A facility-based retrospective cohort study design was conducted in Jinka Hospital, Jinka, from November to March 2016. Adult patients who started HAART from September 2010 to August 2015 in this hospital were recruited for the study. Jinka, the administrative capital of the South Omo Zone in the Southern Peoples, Nations, and Nationalities Region, is located 750 km to the south of Addis Ababa, the capital of Ethiopia. The hospital provides general outpatient and inpatient care, surgical and obstetric emergency services, and acts as the only zonal referral hospital for all health services for the pastoralist and agro-pastoralist population residing in South Omo. Jinka Zonal Hospital started providing ART services to HIV/AIDS patients in October 2005. As of August 2015, a total of 3,199 HIV/AIDS patients have utilized ART services at this hospital, of whom 1,899 are currently on ART.

The records of all HIV-positive adults who were on ART at Jinka Zonal Hospital and who enrolled in treatment from September 1, 2010 to August 30, 2015 were the source population. The records of selected HIV-positive adults who had started ART at the hospital within this period were included in the study.

Ethical clearance was obtained from the institutional review board of the Aklilu Lemma Institute of Pathobiology, Addis Ababa University, and Jinka Hospital officials received a support letter from the university. Retrospective data were collected anonymously and confidentially.

The sample size was estimated using the power and sample size calculation software developed by Dupont & Plummer [25]. To obtain the optimum sample size, two populations were considered, categorized by WHO staging as the main exposure variable for HIV/AIDS-related deaths during the 6-year follow-up period. The calculations were based on the following assumptions: a 5% level of significance (two-sided), a power of 80%, and a ratio of exposed to non-exposed of 2:1. The estimated median survival time was 60 months for the non-exposed group (WHO stages I and II) and 40 months for the exposed group (WHO stages III and IV) [19]. Accordingly, the calculated sample size was 319 (213 in WHO stages III and IV, and 106 in stages I and II). After adding 10% as a contingency modifier, the total sample size was 350 (220 for exposed subjects and 130 for non-exposed subjects).

Simple random sampling was done using commands in Excel (Microsoft Corp., Redmond, WA, USA); all 1,899 records of patients receiving ART in Jinka Hospital were listed in a single Excel spreadsheet, and a random selection of 350 records was made.

The data were collected using a standard checklist on an Excel spreadsheet, incorporating information extracted from electronic and paper-based ART registration and follow-up forms used in the ART clinic. Using the data extraction sheet, information about study participants, such as socio-demographic characteristics, residence, weight, disclosure status, duration on ART (in months), functional status, types of opportunistic infections, baseline WHO clinical staging, baseline and follow up CD4 cell counts, ART regimen type, ART eligibility criteria, substance use, regimen changes and reasons for those changes, and survey endpoints were retrieved from the clinical records of the HIV/AIDS patients by trained ART data clerks. Incomplete clinical records were omitted and were replaced with the record of the next patient on the list. The data collection form was checked for completeness and consistency with each patient’s clinical records by a supervisor, a public health officer by profession, on a daily basis.

The primary outcome variable was time to event in months. HAART-initiated patients were followed until the date of death, loss to follow-up, transferring out, or the end of the study. Individuals who were on HAART, lost to follow-up, or had transferred out at the end of the study period were censored; that is, they were considered to be alive for the time period that they had been under follow-up. The survival time was calculated in months using the time between the dates of treatment initiation and the date of the event (death) or date of censoring. The predictor variables were age, sex, educational level, marital status, occupation, residence, functional status, WHO clinical stage, drug regimen, CD4 count, substance use, and opportunistic infections.

After the Excel data were reshaped and adjusted, the data were transported to SPSS version 20.0 (IBM Corp., Armonk, NY, USA). Descriptive statistics such as median, interquartile range (IQR), and mean and standard deviation (SD) were used to summarize the characteristics of the cohort. The Kaplan-Meier log-rank model was used to estimate the survival time of ART patients based on the independent variables. Bivariate and multivariate Cox proportional hazards regression models were used to identify the predictors of mortality. Variables found to be statistically significant in the bivariate analysis (p<0.05) were included in the multivariable Cox proportional hazards regression model to identify independent predictors of mortality and to estimate the adjusted hazard ratios (aHRs). Both crude hazard ratios and aHRs with 95% confidence intervals (CIs) were reported. Variables that were statistically significant (p<0.05) were considered predictors of mortality among the study participants.

The total extent of follow-up was 1,995 person-years, with an incidence rate of 1.75 deaths per 100 person-years. The mean survival time of patients on HAART was 30.84 months (SD, 19.57 months). The overall survival of patients on HAART was 64.00% (95% CI, 61.85 to 66.21%) at 72 months of follow-up (Figure 1).

The log-rank test was conducted to check for the existence of any significant differences in survival among various levels of the categorical variables considered in the study. Accordingly, the Kaplan-Meier analysis indicated significant evidence of differences in survival times in the categories of past tuberculosis treatment, functional status, WHO clinical staging, baseline CD4 count, initial regimen types, substance abuse, disclosure status, and opportunistic infections (Table 3).

The Kaplan-Meier survival function (Table 3) indicated a significantly higher survival of HAART patients with the baseline characteristics of clinical stage I and II disease (67.12%; 95% CI, 64.34 to 69.89%), unimpaired functional status (66.27%; 95% CI, 64.39 to 68.17%), HIV status disclosure (64.96%; 95% CI, 62.83 to 67.10%), baseline CD4 count of at least 351 cells/μL (68.84%; 95% CI, 65.91 to 71.76), and no substance abuse (68.34%; 95% CI, 66.25 to 70.43).

The lowest survival probabilities were observed for patients with a bed ridden functional status (13.25%), a history of tuberculosis treatment (50.86%), no disclosure of HIV status (52.41%), a baseline CD4 count <350 cells/μL (52.83%), and WHO stage IV disease (43.7%) (Table 3).

To identify independent predictors of mortality and survival, multivariate regression was performed for all predictors found to be significantly associated with survival in the bivariate analysis. According to the Cox survival regression model, all predictors associated in the bivariate analysis were found to be strong predictors of mortality in the multivariate analysis, except for baseline CD4 cell count (Figure 2, Table 4). The HR for death was 5.82 (95% CI, 1.91 to 17.72) in patients who had not disclosed their status, compared to those who had; 1.82 (95% CI, 1.41 to 3.51) in patients with a history of tuberculosis treatment; 2.97 (95% CI, 1.20 to 8.86) and 4.67 (95% CI, 1.30 to 17.27) in patients with an ambulatory and bedridden functional status, respectively, in comparison those with an unimpaired functional status; 24.97 (95% CI, 2.75 to 26.45) in patients with WHO clinical stage IV disease; and 3.72 (95% CI, 1.39 to 9.97) in patients reporting substance abuse. However, the HR for mortality was found to be reduced by 0.20 (95% CI, 0.07 to 0.53) in patients taking the 1e (TDF-3TC-EFV) regimen in comparison to the other five regimens used (Table 4).

The ART retention rate (79.0% after 72 months of follow-up) was significantly associated with occupation and WHO clinical stage, with housewives (adjusted odds ratio [aOR], 3.34; 95% CI, 2.08 to 18.48) and daily laborers (aOR, 2.60; 95% CI, 1.22 to 5.52) more likely to continue ART than other occupation types (Table 5).

The retention rate of 79.0% after 72 months of follow-up in this study is comparable with studies performed throughout most of sub-Saharan Africa, where retention ranges have been reported to range from 57 to 72% [15,26]. Nearly 80.0% (95% CI, 76.7 to 82.1%) of patients were retained in care in the first 3.5 years of ART in Addis Ababa [24], while a value of 64.7% has been reported for in southern Ethiopia [27], which is lower than what was found in the current study. According to Mekuria et al. [24], severe immune deficiency at enrollment in care or at the time of ART initiation, as well as a bedridden or ambulatory functional status at the start of ART, predicted attrition in Addis Ababa. Similar to our findings, a 74.0% retention rate was reported in another large-scale retrospective study done in Ethiopia [21]. Retention was significantly associated with occupation and WHO clinical stage. Housewives (aOR, 3.34; 95% CI, 2.08 to 18.48) and daily laborers (aOR, 2.60; 95% CI, 1.22 to 5.52) were more likely to be retained in the ART program than other occupation types. Patients with WHO stage I (aOR, 6.20; 95% CI, 2.08 to 18.48) and II (aOR, 5.02; 95% CI, 1.94 to 10.53) disease were more likely to be retained in the ART program than patients with an advanced WHO stage of disease.

In this study, the overall survival probability of HIV/AIDS patients on ART was found to be 64.00% (95% CI, 61.85 to 66.21%) at 72 months of follow up, which is slightly higher than was reported in studies performed in Debre-Markos, Addis Ababa, and Cameroon, where the mean survival probabilities were 57% (95% CI, 53 to 60%) [19], 41.00% (95% CI, 39.69 to 42.64%) [28], and 47% (95% CI, 40 to 55%) [29], respectively, but similar to a study carried out in Gonder, Ethiopia, in which 61.4% of the patients survived and 10.4% died [16].

In this 6-year retrospective cohort study, 35 patients died of a total of 350 participants over six years of follow-up, resulting in a total death prevalence of 10.0% and an incidence rate of 1.75 per 100 person-years. Twenty-two (62.9%) of the deaths occurred during the first year of follow-up. The total number of deaths that occurred in this study is much lower than was reported in a study conducted in Debre-Markos Hospital, Ethiopia, where 40.7% of patients died [18]. Moreover, data from Senegal, Malawi, and Tanzania have indicated high rates of HIV-related mortality, ranging from 24.2 to 44.0% [30], in contrast to the present study.

However, our results are similar to the mortality rate of 11.1% exhibited in the Somali region of Ethiopia [22], and with other findings from sub-Saharan Africa, where between 8 and 26% of patients die in the first year of ART, with most deaths occurring in the first few months [31]. Although the immunological and virological responses to ART in patients in sub-Saharan Africa are comparable to the responses in patients treated in high-income countries, early mortality rates in the region have remained very high [31]. The main reasons for early mortality have been found to be baseline CD4 cell counts less than 50 cells/μL, advanced WHO stage, and opportunistic infections. Hence, the earlier diagnosis of HIV infection, timely initiation of antiretroviral treatment, and strengthening of the HIV care continuum should be implemented in order to reduce early mortality among HIV/AIDS patients.

The multivariate Cox regression analysis showed that non-disclosure of HIV status, previous tuberculosis treatment, ambulatory and bedridden functional status, and advanced WHO clinical stage were independent risk factors for early mortality. In our study, HIV status disclosure was found to be a predictor of prolonged survival in HIV patients on ART, in agreement with another study conducted in Addis Ababa, Ethiopia, where disclosure was found to have a protective effect for early mortality (aHR, 0.59; 95% CI, 0.36 to 0.96) [28]. Although future research may be necessary to establish the direct cause-and-effect association between non-disclosure status and early mortality in HIV patients, it is known that disclosure has more positive than negative social outcomes, such as improved social support, stronger family relationships, and reductions in anxiety and depression. Correspondingly, non-disclosure has detrimental health impacts associated with distress, loneliness, and medical non-adherence as a way to hide the presence of disease from others, and these factors may lead to a higher rate of mortality than in those who have disclosed their HIV status.

As in the present study, bedridden functional status (aHR, 5.91; 95% CI, 2.87 to 12.16) and advanced WHO stage (aHR, 7.36; 95% CI, 3.17 to 17.12) were reported to be independent predictors of mortality in the Somali region of Ethiopia [22], and in Addis Ababa among armed forces members [27]. The mortality incidence rate of 1.75 deaths per 100 person-years in the current study is in line with a study carried out in Debre-Markos, Ethiopia that reported 1.90 deaths per 100 person-years [19]; however, this rate was slightly lower than the mortality incidence rate of 2.03 deaths per 100 person-years reported in Southern Ethiopia [9] and East Ethiopia [32].

The current study found a much lower mortality incidence rate than early studies carried out in Ethiopia that reported mortality rates of 15.4 and 16.7 per 100 person-years in ART taking patients [33,34]. This indicates that ART is significantly reducing deaths among AIDS patients in Ethiopia and proves that the ART program is functioning well.

In conclusion, the probability of survival and the retention rate of patients on ART in this study was 64.0% and 79.0%, respectively, at 72 months of follow up, which is moderate and comparable with most results found in sub-Saharan countries. The independent predictors of mortality were non-disclosure of HIV status, previous tuberculosis treatment, ambulatory and bedridden functional status, and advanced WHO clinical stage. Hence, patients should be encouraged to disclose their status to their families or others, and those with an ambulatory or bedridden functional status and/or advanced WHO clinical stage disease should be monitored closely by their clinicians. Although hospital ART services are functioning well, further studies should investigate the causes of early mortality.