Comparison of Mental Health Between Former Child Soldiers and Children Never Conscripted by Armed Groups in Nepal FREE

Armed groups throughout the world continue to exploit children to wage war.¹ The dedicated efforts of humanitarian organizations,² psychosocial workers,³ and former child soldiers^4- 5 have called international attention to this issue. However, in a recent report on the status of child soldiers, Betancourt et al⁶ revealed gaps in crucial areas of research to understand the impact of becoming a soldier on child mental health. First, child soldiers are considered in need of special psychosocial intervention. However, there is a lack of published research comparing the severity of mental health problems and functional status among child soldiers with children living through war who were not conscripted to armed groups^6- 9; unpublished studies of nongovernmental organizations suggest there may not be a difference between these groups.^6,10 Second, despite suggestions of increased psychological distress for girl soldiers,^5,11- 13 prior to the conduct of this study no published epidemiological studies to our knowledge had explored sex differences in the psychological impact of soldiering. Third, child soldiers are assumed to have greater exposure to trauma than nonconscripted children.¹⁴ Yet, major studies of child soldiers have not shown an association between trauma and posttraumatic stress disorder (PTSD).^15- 16 Finally, Betancourt et al⁶ call for studies using validated and cross-culturally appropriate mental health measures, which have been lacking in this field. Researching these issues is crucial to designing the most effective mental health interventions for children in armed conflicts.

In the current study, we worked toward addressing these gaps. Our first objective was to compare the mental health of former child soldiers who have returned home with that of children growing up in active conflict settings but who were not conscripted by armed groups in Nepal, using cross-culturally validated measures of psychosocial well-being. We sought to determine (1) if former child soldiers have more mental health problems than never-conscripted children; (2) if becoming a soldier has a greater impact on girls vs boys when compared with never-conscripted children; and (3) if differential exposure to trauma is associated with mental health differences between child soldiers and never-conscripted children. Our second objective was to describe predictors of mental health outcomes within child soldiers: whether trauma exposure, combat experience, military roles, and other soldier-related variables are associated with mental health outcomes and whether the associations between these factors and mental health outcomes were modified by voluntary vs involuntary recruitment, time since leaving military service, or maintained association with an armed group.

We conducted a matched-pair cross-sectional study to assess the association between soldier status and mental health outcomes of children in Nepal. Children with the main exposure (history of being a child soldier) were matched with unexposed children (children who had never been conscripts of a military group, referred to as never-conscripted children).

The Communist Party of Nepal–Maoists fought a 10-year war with the government of Nepal ending with peace accords in November 2006. The Maoists and the Nepal Army recruited individuals younger than 18 years as soldiers, sentries, spies, cooks, porters, and messengers.^17- 18 During and after the conflict, many former child soldiers returned home. Because of difficulties in accessing this population, we used expert purposive sampling, rather than a probability sample, to identify former child soldiers who had returned home and compare them with children never conscripted by armed groups. The research was conducted during March and April 2007.

We use the term child soldier to reflect the Paris Principles² definition of children associated with armed forces and armed groups, which refers to “any person below 18 years of age who is or who has been recruited or used by an armed force or armed group in any capacity, including but not limited to children, boys, and girls used as fighters, cooks, porters, messengers, spies, or for sexual purposes. It does not only refer to a child who is taking or has taken a direct part in hostilities.” Additional selection criteria included armed group participation for at least 1 month, being younger than 18 years during study enrollment, and having a consenting caregiver. In Nepal, full adult franchise occurs at age 18 years.^19- 20

Experts, comprising human rights groups and humanitarian agencies, provided names and locations of former child soldiers who had returned home. The lists included all of the known child soldiers at the time of the study. Local civil society groups, teachers, and community leaders validated the veracity (eg, based on news media documentation, disruption in school attendance, hospital records) of these child soldier cases. Every child and her/his caregiver on the lists provided by the experts was invited to participate in this study.

Former child soldiers who agreed to participate identified a matched child who had never been conscripted by an armed group. Matching factors were sex, age, level of education, and ethnicity. Human rights groups and community leaders confirmed never-conscripted status. Matching was used to increase the feasibility of identifying a comparison (unexposed) group and to control for potential confounding by the matching factors.

One child soldier reported affiliation with the Nepal Army. This child and his matched counterpart were excluded from statistical analyses to avoid heterogeneity of the study group.

The institutional review board of Emory University (Atlanta, Georgia) approved the study protocol and consent process. Children provided oral assent; caregivers provided oral consent. Caregivers did not participate in the interviews. Because of high illiteracy rates, research assistants read consent forms and questionnaires. Children were provided with a snack during the interview and received a notebook and pen in appreciation of their participation. Caregivers were not offered any incentive for their children to participate. No participants had received psychosocial services prior to enrolling in the study. Participants whose symptoms scores were above the cutoff, who reported suicidal ideation, or who requested services were enrolled subsequently in programs for psychosocial support.

Participants completed 60- to 90-minute interviews with trained research assistants conducted in private locations away from caregivers and others. Demographic characteristics (Table 1) were recorded as categorical (eg, education) or continuous (eg, number of household facilities as a measure of economic status). Field researchers recorded the ethnicity of children based on the child's last name, which indicates caste and ethnicity in Nepal. Researchers classified participants into 3 groups: Brahman/Chhetri (“upper” caste), Dalit (“lower” caste), and Janajati (ethnic minorities) according to Nepal Central Bureau of Statistics categories.²¹ We assessed ethnicity because it has been associated with mental health in Nepal.^22- 23

Standard instruments were used to assess symptoms of depression, anxiety, and PTSD and general psychological difficulties. For these scales, cutoff scores were used as markers for psychological or psychiatric intervention. Validation of instruments (assessing accuracy and cutoff thresholds) was performed with a separate sample of Nepali children (n = 162) comparing questionnaire scores with a standardized indication for psychosocial intervention determined by a trained Nepali psychosocial counselor using the Global Assessment of Psychosocial Disability^24- 25 as an external criterion.²⁶ The 18-item Depression Self Rating Scale (DSRS)^27- 28 assessed depression symptoms over the past week on a 3-point scale (range, 0-36), with a clinically significant cutoff score of 14. The 5-item version of the Screen for Child Anxiety Related Emotional Disorders (SCARED-5)²⁹ assessed anxiety symptoms over the past week on a 3-point scale (range, 0-10), with a clinically significant cutoff score of 4. The 17-item Child PTSD Symptom Scale (CPSS)³⁰ assessed PTSD symptoms over the past week on a 4-point scale (range, 0-51), with a clinically significant cutoff score of 20. Nonclinical, general psychological difficulties over the past week were assessed with the 20-item “total difficulties” 3-point scale (range, 0-40) of the Strength and Difficulties Questionnaire (SDQ), with a clinically significant cutoff score of 16.^31- 32

Based on methods adapted from Bolton and Tang,³³ a rating scale was developed to measure children's functioning in a contextually valid manner (W.A.T. et al, unpublished data, September 2005–January 2007). This method involved qualitative techniques (participant observation and child focus groups) to identify daily activities in the realms of child, family, peer, and community functioning. The final 10-item child Function Impairment tool assessed daily functioning over the past 2 weeks on a 4-point scale (range, 0-30), with higher scores indicating more impairment.

Lifetime traumatic events were assessed with the Kiddie Schedule of Affective Disorders and Schizophrenia (K-SADS) PTSD traumatic event checklist.³⁴ The K-SADS includes a series of traumatic events (experience a car accident, experience other type of accident, fire, witness a disaster, witness a violent crime, victim of a violent crime, confronted with traumatic news, witness domestic violence, experience physical abuse, experience sexual abuse, or other), with specific criteria for each item (eg, for fire, “child close witness to fire that causes significant property damage or moderate to severe physical injuries”).

All instruments underwent a transcultural translation procedure.³⁵ Based on focus groups with Nepali children, no items were added to or removed from the mental health instruments. However, items such as bombing, abduction, and torture were added to the traumatic events checklist and sexual abuse and “hearing traumatic news” were excluded. Sexual abuse was determined to be culturally inappropriate and unsafe to ask of young girls as it could place them in jeopardy of harm from community and/or family members if they were suspected of discussing sexual behavior with strangers. “Hearing traumatic news” was excluded because it was difficult to frame “news” as “traumatic” for these children, and the experience was seen as ubiquitous during the conflict. The final trauma exposure list included experiencing unintentional injury, fire, natural disaster, beating, bombing, abduction, torture, murder of a family member, domestic violence, physical abuse, and witnessing a violent death.

Internal consistency of the instruments was measured in the total sample (n = 282). Test-retest reliability and interrater reliability of the instruments were established in 2 other studies among children in Nepal.²⁶ Instrument properties were sufficient to excellent (DSRS: Cronbach α = .80, Spearman-Brown coefficient for test-retest reliability r = 0.80, area under the curve [AUC] = 0.82, optimal cutoff score for psychosocial treatment indication = 14; SCARED: α = .87, r = 0.84, AUC = 0.64, optimal cutoff = 4; CPSS: α = .91, r = 0.85, AUC = 0.77, optimal cutoff = 20; SDQ: α = .76, r = 0.85, AUC = 0.72, optimal cutoff = 16; Function Impairment: α = .68, r = 0.70, AUC = 0.67, optimal cutoff = 4). Interrater reliability was excellent for all instruments (average intracluster correlation coefficient = 0.972; average κ = 0.891).

Additional survey questions were included based on qualitative research with former child soldiers, never-conscripted children, and adult community members. Child soldiers reported their roles during association with an armed group, such as military regiment (received training for combat and given a rank in the People's Liberation Army), “cultural program” (partaking in cultural performances promoting Maoism), cook (traveling with the military preparing meals), porter (carrying ammunitions, food, clothing, medicine, and other supplies), messenger (carrying information between platoons or between villagers and military personnel), sentry (standing guard, typically at night, often armed with grenades), and spy (gathering information from the opposite armed group, often by posing as a member of the other group, and gathering information from teachers, community leaders, and others to provide to the Maoists). Participants indicated if they were involved in multiple roles. We also ascertained exposure to combat, process of returning home, and whether joining the armed group was voluntary or forced. Interest in participating in political activities was assessed with a 4-point scale (no interest, minimal interest, moderate interest, or intense interest). Although all child soldiers participating in the study were living in the community at the time of the study, based on prior qualitative research we determined that some of these children considered themselves “still associated” with an armed group; ie, living in the community but awaiting orders to return to their duties traveling and participating with the armed group. However, we expected that the majority of the children would identify themselves as former child soldiers “no longer associated” with an armed group; ie, they were no longer participating in any armed group activities and did not plan to return to duty. Thus, we asked each child soldier to identify him/herself as “still associated” or “no longer associated.”

Former Child Soldiers vs Never-Conscripted Children. To determine the necessary sample size for exposed former child soldiers and the matched group of unexposed children, we used a power analysis (α = .05, 95% power) relying on treatment effect studies using the CPSS and DSRS,³⁶ which suggested a minimum of 35 children per treatment group when randomizing children individually (Layne et al³⁷ reported an effect size of 0.78 on depression symptoms, with α = .05 and β = .05). To account for the intracluster correlation, this number was multiplied by 1 + (m − 1) ρ, with m = 30 (average cluster size) and ρ = 0.1 (intracluster correlation coefficient), resulting in an appropriate sample size of 137 per group for child soldiers and matched counterparts. Our final sample for statistical analyses was 141 child soldiers and 141 matched counterparts.

We used logistic regression to model the association between child soldier status and mental health outcomes, using the validated cutoffs to dichotomize each outcome. There were 2 nonnested levels of clustering in the data, matched-pair and village, and 17 matched pairs were from different villages. We used the generalized estimating equation (GEE)–based strategy of Miglioretti and Heagerty³⁸ to obtain empirical standard error estimates for nonnested clustering structures.

We created a series of models for each mental health outcome. To assess possible effect modification of child soldier status by sex, we evaluated the statistical significance of a first-order cross-product term in a regression model that included all potential confounders. To determine the role of the different types of traumatic exposures as possible confounders of an association between child soldier status and mental health outcome, we compared the magnitude of the effect estimate (point estimate of the effect) for child soldier status in an unadjusted model with the magnitude of the effect estimate for child soldier status in a series of models that each included 1 type of trauma. To determine the role of aggregate traumatic exposure as a possible confounder, we compared the unadjusted model for child soldier status with a model that included all 11 types of trauma. Finally, we used a model that included trauma variables and other potential confounders to obtain a fully adjusted effect estimate for child soldier status. In addition to the trauma variables, other variables included current religion, school enrollment, family type (joint families with multiple generations and adult siblings in the same dwelling vs nuclear families), marital status, house ownership, and interest in politics.

Because the explanatory power of the analyses was reduced by dichotomizing the outcome variables rather than using the linear instrument scores, we conducted supplementary analyses using linear regression with the outcomes as continuous variables. The series of models evaluated for each outcome had the same explanatory variables as for the logistic regression analyses. Again, we used the GEE-based method of Miglioretti and Heagerty³⁸ to account for the clustering structure.

Former Child Soldiers. We used logistic regression to model the association between child soldier characteristics and each mental health outcome, with GEE-based empirical standard error estimates to account for the clustering of the soldiers within villages. In addition to the 11 types of traumatic exposure and the aforementioned nontrauma covariates, we examined age at recruitment, military role with the armed group, combat exposure, current association with an armed group, type of recruitment (voluntary or involuntary), duration of military service, and time since returning home. Current association with an armed group was not used in models for depression, anxiety, or psychological difficulties because of low cell counts. Sex, ethnicity, and education (matching factors for the primary study) were also considered as explanatory variables. To determine if trauma exposure, combat exposure, and other aspects of being a child soldier were independently associated with mental health outcomes, we compared unadjusted models for each covariate to fully adjusted models. As with the soldiers vs nonsoldiers analyses, we conducted a series of parallel, supplementary analyses using linear regression. In particular, this allowed us to better describe the exposure variable “current association with armed group,” for which it was not possible to estimate an odds ratio (OR) for some outcomes.

P < .05 was considered statistically significant. Statistical analyses were performed with SPSS version 15.0³⁹ and SAS⁴⁰ software.

Researchers contacted 227 potential participants (all the children on the expert lists) and/or their caregivers. Of those 227 children, 169 (74%) met inclusion criteria. Of the 58 children not included, 32 were older than 18 years of age and 26 had served as armed soldiers for 30 days or less. Of the 169 children meeting criteria, 27 (16%) did not participate: 10 had moved (according to caregiver reports), 5 were engaged in school examinations, and 12 former Nepal Army conscripts refused for fear of retaliation. The matched counterparts comprised 37 siblings, 40 other relatives, and 65 unrelated peers. Of the matched pairs, 125 (88%) were from the same villages. None of the former child soldiers were actively participating with an armed group at the time of the study. However, 22 (15.5%) child soldier study participants reported that they were still associated with an armed group and would return to duty if requested.

Demographic characteristics and trauma exposures of former child soldiers and never-conscripted children are shown in Table 1. Sex distribution did not differ between former child soldiers (53% girls) and never-conscripted children (51% girls) (McNemar χ² test, [cumulative] binomial distribution used, P = .69); however, 2 pairs could not be matched on sex but were matched on all other characterisitcs. Other demographics that did not differ include ethnic distribution (approximately one-third Dalit [P = .69], Brahman/Chhetri [P = .73], and Janajati [P = .99]), religion (P < .001), family type (P = .001), and education level (McNemar-Bowker χ² = 10.97; P = .28). Child soldiers were older (mean age, 15.75 years) than never-conscripted children (mean age, 14.92 years) (paired t test = 5.77; P < .001). School enrollment and marital status differed between groups (McNemar χ² test, [cumulative] binomial distribution used, P < .001 and P = .001, respectively).

All participants experienced at least 1 type of trauma. Exposure to bombing was more common among former child soldiers (n = 80 [56.0%]) than never-conscripted children (n = 29 [20.6%]) (OR, 4.92; 95% confidence interval [CI], 2.91-8.33), as was torture (child soldiers, n = 41 [29.1%] vs never-conscripted children, n = 15 [10.6%]; OR, 3.44; 95% CI, 1.80-6.57) and witnessing a violent death (child soldiers, n = 57 [28.7%] vs never-conscripted children, n = 24 [17.0%]; OR, 3.31; 95% CI, 1.90-5.75). In contrast, exposure to beating did not differ between child soldiers (n = 104 [73.8%]) and never-conscripted children (n = 96 [67.4%]) (OR, 1.36; 95% CI, 0.81-2.28).

More former child soldiers reported symptoms that were above the cutoff scores for each mental health scale compared with never-conscripted children (Table 2), including depression (n = 75 child soldiers [53%] vs n = 34 never-conscripted children [24.1%]), anxiety (n = 65 [46.1%] vs n = 53 [37.6%]), PTSD (n = 78 [55.3%] vs n = 28 [20.0%]), psychological difficulties (n = 55 [39.0%] vs n = 26 [18.4%]), and function impairment (n = 88 [62.4%] vs n = 63 [44.7%]).

In the total sample, former child soldiers had greater odds of being above cutoff scores for mental health outcomes except anxiety (Table 2). There was a statistically significant interaction between sex and child soldier status for PTSD in the logistic regression analysis and for all outcomes in the linear regression analysis. Therefore, we report the sex-stratified results for these analyses. When controlling for all traumatic exposures (adjusted model 1), the OR point estimate decreased for depression, psychological difficulties, function impairment (all children), and PTSD (in boys and girls), and was no longer statistically significant for psychological difficulties, function impairment, or PTSD (in boys). For depression, psychological difficulties, and function impairment, the OR did not change appreciably with additional covariates (adjusted model 2). However, after adjusting for trauma and other covariates, former child soldiers had significantly worse outcomes for depression and PTSD. Exposures to different types of trauma largely explain the observed unadjusted associations between child soldier status and mental health for psychological difficulties, function impairment, and PTSD (among boys); but traumatic exposure only in part explains the association for depression and PTSD (among girls).

In the supplementary linear regression analyses, the association between child soldier status and anxiety was not significant for either sex. For the other 4 outcomes, the association with child soldier status was greater among girls than boys, and adjustment for trauma largely explained the association between child soldier status and mental health outcomes among boys but not among girls. For example, in boys, the regression coefficient for child soldier status on PTSD (CPSS questionnaire) was 6.01 (95% CI, 3.45-8.58) in the unadjusted model, 1.99 (95% CI, −0.41 to 4.39) when adjusting for trauma, and 2.43 (95% CI, −0.22 to 5.08) in the full model (adjusting for trauma and other covariates). In contrast, the regression coefficient for associations between child soldier status and PTSD in girls was 7.96 (95% CI, 4.94-10.97) in the adjusted linear regression model, 5.74 (95% CI, 1.84-9.65) when adjusting for trauma, and 5.93 (95% CI, 1.94-9.91) in the full model. A similar pattern emerged for depression, general psychological difficulties, and function impairment. (Further details from the linear regression analyses are available from the authors on request.)

In models (both logistic and linear) that added 1 type of trauma exposure, the magnitude of the regression coefficient for child soldier status did not change appreciably (not by more than about 5%), indicating that no type of trauma in and of itself played a large confounding role with respect to child soldier status (details available from the authors on request).

For the fully adjusted logistic regression models (adjusted model 2) shown in Table 2, among the 11 trauma covariates (unintentional injury, fire, natural disaster, violent death, beating, bombing, abduction, torture, murder of family member, domestic violence, and physical abuse), exposure to beating was associated with depression (OR, 2.21; 95% CI, 1.01-4.83) and daily functioning (OR, 2.58; 95% CI, 1.45-4.59). Exposure to bombing was associated with depression (OR, 1.93; 95% CI, 1.17-3.20) and psychological difficulties (OR, 2.54; 95% CI, 1.38-4.69). Exposure to torture was associated with anxiety (OR, 1.99; 95% CI, 1.00-3.94), psychological difficulties (OR, 2.35; 95% CI, 1.17-4.71), and daily functioning (OR, 2.10; 95% CI, 1.02-4.30) in all children and with PTSD in both boys (OR, 6.96; 95% CI, 2.08-23.35) and girls (OR, 3.53; 95% CI, 1.17-10.70). All other traumatic exposure covariates were not significant.

Table 3 describes the characteristics of former child soldiers. The sample represented relatively equal groups of children who joined voluntarily (n = 65 [45.8%]) vs forced conscription (n = 77 [54.2%]). More than half of the children were conscripted before reaching 14 years of age. Half of the child soldiers (50.7%) reported having been directly engaged in combat. Thirty-two (22.5%) of the children had been members of armed groups for more than 1 year, and 59 (41.5%) had been returned to the community for more than 1 year. Of the child soldiers interviewed, 22 (15.5%) described themselves as still associated with an armed group, while the majority of children no longer considered themselves active participants. Girls, compared with boys, were more likely to have been in cultural programs (OR, 3.86; 95% CI, 1.86-8.05). In contrast, girls, compared with boys, were less likely to have been in military regiments (OR, 0.37; 95% CI, 0.16-0.86), acting as sentries (OR, 0.44; 95% CI, 0.22-0.88), or active in the armed group for more than a year (OR, 0.41; 95% CI, 0.20-0.82). There was no difference between girls and boys in exposure to combat (OR, 0.65; 95% CI, 0.33-1.26), involuntary association (OR, 1.05; 95% CI, 0.54-2.04), age of recruitment (OR, 0.56; 95% CI, 0.28-1.10), time since returning home (OR, 1.59; 95% CI, 0.82-3.12), or still being associated with the armed group (OR, 0.46; 95% CI, 0.18-1.18).

Results of logistic regression models with fully adjusted effect estimates for the covariates considered in the analysis of the subgroup of former child soldiers are presented in Table 4, Table 5, and Table 6. In the adjusted analyses of child soldiers, female sex was associated with worse symptom scores for depression, anxiety, and general psychological difficulties. Economic status was associated with reduced mental health symptoms on all scales. Among the traumatic events included in this analysis, exposure to beating was associated with worse outcomes for depression, exposure to bombing was associated with depression and general psychological difficulties, and exposure to torture was associated with symptoms of PTSD and general psychological difficulties.

Still being affiliated with an armed group was associated with better outcome scores on all scales. In the logistic regression analysis, continued association had a lower point estimate for PTSD symptoms and function impairment compared with no continued association. For 3 of the outcome variables (depression, anxiety, general psychological difficulties), we did not calculate ORs of still being associated with an armed group because of low cell counts. The adjusted linear regression coefficients of current association with an armed group on outcomes were as follows: for the DSRS (depression), −5.26 (95% CI, −7.48 to −3.04); for SCARED-5 (anxiety), −1.73 (95% CI, −2.84 to −0.62); for the CPSS (PTSD), −6.83 (95% CI, −11.37 to −2.28); for the SDQ (psychological difficulties), −3.74 (95% CI, −6.19 to −1.30); and for the FI (function impairment), −2.69 (95% CI, −4.03 to −1.36). (Further details from the linear regression analyses are available from the authors on request.)

In this study comparing the postconflict mental health symptoms of former child soldiers and matched children who had never been conscripted in Nepal, both groups displayed a substantial burden of mental health and psychosocial problems. The mental health burden among former child soldiers ranged from 39% to 62% of participants depending on type of distress vs 18% to 45% of children not conscripted by armed groups. Child soldiers had worse mental health outcomes (symptoms of depression, PTSD, general psychological difficulties, and function impairment) than the comparison groups, with the exception of anxiety symptoms. The difference in mental health outcomes between child soldiers and never-conscripted children can be explained in part by greater exposure to traumatic events among child soldiers, especially for general psychological difficulties and function impairment.

However, even after controlling for exposure to trauma, child soldier status was associated with poorer outcomes for depression and PTSD. This association of child soldier status with PTSD was twice as strong for girls compared with boys. This suggests that factors such as nontraumatic child soldier experiences or traumatic exposures other than those we assessed may contribute to depression and PTSD, with these factors especially important for girl soldiers. In this study, no single type of traumatic event in isolation explained the association between soldier status and mental health; rather, it was the aggregate traumatic exposure burden. For all children, the traumatic exposures most strongly associated with poor mental health outcomes were beatings, bombings, and torture.

The lack of difference in anxiety symptoms suggests that anxiety may be a generalized response of children living through war and conflict regardless of their status as soldiers or civilians. The results also suggest that being a soldier exposed children to more traumatic events, which is associated with higher rates of symptoms of depression and PTSD, generalized psychological difficulties, and function impairment. These findings are, in part, congruent with other studies, which suggest that the difference between former child soldiers and civilians is concentrated among the soldiers with greater trauma exposure, following a dose-response tendency.^6,10 However, our study differs in finding elevated symptoms of depression and PTSD even after controlling for trauma, especially among girls.

We are thus left with the question of what other aspects of the child soldier experience contribute to the poorer mental health outcomes among soldiers compared with civilians. One possibility is the impact of sexual violence on psychosocial well-being among former child soldiers, as has been observed in Africa.^11- 13 Unfortunately, we were unable to include sexual violence in our analysis because it was deemed culturally inappropriate and potentially unsafe to ask the study participants about such exposures.

Reintegration difficulties when former child soldiers return home are another possible contributor. Communities may fear returned former child soldiers and socially ostracize them.^3,7,41- 42 In Nepal, association with Maoists may lead to perceived violations of Hindu cultural norms (such as carrying dead bodies, eating in other ethnic groups' homes, and both sexes sleeping in the same areas). This could result in maltreatment (eg, stigmatization and abuse) by families and communities when soldiers, especially girls, return home. Reintegration difficulties may warrant further investigation as child soldiers in this study who were still associated with an armed group had better mental health outcomes than former child soldiers who were no longer associated with an armed group. Furthermore, the protective association seen among child soldiers with higher economic status with all outcomes raises the possibility that financial resources may buffer against some of the difficulties met after returning home.

This study has several limitations. For feasibility reasons, we used a convenience sample of soldiers, and soldiers identified their own controls, although we used methods appropriate for matched pairs to minimize the effect of the latter limitation. Another limitation of this study is that it represents a subset of child soldiers: those who returned home. We did not assess child soldiers of the Nepal Army, child soldiers who remained within military cantonments of the People's Liberation Army, child soldiers (especially boys) who fled to India, or those who joined the Young Communist League. Ultimately, our study provides insight into the impact of being a child soldier on mental health, but the findings cannot be applied universally to all groups of child soldiers in Nepal or around the globe.

Because the sample size for the study was chosen with the primary objective in mind, low power is a limitation of the within-soldier analyses. For example, we did not find a significant effect of combat exposure for any of the mental health outcomes, but the detectable OR⁴³ for combat based on the observed prevalence ranged from 2.63 to 2.98. Thus, because the study may have been underpowered to detect such differences, the failure to find associations with combat exposure, military role, duration of association, and other child soldier variables does not conclude that these factors do not influence mental health status. Finally, we did not have information on drug or alcohol use or exposure to sexual violence. Girls associated with the Maoists may have experienced less sexual violence compared with civilian girls because of the Maoist focus on gender equality⁴⁴; similarly, Maoist prohibitions on substance use may have reduced drug and alcohol problems during association.⁴⁵ However, sexual violence and substance abuse should be considered during interventions and for further research because the status of these problems among former child soldiers after they return home is unknown.

The study has several clinical and programmatic implications. First, the greater burden of mental health problems among former child soldiers supports the need for focused programming, which should include, but not consist solely of, interventions to reduce depression symptoms and the psychological sequelae of trauma, especially bombings and torture, as well as incorporate belongingness and income generation. Second, girl soldiers may require focused attention, possibly for factors not addressed in this study, such as problems of sexual violence and reintegration difficulties. Third, the variation in type and severity of mental health problems highlights the importance of screening, including locally developed measures of function impairment, as a base for intervention.⁴⁶ Without screening there is a risk of pathologizing child soldiers as a group rather than providing support to those individuals most impaired. Finally, the presence of mental health problems among never-conscripted children illustrates the need for comprehensive postconflict community-based psychosocial care not restricted only to child soldiers.