Design and procedures

The design of the study was similar to that of the two earlier studies of this cohort. ^{Reference Hauff1} An interview administered in the respondents' home (by A.B.G. and A.B.V.) included a self-report questionnaire available in Vietnamese and Norwegian, and a structured face-to-face interview in Vietnamese. The study was approved by the regional committee for medical research ethics and the Norwegian Social Science Data Services. Written information about the study was provided in Vietnamese and Norwegian.

Sample

All Vietnamese refugees aged 15 years or older who arrived in south-east Norway from transit camps in south-east Asia during an 8-month period in 1982 were invited to take part in the first study (T ₁). Norwegian merchant vessels had rescued the refugees by chance from the South China Sea, and these people were offered resettlement in Norway. Thus the original cohort was a relatively unselected sample from this wave of Vietnamese ‘boat people’. In the current study (T ₃; 2005–6) we sought to interview all surviving members of the cohort remaining in Norway. We were able to collect data from 80 respondents; 24 others had died or moved to another country (Fig. 1). Hence the T ₃ sample represented 57% of the original cohort and 74% of those eligible for inclusion.

Fig. 1

Flow diagram of inclusion and attrition.

The Norwegian normative sample, used for comparison of the Symptom Checklist–90–Revised (SCL–90–R) results, consisted of 466 men and 507 women over 19 years old who were representative of the adult population in Norway. ^{Reference Vassend, Lian and Andersen6}

Assessments

Statistical analysis

Univariate analyses included chi-squared tests, Mann–Whitney tests, t-tests and Pearson correlations. Changes in SCL–90–R scores over time were assessed by linear mixed effects models (R version 2.9.7 for Windows, R package ‘nlme’, R Foundation for Statistical Computing, see www.r-project.org), ^{Reference Pinheiro and Bates13} adjusting for relevant T ₁ variables. The relationships between SCL–90–R scores at T ₃ and baseline (T ₁/T ₂) variables were investigated by two types of multiple linear regression analyses. First, we generated a model identical to the one published for T ₂ data, ^{Reference Hauff and Vaglum2} except that the outcome was the SCL–90–R GSI score at T ₃ (independent variables measured at T ₁ – or, if not assessed at T ₁, at T ₂ – were age, gender, years of education in Vietnam, an additive score of extreme traumatic stress, close confidante upon arrival, high-impact negative life events during the first 2 years in Norway, and separation from close family at T ₂). Second, for a more comprehensive model predicting GSI at T ₃, we included some pre-specified baseline variables from T ₁ and T ₂ together with two indices of mental health status prior to T ₃, namely the GSI score at T ₁ and a change measure (GSI at T ₂ minus GSI at T ₁, with a positive value indicating worsening of symptoms from T ₁ to T ₂). The significance level was set at 0.05. When adjusting for multiple comparisons the Benjamini–Hochberg procedure was used. ^{Reference Benjamini and Hochberg14} All analyses used SPSS version 15 for Windows and R (see above).

Sociodemographic variables

The sample at T ₃ consisted of 12 women and 68 men, with a mean age of 47.5 years (s.d. = 6.8, range 38–70), the respondent group being younger and comprising more men than women compared with the attrition group (further details are given in online Table DS1).

Psychological distress

All but one of the 80 respondents completed the SCL–90–R at T ₃ (the exception was a person with psychosis). There was no significant gender difference in mean GSI scores at T ₃ (women 0.48 v. men 0.49). The mean GSI decreased significantly from 0.81 at T ₁ to 0.49 at T ₃ (difference: s.d. = 0.55, 95% CI 0.29–0.44, P<0.001).

Table 1 shows the GSI mean scores at the three time points compared with the Norwegian normative sample (for subscale scores see online Table DS2). From arrival to follow-up at T ₃ there were significant decreases in both the GSI and the means for all subscales of the SCL–90–R (online Table DS3), with the exception of somatisation and anger/hostility. Eighteen per cent of respondents (n = 14) reached threshold scores at T ₃ (GSI = 1.00), more than twice the percentage of Norwegians (7.2%). For the Vietnamese cohort there was no significant difference between GSI rates at T ₃ and those recorded at T ₂ (25%, n = 20) and T ₁ (26%, n = 20).

Table 1

Global Severity Index scores for the study cohort (n = 79) at the three assessment points compared with the Norwegian normative sample

	SCL—90—R GSI score: mean (s.d.)
	Vietnamesea,b			Norwegianc	Respondents scoring > 1 at T 3, %
	1982	1985	2005-6		Vietnamese	Norwegian
Total	0.81 (0.45)	0.75 (0.55)	0.49 (0.49)		18	7.2
Male	0.76 (0.41)	0.73 (0.53)	0.49 (0.49)	0.32 (0.36)d
Female	1.04 (0.60)	0.90 (0.64)	0.48 (0.49)	0.41 (0.43)

Prediction of psychological distress at T ₃

In the first regression analysis using T ₁/T ₂ predictors and GSI at T ₃, the model being similar to the one published earlier, ^{Reference Hauff and Vaglum2} the included predictors explained 13% of the variance of GSI but no single predictor was associated significantly with the index of distress at T ₃. In the second regression, a greater GSI score at T ₁ and the change score (the difference in GSI scores between T ₂ and T ₁) were both significant predictors of a higher GSI at T ₃ (Table 2).

Table 2

Multiple regression analysis for Global Severity Index scores at T ₃ (n = 79)

Variables from T 1/T 2	Regression coefficient	P	95% CI	R 2
Age at inclusion	-0.001	0.86	-0.02 to 0.01	0.008
Gendera	0.11	0.44	-0.18 to 0.40	0.009
Vietnamese contactb	0.14	0.19	-0.07 to 0.34	0.049
Norwegian contactc	0.09	0.54	-0.20 to 0.37	0.062
Healthd	-0.003	0.98	-0.23 to 0.22	0.10
GSI score at T 1	0.48	<0.001	0.23 to 0.73	0.16
GSI difference (T 2 - T 1)e	0.41	0.001	0.19 to 0.63	0.31

Implications for the treatment of refugees

Our findings may help to clarify the trajectory of symptoms among refugees over a prolonged period of resettlement. In stable resettlement environments, refugee populations tend to manifest high levels of distress over the earlier years, as indicated by the results for T ₁ and T ₂. This is consistent with other studies, ^{Reference Lie4} and supports the notion that a combination of recent trauma and resettlement/acculturation challenges continue to act as stressors during this time. Nevertheless, the T ₃ data indicate that at a population level, and in the context of a stable resettlement environment, ^{Reference Schweitzer, Melville, Steel and Lacherez16,Reference Porter and Haslam17} most refugees have the capacity for symptom improvement over the longer term, although levels of distress may remain higher than among the host population. Symptom improvement during the first 3 years, signified in this study by a reduction in GSI between T ₁ and T ₂, may therefore represent an important predictor of a positive long-term prognosis, in this instance after 23 years.

Prediction of outcomes: implications for early assessment and intervention

A minority of study participants remained psychologically impaired over many years. Trauma might have been a factor initiating some of these reactions, but in the longer term, as indicated, a key predictive index appeared to be the level and trajectory of psychological symptoms during the period of early resettlement. This supports the potential value of serial screening programmes for refugees, a procedure that could be followed by comprehensive assessment of the identified at-risk group to detect those with clinical disorders in need of treatment. Hence, early intervention may be relevant not only to overcome immediate distress but also to avert risk of psychological difficulties in later years. Still, the predictive capacity of the regression analysis was modest, suggesting that other unmeasured post-migration experiences and/or personal vulnerabilities might have a role in perpetuating psychological symptoms over time, ^{Reference Marshall, Schell, Elliott, Berthold and Chun18} or in initiating new disorders during the ensuing two decades.

Limitations

Despite of the strengths of the study, in particular its prospective longitudinal design, important limitations need consideration. Attrition is evident in most refugee studies, ^{Reference Lie4} and compared with the attrition group, included respondents were younger and more often male than female. This, together with the small original sample and the special characteristics of Norwegian society, may limit the generalisability of the findings to other refugee populations. In addition, gender-specific comparisons could not be undertaken because the relevant information was not available for the Norwegian normative data-set.

The follow-up study did not include a clinical diagnostic interview, which limits comparisons with other studies of refugees. ^{Reference Roth, Ekblad and Agren19,Reference Hinton, Chau, Nguyen, Nguyen, Pham and Quinn20} Although a study among Vietnamese refugees has shown reasonable concordance between a clinical interview (the Structured Clinical Interview for DSM–IV Axis I disorders) and the HSCL–25, ^{Reference Smith Fawzi, Murphy, Pham, Lin, Poole and Mollica21} a comparable study among Cambodians indicated that a clinical interview yielded a more conservative prevalence rate. ^{Reference Silove, Manicavasagar, Mollica, Thai, Khiek and Lavelle12} A recent meta-analysis of the entire refugee mental health field showed a regular pattern in which questionnaires (often derived from the Symptom Checklist) on average returned a 10–13% higher prevalence than interviews when other methodological factors (sample size, approach to sampling) were taken into account. ^{Reference Steel, Chey, Silove, Marnane, Bryant and van Ommeren22} These observations caution against inferring that all Vietnamese who scored above threshold in our study could definitely be assigned to the ‘cases’ group. Further examination would be needed (and warranted) to specify accurately those in need of treatment.

We acknowledge the risk of reduced semantic equivalence arising from the use of an early version of the Vietnamese translation of the SCL–90–R, ^{Reference Flaherty, Gaviria, Pathak, Mitchell, Wintrob and Richman23} but we considered it important to use the same translation for longitudinal comparisons. Other studies among Vietnamese have used a culturally specific measure of mental health, the Phan Vietnamese Psychiatric Scale (PVPS), in conjunction with an international instrument. ^{Reference Phan, Steel and Silove24} This approach increased the overall rates of detected mental health disorders, suggesting the possibility that cases may be overlooked when Western-derived instruments such as the SCL–90–R are used alone.

Clinical implications

Although self-reported psychological distress among Vietnamese refugees decreased significantly over time, even after 23 years of resettlement a greater proportion of this cohort reached threshold SCL–90–R scores compared with the native Norwegian group. Therefore, this population should still be considered to be at increased risk of mental health problems, justifying special attention from psychiatric services. The longitudinal data support the importance of screening refugees in the early years of resettlement, since elevated levels of psychiatric symptoms during that period appear to indicate long-term risk.