Mental Health, Social Functioning, and Attitudes of Kosovar Albanians Following the War in Kosovo FREE

In late February 1998, clashes in Kosovo between Serbian police forces and members of the Kosovo Liberation Army intensified.¹ Serbian forces burned homes and killed dozens of ethnic Albanians in these raids. As a result of the fighting, thousands of ethnic Albanians were displaced from their homes in Kosovo; many took refuge with host families, while a smaller proportion (several thousands) fled to the hills and forests.¹ By the time North Atlantic Treaty Organization (NATO) operations began against Serbia on March 24, 1999, about 260,000 people had been displaced within Kosovo and 199,000 had fled to other countries.² It is estimated that as result of this conflict, more than 800,000 people became refugees in neighboring countries (mainly Albania, Montenegro, and the former Yugoslav Republic of Macedonia), as well as secondary countries of asylum in Europe, the United States, and elsewhere. On June 9, 1999, an agreement between NATO and Serbia was reached, and the following day NATO halted its bombing campaign.

As the Serbian troops began to pull out of Kosovo, the nearly 750,000 Albanians from Kosovo who had been living in refugee camps in Albania, Macedonia, and Montenegro began to return to Kosovo.² On their return, the displaced Albanians had to come to terms with the destruction of their homes and property, missing family members, and the traumatic experiences of violence, rape, and persecution. The full psychological impact of such emergency situations is a neglected issue.³ However, recent epidemiological studies in Bosnia⁴ and studies among Cambodian refugees living on the Thai border⁵ and in the United States have shown that psychiatric morbidity is much higher in populations that have experienced war, persecution, and mass violence.^6- 7

To estimate the prevalence of psychiatric morbidity and to identify specific vulnerable populations, the Centers for Disease Control and Prevention (CDC) and the Institute of Mental Health and Recovery in Kosovo, in collaboration with Doctors of the World, conducted a mental health survey among ethnic Albanians in Kosovo from August 20 to October 7, 1999. The survey focused on the period of August 1998 through August 1999, when most of the intense violence took place.

Assuming a true prevalence of 20% of mental health–related problems⁸ and a cluster sample design effect of 2, we estimated that a minimum of 1135 adults aged 15 years or older would be required for a 95% confidence interval (CI) to detect a prevalence between 15% and 25%. On the basis of available household size and age distribution, we estimated that a minimum of 504 households would need to be surveyed. The number of households targeted was increased to 600 to compensate for refusals and absent adults and to obtain estimates for various subgroups of the population.

We conducted a 2-stage, 30-cluster sample survey using the 1991 Kosovo census as a primary sampling frame. Because these data did not reflect population movements before and during the ethnic conflict, additional data sources were used to adjust the 1991 population figures. These sources were village surveys from the United Nations High Commissioner for Refugees and food distribution population estimates from Action Against Hunger (a nongovernmental organization), both reflecting information collected during the weeks before our survey. The primary sampling frame consisted of all villages and cities listed in the 1991 census, excluding those that were predominantly populated by Serbs (≥70% Serb population) and those that had a population of less than 100 Albanian inhabitants. The sampling frame was stratified into urban (cities with a population >10,000) and rural areas. Using this sampling frame, we estimated the total ethnic Albanian population in Kosovo to be 1.6 million. With probability proportional to population size, we selected 15 clusters from the rural and 15 from the urban frame in the first sampling stage. In the second stage of sampling, 20 households were randomly selected within each chosen cluster (20 households from each of 30 clusters for a total of 600 households) using an appropriate method designed for the Expanded Programme on Immunization and adapted to the particular field conditions.⁹

Identification of cluster samples differed for urban centers and rural villages. No maps were available for the villages, and many villages were spread out over a large geographic area. We drew maps of each cluster, which were then divided into segments of approximately equal populations. We then randomly chose a single segment by first numbering all segments and then blindly drawing a segment number from a bag containing all numbers. In the cities, Kosovo Force (KFOR) offices usually had aerial or other maps available. In these cases, we superimposed a grid to partition the map into neighborhoods. The neighborhoods were numbered, and then a number was blindly chosen to randomly select a neighborhood for our survey.

After a segment or neighborhood was chosen, the first household to be surveyed was chosen randomly as follows. Households were mapped and numbered in a random direction from the center to the edge of the segment, chosen by spinning a bottle. The first household was chosen by blindly drawing a number from a bag using the same method described above. The next house was selected to be the closest house to the left, as the interviewer exited the house just surveyed. This process was repeated until 20 households were surveyed, or until the team leader decided it was time to leave for security reasons.

We interviewed all adult members of the household present. To ensure as much privacy as possible, we encouraged people to complete the questionnaires in separate rooms, and men and women interviewers paired up with same-sex interviewees to help them complete the questionnaires. A security curfew at dusk imposed by KFOR prevented interview teams from coming back to survey adults not present during the day. Because of the ongoing threat of land mines, KFOR considered access to some remote houses unsafe. These homes had to be excluded from our sample and replaced by the closest accessible household.

Native Kosovar Albanian survey team members had 3 days of training on general survey objectives, safety precautions, procedures for proper household selection (including randomly selecting the first household and handling special situations), and interviewing techniques (understanding the questionnaires and addressing sensitive topics). All members of the survey team were closely supervised for the first 2 days, and they continued to receive daily supervision and instruction until the survey was completed. Interviewers were instructed to refer participants who appeared to be in obvious distress to community mental health services where available. A list of these services was procured from the nongovernmental organization coordinating office at the United Nations Mission in Kosovo.

The study protocol was reviewed by a CDC institutional review board representative and informed consent was obtained verbally from all participants (with communication occurring in the potential participant's native language). The study protocol was also reviewed by Doctors of the World for ethical considerations.

All instruments used in this survey were designed as self-report questionnaires, but because of a high percentage of illiteracy, especially in rural areas, questionnaires frequently had to be read aloud. Because of the need for expediency in collecting data, interviewers were instructed to read the questionnaires only to those who were illiterate, and to provide assistance if needed to those who completed the questionnaire themselves. We used 3 screening tools to assess mental health problems and social dysfunction: the General Health Questionnaire-28 (GHQ-28),^10- 11 the Harvard Trauma Questionnaire (HTQ),¹² and the Medical Outcomes Study 20 (MOS-20).¹³ We chose these instruments to obtain information on common, nonspecific psychiatric problems, to gather information on specific psychiatric syndromes such as posttraumatic stress disorder (PTSD) and related traumatic events, and to get a broad understanding of the level of social functioning and disability in this population.

The GHQ-28 is used as a community screening tool and for the detection of nonspecific psychiatric disorders among individuals in primary care settings.¹¹ A higher mean score on the GHQ-28 represents poorer mental health status (score range, 0-28). The GHQ-28 is composed of 4 subscales (score range, 1-7): somatization, anxiety, social dysfunction, and depression. The HTQ combines the measurement of trauma events (part I) and symptoms of PTSD (part II), selected from the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV).¹⁴ We defined the occurrence of PTSD symptoms according to a scoring algorithm proposed by the Harvard Refugee Trauma Group,^4,12 on the basis of DSM-IV diagnostic criteria. The MOS-20 consists of 20 items on 6 different scales that assess physical functioning, bodily pain, role functioning, social functioning, mental health, and self-perceived general health status. We scored the MOS-20 as recommended in the user's manual; each raw score was transformed to fit a 0-to-100 scale using a standard formula,^13,15 with the higher scores on this scale representing better functioning. All 3 tools have been extensively validated in many countries and cultures and in many disease settings.^16- 18

To assess the effect of broadly defined demographic characteristics on mental health status, we collected demographic information including age, sex, education level, and marital status. We added additional questions specific to the Kosovar Albanian population on feelings of hatred and a desire for revenge. All questionnaires were translated into Albanian and back-translated to English to ensure cultural appropriateness of the instrument and accuracy of the translation. A team of Albanian translators including a psychiatrist, a psychologist, and a primary care physician from the Institute for Mental Health and Recovery did the translation and adaptation of the screening tools.

We adjusted prevalence estimates and CIs for cluster sampling and stratification using Epi Info version 6.4.¹⁹ Regression analyses were performed using SUDAAN, release 7.5.2 (Research Triangle Institute, Research Triangle Park, NC). For continuous variables, we used multivariate linear regression models to assess the effects of exposure on outcome and multivariate logistic regression models to analyze dichotomous outcomes. When the exposure variable had more than 2 levels (eg, displacement), we made multiple comparisons of the responses between pairs of the different levels using single df contrasts. When the exposure variable had a natural ordering (eg, age, education, number of traumatic events), we did a test for linear trend. All P values were derived from adjusted Wald F tests based on these regression models, and P<.05 was considered statistically significant. All analyses were adjusted for stratification and the clustered design, and were weighted to account for unequal selection probabilities among the individual respondents.

A total of 558 households, consisting of 1358 adults aged 15 years or older, were included in the survey (mean [SD] household size for all ages, 7.3 [3.5] persons). This is smaller than the target number of 600 households since logistical and time constraints prevented the completion of 20 surveys in some villages. However, 558 households is still greater than the 504 households deemed needed from sample size calculations.

Demographic characteristics are summarized in Table 1. Of the adults surveyed, 62.3% were women, 55.8% lived in a rural area, 59.5% had completed only primary school or less, 67.3% were married, and only 15.1% were currently employed. Nearly 41% of participants reported having a chronic illness (diagnosis by a medical professional of hypertension, diabetes, cardiovascular disease, kidney disease, asthma, epilepsy, cancer, or major injury such as loss of a limb), and 1.7% reported having received a diagnosis by a physician of a previous mental illness, such as schizophrenia or bipolar disorder, before the conflict.

The exposure to traumatic events, including displacement, is summarized in Table 2. High percentages of respondents reported having personally experienced traumatic events. For example, 66.6% reported being deprived of water and food, 66.5% reported being in a combat situation, and 61.6% reported being close to death. Furthermore, 39.4% of participants reported experiencing 8 or more of the traumatic events listed; 56.2% had fled to another country as refugees during the past year, 25.6% had been internally displaced within Kosovo, and only 18.2% remained in their homes during the war. In all analyses, the traumatic events were equally weighted since we had no resources for in-depth questioning needed to provide additional information.

Estimated mean scores on the GHQ-28 and the MOS-20 and the prevalence of PTSD symptoms from the HTQ are shown in Table 3, along with 95% CIs adjusted for stratification and cluster design effects. These figures represent estimates of the population indicator measured by each test for the adult Albanian population living in Kosovo at the time of this survey.

For the GHQ-28, the estimated mean total score based on a possible 28 questions was 11.1 (95% CI, 9.9-12.4). A higher mean score signifies a greater number of symptoms. The mean scores for somatic symptoms and for anxiety and insomnia were higher compared with the mean scores for social dysfunction and depression.

The estimated MOS-20 mean scores are shown on a scale of 1 to 100, with a higher score representing better functioning. In general, respondents tended to score highest on physical functioning and role functioning and lowest on the mental health and social functioning components (Table 3). We compared scores on the MOS-20 with scores of a US general population^14,20- 21 (data for the Albanian Kosovo population before the conflict are not available). The mean scores for mental health (29.6) and social functioning (29.5) were strikingly lower for the Kosovar Albanians than for the US population (74.7 and 83.3, respectively). However, there were no great differences between the 2 populations in the measures of general health, physical functioning, bodily pain, and role functioning. The estimated prevalence of PTSD symptoms in this population of Kosovar Albanians was 17.1% (95% CI, 13.2%-21.0%).

Questions regarding hatred toward the Serbs and desire for revenge revealed that high percentages of both men and women (>88% among each) had strong feelings of hatred, defined as a response of "extreme hatred" (men, 60% [n = 288]; women, 55% [n = 464]) or "a lot of hatred" (men, 29% [n = 142]; women, 35% [n = 271]). The proportions of people having strong feelings of revenge were lower (> 43% for both men and women), but still very high. Strong feelings of revenge were defined as a response of feeling revenge "all the time" (men, 35% [n = 159]; women, 23% [n = 192]) or "a lot of the time" (men, 16% [n = 92]; women, 20% [n = 166]). Of those men and women who had feelings of revenge ("all the time," "a lot of the time," or "sometimes"), 44.2% of men (n = 177) and 33.3% of women (n = 197) said they would definitely act on those feelings, and only 17.3% of men (n = 71) and 26.2% of women (n = 184) said they would not act on those feelings.

Table 4 summarizes the univariate analysis of the effect of selected demographic factors and exposure to trauma on the mental health and social functioning outcomes. We present the results of the GHQ-28 total score, estimated prevalence of PTSD symptoms, and MOS-20 social function scale as outcome measures in relation to various demographic and trauma experience measures. P<.05 was considered significant for univariate and multivariate analyses. Being older, being currently unemployed, being widowed, having little education, reporting a previously diagnosed psychiatric illness, and reporting a previous diagnosis of a chronic health condition were associated in this analysis with a high (eg, worse) GHQ-28 score, indicating nonspecific psychiatric morbidity. Similarly, living in a rural setting, being currently unemployed, being older, having little education, and reporting having received a diagnosis of a chronic health condition were associated with a low (eg, worse) social functioning score. Finally, HTQ results indicate that being female and having received a diagnosis of a chronic health condition were associated with PTSD symptoms.

Most traumatic event variables (forced separation from family, murder of family or friend, and increasing number of traumatic events) but not rape were associated with a worse score in the 3 measured mental health outcomes, with the exception of forced separation for social functioning. The association between rape and psychiatric morbidity and social functioning may be difficult to observe here because of the relatively small number of reported rape cases.

Since we had identified 2 different groups of explanatory variables, demographic and exposure, we treated these differently using a multivariate analysis. First, the effect of each demographic variable on the mental health outcomes was adjusted for all other variables, both demographic and exposure (Table 5).

Subpopulations at risk (statistically significant as measured by the multivariate analyses) for psychiatric morbidity as measured by GHQ-28 scores were those aged 65 years or older, those with previous psychiatric illnesses, and those with self-reported chronic health problems. In the multivariate analysis, employment, location, sex, marital status, and education were not statistically significant risk factors for psychiatric morbidity. Subpopulations at risk for poor social functioning, as measured by the MOS-20, were people living in rural areas, those currently unemployed, and those with chronic health problems. There was no significant decrease in social functioning with increasing age or education status when adjusted for all other variables. Women and persons with a previous psychiatric illness had a significantly higher estimated prevalence of PTSD symptoms.

To analyze the effect of exposure variables on mental health outcomes, we performed a second multivariate analysis for which all P values for the relationship between each exposure variable and each outcome measure were adjusted for all demographic variables, previous psychiatric illness, and chronic health condition (Table 6). People who were internally displaced tended to have higher total GHQ-28 scores than refugees (P = .03) or those who did not move (P = .009). However, there was no significant difference in the total GHQ-28 scores between refugees and those who did not move (P = .50), and the displacement seemed to have no effect on significance for MOS-20 social functioning scores or the prevalence of PTSD symptoms, when adjusted for the effects of the demographic variables.

There was a significant linear increase in total GHQ-28 scores (P<.001), a significant linear decrease in MOS-20 social functioning scores (P = .02), and a significant linear increase in the prevalence of PTSD symptoms (P<.001) with increasing numbers of trauma events (Table 6). Specific traumatic events seemed to be closely related to specific mental health conditions. People experiencing forced separation from family or murder of a family member or friends had significantly higher total GHQ-28 scores and significantly higher prevalence of PTSD symptoms than people without these experiences. People experiencing murder of a family member or friend also had significantly lower MOS-20 social functioning scores.

A rape experience seemed to have no effect on GHQ-28 scores, MOS-20 social functioning, or prevalence of PTSD symptoms, although, as stated earlier, a relationship may be difficult to observe due to the relatively small number of reported rape cases.

There was a high prevalence of traumatic events (Table 2) among the Kosovar Albanians, and large numbers appear to have experienced multiple traumas. Higher levels of PTSD symptoms, an increase in nonspecific mental morbidity as measured by the GHQ-28, and a decrease in social functioning were associated with higher levels of cumulative trauma. These relationships remained even after adjusting for the effects of demographic variables, previous psychiatric illness, and other chronic health conditions. Our results are consistent with those of other studies.^22- 24 Although the 4 subscales of the GHQ-28 provide information on types of symptoms, they have not been designed to make a psychiatric diagnosis. They do, however, give information on the mean scores for somatic, anxiety, social dysfunction, and severe depression symptoms (Table 3). It has been shown in other studies that the 4 subscales are not independent from each another.¹¹ In our study, the mean scores for somatic symptoms and anxiety and insomnia were higher than those for social dysfunction and severe depression. It is possible that in this culture depression is more likely to be expressed as somatic and anxiety symptoms. Alternatively, despite the traumatic events experienced by many people by the time of the survey, there may have been a genuine sense of hope and optimism because the war had ended, and people were rebuilding their homes, lives, and country.

The optimal threshold score to determine prevalence of psychiatric morbidity from the GHQ-28 has not been established for this population. Although we found that the GHQ-28 was well accepted and easy to administer, the interpretation of the results for prevalence estimates is not straightforward unless an optimal cutoff score is established for the specific population. Goldberg et al²⁵ have suggested that a mean score will provide a rough guide to the best threshold; however, this would always result in a general psychiatric morbidity prevalence of approximately 50%. Adopting a similar method with a conservative cutoff score of 11/12 out of 28 (so that those answering positively to 12 questions would be considered a "case"), we found an estimated prevalence of nonspecific psychiatric morbidity of 43%. In studies of general populations in 15 different countries, the highest cutoff score found was 6/7.^26- 28 However, no cutoff scores have been published for refugee populations or those recently exposed to war, where it is likely that the prevalence of nonspecific psychiatric morbidity is much higher than in general populations.

A similar type of cutoff score is needed to estimate the prevalence of psychiatric morbidity using the MOS-20 in refugee populations. In the US population, a cutoff score of 52 (range, 0-100) was established based on studies of the relationship between mental health and clinical measures of the probability of any psychiatric disorder.¹³ Using the same cutoff score for the Kosovo population would result in an estimated prevalence of psychiatric disorder of 83.5% vs 13.2% in the US population.²⁰ Further clinical validation of the GHQ-28 and the MOS-20 is under way to establish the best thresholds for the Kosovar population. The estimated prevalence of PTSD symptoms (17.1%) is somewhat lower than the reported PTSD figures (26.3%) for Bosnian refugees living in Croatia.⁴

The findings from the GHQ-28, MOS-20, and HTQ confirm earlier anecdotal reports that while the general health status of the Kosovo population remained fairly stable, mental problems related to the war situation are common. This is in line with other findings in refugee camps and war/conflict situations.^3- 7 No baseline general mental health status data from before the war are available for Kosovo. However, in our survey, self-reporting of previous mental illness (1.7%) correlated with findings in other populations.²⁹

We identified several subpopulations at risk for poor mental health status and social functioning and we also attempted to identify mitigating factors. In general, Kosovar Albanians younger than 35 years old, in good physical health, and without previous psychiatric illness appear to have been protected from war-related psychiatric morbidity. Future research will have to determine whether there are other protective factors that could be influenced by policy (eg, adequate housing, social and community support). Social functioning was significantly lower among the population in rural areas; however, location did not seem to have the same effect on general mental health. It is possible that the extensive disruption of the civic infrastructure in the rural areas made it harder to function socially than in cities, but closer family ties in these areas mitigated mental health problems. Not unexpectedly, people with previous psychiatric illness had worse mental health outcomes, including higher levels of PTSD symptoms, than did those without such illness. Similarly, indication of a previously diagnosed chronic health condition was associated with general psychiatric morbidity and social functioning but not PTSD.

As measured by the GHQ-28 scores, people who were internally displaced had worse mental health status than did refugees and those who never moved. In fact, a subsequent analysis revealed that on the average, those who did not move experienced a mean (SE) of 5.36 (0.53) traumatic events, while refugees experienced an average of 6.87 (0.33) and those internally displaced an average of 8.02 (0.56). This difference was statistically significant (P = .01). Virtually all people who were internally displaced were being persecuted, and as a result of this suffered continuous trauma. People who became refugees faced similar traumatic events, but usually of shorter duration because they were able to escape to other countries. It can be hypothesized that people who never moved away from their homes were able to stay because they happened to be in relatively safer areas and thus experienced less trauma.

There are a number of limitations to this study. Women were overrepresented in our sample probably because they were more likely to be at home during the daytime (data from other sources³⁰ indicate that the male-female ratio in Kosovo is close to 1). People who were employed during the time of the survey were less likely to be home during the day. Because of security curfews it was not possible to return to homes and interview those who were absent during the day. There is a possibility that some people who were the most stressed, because they were living in the most dangerous areas, were excluded from our study. However, if at all, this exclusion happened very seldom and would have resulted in underreporting of mental morbidity. Our study might be somewhat limited in statistical power since resources were available to sample only 30 clusters. However, the potential reduction in statistical power may have been moderated by the use of a stratified design.

Because no structured clinical interviews were performed, it is unclear to what extent self-reported symptoms of PTSD and nonspecific psychiatric morbidity, in the HTQ and the GHQ-28 respectively, would match clinical diagnosis. It is possible that cross-cultural differences could have influenced the results of this study. Even though the screening instruments used were created and validated in developed nations similar to Kosovo, the instruments were not specifically validated for this society. However, the GHQ-28 has proven to be a reliable instrument in a wide variety of cultures. The HTQ traumatic events section was specifically adapted for the Kosovo situation.

Although not traditionally part of a mental health survey, the questions regarding feelings of hatred and a desire for revenge give a poignant picture of all-too-common emotions in this setting. These findings underscore the challenge faced by the interim government of the United Nations Mission in Kosovo as it to seeks to establish reconciliation among different ethnic groups.

Whether measured by the prevalence of nonspecific psychiatric morbidity (43%), social dysfunction, or prevalence of PTSD symptoms (17.1%), our study demonstrates the severity of mental health problems among Kosovar Albanians.

When we conducted this survey the war had just ended. The wounds of war were still fresh, including the events that had shaken the lives of hundreds of thousands of people. Violence and acts of revenge continue in Kosovo. On the basis of the results of our survey, these incidents are not surprising. Mental health problems related to the psychological trauma of war and conflict situations are a major public health concern. The high rates of poor mental health status among those internally displaced and refugees who have returned to Kosovo also raises concern for the mental health status of those who remain in countries of asylum and resettlement.

Until the psychological and social effects of the war and persecution in Kosovo are evaluated over time, we must exercise caution in basing future predictions on the results of our survey. Follow-up studies and monitoring of mental health problems to determine long-term effects of multiple, prolonged, and severe traumatic events among the Kosovar Albanian population will provide more accurate data for policy recommendations.