Previous studies of depression in Haiti have been limited to specific populations and convenience samples. Moreover, to date, all published studies of depression in Haiti reflect pre-earthquake levels of severity. Among a convenience sample of 258 hospital patients in Haiti, Martsolf (2004) found that 54% met the criteria for major depression using the Center for Epidemiologic Studies – Depression scale 28 . Further research among people living with HIV in Haiti found that eight of the 15 depressive symptoms on the Hopkins Symptom Checklist were endorsed as “quite a bit” or “extremely” by 45-75% of individuals 29 . These psychometric instruments were not validated for use in Kreyòl with Haitian populations, and therefore one cannot confirm that these individuals suffered from clinical depression. Nevertheless, these findings make an important contribution by suggesting a high burden of depressive symptomatology among medically ill Haitians.

In these studies, identified risk factors for depression included: emotional and physical abuse, low socioeconomic status (SES), psychological symptoms of other individuals in the household, and being HIV-positive 29 30 . More research is needed using representative samples to accurately describe the population risk profile of depression in Haiti.

Local cultural and religious factors may interact with conceptualizations of depression, suicidal ideation, and completed suicide in Haiti 31 32 . Idioms of distress that reflect psychosocial complaints among lay persons are often interpreted as other biomedical maladies in healthcare settings (e.g. anemia, gastroesophageal reflux disease) in Haiti 33 . Socio-cultural differences in the understanding of mental distress and suicidal behavior necessitate the study of culturally-specific risk factors and the use of culturally-adapted psychometric screening tools in order to build appropriate prevention and treatment services 21 23 25 28 34 . Yet, the lack of appropriate mental health services and training for health workers may be an even greater challenge than cultural differences for the treatment of mental disorders. For example, lay individuals’ perceptions of biomedical health worker competence addressing mental health issues – rather than differences in cultural explanatory models – may be the dominant determinant of why religious leaders, such as Protestant pastors and Vodou priests, are often preferred over nurses and doctors for psychosocial care in rural Haiti 35 .

Suicide is a major public health issue, and rates of suicide have increased 60% over the past 45 years, with 85% of suicide deaths occurring in LMICs 19 . There is a dearth of research on suicidal behavior and associated factors in Haiti. Risk factors for suicide documented in other LMICs include: female gender, living in a rural area, young or old age, having low SES, and holding religious beliefs that sanction suicide 36 37 . Also, in India, Taiwan, and China, recent stressful life events, including familial problems and social change 38 39 , as well as previous suicide attempts 40 41 , have been associated with suicidal behavior. Other studies in Asia have found that young women carry a disproportionate burden of suicide and that there is a weaker association with mental illness than in high-income countries 28 42 43 44 45 46 47 .

Worldwide, there are contrasting findings on the impact of natural disasters, such as earthquakes, on suicide-related behaviors 48 49 50 . Most research has been conducted in developed countries and suggests that suicide rates decrease following natural disasters 48 49 . However, in one study following Hurricane Katrina in the US, suicidal ideation rates were shown to increase 50 .

Globally, only 17% of individuals in low-income countries present to providers in the year prior to suicidal behavior 51 . In Haiti, access to biomedical healthcare is especially low in rural areas. Approximately half of Haitians lack access to primary healthcare, and in rural areas mental health services are virtually nonexistent 52 . Furthermore, of the few psychiatrists practicing in Haiti, most are living and working in urban areas 53 , which precludes access for most rural Haitians with depression and/or suicidal ideation.

The present study represents one of the first epidemiologic studies of depression symptomatology, suicidal ideation, and associated risk factors using a culturally-adapted instrument among a representative community sample in Haiti’s Central Plateau. This study aims to inform the development and scale-up of sustainable mental health and psychosocial services in Haiti.

We conducted a cross-sectional, zone-stratified household survey over six weeks in May-June 2011 in the rural Central Plateau of Haiti. Participants were selected by a stratified design using a modified version of the WHO’s Expanded Program on Immunization “random walk” protocol 54 . Surveys took place in the communal section of La Hoye, with the stratification factor being governmentally-determined zones within the area. Four of the 17 zones were not included in the sampling frame because they had a low population density and were difficult to safely access during the rainy season.

In the study area, livelihood is largely sex-differentiated: males tend to travay la tè (work the land), while females fè kòmes (do commerce) and care for the family and home 32 . The majority of the population cannot read or write 55 . While everyone speaks Kreyòl, a small minority also speak French (these individuals tend to have higher levels of education) or Spanish, due to the proximity to the Dominican Republic.

Regarding impact of the 2010 earthquake, the Central Plateau sustained little direct damage. However, many areas of the Central Plateau saw significant increases in population due to internally displaced persons. Additionally, many residents of the Central Plateau had family members killed or directly impacted by the earthquake, and many housed displaced family or friends. Despite these devastating and far-reaching impacts, participants made it clear that the earthquake represents one among many challenges that impact mental and physical health in the Central Plateau.

The communal section of La Hoye was chosen as the research site because of its proximity to a non-governmental organization (NGO) that was interested in providing mental health services in the future. The NGO partners with American medical schools and Haitian healthcare personnel to provide year-round medical care in several communal sections in the Central Plateau. While there are three hospitals within a two-hour drive from the research site, the time and resources required to reach these healthcare facilities by foot, horseback, or motorcycle puts them firmly out of reach for the majority of rural Haitians. Instead, they rely on small clinics, many run by NGOs.

At the time of the research, one town in the research setting had a small clinic with one doctor, one nurse, two auxiliary nurses, and one lab technician. Approximately once per week, medical staff conducted mobile clinics, visiting each zone in the communal section once every three months. During the single day of healthcare provision for the zone, dozens of patients would line up, with several often being turned away at the end of the afternoon when rains began. Rather than utilizing formal biomedical care, people could also purchase medicines and herbal remedies at local markets or draw upon a range of traditional supports, including hougan-s ¹This paper utilizes the standard convention of adding –s to indicate plural Kreyol words, rather than –yo, the plural indicator in Kreyol. and manbo-s (Vodou priests and priestesses), doktè fey-s (herbalists), and religious leaders 56 57 .

Our survey consisted of demographic information, daily stressors (such as water insecurity), traumatic experiences associated with the earthquake (such as injury), general trauma (such as death of a family member), social support, care seeking behavior, and explanatory models of distress. Demographic information included a SES scale ranging from 0 to 9, representing a count of the number of the following items possessed: tin roof, bicycle, motorcycle, television, radio, cement house, latrine, electricity, and/or a telephone. Assessing household possessions can provide a more objective measure of SES than estimated household income in subsistence economies or when all household members are not included in financial decision making 58 59 . Additional file 1: Table S1 includes a detailed description of all study variables.

Depression symptom burden was assessed using a culturally-adapted Kreyòl version of the Beck Depression Inventory (BDI) 60 . The BDI is a self-report measure of depression symptoms assessed over the past two weeks, with 21 questions answered according to a 0 to 3 severity rating, and an overall range from 0 to 63 61 . The BDI has been adapted and validated for use in a variety of cultural settings 62 63 64 65 66 67 .

The BDI was adapted using a standardized approach for transcultural translation that evaluates semantic, content, construct, and technical equivalence 24 26 . This process involves five steps: translation by bilingual lay individuals, review by bilingual medical professionals working from the local setting, focus group discussions with lay representative members of the target group, back-translation by a bilingual individual blinded to the original instrument, and pilot testing. At each of the five stages, all items are individually assessed for comprehensibility, acceptability, and relevance. These assessments are used to make adjustments to the translation at each stage.

In Haiti, the BDI underwent the process described above using six focus groups comprised of lay men and women 59 . Examples of changes made to the BDI arising from qualitative research include items for “loss of interest” and “indecisiveness.” Loss of interest in the literal translation was interpreted as disinterest in helping others. To capture the construct of anhedonia, the translation was changed to “I am not on it” (m pa sou sa), which more closely approximates lack of interest and desire to engage with people and activities. “Indecisiveness” was also changed because the original translation was interpreted as not having the power to decide, rather than not having the concentration and motivation to make decisions. The item was reworded to incorporate a sense of confusion (pa fouti ka pran yon desisyon).

The adapted BDI was piloted among a sample of 31 lay participants to assess internal reliability. The Cronbach’s alpha for the pilot sample suggested high reliability (α = 0.89). Of note, the items for sleep difficulties and changes in eating appeared to discriminate poorly between individuals scoring higher or lower on the depression scale, as these items were highly endorsed by nearly all respondents. This phenomenon has been observed in other settings with high levels of gastrointestinal pathology or other physical medical problems, such as rural Nepal 25 62 68 69 . Similarly, medically ill and geriatric populations may have somatic complaints that elevate BDI scores in the absence of clinical depression 63 70 . While these items were not removed, they suggest that total BDI scores may be inflated relative to Western settings. Ultimately, a diagnostic validation study of the Kreyòl BDI is needed to determine the cut-off score with the strongest psychometric properties.

One of four trained research assistants administered each survey verbally in Haitian Kreyòl, recording responses on paper. Research assistant training consisted of two days of didactic training on survey delivery, sampling, and basic epidemiologic methods, as well as three days of pilot survey data collection to ensure quality delivery of the BDI. On-going feedback and training took place throughout the period of data collection to ensure appropriate survey implementation. Finalized surveys took approximately one hour to complete.

Four research assistant days were spent in each of the 13 zones, beginning at a central zone location and proceeding in opposite directions, sampling every house encountered. A household was defined as a group of people sleeping in the same lakou (compound). Individuals self-identifying as household visitors were ineligible. Because data are not available regarding the precise age structure of the Central Plateau, research assistants rotated asking to speak to a member of a given household aged 18–30, 30–50, or 50+, with individuals under 18 being ineligible. If a person of the preferred age range was not present, the research assistant chose the person closest in age to the preferred age category. Additionally, research assistants alternated recruiting male and female participants. Surveys were double-entered into Excel 2007 71 each day, and data entry errors were fixed by crosschecking for consistency using Excel Compare v2.4 72 .

Scores on the 21-question culturally-adapted Kreyòl BDI were summed to create an index of increasing symptoms of depression, with potential scores ranging from 0 to 63. Linear scores were employed rather than a categorical cut-point because the Kreyòl BDI has not yet undergone clinical validation. For individuals with missing or no response on BDI questions, we used the individual mean score on completed BDI questions and multiplied this number by 21. A sensitivity analysis showed that using group mean imputation did not meaningfully change associated factors or magnitudes. Suicidal ideation endorsement was defined as responding to BDI item #9 with either: (1) “I have the idea to kill myself but I will not do it,” (2) “I would like to kill myself,” or (3) “I will kill myself if I get the chance”.

We used SAS 9.3 73 for statistical analyses. Associations were evaluated for statistical significance at α = 0.05 using two-tailed tests. Multivariable linear and logistic models were built, with the outcomes being continuous BDI scores and endorsing suicidal ideation, respectively. Variables considered for inclusion into each model were: age, gender, marital status, education, religion, distance to and type of work, distance to and type of drinking water, care seeking behavior, household size, SES, trauma related to earthquake, general trauma, reported household mental illness, number of children, months without enough food, stigma towards mental illness, explanatory models of distress, alcohol use, care and household help, BDI score minus item 9 (for logistic model only), and endorsing suicidal ideation (for linear model only).

Backward elimination procedures (α = 0.05) were used to arrive at final models. To examine differences between genders, associated factors, and magnitudes, male-only and female-only linear regression models were built in addition to a combined gender model. We a priori chose to analyze BDI scores separately by gender because of differences in risk factors for common mental disorders between men and women observed cross-culturally 74 75 76 77 78 . For logistic regression, due to limited events, only a combined gender model was built and variables were screened prior to backward elimination. Only variables with at least one event among individuals endorsing suicidal ideation and a bivariate p <; 0.20 were entered into the logistic model for backward elimination.

To establish significance of individual predictors, t-tests were used for linear regression, and Wald chi-square tests were used for logistic regression. For significance of group predictors (distance to work, education and marital status), chunk F-tests were used for linear regression, whereas likelihood-ratio tests were used for logistic regression. The fit of each final linear model was evaluated using residual diagnostics, partial residual plots, Cook’s distance, leverage values, and variance inflation factors. Hosmer and Lemeshow’s goodness-of-fit test and the area under the receiver operating characteristic (ROC) curve were used to evaluate fit for the logistic model.

This project was approved by the Institutional Review Board of Emory University (expedited review, #IRB00042396) and the Haitian Ministry of Health. Prior to asking survey questions, research assistants completed an informed consent process with each participant in Kreyòl. Because the majority of rural Haitians are not literate, verbal consent was used. Regarding management of high risk participants in the study, an American licensed clinical social worker from our research team conducted a follow-up diagnostic interview with anyone responding with a 2 or 3 on the adapted BDI suicidal ideation item to assess level of distress and refer those who endorsed active suicidal ideation.

The mean score on the imputed BDI was 20.4 (CI: 19.3-21.5; median: 18), with 41.7% scoring 20 or greater and 22.7% scoring 29 or greater. Of the 408 participants, 6.13% (N = 25) endorsed current suicidal ideation, and all endorsing active suicidal ideation were confirmed as suicidal through follow-up diagnostic interviews.

For the combined, male-only, and female-only linear models, each final model revealed no significant lack of fit or multicollinearity and achieved R-squared values of 0.40, 0.49, and 0.34, respectively. For the final logistic model, Hosmer and Lemeshow’s goodness-of-fit test revealed no significant lack of fit (p = 0.75), and the model achieved excellent discrimination, with the area under the ROC curve being 0.86. Internal consistency of the locally-adapted BDI was high, achieving a Cronbach’s alpha of 0.86.

Older age was associated with increased BDI scores for both genders. Each 10-year increase in age was associated with an increase of 1.1 BDI points for men (p = .009) and 1.8 points for women (p = .002); (see Table 2). Women, on average, scored 2.1 points higher than men on the BDI (p = .031). In the combined gender model, traveling over one hour to a doctor was associated with scoring 2.4 points higher on the BDI (p = .012). However, distance to formal healthcare services was not significantly associated with BDI scores in male-only or female-only models. Endorsing suicidal ideation was associated with an increase of 14 points on the BDI for men (p <; .001) and 7.7 points for women (p = .009). Collapsing across genders, having a death in one’s family and having had a life-threatening illness were associated with a 2.7 point (p = .013) and 2.6 point (p = .006) increase in BDI scores, respectively.

*P <; .05 (T ₁ ).

^† P <; .001(T ₁ ).

n.s. = Eliminated through backward selection for given model.

^a Of 408 total respondents, excludes 1 with missing data for education, 1 with missing data for distance to work, 3 with missing data for household mental illness, 1 with missing data for months without food, 5 with missing data for time to doctor, 1 with missing data for age, and 1 with missing data for education and marital status.

^b Of 206 total males, excludes 1 with missing data for distance to work.

^c Of 202 total females, excludes 2 with missing data for education, 1 with missing data for household mental illness, and 1 with missing data for age.

^d Variables considered for inclusion: age, gender, marital status, education, religion, distance to and type of work, distance to and type of water, care seeking behavior, household size, SES, trauma related to earthquake, general trauma, reported household mental illness, number of children, months without enough food, stigma towards mental illness, explanatory models of distress, alcohol use, care and household help, opinions on mental illness causation, and endorsing suicidal ideation.

^e Construct p-values for distance to work were p <; .001 (combined), p <; .001 (males only), and p = .005 (females only).

^f Construct p-values for education were p = .035 (combined), non-significant (males only), and p = .012 (females only).

^g Construct p-values for marital status were p = .002 (combined), non-significant (males only), non-significant (females-only).

Distance to work was strongly associated with BDI scores for both genders (p <; .0001 men; p = .005 women). Compared to those who do not work, traveling over one hour to work was associated with a 12.2 point increase on the BDI (p <; .001) for men and a 6.8 point increase for women (p = .002).

Marital status was associated with BDI scores only in the model collapsed across genders (p = .002). Compared to those who are single, being divorced was associated with a 7.5 point increase on the BDI (p = .031).

Among men only, responding that the last time they were sick they went to the hospital was a protective factor, associated with scoring 5.0 points lower on the BDI (p = .007). Each three-month increment that men stated they did not have enough food for their family was associated with scoring 2.0 points higher on the BDI (p = .006). Having more people come live in one’s house after the earthquake was associated with men scoring 3.8 points higher on the BDI (p = .002). Stating that disasters are the primary cause of stress, sad heart, or sadness was associated with men scoring 7.8 points higher on the BDI (p <; .001).

For women only, each additional item possessed on the SES scale was associated with scoring 1.5 points lower on the BDI (p = .008). Reporting that someone in their house suffers from stress, sad heart, or sadness, that a family member died in the 2010 earthquake, or that spirits can cause stress, sad heart, or sadness was associated with scoring 4.1 (p = .008), 3.5 (p = .019) and 5.9 (p = .009) points higher on the BDI, respectively. Level of education was significantly associated with BDI scores for women only (p = .012). Compared to those with no formal education, having more than a primary school education was associated with scoring 7.7 points higher on the BDI (p = .008).

Score on the BDI (minus item #9) was the factor most strongly associated with suicidal ideation. For each ten point increase on the BDI, the odds of endorsing suicidal ideation increased 2.5 fold (p <; .0001; see Table 3). Individuals who use alcohol were 3.3 times as likely to endorse suicidal ideation (p = .012). Individuals who reported ever having gone to a Vodou priest were 3.2 times as likely to endorse suicidal ideation (p = .037). Last, individuals reporting that they did not have someone to care for them if sick were 5.5 times as likely to endorse suicidal ideation (p = .043).

*P <; .05 (Wald Χ²).

^† P <; .001 (Wald Χ²).

^a From 408 total respondents, excludes 2 with missing data on alcohol use, and 1 with missing data on whether they have care if sick.

^b Variables considered for inclusion: age, gender, marital status, education, religion, distance to and type of work, distance to and type of water, care seeking behavior, household size, SES, trauma related to earthquake, general trauma, reported household mental illness, number of children, months without enough food, stigma towards mental illness, explanatory models of distress, alcohol use, care and household help, and BDI score minus suicidal question (item #9).

The goal of this study was to elucidate patterns of burden for depression symptomatology and suicidal ideation in rural Haiti in order to identify potential actions for referral, treatment, and prevention. Regarding referral, having visited a Vodou priest was associated with suicidal ideation, whereas having visited a biomedical provider was not. This suggests that persons with suicidal ideation are likely to have visited a Vodou practitioner. Therefore, referrals from Vodou priests and other community and mental health resources may be ideal to support those with current thoughts of ending their life. That said, partnerships with traditional healers can have negative outcomes as well, such as being more expensive than biomedical treatment 108 . Religious and traditional healers may also worsen stigma based on the explanatory models they invoke 109 110 . These risks should be taken into account when developing referral systems in Haiti.

Our research indicates that natural disasters, such as earthquakes, can have far reaching psychological effects, including on individuals not directly in the earthquake-affected region. Therefore, depression screening programs should attempt to reach those who are indirectly affected, such as those accepting displaced persons and those with relatives who died in the disaster. For example, individuals within households that hosted displaced persons could all undergo screening, as hosts are also at risk of increased symptoms of depression. An additional screening group could be those who have suffered recent losses such as widows and those with a death in the family. Individuals who deal with these groups, such as religious leaders who assist in funerary rites, could be a source of referrals. Rather than medical professionals, religious leaders may be better positioned and able to distinguish between culturally-appropriate grief expressions versus more pathological and impairing psychiatric sequelae.

For treatment, while psychiatric services need to be improved in general, there are specific populations who would particularly benefit from improved mental healthcare. One such group is substance abusers who are at risk of suicidal ideation and likely suicide attempts. Packages of care for substance abuse are available, and trials have been conducted in a range of LMICs 111 . These could be piloted and optimized for Haiti’s Central Plateau. Cross-cultural studies of substance abuse treatment have shown that involvement of the family is extremely important, and this should be emphasized in Haiti. Another important treatment group is the medically ill. In our study, persons with medical problems were at increased risk for depression symptoms, a finding that is consistent with global literature on medical illness and mental health 112 . Therefore, interventions that target depression, anxiety, and demoralization among the medically ill could significantly lower the burden of mental illness 113 114 115 , while also improving medical outcomes 112 .

The finding that depression symptom burden was higher among women may be associated with reproductive life events 77 . In Pakistan, community-based psychotherapeutic programs have successfully reduced the burden of maternal depression 116 . Similar interventions would be feasible in Haiti given the low level of technical psychiatric knowledge needed for implementing such a program 117 . Further research is needed to elucidate gender-specific causes and effects of depression in Haiti.

For prevention of mental illness, structural interventions to reduce poverty are likely indispensible. Poverty and mental disorders easily become vicious cycles; therefore, reducing poverty will help prevent mental disorders and reduce barriers to recovery among those who already suffer from psychological distress 118 119 . Food security is one component of poverty reduction that has shown a strong association in this population and thus should be targeted. In addition, improving the quality and availability of medical care also will likely prevent some of the current burden of depression. This medical care should include not only clinical services, but also community outreach for persons with current medical conditions who may need home health assistance. The observed relationship between reported lack of a caregiver when sick and suicidal ideation points toward the need for community-based psychosocial support to reach out to ill persons lacking family support.

Ultimately, there are a number of current models for improving mental healthcare in low-income settings 21 120 121 . An additional key will be to address the policy barriers to implementing these evidence-based practices in Haiti.

Our study has several limitations. First, compared to the 13 zones sampled, the four zones excluded due to inaccessibility could have higher depression symptomatology and suicidal ideation due to further isolation from biomedical, mental health, and psychosocial services. Second, due to lack of an up-to-date census or list of households, we had to rely on the WHO’s “random walk” protocol. Nevertheless, the “random walk” protocol has been shown to give accurate and precise results even when compared to a randomized probability sample based on a household census 122 . Third, since our BDI instrument has not been clinically validated, we cannot establish specific cut-offs to determine the absolute prevalence of major depression. Regardless, the use of a culturally-adapted BDI is a major strength of our study and aids in ensuring valid determination of risk factors, even if absolute prevalence cannot be estimated 123 . An additional strength is that all reported cases of active suicidal ideation were confirmed through follow up by a licensed clinical social worker.