Effects of recurrent violence on post-traumatic stress disorder and severe distress in conflict-affected Timor-Leste: a 6-year longitudinal study

Methods We assessed 1022 adults (600 from a rural village, 422 from an urban district) exposed to mass confl ict during the Indonesian occupation after independence in 2004, and again in 2010–11, following a period of internal confl ict. We took a census of all adults living at the two sites. The survey included measures of post-traumatic stress disorder, severe distress, traumatic events, poverty, ongoing confl ict, and injustice.


Introduction
After mass confl icts, countries are at increased risk of recurrent episodes of communal violence and upheavals that adversely aff ect social, economic, and political development. [1][2][3] However, no studies have assessed the eff ect of repeated communal violence on mental health of citizens in these settings.
Generally, symptoms of post-traumatic stress remain high in the period soon after mass confl ict (up to 3 years), with a gradual improvement in mental health in the years afterwards. [4][5][6][7] Recurrent confl ict might increase the prevalence of mental disorders for several reasons. Repeated exposure to trauma increases the risk of post-traumatic stress disorder and other common forms of mental distress in a well established dose-eff ect relationship. 8 Furthermore, the nature of the confl ict-and what it means to the aff ected citizens, for example, a sense of persistent injustice-might be instrumental in determining mental health responses. 9,10 Timor-Leste exemplifi es a country that has had recurrent confl icts that diff er in their characteristics and implications for the society as a whole. The Indonesian occupation  leading up to the humanitarian crisis of 1999 was associated with extreme trauma and human rights violations, including torture, murder, and mass displacement. 11 Yet our study of a rural village and an urban district in 2004 yielded low prevalence of posttraumatic stress disorder and severe distress compared with other post-confl ict settings, which we tentatively attributed to the general sense of optimism that followed national independence in 2002. 12 A period of internal violence occurred in 2006-07, involving warring factions from within the country. 14 Tensions persisted even after the UN contained the violence, with members of opposing groups continuing to live side by side in the same villages. Observers noted that the internal confl ict had a detrimental eff ect on the society as a whole, undermining the sense of optimism that had followed independence. 14 The loss of community cohesion and continuing feelings of insecurity after the internal confl ict might have created a general vulnerability to mental disorders in the community as a whole.
Persistent feelings of injustice related to human rights violations might be an important factor leading to mental disorders in post-confl ict populations. 15 We anticipate that successive periods of confl ict in Timor-Leste in which human rights traumas were common would lead to a mounting sense of injustice among survivor populations, particularly because most of the perpetrators have not been brought to account. 9,16 Debate continues as to the relative importance of poverty and psychological trauma in shaping adverse mental health outcomes in people living in post-confl ict settings. 17,18 Poverty and trauma might exert diff erent eff ects on mental health outcomes-eg, poverty is associated with symptoms of depression, whereas trauma is associate with post-traumatic stress disorder. 19 Establishing how each of these determinants contributes to mental health is important for the development of strategies for prevention and intervention of mental health disorders.
We did this study to test several related hypotheses: fi rst, that post-traumatic stress disorder and severe distress would be more common after the period of internal confl ict in Timor-Leste compared with baseline. Second, that these mental disorders would be associated with wide-ranging psychosocial disabilities. Third, that post-traumatic stress disorder would be linked to potentially traumatic events related to violence and ongoing tensions, whereas ongoing poverty would be most relevant to severe distress (including symptoms of depression). And fi nally, that distressing preoccupations with injustice spanning several historical periods would be associated with mental health symptoms. 11,15,20,21

Study design and participants
We did this longitudinal cohort study in two locations in Timor-Leste. In a baseline study, we sought to measure the prevalence of common mental disorders 4 years after the cessation of confl ict in Timor-Leste with the aim of following up the sample to assess how symptoms changed over time. 3 months before our baseline study, the Timorese Government did a pilot census in a defi ned district of the capital, Dili, and a rural village, Hera, 1 h away by car. These locations were chosen because together they included populations with varied socioeconomic characteristics typical of Timorese society as a whole. Each location (suco) consists of contiguous hamlets (aldeias) that fall under the overall administration of one chief (chefe). Global positioning system coordinates and aerial maps used in the government pilot census enabled us to identify all households in the two locations, including those in forested or remote areas. Both locations were severely aff ected by confl ict, violence, and displacement during the Indonesian occupation and the subsequent internal confl ict. Details of the 2004 survey have been published. 12,22 We used the same method in a 2010-11 follow-up survey with the addition of the measures described here. The study was approved by the ethics committee of the University of New South Wales, the Minister of Health of Timor-Leste, and the chiefs of each village.

Procedures
We did a census to enable data analysis of both the individual (in the present report) and the household (the intended subject of future reports). At baseline, we sought to interview all adults aged 18 years or older residing in every dwelling in the two locations. In the follow-up study (June, 2010-June, 2011), we aimed to interview all participants surveyed in the baseline survey. In both surveys, we visited households up to fi ve times to maximise participation. At follow-up, we traced those who had moved by asking families, village chiefs, and other informants. We did a longitudinal analysis of participants who had suffi ciently complete data at baseline and follow-up. The fi eld team consisted of 18 Timorese workers with previous survey experience or public health degrees, supervised by an Australian project director (BL) and a Timorese incountry manager. The interviewers received 2 weeks of training followed by 2 months of fi eld testing at geographical locations separate from the two villages participating in the main study. Interviewers were assessed serially in pairs until they achieved a consistent 100% interrater reliability for application of symptom measures. We have previously described the comprehensive capacitybuilding programme off ered to the East Timorese fi eldworkers, the eff ect on morale contributing to the high response and retention rate in the team. 23 We used the Harvard Trauma Questionnaire, 24 which assesses 16 symptoms of post-traumatic stress disorder, each scored on a four-point frequency scale: the summary score (1-4) is the mean score for the 16 symptoms. We also used the Kessler-10 questionnaire, 25 which consists of ten items each scored on a fi ve-point frequency scale primarily measuring depression but also including anxiety and somatic symptoms. A score of 30 or more represents severe distress that in other settings corresponds with the presence of mental disorder in general. 25 We did a clinical recalibration study by comparing the Harvard Trauma Questionnaire and Kessler-10 with a gold standard clinical interview to derive culturally appropriate diagnostic thresholds for post-traumatic stress disorder and severe distress. We included 97 people who participated in the follow-up study, stratifying the sample to include a range of symptom scores on both questionnaires. Because the Kessler-10 questionnaire predominantly includes symptoms of depression, we compared the measure with the module for major depressive episode of the Structured Diagnostic Interview for the Diagnostic and Statistical Manual IV. 26 Fieldworkers readministered the questionnaires and Australian trained psychologists (including BL) assisted by interpreters applied the modules for post-traumatic stress disorder and major depressive disorder of the Structure Clinical Interview for the Diagnostic and Statistical Manual IV. 26 Fieldworkers were masked to the results of psychologists' assessments and vice versa.
The Harvard Trauma Questionnaire score of 2·2 provided the highest overall level of convergence with the post-traumatic stress module from the Diagnostic and Statistical Manual: area under the curve 0·82 (95% CI 0·71-0·94), sensitivity 77·3%, specifi city 87·5%, overall correct classifi cation 83%. For the Kessler-10 questionnaire, the international cutoff score of 30 or more for extreme distress provided the highest level of convergence: area under the curve 0·79 (95% CI 0·67-0·91), sensitivity 92·3%, specifi city 66%, correct classifi cation 71%. The lower specifi city for the Kessler-10 questionnaire is probably a result of the inclusion of anxiety and somatic complaints, in addition to depression, within the measure of symptoms.
We assessed the 16 potentially traumatic events listed in the Harvard Trauma Questionnaire at two points: at baseline (2004), which captures eff ects of the Indonesian occupation and subsequent humanitarian emergency (in 1999) and the period surrounding independence in 2002, and at follow-up (2010-11), including the eff ects of the internal confl ict (2006-07). A previous factor analysis 8 of the baseline data yielded fi ve potentially traumatic event domains: human-rights trauma (including torture and arbitrary imprisonment), witnessing murder (of family or others), natural disaster (fl oods and earthquakes), severe deprivation of health care, and confl ict-related trauma (fi re, lack of shelter, and insuffi cient food and water). Subscale scores were derived by adding endorsed items (1=yes, 0=no) for each factor.
To assess how poverty and family or community confl ict were related to mental health disorders, we made an inventory of contemporary daily living diffi culties on the basis of extensive community consultations and refi nement of items during piloting. Responses on a fi vepoint visual analogue scale were dichotomised into serious or very serious problems (scoring 1) or not serious or less serious problems (scoring 0). Two composite scales (each scored 1-6) were derived for poverty (insuffi cient food [two items], lack of money for school fees and to meet traditional obligations to family, poor shelter, unemployment) and family or community confl ict (with spouse, children, extended family, young people, and the wider community).
At follow-up, participants were asked to identify the worst injustice they had experienced for each of three defi ned historical periods: the Indonesian occupation and humanitarian emergency (1975-99), the internal confl ict (2006-07), and in their present lives (2008-10). Participants who reported relevant events were asked to say whether they had ongoing distressing preoccupations of injustice (no events scored 0, distressing preoccupations for one historical period scored 1, distressing preoccupations for two or three historical periods scored 2). Scores of 2 and 3 were confl ated because of small numbers of people who scored a 3.
For disability, two items in the World Mental Health Survey assessed days-out-of-role related to ill-health in the past month (categorised as 0, 1-4, and ≥5). 27 The Work and Social Adjustment Scale applied at follow-up is a self-report measure of functioning that assesses diffi culties in doing housework, study or work, caring for family, and social activities. 28 All measures were translated into Tetum, the most widely spoken language in Timor-Leste. Minor inconsistencies were addressed during piloting and the fi nal versions were translated and back-translated with a standard approach. 29 Interviews usually lasted an hour and were done in participants' homes. We checked forms for missing data immediately after the interview and gaps were rectifi ed. Each electronic data entry was checked at least once for completeness and accuracy against the paper questionnaires.

Statistical analysis
We did fi ve analyses. First, we generated descriptive data (frequencies, means) with SAS (version 9.2). Second, we calculated intraclass correlations to assess clustering of post-traumatic stress disorder and severe distress within households. Correlations were low (<0·05), indicating  30 Fourth, we used χ² tests to test for associations between mental health and disability at follow-up. Fifth, we applied multiple hierarchical logistic regression models to assess predictors of post-traumatic stress disorder and severe distress at follow-up, each regression controlling for the corresponding baseline disorder (baseline post-traumatic stress disorder for prediction of follow-up post-traumatic stress disorder, baseline severe distress for prediction of follow-up severe distress). Six variables or sets of variables were entered into each regression in the following order: (1) baseline post-traumatic stress disorder or psychological distress, (2) sociodemographic variables, (3) fi ve potentially traumatic events during the Indonesian occupation (measured at baseline), (4) fi ve potentially traumatic events during the internal confl ict (measured at followup), (5) poverty and ongoing family or community confl ict (measured at follow-up), and (6) distressing preoccupations with injustice. The hierarchical method ensured that the eff ects of earlier variables were taken into account in assessing the additional contributions of subsequent sets. Multivariate tests of association are presented as adjusted odds ratios (ORs) with 95% CIs, the level of signifi cance set at p<0·05. We used the SURVEYLOGISTIC procedure in SAS to adjust for design eff ects. Because the data were derived from a complete census of the target population, we did not apply post-stratifi cation weighting.

Role of the funding source
The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had fi nal responsibility for the decision to submit for publication.

Results
At baseline, 1247 (80%) of 1554 adults in the two locations completed interviews. Of respondents who were still eligible at follow-up, 1038 (549 women) were interviewed (89%), including 420 of the 524 baseline urban and 618 of 747 baseline rural participants. Exclusion of records with extensive missing data yielded a fi nal sample for analysis of 1022 people (fi gure). More people lived in the rural location than in the urban location at both baseline and follow-up (table 1). Mean age increased from 33·9 years (SD 12·0) at baseline to 41·5 years (SD 13·5) at follow-up. A greater proportion of people were married and the level of education increased between baseline and follow-up (table 1). Exposure to each of the fi ve potential traumatic event domains was greater during the Indonesian occupation than for the period of internal confl ict (table 1). The prevalence of post-traumatic stress disorder and severe distress increased over the course of follow-up (table 1).
In the fi nal hierarchical logistic regression model, the signifi cant predictors of post-traumatic stress disorder were: being a woman, human rights trauma and exposure to murder during the Indonesian occupation, confl ictrelated and human rights trauma during the period of internal confl ict, ongoing family or community confl ict from 2008 to 2010, and distressing preoccupations with injustice for one, two, or three historical periods (  Severe distress at baseline was not signifi cantly associated with severe distress at follow-up. Signifi cant predictors of severe distress were health stress, exposure to murder, and natural disaster during the Indonesian occupation, confl ict-related trauma during the internal confl ict in Timor-Leste, ongoing poverty during 2008-10, and preoccupations with injustice for two or three historical periods (table 3).

Discussion
Our fi ndings are novel in showing that a period of internal violence was associated with a major escalation in the prevalence of post-traumatic stress disorder and severe distress among a population previously exposed to prolonged confl ict (panel). Specifi cally, we recorded a seven-times increase in post-traumatic stress disorder and an almost three-times increase in severe distress at follow-up compared with baseline.
Strengths of our study include the large sample size and high response and retention rates. In accordance with best practice, we recalibrated symptom measures against a gold standard clinical interview. 31 A limitation of the study is that the sample is restricted to two locations. The study should be replicated in other parts of Timor-Leste and in post-confl ict populations worldwide to assess whether the fi ndings can be generalised. The census method of sampling includes several members of the same household. Nevertheless, clustering by dwelling proved to be negligible in relation to our key mental health indices. Moreover, we corrected for design eff ects in the regression analyses.
We restricted the number of symptom measures to keep participant interviews short. However, in other settings, distress can act as a proxy indicator of mental disorder in general. 25,32,33 Inaccuracy in recallparticularly for distant events-might aff ect responses to inquiries about previous potentially traumatic events. We limited the period of recall by assessing potentially traumatic events for the Indonesian occupation at baseline and for the internal confl ict at follow-up. The multidimensional measure of disability was only applied at follow-up, precluding an assessment of change of disability over time. Although we applied the measure of injustice as a predictor of post-traumatic stress disorder and severe distress, we acknowledge that distressing preoccupations with injustice might be thought of as a psychological outcome in its own right.
Recurrent internal confl ict is common in post-confl ict countries and hampers socioeconomic development. 1 Our study is the fi rst to show its specifi c adverse eff ect on the mental health of the population. 3 The size of the eff ect on symptoms is not fully explained by a simple linear dose-eff ect relation between cumulative exposure to trauma and mental disorder, given that potentially traumatic events were rarer during the internal confl ict than during the preceding Indonesian occupation. It therefore seems likely that the overall eff ects of the internal confl ict on the morale, sense of communal cohesion, and security of the population created a general underlying vulnerability to psychological distress. This underlying vulnerability was probably aff ected by the specifi c risk factors we measured, increasing the prevalence of mental disorder at follow-up.
Preoccupation with injustice was associated with mental disorder, particularly when events from two or more  historical periods were involved. The quadrupling of risk for post-traumatic stress disorder associated with injustice for two or three periods was particularly noteworthy. Although a truth and reconciliation commission was formed in Timor-Leste in 2002, qualitative data cast doubt on whether the process adequately addressed feelings of injustice among many survivors. 34 Exposure to further human rights violations-in this instance enacted by compatriots during the internal confl ict-probably exacerbated the sense of injustice, particularly since few perpetrators from either period of confl ict have faced prosecution. Our fi ndings suggest that repeated exposure to human rights violations might result in the accumulation of distressing preoccupations with injustice that in turn contribute to the burden of mental disorder in postconfl ict societies. 8 Our fi ndings help to resolve ongoing debates about the role of poverty or ongoing hardship and past trauma in generating mental disorder in post-confl ict settings. 17,19 In our study, poverty was associated with severe distress including symptoms of depression, whereas cumulative human rights trauma and ongoing family and community  tensions were linked to post-traumatic stress disorder. Nevertheless, associations between socio economic underdevelopment, recurrent confl ict, and mental disorders are probably complex in countries like Timor-Leste. Observers in the country have noted that communal frustrations with the slow pace of economic development played an important part in generating the divisions leading to the internal confl ict. 14 The ensuing violence and destruction in turn exacerbated conditions of insecurity and poverty which, as shown here, contributed to an increase in mental health-related disability. 1,3,35 A complex recursive relation seems to exist between poverty and internal violence: the two interact in a manner that increases vulnerability to mental disorder as countries emerge from major periods of mass confl ict. Our fi ndings show the need for a multitiered approach to advancing recovery and reconstruction in post-confl ict settings. Clinical services should be provided for people with severe and persisting psychological reactions. On a wider scale, an integrated plan of recovery is needed to alleviate poverty, ensure political and social stability, reduce communal tensions, and off er survivors genuine redress for past and ongoing injustices.

Contributors
SR and DS had the idea for the study, applied for funding, designed the study, trained and supported fi eld staff , collected, managed, and analysed data, and wrote the paper. ABZ designed the study, established in-country partnerships, analysed and interpreted results, commented on the draft report, responded to reviewers, and approved the fi nal version. ZS designed the study, analysed and interpreted data and prepared the report. NT organised fi eld staff and logistics. RB had the idea for the follow-up study, supervised fi eld staff , and analysed data. AN designed the study, collected and interpreted data, and wrote the report. BL was research director for the project from 2011 to 2012, she designed the study design, collected and interpreted data, and wrote the report. TC searched the published work, designed the study, collected, analysed, and interpreted data, and translated documents from English to Tetum, and made writing suggestions. LLS managed the Timorese team, traced and recruited participants, and liaised with stakeholders.

Declaration of interests
We declare that we have no competing interests.

Interpretation
Our study is the fi rst longitudinal survey of the eff ect of a period of recurrent violence on the mental health of a population previously exposed to mass confl ict. The prevalence of mental disorders increased over the course of 6 years of follow-up. Predictors of post-traumatic stress disorder at follow-up included repeated exposure to human rights trauma, ongoing tensions or confl ict in the community, and cumulative preoccupations with injustice. Conditions of poverty were associated with severe distress. Our fi ndings suggest that a period of internal confl ict can result in a major escalation in mental disorders in post-confl ict countries. Mental health services should be enhanced in these settings to meet the increased needs. Prevention of communal violence, alleviation of poverty, and addressing past and ongoing injustices in post-confl ict populations might help to both avert recurrent confl ict and prevent an increase in mental disorders.