Prevalence and correlates of depressive symptoms among adults living with HIV in rural Kilifi, Kenya

This cross-sectional study was conducted in Kilifi County – coastal Kenya, between February and April 2018, through the Centre for Geographic Medicine Research (CGMR). Kilifi County consists mostly of a rural population. Majority of its residents are poor (71.4% live below poverty line), lack adequate formal education (only 7.1% have attained secondary education) and earn a living mainly through subsistence farming or fishing [16, 17]. Adult HIV prevalence is estimated at 5%, almost near the national average prevalence of 6% [18]. The socio-economic disparities in this setting may have a role to play in HIV infection and prognostic patterns. A public hospital – Kilifi County Hospital – from where participants were recruited, has a specialized HIV care and treatment clinic that was established in 2003. This HIV clinic provides comprehensive care services (including medical reviews, counselling, family planning, nutritional care, and antiretroviral dispensary) to PLWHA and serves as a research facility. About 60 patients are seen daily. To date, the clinic has enrolled over 9000 patients of all ages.

This work is part of data collected from a larger project looking at different outcomes in adults living with HIV, including health-related quality of life and mental health. Participants were patients attending an HIV care and treatment clinic at Kilifi County Hospital. To be included in the study, participants had to be adults, 18 to 60 years old, with confirmed HIV positive status, on cART, and providing consent for participation. We excluded participants who did not meet the inclusion criteria above. We did not include elderly individuals living with HIV (above 60 years) because of the increased likelihood of illnesses associated with advancing age, that may have an impact on the quality of life [19]. Eligible participants were also excluded if they could not comprehend and/or communicate in the national language, Kiswahili, which was used during the administration of all study instruments.

Selection was done through consecutive sampling of eligible patients on arrival at the clinic until the required sample size was achieved. Sample size calculation was determined using the single population proportion formula. With precision set at 3.5% around a previously reported prevalence of 15.5% for depressive symptoms among adults living with HIV [20], a sample of 410 participants or more would give reliable prevalence estimates. To cater for a 10% non-response or systematic missing of data, non-contact, or other factors that tend to reduce the final sample size, we considered a sample of 450 as sufficient. This sample is > 95% powered to detect a proportion of ≥0.10 given a null proportion of 0.05 at 5% level of significance.

For all participants, study instruments were interviewer-administered via android tablets, in the same order, and under the same administration environment. Research assistants received a 4-day training in research ethics and good interviewing techniques (with role plays) and were taught how to administer the tablet-based questionnaires. Questionnaire administration was done in a private and quiet room within the clinic setting, and the entire interview session lasted between 30 to 45 min.

A researcher-developed sociodemographic questionnaire was used to capture participant’s age, sex, marital status, educational level, employment status, and people with whom they were living with. Additionally, an asset index form previously used in this setting [10] was used to gather information on disposable assets owned by participants as an indicator of their socioeconomic status.

During the face-to-face interviews, participants self-reported about current history of smoking, khat or alcohol use, and presence of any current opportunistic infection or chronic illness (that they were made aware of by their clinician) using “0” (no) or “1” (yes) response options.

In the interviews, participants were also asked about disclosure (if they had disclosed their seropositive status or not), their satisfaction with the current level of care (satisfied or not) and their perception on clinic accessibility (if current HIV-clinic was accessible or not). In responding to the latter item, participants were asked to consider the distance between where they were living and the location of the HIV-clinic.

A clinical record form was used to extract participant data from the clinic’s medical record database on: dates of HIV-diagnosis and cART initiation, most recent cART regimen, World Health Organization (WHO) clinical staging, viral load, CD4 cell count, height and weight (for Body Mass Index [BMI]) calculation). We used patient-unique clinic numbers to access the medical records of our study participants. All participants gave informed consent for this clinical information to be retrieved from their clinic medical records.

The 12-item HIV stigma scale [27] was used to assess patient perceived HIV-related stigma under four dimensions of i) personalised stigma; ii) disclosure concerns; iii) negative self-image; and iv) concerns with public attitudes. Items on this scale are rated as “1” (strongly disagree), “2” (disagree) “3” (agree) and “4” (strongly agree). A total score derived from summated item scores ranges between 12 and 48. Higher scores indicate a greater level of perceived HIV-related stigma. In its initial validation, Cronbach’s alpha was > 0.7 [27]. In the present study Cronbach’s alpha was 0.81 (95% CI: 0.78, 0.83).

Frequencies (with percentages) and means (with standard deviations) were used to describe sample characteristics. Departure from linearity in continuous variables was checked through visual inspection of the frequency histogram with normal overlay graph. Prevalence was presented as percentages. Univariate logistic regression analysis was used to assess the crude association between sample characteristics and positive depression screen (PHQ-9 cut-off score ≥ 10). Multivariable logistic regression analysis was used to investigate factors independently associated with a positive depression screen. Since there is no universal consensus in selecting predictor variables, we included all variables with p ≤ 0.20 from univariate logistic regression analysis in multivariable analysis. Bursac et al. [28] suggest that p-values less than 0.25 in univariate logistic regression may indicate some reasonable association with the outcome. Backward stepwise logistic regression models were fit in the multivariable model building process, removing all variables with p > 0.05 (one at a time). Age and sex were included regardless of the p-value as these have been reported to be associated with depressive symptoms in other studies from SSA [4, 6, 13, 14]. Regression diagnostics for correlation of covariates were performed, with no collinearity problems identified. For the final model, Hosmer-Lemeshow goodness of fit statistic (p-value > 0.05) was considered a well-fitting logistic regression model. All analyses were conducted in STATA (V.14.0) statistical software package [29] with p < 0.05 considered statistically significant for all tests of hypothesis.

In total, 512 adult patients attending an HIV care and treatment clinic at Kilifi County Hospital were approached to participate in this study. Forty-four patients declined to participate after the study was explained to them in details. Eighteen patients were excluded because they were over 60 years (n = 11) or were unable to comprehend and/or communicate in Kiswahili (n = 7). Consequently, 450 patients, hereafter referred to as study participants, were consecutively recruited.

Table 1 presents sample characteristics aggregated by depressive symptom. Overall, the mean age of the participants was 42.7 (SD = 9.7) years. Most study participants were female (79.1%), had attained primary level education (53.1%), were unemployed (59.8%) and lived with a family member (82.4%). Only 12.7% of study participants were never married. A small proportion of study participants currently smoked (9.8%), used khat (4.4%) and were taking alcohol (11.4%). Chronic illnesses comorbid with HIV was present in 8.2% of the study participants, hypertension being the commonest (n = 20). The mean BMI of study participants was within the normal range (mean [SD] = 22.4 [4.8]). Most study participants were on first-line cART regimen (95.3%), and in stage 1 of the WHO clinical staging (93.7%). Only 6.0% of study participants had not disclosed their seropositive status. Less than a quarter of study participants (18.4%) had an opportunistic infection at the time of interview. Most participants found the HIV care and treatment clinic easily accessible (86.2%) and were generally satisfied with the current level of care they were receiving (95.8%).

Using PHQ-9 cut-off score of ≥10, which has been shown to maximize on sensitivity and specificity for depression screening [23]; the overall prevalence of depressive symptoms was 13.8% in this sample. In terms of severity of depressive symptoms in the sample, 23.8% had mild, 9.6% had moderate, 3.6% had moderately severe, and 0.7% had severe depressive symptoms (Table 2).

Tables 1 and 3 present the results following univariate and multivariable logistic regression analyses for factors associated with positive depression screen among adults receiving cART in Kilifi. Even though viral load was associated with positive depression screen at p = 0.20 in univariate analysis (Table 1), we did not include this variable in the multivariable logistic regression model because many patients had missing values (n = 145). For all our study participants, we did not find any current records of CD4 cell count. Few participants had baseline CD4 data (when they were being enrolled to the clinic). Therefore, logistic regression analyses did not include CD4 cell count information.

In the univariate logistic regression analyses, increasing perceived HIV-related stigma, being separated or divorced or widowed, on second-line cART regimen, having any current chronic illness or opportunistic infection and clinic inaccessibility were significantly associated with an increased risk for positive depression screen (p < 0.05; Table 1). Having no education or secondary education (relative to tertiary), living with a relative/friend (relative to family), and increased duration since HIV diagnosis were all reasonably associated with a positive depression screen at p ≤ 0.20. With increasing BMI, the risk of positive depression screen was 6% less likely at p = 0.11 (Table 1). All these variables were added to the multivariable logistic regression modelling in addition to age and sex (forced variables).

In the final multivariable analysis, factors that were significantly associated with positive depression screen among adults receiving cART in Kilifi included: having a current chronic illness (p < 0.001), being on second-line cART regimen (p < 0.001), clinic inaccessibility (p = 0.02), and increasing perceived HIV-related stigma (p < 0.001) (Table 3). The cross-validated mean area under the curve (cvMean AUC) for the final multivariable model was 0.75 (bootstrap bias corrected 95% CI: 0.66, 0.80).