Study design and setting

This paper reports baseline findings of a 5-year prospective open cohort study involving a representative sample of conflict-affected people living in the former Gulu district, which was later sub-divided to create Amuru, Gulu, and Nwoya Districts, Northern Uganda. Baseline field activities were conducted between November 2011 and July 2012.

Sample size calculation

The required sample was calculated using the formula: \( n={\left(\frac{z}{m}\right)}^2\widehat{p}\left(1-\widehat{p}\right)\times \frac{deff}{R} \) [36]. This gave a total sample size adjusted to 2400 to take care of possible non-response (R) and sampling design effect (deff), assuming a design effect (deff) of 2 and a response rate (R) of 98.7% based on previous studies that were conducted in rural South West Uganda, ‘m’ being the level of precision desired. This sample size would produce a two-sided 95% confidence interval (Alpha = 0.05) with a width ranging from 2.0 to 2.3% around the point estimate when the overall prevalence of HIV infection ranges between 6.4 and 9.1% in all communities combined.

Selection of study communities and population

A two-stage stratified sampling method was used to randomly select three study communities in each district, one from each settlement category. All communities in the three districts were listed; all major settlements were mapped and categorized as either permanent, transient or displaced. Settlements created to accommodate IDP during the war were categorized as displaced communities; Settlements created to accommodate IDP returning to their overgrown gardens and destroyed houses were categorized as transient communities and settlements which were there before the war and residents were never displaced were categorized to as Permanent. Amuru district had 10 communities (6 permanent, 2 transient, 2 displaced); Gulu district had 16 communities (10 permanent, 2 transient, 3 displaced, 1 pilot community) and Nwoya district had seven communities (4 permanent, 3 transient, 0 displaced). A total of eight study communities were selected from 32 listed communities excluding the pilot community. One community was purposively selected for the purposes of piloting questionnaires and survey tools. Information obtained from residents of this pilot community were not included in the study analyses but was used to identify sensitivities individuals or communities had to some questions and to inform changes to the final survey tools. Communities with large population sizes (>250 adults) like Awach and Layibi were sub-divided by villages\divisions which were randomly selected to represent these communities and contributed household numbers proportional to size of the communities they represent as compared to the populations of other selected communities to fit within the estimated study sample size of 2400 individuals. Only two communities were selected in Nwoya district because it did not have any categorized as ‘displaced’. All households in the selected communities were mapped and a household census was completed to establish the number and the demographic characteristics of all individuals (N = 6375). A “take all” approach was used to survey all consenting individuals aged 13–49 years who had been residents in the selected communities for the last month. We used the 13–49 year age-group for two main reasons. 1) To be able to compare our prevalence estimates with the Uganda AIDS Indicator Survey 2011 and 2) to increase our power of estimating HIV incidence since HIV-infections mostly occur in this age range.

Data collection procedures

The study team was trained on the key components of the study protocol including the study objectives, consenting procedures, administration of the study questionnaire, counselling and the referral process for those respondents found to have HIV, syphilis, PTSD or depression.

Laboratory testing

Two ELISA tests, Vironostika HIV Uni-Form II plus O (Biomerieux SA, Marcy l’Etoile, France) and Murex HIV-1.2.O (Diasorin S.P.A, Dartford, United Kingdom) were used in parallel to test for HIV infection at the Uganda Virus Research Institute IAVI-laboratory. The Western Blot (Genetic Systems, Bio-Rad Laboratories) was used as a tiebreaker if the ELISA results were discordant. Indeterminate Western Blot results not resolved by the time of this analysis were excluded. Samples screening positive for syphilis using the rapid plasma regain (RPR) test were subjected to a confirmatory treponema pallidum haemagglutination test (TPHA).

Data collection tools

Blood specimen collection was used to determine HIV prevalence and was linked with a questionnaire assessing socio-demographic characteristics, conflict-related experiences and sexual behaviours. Questionnaires were translated into local language (Luo) through a process of forward and backward translation by an experienced team of health professionals working independently. Where there was a wide disparity, translations were discussed and revised to bring out the intended meaning. Given that some questions may elicit memories of significant trauma and victimization, participants who requested care were referred to Trauma Clinics in the study area. Part I and IV of the Harvard Trauma Questionnaire (HTQ) were used to measure post-traumatic stress disorder (PTSD). PTSD scores were calculated using the sum of all answered items in Part IV divided by the number of answered items. Scores ≥2.0 were classified as meeting the criteria for screening positive for PTSD [37–39]. The Hopkins Symptom Check List-25 (HSCL-25) was used to measure major depressive disorder (MDD). Participants with a mean score of ≥1.75 on Part II were classified as meeting the criteria for screening positive for MDD [37–39]. Luo versions of the HTQ and HSCL-25, as developed and validated by Roberts et al. for use in Gulu district, Uganda, were adopted [37, 38]. Calculation of scores on the HTQ and the HSCL was carried out using published guidelines for these instruments. Experiences of war events in Northern Uganda were collected using a 15-item War Trauma Experience Check-list (WETC-15) instrument. Scores on the WTEC-15 were dichotomized; ≤12 vs. >12 severe traumatic events based on a previous study by Roberts [21].

Statistical analysis

Data were entered in duplicate using Microsoft Access software and analysed using SAS 9.4 software (SAS Institute, Cary, NC, USA). Unadjusted and multivariable logistic regression models were used to determine variables independently associated with HIV infection. All factors including socio-demographic, medical, laboratory, psycho-social and war-related trauma associated with HIV sero-positivity at a ≤0.1 level of significance in the unadjusted models were entered into a multivariable model that adjusted for differences in age, gender, marital status, religion, education, and district of residence. A stepwise regression procedure that dropped, at each step, variables that did not reach the <0.1 level of significance was used to develop the multivariable models excluding factors considered to be on the causal pathway and including factors that were associated with HIV sero-positivity at p < 0.05 in the final model for each potential risk factor. A likelihood ratio test was used to check for interactions and determine the model that best fits the data. Unadjusted and adjusted odds ratios of prevalent HIV infection are reported. To account for the additional variance due to the complex two-stage sampling study design that included stratification by the three settlement types and the three study districts, and the unequal selection probabilities and clustering, survey analysis procedures in SAS were used. Communities were the primary sampling units (PSU) and the stratification variables were the settlement type and the district. The sampling weights were based on the selection probabilities at each level of selection and on the proportion of survey participants consenting for interview and HIV testing. For independent categorical variables, we included a missing value category to minimize list-wise deletion of observation in the models. All 61 indeterminate HIV test results (Outcome variable) were considered missing, and were excluded from the analysis.

Ethics, consent and permissions

Ethical approvals were obtained from the University of British Columbia-Providence Healthcare Research Ethics Board (Canada), Makerere University College of Health Sciences-School of Public Health Science Ethical Committee, Uganda Virus Research Institute-Science and Ethics Committee, and Uganda National Council for Science and Technology. The Office of the President of Uganda issued a letter, which was signed by the Resident District Commissioner in each district and from all participating district authorities. Informed written consent was also obtained from all eligible study participants aged 13–49 years after explaining the objectives and procedures of the study. Parental/legal guardian’s written consent for <18 years with minor’s assent were also obtained. Every participant was properly counselled and consented before being asked to sign or put a fingerprint on the consent form if they were not able to write. Study participants were assured of confidentiality before the start of each interview. All data collected was kept under lock and key only accessible to the research team.

Study population

From a household census of 6375 residents, 2976 individuals were not surveyed because they were less than 13 or greater than 49 years old, 434 were non-residents and 11 were absent or had out-migrated between the time of conducting the household census and the survey. A total of 2954 individuals were eligible, of whom 2449 (82.9%) consented to participate (Amuru district = 732, Gulu district = 1159 and Nwoya district = 558). A total of 61 records with indeterminate HIV Western Blot results were excluded from any analysis modelling HIV (Fig. 1).

Socio-demographic characteristics of the respondents by HIV sero-status

HIV infection and mental health

Socio-behavioural, sexual histories and STI related risk factors for HIV

There was no significant difference in HIV infection between participants who had ever been abducted during the war and those who had never been abducted (p = 0.578). Giving or being given money or gifts in exchange for sex in the past 12 months only attained borderline significance (p = 0.093); Participants who stayed in a child-headed household (p = 0.274); and men who were circumcised were less likely to have HIV but this effect did not attain statistical significance in the multivariable model.

Sexual experiences at first sexual debut

The median age at first sexual debut among 1955 study participants who reported ever having sex was similar between those with and those without HIV infection (16 years, IQR: 14, 18) but the number of lifetime sex partners was significantly higher in participants with HIV infection (median = 3: IQR: 2–5) than those without HIV infection (Median = 2: IQR: 1–3), p < 0.001.

HIV prevalence

Findings from this study highlight the enormity of the problem of HIV infection among people residing in Gulu, Nwoya, and Amuru districts of Northern Uganda. HIV prevalence overall was alarmingly high at 12.2%. In contrast, a national AIDS survey conducted in 2011 estimated HIV prevalence in the whole mid Northern region of Uganda to be just over eight percent [40]. Observed estimates of HIV prevalence in ‘Cango Lyec’ are markedly higher than the national averages, particularly among women. High prevalence of HIV infection is deeply concerning and demonstrates the continuing crisis of the epidemic in this post-conflict setting.

We observed that district of residence was an important risk factor for HIV infection, with participants residing in Gulu district at least two times as likely to be living with HIV, compared to those in Amuru district. Those living in Nwoya district were also 1.67 times as likely to be living with HIV compared to those in Amuru district. Early in the conflict, the Ugandan government was reluctant to declare Northern Uganda an ‘emergency situation’ and HIV/AIDS funding for the conflict-affected districts remained stagnant. In 2005, the Uganda AIDS Commission and Office of the Prime Minister described the situation in districts impacted by the war as one of poor access, uneven distribution and poorly linked HIV care, treatment, and referral services [41]. With cessation of hostilities, the Government of Uganda announced an official shift in policy towards development-related activities. As a result, many non-government organizations engaged in relief efforts that supported HIV prevention and care closed down operations, leaving significant gaps [42]. Currently, more than 90% of the 2 million internally displaced people in the region have now migrated back to their war-ravaged ancestral villages. Although the conflict is over, district officials remain concerned that inadequate health-care infrastructure and health worker attrition has alarming consequences for HIV prevention as the previously encamped resettle [42]. Our findings clearly indicate the potential for rapid progression of HIV infection in Northern Uganda requiring an aggressive, evidence-based response to HIV prevention that takes into account post-conflict realities.

Gender

HIV prevalence was much higher among women (14.6%) compared to men (8.5%) in this study (p = 0.001). In the multivariable analysis, we observed that young women aged 18 to 34 are at considerably higher risk than young men of the same age, with a 1.64 increase in odds of being HIV positive, after controlling for marital status, district, religion and age. However, the difference was not statistically significant among 35+ year olds. The impact of gender-based disparities in HIV risk were further illustrated by our finding that participants who lived in a female-headed household were more than two and a half times likely to be living with HIV. As conflict-affected young women navigate new environments outside of the bush and IDP camps where they have spent large portions of their lives, district officials are concerned that they face new and undocumented vulnerabilities [25]. With prevailing peace, trade between Northern Uganda and South Sudan has expanded, leading to cross border movement of truckers, agricultural traders, and cattle-loaders into the previously isolated region. These trends have been accompanied by increased transactional and subsistence sex along the Kampala-Juba highway. Young women displaced by war who experienced multiple traumas and never learned agricultural skills may be transitioning into sex work along this new corridor and in rural areas, resulting in the emergence of HIV and STI transmission hotspots. District leaders have raised concerns that families moving away from trading centres back to their ancestral homes leave daughters behind to be closer to schools without adequate socioeconomic support, exposing them to predation. In addition, it has been highlighted that young women have been working in strip/sex clubs and exchanging sex for essential goods, such as sanitary napkins. It is clear from our current work that with the restoration of relative security in Northern Uganda, young women are just as vulnerable to HIV infection, predation, and gender-based violence as they were at the height of the conflict [43]. The drivers of risk have changed, but the vulnerability remains the same.

War trauma, psychiatric disorder and HIV

This study identified powerful associations between HIV prevalence and war trauma experiences and the psychiatric problems of PTSD, MDD and suicidal ideation among conflict-affected populations. We observed that participants living with HIV were two and a half times as likely to report 12 or more traumatic events. Participants living with HIV were 1.44 times as likely to have probable PTSD, nearly two times as likely to have probable MDD and nearly two times as likely to report suicidal ideation within the last 2 weeks. PTSD may enhance HIV vulnerability by promoting high-risk behaviour, reducing immune function, and interfering with medication adherence [44–46]. However, the relationship between HIV and MDD is complex and bidirectional where persons with MDD are at increased risk for HIV infection, and persons with HIV are at increased risk of developing MDD [47]. Delineating which of these associations is at play in the post-conflict situation of Northern Uganda will be answered by the follow-up component of this study. There is need for more studies in order to understand the link between mental health and HIV vulnerability in the context of war and forced migration [48–50].

Of note, contrary to the speculation that the experience of abduction and its’ associated experience of sexual abuse exacerbated vulnerability to HIV, there is no evidence to suggest that people with a history of abduction are more likely to be HIV positive in this study [48, 51]. However, former abductees in the ‘Cango Lyec’ study were nearly three times more likely to meet the criteria for probable PTSD which in this study has been associated with HIV infection. Anecdotal observations from the region indicate that former abductees experience significant levels of stigma on returning from the bush including the belief that they are HIV infected. Could this stigma and isolation explain the lack of association between abduction and HIV infection? This and many other questions call for further studies in order to understand the complex relationship between war experiences, war related psychiatric disorder and vulnerability to HIV infection. The positive association observed in this study between HIV infection and war related psychiatric disorders calls for the integration of mental health care in HIV prevention and treatment programs in conflict and post-conflict communities.

Sexually vulnerability

Consistent with existing research, this study observed increased likelihood of HIV among participants with other STIs. It is of great concern that among ‘Cango Lyec’ participants, baseline prevalence of self-reported genital ulcer disease (GUD) in the last year was nearly 10%. Indeed, participants who reported GUD in the past year were more than four times as likely to be living with HIV at baseline. Among participants who tested positive for active syphilis, risk of HIV increased nearly four-fold. Both ulcerative and non-ulcerative STIs have been shown to increase transmission of HIV infection [49, 50, 52, 53]. In a study of monogamous HIV-discordant couples in Rakai, Uganda, risk of HIV transmission was approximately double if one partner had GUD. When both partners had GUD, the risk was nearly four-fold higher [54]. Further, a study of young conflict-affected people aged 15–29 residing in transit camps in Northern Uganda observed that ever having had an STI was associated with a four-fold increase in HIV risk among young women, but was not significantly associated with HIV risk among young men [55]. Active syphilis and other GUD in conjunction with low prevalence of male circumcision is worrisome and may be a warning sign for a rise in new HIV infections in the aftermath of war [56]. These findings highlight the importance of post-conflict HIV prevention interventions that support access to STI diagnosis and treatment.

Compared to those who were single, those who were currently or had ever been married were significantly more likely to be living with HIV in this study. These findings are consistent with other research demonstrating that many new HIV infections in sub-Saharan Africa occur in stable partnerships [57]. However, HIV prevention programs continue to focus on reducing sex and increasing condom use with casual partners.

In addition, participants who reported consistent condom use in the last 3 months were two times as likely to be living with HIV. These findings and the observed extremely low rates (11%) of consistent condom use are concerning, as elsewhere in Uganda [58]. More work must be done to help men and women, including those living with HIV, negotiate condom use with all partners.

Strengths and limitations

Significant strength of this study includes the ability to determine the HIV and syphilis status using laboratory tests. Efforts were made to enrol all eligible members of the study population. Unfortunately, some individuals were repeatedly absent or refused to participate in the survey. Some of these individuals were more likely to be mobile individuals who have been shown to have higher risks of infection. Reasons for refusing to participate in the survey were not given; some may have refused because of known HIV sero-positive status, we cannot completely rule out selection bias. In addition, the study relies on self-reported behavioural data, there is therefore potential for recall bias, socially desirable reporting, and misclassification of exposure. Responses to historical questions may be influenced by the participant’s ability to recall event(s). Social desirability may lead to an underestimation of some HIV risk behaviours. The effect of memory on these study variables is difficult to assess. Due to the cross-sectional nature of this analysis we are unable to identify cause-effect relationships and temporal sequences. Finally, while the HTQ and HSCL-25 have been demonstrated to be both reliable and valid in this setting, they are screening tools and not diagnostic, which may lead to a conflation in levels of probable PTSD and MDD. On the other hand, ill people might have been more inclined to comply in the expectation of receiving treatment. Despite these limitations, we are confident that due to our recruitment methods and rigorous eligibility criteria our sample is representative of people residing in study communities.