Association of Torture and Other Potentially Traumatic Events With Mental Health Outcomes Among Populations Exposed to Mass Conflict and Displacement:  A Systematic Review and Meta-analysis FREE

Mass conflict and displacement continue to affect large numbers of people worldwide.¹ In 2007, the United Nations High Commissioner for Refugees documented 16 million refugees, 26 million internally displaced persons, and 12 million stateless persons.² Yet controversy about the mental health needs of conflict-exposed populations persists,³ a problem that is magnified by the wide variability in rates of the most commonly studied psychiatric conditions among affected populations, namely, posttraumatic stress disorder (PTSD) and depression. Prevalence rates for PTSD have ranged from 0% in a conflict-affected region of Iran⁴ to 99% in Sierra Leone⁵ and between 3% and 86% for depression across surveys.^6- 7 Identifying the potential methodological and substantive factors that are associated with this variation is vital to determining with any accuracy rates of mental health problems of conflict-affected populations.

Methodological factors, particularly the approach used for sampling and diagnosis, appear to exert a large influence on the prevalence rates yielded by surveys.⁸ An outstanding question, however, is whether exposure to torture accounts for a significant variation in prevalence, even when methodological factors are taken into account. Torture is a common form of human rights violation, being documented in more than 132 countries worldwide.^9- 10 Yet controversy persists as to the extent of the adverse psychological consequences of torture,^11- 12 a contentious issue that extends to other potentially traumatic events (PTEs).^13- 14 A small body of studies focusing specifically on torture survivors has observed high rates of PTSD and depression, but most of these investigations have been based on convenience samples and nonrandom case-control designs.^9,15- 17 There is value therefore in examining the larger body of epidemiologic surveys in which torture and other PTEs^18- 19 are included in the factors assessed to gauge the relative importance of the former type of abuse on mental health among populations exposed to mass conflict.

This study represents the first meta-regression focusing on the largest to date set of epidemiologic surveys in the refugee and postconflict mental health field. The study goals were (1) to assess the influence of methodological factors and substantive population risk factors in accounting for variation in rates of PTSD and depression across surveys and (2) to evaluate the importance of substantive factors on the prevalence of reported PTSD and depression, namely, levels of reported exposure to torture and other PTEs, sociodemographic characteristics, country of origin, time since conflict, security of residency, and the level of terror in the recovery environment, after controlling for methodological factors.

We searched MEDLINE and PsycINFO to identify published articles reporting the prevalence of depression or PTSD among refugee, conflict-affected populations, or both.²⁰ Populations were identified as being from conflict-affected countries if the country was subject to armed conflict or widespread organized violence. The search applied optimum mental health search criteria by combining Medical Subject Headings (MeSH) with text-based search terms²¹ and followed the Meta-analysis of Observational Studies in Epidemiology reporting guidelines.²² Mental health content was identified using the MeSH headings for mental health and mental disorders; articles with refugees were identified using MeSH and text key words for refugees; MeSH headings were also used to identify articles indexed by war, genocide, holocaust, terrorism, or torture. Because of minor inconsistencies in database key words, we generated individualized search criteria for MEDLINE and PsycINFO. In addition, we searched PILOTS,²³ a bibliographic index providing a comprehensive list of articles and other publications on posttraumatic stress applying the following key words: refugees, asylum seekers, displaced persons, internally displaced person, genocide, holocaust, persecution, torture. Inclusion criteria were articles published between 1980 and May 2009 in the English language; surveys assessing the prevalence of PTSD, depression, or both among displaced, conflict-affected populations, or both; surveys of adults 18 years or older; and a sample size of 50 or more. We excluded countries at war in which the main arena of combat was external to the national borders. We also did not include countries exposed to single or multiple (but geographically circumscribed) human-instigated disasters and terrorist attacks in high-income countries because in those settings the general civilian population is not exposed to a wide range of traumas associated with war and rates of torture are likely to be negligible. Some studies needed careful consideration in relation to our inclusion criteria. In particular, studies involving Palestinian populations residing in Gaza and other parts of Palestine were eligible given the high level of combat and organized violence that has occurred in civilian settings in those locations. Studies undertaken in Israel proper were excluded, however, because they focused primarily on members of the military or on a single category of trauma, usually exposure to terrorist attacks, affecting a minority of the population.²⁴ Other exclusion criteria were combatants in armed forces, clinic samples unless recruited by screening of the target population, and populations exposed to PTEs 25 years or more before the survey.

The Figure shows the relevant articles extracted from MEDLINE, PsycINFO, and PILOTS, yielding a total of 3541 nonduplicate articles. After an initial screen of the titles and abstracts and a subsequent screen of text of remaining eligible articles, 161 articles remained, providing data for 181 surveys (eTable 1; eTable 2 presents reasons for exclusion of omitted articles; and eTable 3 provides bibliographical references for all articles). For prospective surveys, the baseline data set was used except in 2 instances^25- 27 in which superior assessment tools were applied in the second survey. Where nonconflict affected populations were included as comparison groups, only data from the conflict-affected sample were extracted.

Data extraction included methodological factors and population risk factors (or substantive factors). The methodological data extracted from each survey included sample size, sampling method (probability survey in which every member of the target population had a known likelihood of inclusion; census survey in which all members of the target population were surveyed either by consecutive recruitment or by total population survey; special population survey in which participants were identified because they shared a common feature, such as having been a political prisoner; and mixed sampling survey including partially representative sampling, linkage sampling, convenience sampling, or volunteer samples), response rate, time period of symptoms (point prevalence [current-1 month]; period prevalence [6-12 months or lifetime]), and type of measure (self-report questionnaire or diagnostic interview). We corresponded with the authors of 19 surveys to obtain additional data, with information being provided by 15 authors. The substantive factors extracted included reported prevalence of PTSD, depression, or both; sex (% female); source country, country in which the survey was undertaken; the year(s) of data collection; time since resettlement in another country or since the cessation of major hostilities as determined by information provided by the study authors or by reference to the Political Terror Scale (PTS)²⁸; and residency status (not displaced, displaced internally or externally to the source country, living in a refugee camp, or resettled in a high-income country).

Where data were available, we recorded the percentage of respondents who reported torture. Because the list of other PTEs assessed varied across surveys, we derived an adversity ratio (the average number of PTEs endorsed divided by the total number of PTEs assessed). Level of political violence in a source country was based on the PTS.²⁸ PTS provides an annual assessment of the level of state terror and political violence on a 5-point rating scale based on data collected by Amnesty International and the US State Department country reports. Scores of 4 or higher designate countries in which violence has been rated as affecting large sections of the population, and where reports of murders, disappearances, and torture are common. PTS data were available from 1976 onwards.

We used SAS version 9.1.3²⁹ to conduct the meta-regression analyses, examining sequentially the association of methodological factors and substantive factors with reported rates of PTSD and depression. The rates of PTSD and depression were transformed into logits. The Delta method was applied to compute within-study variance, namely, var(logit) = 1/case + 1/non_case.³⁰ Interstudy heterogeneity was examined using the Cochran Q test and I² index.³¹ We applied a mixed model with fixed and random-effects components. For the PTSD model, we identified 69% power (based on n = 144, t = 2.21; odds ratio [OR], 1.62; P < .01) to generate a minimally significant OR (1.48 or P < .05); for depression, 87% power (based on n = 117, t = 2.93; OR, 1.91; P < .001) was identified for a minimally significant OR (1.51 or P < .05). The within-study variance was specified using the “parms” statement.²⁹ Tables present the additive variance for each model and the Wald-based ORs and 95% confidence intervals (CIs). Regression logits were back-transformed and expressed as absolute prevalence estimates (ie, they were not relative to the reference group). Adjusted prevalence rates have only been reported for variable strata with 5 or more surveys. Population attributable risk (PAR) percentages were calculated according to the formula:

where pi = proportion of population in n categories (i), ORi = odds ratio in each n category (i), and (l) = low exposure category. The PAR% can be interpreted as the percentage of the observed difference of PTSD or depression between the high and low categories that is attributable to the designated methodological or substantive risk factor.

Meta-regression models were calculated as follows. Step 1: For estimates of PTSD and depression, univariable mixed-effects models were calculated to identify methodological determinants of between-study variance. Next, a forward selection procedure including all statistically significant methodological variables was used to construct a multivariable mixed-effects methodological model. This was used as the base methodological model in examining substantive predictor models. Step 2: Meta-regression models examining the association between interstudy variance in PTSD and depression and each substantive factor were next calculated adjusting for the base multivariable methodological model. Step 3: A series of regression models were then calculated to provide average weighted prevalence estimates for PTSD and depression associated with the stratification of methodological and substantive factors.

To ensure adequate power, we dichotomized all variables to produce a series of 2 × 2 cross classifications as follows: (a) methodological: (i) sampling method by sample size; (ii) sampling method by measure; and (iii) measure by sample size; (b) methodological and substantive: (i) torture by sampling method; (ii) torture by sample size; (iii) PTE ratio by sampling method; and (iv) PTE ratio by sample size; and (c) substantive: (i) torture by time since exposure; (ii) torture by PTS level; (iii) PTE ratio by time since exposure; and (iv) PTE ratio by PTS level. As with the substantive factors examined in step 2, all of the substantive models listed in (c) were adjusted for the baseline methodological model identified in step 1.

The 161 articles yielded diagnostic information for 181 surveys that included 81 866 persons from 40 source countries. Study characteristics are shown in eTable 1. A total of 122 surveys (n = 62 069) were undertaken in low- and middle-income countries and 59 surveys (n = 19 797) with refugees and asylum seekers residing in high-income countries. Twenty-five surveys (n = 9856) were undertaken during the 1980s, 73 (n = 29 684) in the 1990s, and 83 (n = 42 326) in the 2000s. Sample sizes ranged from 50 to more than 4000 (median, 201). Reported response rates ranged from 27% to 100% (median, 90.3%); 56 surveys did not report response rates. Reported prevalence rates for PTSD and depression varied widely, from 0% to 99% for PTSD and 3% to 85.5% for depression.

The population prevalence of reported torture was recorded in 84 surveys (n = 42 626), with an estimated prevalence of 21% (95% CI, 17%-26%), an estimate that remained unchanged after excluding surveys specifically targeting torture survivors (n = 7). Population exposure to 1 or more other PTEs was recorded in 120 surveys (n = 55 101), with a weighted average adversity ratio of 0.29 (interquartile range [IQR], 0.16-0.42) (on average, 29% of the total number of PTEs assessed were endorsed across surveys). PTS scores were available for 125 surveys (n = 66 862), with an average weighted PTS score of 4.0 (IQR, 3.0-5.0) for source countries at the time each survey was undertaken. Forty-six surveys (37%; n = 22 582) had PTS scores higher than 4.0, indicating that political violence and terror were pervasive, affecting the majority of the population; 36 surveys (29%; n = 17 715) had PTS scores between 3.1 and 4.0, indicating reports of extensive civil and political rights violations directed against large sections of the population; and 43 surveys (34%; n = 16 528) had a PTS score of 3.0 or lower, indicating reports of political violence and imprisonment targeted against specific subsections of the population.

A total of 145 surveys (n = 64 332) reported PTSD prevalence estimates based on 21 different measures eTable 4), with the Harvard Trauma Questionnaire³² the most commonly used instrument, followed by the Composite International Diagnostic Interview³³ and the PTSD Checklist.³⁴ Rates of PTSD showed large intersurvey variability (0%-99%) and heterogeneity (Q = 6119, df = 144, P < .001; I² = 97.6%), necessitating the use of random-effect regression models.³¹ The eFigure provides a funnel plot of prevalence by sample size indicating a pattern of asymmetry, with the largest survey samples returning the most conservative prevalence estimates.³⁵ The funnel plot does not support the likelihood of publication bias with the distribution of prevalence estimates across smaller surveys showing no evidence of reduced reporting in the low-prevalence spectrum. The random-effects model returned an overall weighted prevalence of reported PTSD across the 145 relevant surveys of 30.6% (95% CI, 26.3%-35.2%) (Table 1).

Methodological Factors. Table 1 shows the methodological factors associated with PTSD prevalence. Sample size accounted for 9.1% of the variance across surveys. The 17 surveys sampling 1001 or more persons produced the lowest PTSD prevalence estimates (15.7%; 95% CI not available as n > 1000 surveys indexed as reference group), with the rates increasing with progressively smaller samples. For example, the 37 surveys with 100 respondents or less produced a weighted PTSD prevalence estimate of 39.4% (95% CI, 24.3%-56.7%). The PAR comparing large (n > 1000) and small surveys (n < 100) suggested that 43.0% of the observed change in prevalence in the latter group could be attributed to sample size.

Sampling method also emerged as relevant, accounting for 4.4% of intersurvey variance; probability sampling (26.6%) yielded smaller prevalence estimates than surveys applying mixed, nonprobabilistic sampling approaches (37.2%; 95% CI, 26.7%-49.1%). Moreover, PTSD assessments based on self-report questionnaires (34.6%; 95% CI, 25.6%-44.8%) produced estimates 10 percentage points higher than those applying diagnostic interviews (24.6%) with a PAR of 24.9%. Higher prevalence estimates were obtained by surveys reporting point prevalence (32.9%) of PTSD vs period prevalence (17.0%; 95% CI, 10.0%-27.6%). Prevalence rates were not influenced by the response rate or calendar year of survey. A final multivariable meta-regression model applying forward selection criteria included sample size and type of measure and accounted for 12.9% of the intersurvey variance in PTSD prevalence.

Substantive Factors Adjusting for Methodological Factors. Sex (% female), site of survey (high-income country or low- and middle-income country), and residency status were not associated with PTSD prevalence rates, after adjusting for methodological factors. Seventy-seven surveys included data for reported exposure to torture. Torture accounted for the greatest intersurvey variance in reported PTSD (23.6%), after adjusting for methodological factors. Surveys undertaken in which 40% or more of the sample had reported a history of torture returned a weighted average prevalence for PTSD of 46.2% (95% CI, 33.0%-60.0%), 4 times (OR, 4.03; 95% CI, 2.31-7.04) the rate of that obtained from surveys with the lowest rates of torture (≤19% of the sample). The PTE adversity ratio could be calculated for 107 surveys and accounted for 10.8% of the method-corrected variance of PTSD. Reported PTSD rates increased from 11.1% for surveys with adversity ratios of 0 to 0.19 (<20% of PTEs endorsed) to 35.5% (95% CI, 20.8%-53.5%) for studies with an adversity ratio of at least 0.40.

The PTS index for the source country was available for 107 surveys and accounted for 3.5% of the variance in PTSD rates. The weighted average prevalence of reported PTSD was 38.5% (95% CI, 28.6%-49.4%) for samples from countries with a PTS score of more than 4.1 compared with 28.1% for countries with a score of 4.0 or lower. Surveys conducted when the primary conflict was ongoing or had ceased for less than 1 year (n = 49) had a PTSD prevalence of 39.9% compared with those conducted subsequently (2-3 years, 22.1%; 3-5 years, 27.0%; ≥6 years, 22.3%). Of the source countries and regions, Vietnam returned the lowest PTSD prevalence estimate (10%), which differed significantly from rates obtained in surveys with populations from or residents in Cambodia (30.3%; 95% CI, 10.6%-61.3%), Bosnia (28.3%; 95% CI, 11.3%-55.0%), Kosovo (31.6%; 95% CI, 11.9%-61.3%), and 16 countries in Africa (33.5%; 95% CI, 14.2%-60.7%).

Serial Combinations of Methodological and Substantive Factors. We examined PTSD prevalence estimates associated with the serial combinations of methodological factors (Table 2). Sampling method was categorized into models using more scientifically rigorous approaches (probability and census-based) in contrast with convenience or mixed-sampling methods. Similarly, sample size was categorized into large (≥500) and small (<500). The surveys (n = 68) applying nonrandom sampling in combination with a sample size of less than 500 returned an average prevalence rate of 37.0% (95% CI, 26.3%-49.3%) compared with other combinations that returned more conservative rates of PTSD. The 2 models examining sampling method and sample size as a product of measurement type identified a subset of large, well-designed surveys applying diagnostic interviews with lower rates of PTSD (15.4% and 13.2%, respectively), approximately half the prevalence rates yielded by other corresponding method combinations. The PAR percentage was more than 50% in both of these models, suggesting that 50% of the increase in prevalence is attributable to method factors.

The next set of models examined torture prevalence and then PTE ratio by sample size and sampling method (Table 2). A consistent pattern emerged across all 4 models examined, indicating that surveys with populations recording high levels of exposure to torture (≥40%) or cumulative exposure to PTEs (ratio ≥0.3) returned substantially higher PTSD prevalence estimates than those surveys recording lower exposure to either of these 2 factors, irrespective of the sample size or sampling method used.

Further analyses examined for an association of torture and PTE according to the time lapsed since cessation of major hostilities or resettlement and the country level of political terror at the time of the survey (Table 2). The cross-product of torture and time since conflict accounted for 21.1% of interstudy variance in PTSD. The trend line (OR, 1.61; 95% CI, 1.30-2.01) supported a significant linear increase in risk in the expected direction. Samples with the highest exposure to torture (≥40%) that were proximate in time to the conflict (<3 years) returned the highest weighted prevalence of PTSD (53.5%; 95% CI, 36.8%-69.5%). There was an incremental reduction in prevalence for the other derived categories: high torture but extended time (44.4%; 95% CI, 27.7%-62.5%); low torture, recent exposure (26.6%; 95% CI, 17.0%-38.9%); and low torture, extended time (22.3%). A similar pattern emerged for the cross-product of time × PTE adversity ratio, which accounted for a smaller (11.1%) yet significant proportion of interstudy variance in PTSD.

The likelihood of a survey recording higher PTSD prevalence also showed a linear increase when examining the cross-product of torture and the PTS index of country level violence, ranging from 19.2% for surveys conducted in populations recording lower torture rates in countries with lower levels of political terror to a high of 53.5% (95% CI, 36.8%-69.5%) in the greatest torture-exposed populations from countries with the highest ongoing political terror. There was an average 69% (OR, 1.69; 95% CI, 1.37-2.10) increase in PTSD odds across the 4 strata assessed. A similar pattern emerged for cumulative PTE exposure and political terror.

We examined the level of within-strata heterogeneity using the Cochran Q test and I² index for the cross-products shown in Table 2. Across all strata reported, the Cochran Q test remained statistically significant with the I² index ranging between 93.1% and 99.6%, indicating substantial interstudy heterogeneity in reported PTSD prevalence.

A total of 117 surveys (n = 57 796) reported the prevalence of depression. Rates of depression ranged from 3.0% to 85.5%, reflecting a substantial degree of heterogeneity (Q = 10 793, df = 116, P = .001; I² = 98.9%). The most common assessment tool was the Hopkins Symptom Checklist,³⁶ followed by the Composite International Diagnostic Interview³³ and the Structured Clinical Interview for DSM³⁷ (eTable 4). The eFigure provides a funnel plot for depression prevalence by sample size. As with PTSD, there was evidence of asymmetry with the largest survey samples returning conservative prevalence estimates.³⁵ There was no suggestion that the distribution of prevalence estimates across smaller surveys represented underreporting in the low-prevalence spectrum. Table 3 presents the results from the random-effects regression models, yielding an adjusted weighted prevalence for depression of 30.8% (95% CI, 26.3%-35.6%) across the 117 surveys.

Methodological Factors. Sample size accounted for 14.6% of the total interstudy variance in the depression prevalence. Surveys based on at least 1001 respondents reported substantially lower depression estimates (13.3%) than those with smaller numbers, resulting in a PAR of 55.1% (Table 3). Surveys using nonprobabilistic sampling strategies produced the highest prevalence estimates (46.9%; 95% CI, 35.0%-59.2%) compared with probability samples (27.3%) and small census surveys (21.2%; 95% CI, 12.3%-34.1%). Surveys that used a self-report questionnaire yielded higher prevalences (36.7%; 95% CI, 27.3%-47.2%) than those applying diagnostic interview (23.2%; OR, 1.91; 95% CI, 1.24-2.95). With regard to diagnostic time frame, point prevalence (34.8%) was associated with a higher prevalence of depression than period prevalence (OR, 0.40; 95% CI, 0.24-0.68; prevalence, 17.6%; 95% CI, 11.2%-26.5%). The multivariable method model, including sample size, sampling method, and measure type, accounted for 27.7% of interstudy variance in depression estimates.

Some measures, such as the General Health Questionnaire³⁸ and Self Reporting Questionnaire-20,³⁹ although including the construct of depression, also measure other elements of mental disorder, a potential source of interstudy heterogeneity. We recalculated the depression model limiting the analysis to the 97 surveys that used measures designed specifically to assess depression. That analysis failed to alter the level of heterogeneity (Q = 9239, I² = 99%) or interstudy variance accounted for by methodological factors (31.3%); therefore, further analyses included the larger sample.

Substantive Factors Adjusting for the Methodological Factors. Variables that did not affect the prevalence of depression after adjusting for the methodological factors included sex, place of survey, year of survey, and the PTS rating of county-level violence. The PTE adversity ratio (Δ total R² between base methodological model and base model + substantive factor [ΔR²] = 22.0% across 75 surveys) and the rate of torture reported in a survey (ΔR² = 11.4% across 55 surveys) were significantly associated with prevalence of depression after adjusting for the methodological factors. Surveys with a PTE ratio ranging between 0 and 0.19 returned a weighted depression prevalence of 13.7%, an estimate that increased to 34.8% (95% CI, 23.7%-48.0%) and 40.0% (95% CI, 28.4%-52.9%) in surveys with a PTE adversity ratio of more than 0.30 and 0.40, respectively. Surveys in which 40% or more of the respondents reported torture returned a weighted depression prevalence of 49.6% (95% CI, 34.2%-64.9%), higher than in surveys in which less than 20% of respondents reported exposure to that abuse (28.1%).

Reported rates of depression also showed a pattern of gradual decline with increasing time since the conflict or resettlement to a safe third (mainly western) country (OR, 0.80; 95% CI, 0.69-0.93). Populations located in environments free from conflict for at least 6 years had significantly lower rates of depression (19.2%; 95% CI, 13.0%-27.5%) compared with populations in which the primary conflict was ongoing or in which hostilities had ceased for less than 1 year (34.7%). Residency status was also associated with depression, with displaced populations (38.2%; 95% CI, 27.2%-50.5%) and those residing in refugee camps (39.6%; 95% CI, 25.5%-55.5%) having higher rates of depression than refugees resettled to high-income countries (24.1%). In keeping with the PTSD data, the 15 surveys undertaken among populations originating from Vietnam returned the lowest depression rates (15.7%).

Serial Combinations of Methodological and Substantive Factors. Nonrandom sampling, small sample sizes, and use of self-report questionnaires were all individually associated with higher prevalence rates of depression (Table 4). Surveys applying nonrandom sampling and small sample size, or either of these methodological factors in combination with self-report questionnaires, consistently returned the highest prevalence estimates, ranging between 38.9% (95% CI, 21.7%-59.3%) and 43.4% (95% CI, 29.5%-58.4%). The most conservative depression prevalence rates were recorded in the 8 surveys applying diagnostic interviews to samples of more than 500 persons (8.1%; 95% CI, 4.0%-16.0%), followed by the 26 surveys combining diagnostic interviews with representative sampling frames (17.3%; 95% CI, 11.9%-24.5%).

There were insufficient surveys recording high levels of exposure to torture (≥40%) and large sample sizes or probability surveys to generate stable prevalence estimates (n = 4). With respect to cumulative exposure to PTEs, surveys recording a high exposure level (ratio ≥0.3) returned higher depression prevalence estimates than surveys recording lower exposures did, irrespective of the sample size or sampling method used.

We next examined the combined effects of torture and time since conflict, a product that accounted for 20.5% of interstudy variance in the prevalence rates of depression, adjusting for the methodological model across the 53 surveys reporting both sets of data. The trend line supported a significant linear increase in risk across the derived time by torture categories. Samples with the highest exposure to torture (≥40%) had higher rates of depression than those with lower rates of torture exposure, regardless of time since conflict. The importance of time was particularly notable with respect to surveys in the lower torture category (<40%) in which rates were 34.4% (95% CI, 23.0%-48.0%) when the survey was undertaken proximate in time to the conflict (<3 years) in contrast to an average rate of 20.2% for surveys undertaken at least 3 years after resettlement or the cessation of major hostilities. The comparable analysis for the PTE adversity ratio × time also yielded a significant linear trend, accounting for 21.2% of interstudy variance in depression. Surveys with populations endorsing less than 30% of PTEs in which resettlement or cessation of major hostilities occurred 3 years or more previously reported substantially lower rates of depression (15.6%) compared with surveys recording a PTE ratio of 0.3 and above. Due to the absence of an association between PTS and interstudy variance in depression, we did not undertake further analyses using that variable.

We also calculated the within-strata heterogeneity using the Cochran Q test and I² index for the cross-products shown in Table 4. Across all strata, the Q test remained statistically significant with I² index ranging between 92.5% and 99.6%, indicating substantial interstudy heterogeneity in depression prevalence.

Our study addressed a range of issues in the field of refugee and postconflict mental health. First, there has been a need to explain the large differences in rates of PTSD and depression that have emerged from epidemiologic surveys. Second, controversy has persisted about the association of torture and other PTEs with mental disorder across culturally and ethnically diverse populations.^11,40

Two previous quantitative reviews^8,41 and a series of qualitative reviews^15,42- 44 have been undertaken in the field. Fazel et al⁸ analyzed 20 surveys focusing specifically on refugees resettled in high-income countries and found that methodological factors, including sample size and type of diagnostic measure used, influenced prevalence rates, as did contextual influences including length of resettlement. Porter and Haslam⁴¹ and Porter⁴⁵ derived an effect size index of psychological distress from the heterogeneous outcome measures used in a subset of 59 surveys that compared a displaced population with a nondisplaced control group. Rates of distress were influenced both by methodological as well as ecological-social variables, including restricted economic opportunity, insecure housing, and location of residency. Our review incorporates a larger body of surveys comprising a population of 81 866 persons from 40 source countries and included an assessment of the association between torture and other PTEs and the prevalence of PTSD and depression.

The surveys reviewed suggest that torture is endemic in countries affected by pervasive conflict, with respondents in 29 of 40 source countries reporting exposure. In total, 21% of participants in 84 surveys reported personal experiences of torture. The analysis shows that torture and other PTEs are associated with mental disorder across these settings. After adjusting for methodological factors, reported torture accounted for the highest intersurvey variance of PTSD among the substantive factors (23.6%). Torture also emerged as an important source of variance in depression rates, accounting for 11.4% of the method-adjusted interstudy variance.

Cumulative exposure to PTEs emerged as a high source of intersurvey variance for PTSD and depression but was most strongly associated with the latter outcome. For depression, the PTE adversity ratio was the strongest substantive factor (22% of method-adjusted interstudy variance). The PTEs experienced by refugees and other conflict-affected groups often involve multiple losses and deprivations that, in lesser form in civilian settings, have been reported to be associated with depression.^46- 47 In addition, the findings consolidate those of individual surveys showing a dose-response effect for PTEs in relation to both depression and PTSD.^48- 49 There was a tendency toward reduction in mental health risk associated with cumulative PTE exposure as the length of time since conflict or resettlement increased.⁵⁰

PTSD increasingly is conceptualized not simply as a condition triggered by life-threatening PTEs but one that is shaped by conditions of ongoing threat or insecurity.^51- 52 To our knowledge, our study is the first to include a society-wide index of terror, the PTS,²⁸ that is independent of the reports of individual respondents. The association yielded, albeit modest, between political terror and the prevalence of PTSD therefore provides confirmation of the association between the general state of political violence in a country at the time and mental health identified by individual self-reports.

The results also support the notion that broader ecological-social factors interact with personal exposure to PTEs in shaping mental health responses.^18,53 Populations that were displaced within or external to the source country or living in a refugee camp had higher rates of PTSD than those that were permanently resettled in another country. These data may inform policies and strategies that focus on improving conditions in the recovery environment in mitigating the mental health effects of mass conflict and displacement.⁵⁴

Methodological characteristics of surveys accounted for 13% and 28% of the interstudy variance in PTSD and depression, respectively. These findings provide guidance for future researchers in designing surveys in the field. Nonrandom sampling, small sample sizes, and use of self-report questionnaires all tend to generate higher prevalence rates. Self-report questionnaires generate point prevalence estimates, explaining the seemingly anomalous finding that 12-month and lifetime prevalence estimates (based on diagnostic interviews) yielded lower prevalence rates.

The weighted prevalence estimates derived from the subset of methodologically robust surveys may provide the most accurate indicator of PTSD and depression rates among conflict-affected populations included in our review. These rates ranged between 13% and 25% for PTSD and are broadly comparable with the World Health Organization's estimate of 20% for the median prevalence of mental disorder that may be expected in societies exposed to humanitarian emergencies.⁵⁵ An important corollary, however, is that the rate of disorder will depend on the population exposure to the identified risk factors which, as indicated by our analysis, exert a substantial effect on prevalence. Such projections in turn need to be moderated by the recognition that there remains substantial intersurvey heterogeneity in reported prevalence estimates even for surveys using similar methodologies among populations exposed to comparable levels of identified risk factors.

A number of potential limitations exist in our analysis. We have not examined the extent to which the instruments used in the surveys were culturally validated for the specific populations in which they were applied. Shown previously,⁵⁶ criterion validity has rarely been tested in population-based refugee samples. There is an ongoing debate as to whether measures of common psychological reaction patterns such as depressive symptoms measure disorder or distress.^14,57 Although many other mental health variables are salient,⁵⁴ our review focused only on depression and PTSD, because these are the outcome indices that have been studied sufficiently. The assessment of torture across the majority of surveys reviewed was reliant on self-reports, most commonly based on endorsement of the torture item on the Harvard Trauma Questionnaire,³² leaving open the possibility of variation in personal understandings of the term. The effect of cumulative exposure to PTEs was assessed by the ratio of the population PTE count by the total number of PTEs assessed. The index provides a relative measure of PTE exposure that may obscure the absolute number of PTE events reported. Moreover, the index does not assess the subjective component of PTE exposure.⁵⁸ Our review was limited to surveys conducted with displaced, conflict-affected populations, or both. In defining this group of surveys, we excluded the subset of studies focusing on terrorist-affected populations in high-income countries. Thus, we cannot generalize our findings (eg, the prevalence rate of torture identified) to other countries affected by different patterns of violence.

Our review also was limited to surveys published in the peer-reviewed English-language literature. Funnel plots did not provide evidence of publication bias.³⁵ Publication bias may be less of a threat to the validity of meta-analytic surveys focusing on observational surveys than it is to reviews of randomized controlled trials.^59- 60 Nevertheless, there was asymmetry across studies, with larger samples returning more conservative prevalence estimates. The nationally representative population surveys^61- 64 from conflict-affected populations all returned low PTSD prevalence estimates, suggesting that the effects of conflict and displacement are not evenly distributed across societies. Finally, the meta-analytic procedures used are limited to the examination of predictors of interstudy variability only and not within-survey variation.⁶⁵ Initial testing of the present data identified substantial interstudy heterogeneity in prevalence estimates, necessitating the use of random-effects models. The regression models therefore applied more conservative error parameters, which may underestimate the strength of association of some variables.

The results of our analysis indicate that variation in prevalence rates of reported PTSD and depression across surveys can be explained both by methodological factors (sampling and the choice of diagnostic instruments) and substantive risk factors. After adjustment for methodological factors, torture emerged as the strongest substantive factor associated with PTSD and cumulative exposure to PTEs was the strongest substantive factor associated with depression.

Considered as a whole, our findings should encourage researchers to use more methodologically rigorous and valid survey designs when conducting mental health assessments in postconflict settings. These findings should also assist health professionals in their efforts to advocate against human rights abuses and the implementation of international treaties prohibiting the use of torture, to highlight the mental health needs of affected populations, and at an international level to encourage policy makers to ensure that refugees and conflict-affected societies worldwide are assured of secure and supportive recovery environments.⁵⁴