HADStress: A Somatic Symptom Screen for Posttraumatic Stress Among Somali Refugees

Between 1990 and 1992, uprisings occurred in several countries of East Africa. Each nation's upheaval involved certain unique aspects. In Somalia, clans that were out of governmental, commercial, and artistic power rose up against clans holding power. This action involved sudden, coordinated invasion into government offices, commercial enterprises, and other public institutions, with summary executions of adult men. Consequently, many men fled the country, leaving their families behind. Over time, clan violence involved fewer men and more women, who either remained in their homes or were fleeing the country when they were apprehended. This study was planned in 1998 and 1999; data were collected from 2000 to 2003 (about ten years after the original traumatic experiences). This study was approved by the institutional review board at the University of Minnesota.

Study participants included 622 Somali refugees living in the Minneapolis-St. Paul metropolitan area. A nonprobability sample was obtained because of the impossibility of identifying all potential participants to conduct a random sample. A detailed description and demographic evaluation of this representative sample have been published previously ( 16 ). The method of sampling Somali refugees depended heavily on targeted cluster sampling (that is, neighborhoods and demographic clusters). Based on data drawn from refugee resettlement records, birth records, and registered school children, the sample included approximately 8% of Somali adults in the community. The participation rate of those contacted was 97% ( 14 ). Study participants varied in their time of residence in the United States, ranging from a few days to several years. However, most of them had been in the United States for two or three years (mean±SD of 3.5±3.0). Among the 622 participants, 323 (52%) were male and 299 (48%) were female. The mean age was 37.1±15.3 years. Nineteen (3%) of the participants were living outside of Somalia during 1990–1992; they were attending school or employed in other countries.

Two symptoms, headache and sleep disturbance, were selected for the HADStress screen because of their common prevalence around the world in clinical samples ( 15 ). In addition, the prevalence of these symptoms has been shown to be dramatically different in "healthy" populations and in traumatized populations with PTSD or other psychiatric disorders ( 6 ). Dizziness was selected because it is not a common complaint in primary care settings ( 15 ) or even among combat veterans ( 6 ), but 22% of 966 refugees fleeing to Israel reported dizziness ( 17 ). Dizziness is also the most common cardiovascular symptom among women with somatization disorder ( 18 ). Headache and dizziness could both be related to brain damage or to emotional distress, whereas sleep problems and appetite change tend to be related to emotional distress. These four symptoms did not have culturally specific meanings in East Africa, unlike, for example, the term "pain in the heart," which can serve as an expression of loss or sadness in East Africa ( 19 ). All four symptoms occurred within the broader Somali lexicon of somatic symptoms ( 20 ). Although translations, linguistic equivalence, and cultural connotations can affect endorsement of symptoms, variability in translation of somatic symptoms tends to be minimal compared with that of emotional terms, such as anxiety or depression ( 21 ). All study participants completed the screen, and we received no complaints about these queries. Participants were also asked about whether they associated physical symptoms with thinking about their torture experience.

The PTSD Checklist (PCL) was used to assess the severity of posttraumatic stress symptoms ( 22 ). The PCL is a 17-item self-report Likert scale, with responses ranging from 1, not at all, to 5, extremely. Total possible PCL scores range from 17 to 85, with scores ≥50 reflecting more severe posttraumatic symptoms. This instrument was translated into the Somali language by speakers fluent in Somali and English and was then back-translated into English by another group also fluent in Somali and English. Discrepancies were addressed, and the process was repeated until the back-translation accurately reflected the PCL items. We then conducted a pilot study of the translated version in a sample of Somalis to assess the instrument's performance. In this pilot study, participants were queried about any difficulty in understanding and responding to the PCL items.

Items on torture were collected during interviews and pilot studies in the first year of the study, before the survey data collection began. Thus each item was presented to each study participant. Items were compiled in a questionnaire that included 74 types of torture: 41 types of physical torture, 19 types of psychological torture, six types of sensory torture, and eight types of sexual torture (two of which were gender specific). Some items could be classified under more than one category; questionable items were included in the category with which they were most often associated.

Trauma was categorized as torture if it occurred while the person was detained by and under the control of the authorities before they fled Somalia ( 14 ). Miscellaneous or "other" torture items received few endorsements by the 622 study participants, suggesting that the questionnaire reflected most of the torture experienced by this group. [An appendix showing the types of torture and the number of persons in our sample who experienced each type of torture is available as an online supplement at ps.psychiatryonline.org .]

The sample was also asked about nontorture trauma events. These events occurred during the same time frame, but they did not occur while Somalis were detained by officials. These events were related to civic unrest, armed conflict, and refugee flight out of the country (for example, insufficient food or water, engaged in combat, exposed to artillery barrage, assault, rape, attack by wild animals, saw others raped and murdered, loss of home, and loss of crops or other property).

In the descriptive analysis of the data, we used phi and Cramér's V tests to compare the somatic symptoms with one another and with scores on the HADStress screen. Spearman rho correlation coefficients were used to compare the number of torture types with the total number of torture types in each of the four subcategories (that is, physical, psychological, sensory, and sexual torture). Spearman rho correlation coefficients also compared the total number of torture and nontorture trauma items.

Univariate comparisons were analyzed with one-way analysis of variance and chi square tests. Using the Bonferroni correction (.05/5), a significance level was set at .01.

Because the HADStress screen is essentially a count variable with a Poisson distribution, a negative binomial regression model was used to examine the relationship between independent predictors (PCL score, gender, age, number of torture trauma types, and number of nontorture trauma types) and the HADStress score. Independent variables that showed a significance of .10 or less were entered simultaneously into the negative binomial regression. Probability of .05 or less was considered significant for the negative binomial regression.

Analysis of variance was undertaken to assess the group differences of HADStress scores with the PCL scores as the dependent variable. Post hoc comparisons were computed (Tukey-B test) to test for significant differences between individual scores on the HADStress screen and scores on the PCL.

Somatic symptoms. Among the 622 Somalis in our sample, 247 (40%) reported headaches, 100 (16%) reported appetite changes, 65 (10%) reported dizziness or faintness, and 106 (17%) reported sleep problems. The number of somatic symptoms reported by these 622 Somalis followed a skewed pattern. For example, 322 Somalis (52%) reported none of the four symptoms, 163 (26%) reported one symptom, 55 (9%) reported two symptoms, 43 (7%) reported three symptoms, and 29 (5%) endorsed all four symptoms.

Only 22 of the 622 Somalis (4%) reported that they had any physical symptoms in association with thinking about their torture experience. Thus, among most individuals with one or more of these symptoms, there was no conscious link to torture. The four study symptoms were associated with one another and with the total scores on the HADStress to a highly significant extent. As shown in Table 1, each of the symptoms was highly correlated with the others and with the HADStress scores (p<.001). On the basis of the effect size criteria detailed by McCarthy and colleagues ( 23 ), each of the four somatic symptoms had a "very large" effect (.722 to .876) on the HADStress scores.

Torture and other traumatic experiences. In our sample of 622 Somalis, 431 (69%) reported that they had experienced one or more types of torture. The sample indicated that they had experienced a mean±SD of 2.1±2.8 physical torture items (range=0–22, skewness=2.6), 1.3±2.0 psychological torture items (range=0–13, skewness=2.0), .5±1.4 sensory torture items (range=0–9, skewness=3.6), .1±.6 sexual torture items (range=0–4, skewness=5.5), and 4.0±5.7 total torture items (range=0–45, skewness=2.7). The Spearman probabilities between the four categories and the total score were all significant at p<.001 (with correlation coefficients ranging from .38 to .90). A total of 546 Somalis (88%) reported that they had experienced one or more nontorture trauma events (mean of 5.5±4.5 events, range=0–19, skewness=.7). The total number of torture types and total number of nontorture trauma types per individual were significantly correlated with each other (Spearman ρ =.64, p<.001).

Table 2 shows the univariate associations between independent predictors and the HADStress scores. Analysis showed that higher mean HADStress scores were associated with being older (p<.001), being female (p<.001), a higher number of torture types (p<.001), a higher number of nontorture trauma types (p<.001), and a higher PCL score (p<.001). Persons who reported having ever been tortured while in Somalia had higher HADStress scores (tortured: N=431, score of 1.4±1.4; not tortured: N=191, score of .5±.8).

All five variables (PCL score, gender, age, number of torture trauma types, and number of nontorture trauma types) were then entered together in the negative binomial regression model. As shown in Table 3, the PCL score was the most effective variable in predicting the HADStress score. Gender was strongly associated with HADStress scores, with women reporting more symptoms than men. Older age also predicted higher HADStress scores. The number of torture types and the number of nontorture trauma types were not significant and were thus excluded from further analysis.

Post hoc tests showed that the HADStress scores were associated with PCL scores. Persons with HADStress scores of 0 and 1 had similar PCL scores (PCL scores of 25.15±9.7 and 26.96±9.6, respectively), persons with HADStress scores of 2 and 3 also had similar PCL scores (PCL scores of 40.28±13.6 and 43.93±14.2, respectively), and persons with a HADStress score of 4 had a mean PCL score of 51.07±14.8. All three groupings of HADStress scores differed significantly from each other.

Some tortured refugees are reluctant to endorse having been tortured for a variety of reasons, ranging from distrust ( 24 ) to lack of familiarity with mental health services ( 2 ). Somatic symptoms appear to be a useful means of screening for the existence of posttraumatic stress symptoms, whether from torture or nontorture trauma. The HADStress screen can serve as a useful entry point for detecting posttraumatic stress symptoms among people who are reluctant to discuss their trauma experience with strangers. The strong association of somatic symptoms with current posttraumatic symptoms, and not simply with past trauma, on the generalized linear model demonstrates the clinical utility of this screen in identifying people who may be in need of additional clinical care.

Similarly, other investigators have observed higher rates of somatic symptoms in traumatized populations, even years after the trauma occurred. For example, a study of 21,244 treatment-seeking Gulf War combat veterans revealed that healthy but physically traumatized veterans (coming for routine check-ups with no known psychiatric diagnoses) reported higher somatic symptom rates a decade after combat exposure ( 6 ). These Gulf War veterans reported headache in 10% of cases and sleep disturbance in 9%.

Problems often result when patients with somatic complaints seek medical help in venues that do not recognize the origins of the patients' distress ( 25 ). Failure to recognize and treat somatization in primary care settings can lead to disappointed patients, frustrated clinicians, and greatly increased health care costs—although treatment for somatization can be effective in these settings ( 26 ). Delayed recognition and care has resulted in higher rates of comorbid psychiatric disorders and wasteful expenditure of health care dollars ( 17, 27 ). Complicating the diagnostic issue is the fact that groups with high rates of somatization have often experienced greater-than-normal rates of brain injury, which can mimic psychogenic somatization ( 28 ).

Economics, statistics, and practicality must guide the determination of a cutoff score on the HADStress screen. When a PCL score of 50 was used as indicative of severe symptoms, over half of those with a HADStress score of 4 could be expected to meet diagnostic criteria for PTSD ( 22 ).

In a clinical setting, with a symptomatic patient seeking care, the lower cutoff score of 2 would be more practical for several reasons. First, it would exclude the 78% of people in our sample who had a score of 0 or 1 on the HADStress scale and were thus unlikely to have increased posttraumatic stress symptoms, regardless of their trauma history. Second, within a few minutes the clinician could ascertain whether the patient's care warrants a more time-consuming evaluation, perhaps through evaluating the patient over several visits to establish sufficient rapport to elicit a relevant trauma history. Third, clinical screening differs from general population screening in that those in the clinical setting are more apt to be experiencing discomfort or disability than those in the general population. Thus, a lower cutoff score for a screening instrument is often employed in clinical settings as compared to screening in the population at large. Fourth, treatment for PTSD and for other common comorbid conditions, such as mood, anxiety, and substance use disorders, is most effective when undertaken in early stages, before patients are isolated or disabled or require inpatient care ( 29 ).

The situation in a public health setting, where most people are not currently seeking treatment, differs from the clinical setting. By using a cutoff score of 2 on the HADStress scale, about 1,600 of the adult Somali population in the Minneapolis-St. Paul metropolitan area at the time of the study (estimated adult population of 7,272) would warrant clinical evaluation. If only people with scores of 4 were evaluated, 360 of the entire Somali population in the metropolitan region would warrant clinical evaluation. This is a ratio of 4.4 to 1. From an economic perspective, using a cutoff score of 2 to determine which Somalis would require extra care would be many times more expensive, with a lower detection rate. In sum, a cutoff score of 2 may be highly relevant in a clinical setting but unrealistically expensive in a public health setting.

The association of more somatic symptoms on the HADStress scale with female gender could be due to the higher exposure of Somali women to torture, compared with Somali men, rather than to gender alone. It is also likely that the association between older age and a higher HADStress score was due to the delay between the trauma exposure in Africa and data collection in the United States. Some people who met the age criteria for inclusion in this study (that is, age 18 and above) were still children in the early 1990s, when the traumatic events were occurring in Africa. Thus the higher HADStress scores among older Somalis and Somali women may be due to historical-environmental factors, rather than to inherent demographic factors.

Each refugee group possesses characteristics common to all refugees (for example, feeling pushed out of the homeland and acculturation experiences). Likewise, every refugee group has unique characteristics (such as the level of trauma and subgroups targeted for torture). Thus the experience of this group cannot be extrapolated to all refugee groups.

In addition, we focused on only four somatic symptoms. Information about more symptoms might have been informative but would take longer to elicit. Effective screens include relatively few items. Because this study involved only one ethnic group, replication in another ethnic group should be undertaken.

This study was supported by grants R01-MH59579 and 1R01-MH59570 from the National Institute of Mental Health. The authors acknowledge the valuable contributions in advising, translating, recruiting, and interviewing by the following: Abdullahi Gas, M.D., Osman Ahmed, M.D., Israel Gobena, J.D., Johara Mohammed, Elizabeth Namara, Abulqadir Omar, M.D., and Nadifa Osman, M.S.

The authors report no competing interests.