The study took place in November 2006 in Gulu and Amuru districts of northern Uganda (Amuru district was separated out from Gulu district in July 2006). The two districts contain an estimated 650,000 IDPs which is approximately 40% of all IDPs in Uganda. Up to 80% of the districts' population live in camps which range in size from 1,100 to almost 60,000 [9
]. A cross-sectional survey design was used. The sampling population included adult (≥18 years old) male and female IDPs. IDPs were defined as people living in the officially recognised IDP camps in Gulu and Amuru districts. The study used the SF-8 as an outcome measure (with the findings presented elsewhere) and the sample size was calculated to detect associations of independent variables on this continuous outcome variable. The sample size required adequate power (80%) to detect conceptually important differences (0.8 standard deviation) in the health outcomes of different respondent groups within a multivariate analysis with a significance level of 5% with the size of the 'rarest' sub-group of respondents at 5% [11
]. Due to the cluster sampling method used, a design effect of 2.0 was included which doubled the required sample size [12
]. The expected proportion of unusable questionnaires was set at 10%. The resultant sample size required was calculated to be a minimum of 1080.
An adapted multi-stage cluster sampling method was used as random and systematic sampling methods were not feasible in the IDP camps due limited data on the population and the unsystematic layouts of the camps [12
]. The first stage was to randomly select the camps from which the clusters would be chosen. The sampling frame was a list of the total population of IDPs living in all the 65 officially recognised IDP camps in Gulu and Amuru districts [10
]. The data for this list was collected by the World Food Programme in August 2006 and considered to be the most accurate IDP population data available in the two districts. 32 clusters were chosen rather than the more common use of 30 clusters to reduce the design effect [13
]. The 32 clusters were selected using the probability proportional to size technique. This used the World Food Programme list of the 65 officially recognised IDP camps in Gulu and Amuru districts, with a corresponding running cumulative population size for each camp. Clusters were then allocated to camps proportionally to the camp population sizes following the probability proportional to size technique to ensure self-weighting [12
]. The 32 clusters were allocated to 28 IDP camps using this process. The total population living in the 28 selected camps was 452,702.
Due to the large population sizes of the selected camps, a second stage was used to randomly select administrative zones within the sampled IDP camps as second stage units to act as individual clusters. These were existing zones established by the camp authorities and were estimated to be of similar population sizes based upon discussions with camp and zone leaders, and so self-weighting was maintained. The third stage consisted of randomly choosing individuals from the selected clusters. As the clusters were already self-weighted, the same number of individuals were chosen from within each of the selected clusters. The Expanded Programme on Immunisation (EPI) method was used to randomly select households for this stage and one individual was then randomly selected from the eligible individuals within the household [13
Two study staff conducted stage 2 and 3 of the sampling process through a pre-visit the day before the actual data collection. In stage 3, if the randomly selected person was not present, another adult member of the household or a neighbour were asked to inform the selected person to attend the data collection visit. The random selection process was emphasised to avoid accusations of favouritism or risk of stigmatisation against the selected person. It was advised that replacements would not be accepted and precautions were taken to reduce risk of replacements. If a household had been deserted for more than 1 month, EPI methods were followed to select another household.
A questionnaire was developed consisting of items on the demographic and socio-economic characteristics of the respondents, and existing health measurement instruments to measure trauma exposure, PTSD and depression. A slightly adapted version of the original Harvard Trauma Questionnaire (HTQ) was used to identify exposure to trauma events [16
]. This consisted of 16 questions on life-time exposure to traumatic events with a 'yes/no' response. PTSD was measured using 30 questions on trauma symptoms with a 4 point severity scale and a recall period of 1 week. The first 16 items are based upon the Diagnostic and Statistical Manual for Mental Disorders, Fourth Edition
(DSM-IV), and the remaining 14 items developed specifically for conflict-affected persons [16
]. Mean PTSD scores ≥2.0 were considered significant for meeting symptom criteria of probable PTSD based upon the instrument standards [18
]. Scores for symptoms of probable depression were measured using the 15 depression items from the Hopkins Symptoms Check List-25 (HSCL-25) [18
]. This also has a 4 point severity scale and a recall period of 1 week. Mean depression scores ≥1.75 were considered significant for meeting symptom criteria of probable depression based upon the instrument standards [18
]. The 15 depression items are consistent with the depression items in the DSM-IV [17
]. The reliability and validity of the HTQ and HSCL-25 have been tested and proven for use with displaced persons in a number of countries [16
]. The questionnaire was translated and delivered in Luo, the main language of Gulu and Amuru districts. The translation followed recommended guidelines, and involved forward and back translation, and detailed review by the study team [16
A team of 15 data collectors were recruited for the survey (8 men and 7 women) who were all from the Acholi region of northern Uganda, spoke fluent Luo and English, and had experience of data collection in IDP camps in northern Uganda. Six days training was provided for the study. The data collection took place between 6 and 27 November 2006. Each interview took between 35 and 45 minutes approximately. A consent form was used to ensure informed consent and clarify that no direct benefit could be expected from participating in the study. All data collected was confidential and anonymous. Ethical approval for the study was provided by the Ugandan National Council for Science and Technology, Gulu University, and the London School of Hygiene and Tropical Medicine. As some of the questions were on traumatic experiences and mental distress, referral information for support on mental health was provided. One of the study team was a psychiatrist and one of the team leaders was a double trained Clinical Psychiatric Officer/Mental Health Nurse who could offer advice if required. Supervision and quality control were provided by the 3 members of the study team and 2 team leaders.
Two data entry clerks were used to enter the data into SPSS, version 14.0 (SPSS Inc, Chicago, USA). Each questionnaire was cross-checked by project staff and analysis conducted of the dataset to check for inconsistent data entries. Analysis was performed using STATA version 9.2 (Stata Corporation, College Park, Texas, USA) and adjusted for the clustered design. The Cronback α for internal consistency reliability was tested and estimated at 0.86 for the PTSD scale and 0.83 for the depression scale, above the generally accepted minimum threshold level of ≥0.70 for an internal reliability coefficient [25
]. Multivariate logistic regression was applied to produce odds ratios (OR) of associations between independent demographic and trauma exposure variables with outcomes of PTSD and depression and adjusted to address the influence of the other significant variables. Based upon the cut off levels given in the instrument guidelines, the outcome of PTSD was dichotomised into respondents exhibiting or not exhibiting signs of PTSD (cut off ≥2.0), and the outcome of depression dichotomised into respondents exhibiting not exhibiting signs of depression (cut off ≥1.75) [18
]. Continuous independent variables were also categorised for the analysis. All trauma exposure variables were included in the multivariate analyses. All demographic variables which were statistically significant (P
< 0.05) following a univariate analysis to test for strength of association were included for the multivariate analysis. The associations in the multivariate analysis which were statistically significant (P
< 0.05) using a backward elimination regression approach were included in the final results. Separate regression models were used for the association of independent trauma events and the cumulative events on the outcomes of PTSD and depression. Statistical interaction between independent variables was tested but none were significant (P