset large enough to explore both population-based estimates and PTE-restricted patterns, and in a Nepalese sample where PTE were assessed comprehensively and with a high response rate using the established CIDI instrument. Several limitations should be noted. First, generalizability to other settings may be limited given unique trauma exposures in Nepal, such as the 2015 earthquake (Goda et al., 2015). At the same time, Nepal’s agrarian population is similar to the majority of the world’s population living in Asia and Africa. In addition, given the dearth of such studies from LMICs, and the increased population-level risk of individuals within LMICs, we provide novel data on PTE exposure profiles that may more accurately reflect the realities of individuals in LMICs than those from similarly unique high-income contexts that dominate the literature. Second, while LCA methods allow researchers to explain variability within populations, there is much debate about how to select final class solutions using model fit statistics (Nylund et al., 2007). Since each fit statistic has limitations and various researchers have used different fit statistics to select final class solutions (Nylund-Gibson and Hart, 2014a), our study used an iterative approach combining information from multiple fit statistics to guide our final class selection decision. Some researchers have recommended exploring plausibility of models in addition to fit statistics (Nylund-Gibson and Hart, 2014a). Model entropy values for the population-based sample are low overall, suggesting poor separation of classes in these models, but not the PTE-restricted sample. Third, using the natural disaster class as the reference group in the trauma-exposed sample was informed by the fact that many individuals were exposed to the Nepal earthquake, and thus represents a more ubiquitous event, but limit the interpretation of the findings between classes. Fourth, the timing of exposure to PTEs and onset of mental health conditions was not ascertained specifically. This leaves open the possibility that onset of psychiatric disorder could have preceded PTE exposure, especially for those individuals endorsing multiple PTEs. This issue was somewhat attenuated by considering lifetime and 12-month psychiatric disorders separately and the brief sensitivity analyses that explored age at incident PTE and age at incident psychiatric disorder onset.

5. Conclusion

This is the first study to directly compare polytraumatization profiles between a population-representative sample and the more commonly employed PTE-restricted subsample. This is one of the few studies to document and explore latent classes of PTE in LMICs. Understanding profiles of lifetime trauma exposure, and their correlates and potential consequences, is challenging but important to improve treatment and preventive services, especially in resource-constrained contexts.