In 2021, approximately 38.4 million people worldwide were living with HIV. Since the beginning of the epidemic, 84.2 million people have become infected with HIV, and 40.1 million have died from AIDS-related illnesses [1]. At the end of December 2021, around 28.7 million people were accessing ART, up from 7.8 million in 2010 [2]. According to the World Health Organization (WHO), the African region continues to be the most severely affected, accounting for nearly two-thirds of people living with HIV [3]. In Ethiopia by 2021, nearly 610,000 people were living with HIV/AIDS. The overall ART coverage is 78% of which adult ART coverage accounts for 76% [4].

ART lowers mortality and morbidity rates in HIV patients while improving their quality of life. The benefits also include preventing HIV transmission by suppressing HIV replication in people who already have the virus. Despite a significant reduction in HIV-related deaths as a result of widely available ART treatment, success remains critically dependent on regular patient follow-up in care [3,5,6].

The goals of ART are to lead and effectively suppress viral replication in order to prevent the development of HIV drug resistance, treatment failure, and lost follow-up. Adherence to ART and retention in HIV care program are critical for achieving optimal health outcomes in HIV patients [7,8]. Loss of follow-up from ART service has a pronounced negative impact on ART treatment outcomes. It can reduce the immunological benefits of ART and increase AIDS-related morbidity, mortality, and hospitalization, resulting in serious consequences such as drug toxicity, treatment failure due to poor adherence, and drug resistance [9,10].

A competing risk is an event that either impedes or modifies the occurrence of the primary event of interest. Classical survival analysis assumes non-informative or independent sampling (censoring), which means that subjects who remain in a given study at any given point in time have a similar future risk of event occurrence as those who are no longer being followed, as if subjects lost to follow up were random and thus non-informative [11,12]. Traditional survival analysis becomes unsuitable when the event of interest is neither censored nor observed, and as a result, the cumulative incidence of an event may be overestimated in the presence of competing risks. Because the assumption of non-informative censoring may be violated and event probability estimation is interpreted as occurring in a setting where competing events do not occur [13,14].

According to competing risk regression model results, the cumulative incidence in different countries was 7.9 per 100 person-year, 5.6 per 100 person-year, 11.6 per 100 person-year, and 10.9 per person-year, respectively [9,14-16]. Similar studies conducted in different parts of Ethiopia based on classical survival analysis reported the cumulative incidence ranging from 3.7 per 100 person-year up to 26.6 per 100 person-year [17-24].

Different studies showed that LTFU is associated with baseline sociodemographic factors like male sex [8,25-31] those whose age is younger 15-24 years [7,14,16,27,28,32,33] and those not having a committed partner [34], patients with no education [27,35,36], and unmarried [27,36], and clinical and treatment-related factors like CD4 count below 200 cells [8,19,37,38], BMI<18.5 kg/m² [8,25,32], WHO stage I or II [8,14,15,38-41], anemia [42], fair/poor level of adherence and WHO stage III or IV [17,18,24,28,35], don’t take INH prophylaxis [16,17,22], BMI<18.5 kg/m² [21,35], no CPT prophylaxis [16,43], ambulatory functional status [16,17,19,44] were significant predictors for LTFU.

Although many studies have been conducted on LTFU and its determinants, valid estimates of incidence and predictors of LFTU can be obtained if death and transfer out are considered as competing events (rather than counting those as censored), especially in poor clinical settings where death is common and alters the probability of the occurrence of LTFU. However, in most of these studies, death and transfer out, which are competing risks of LTFU, are frequently ignored, which can lead to misleading results. As a result, the purpose of this study was to estimate the incidence rate and identify the predictors of LTFU in Jigjiga city Governmental Hospitals ART Clinics, Eastern Ethiopia, by considering death and transfer out as competing events.

Study design and Setting

An institution-based retrospective follow-up study was conducted in Jigjiga city public hospitals in eastern Ethiopia between January 1, 2015, and December 31, 2021. Jigjiga is the capital city of the Ethio-Somali region, which is located 630 kilometers from Addis Ababa, Ethiopia's capital city. The study was conducted in three governmental hospitals in the city; these hospitals housed the city's only ART Centers. All of these hospitals offer HIV/AIDS interventions, such as free diagnosis, treatment, and patient follow-up. The research was carried out between July 1 and July 30, 2022. During the study period, there were a total of 2016 ART-attending patients in these hospitals, with 1816 of them being adults over the age of 15.

Population

All adult HIV-infected patients (Age ≥ 15years) who initiated ART at Jigjiga public hospitals (Karamara General Hospital, Sheik Hassen Yeberre Referral Hospital, and Ablelle Primary Hospital) from January 01, 2015, to December 31, 2021, were our study population. Adult HIV-positive patients who had at least one follow-up visit after starting ART were included in the study. HIV patients who transferred in with incomplete baseline data, whose medical charts were unavailable during the data collection period, and those with unknown ART initiation dates were excluded from the study.

Sample size determination and sampling method

The minimum required sample size (842) was calculated using Stata 14 software's survival analysis formula power cox command. The study participants' records were first filtered from the ART database based on their entry time to the follow-up, age, and inclusion criteria, and then we used stratified proportionate random sampling (among health facilities). Finally, we used SPSS Software to select the final sample size using a simple random sampling technique (computer-generated Random Sampling) (Figure 1).

Variables definition

The primary outcome variable was time to LTFU, with death and transferred out considered as competing events. The predictor variables assessed were baseline socio-demographic factors like (sex, age, marital status, educational status, occupation, residence), baseline clinical and treatment-related factors like (BMI status at baseline, History of Tuberculosis (TB), Opportunistic infections, Hemoglobin level, Baseline CD4, Viral load, WHO clinical stage, Partner HIV status, Functional status, Co-morbidities) and ART adverse event, Regimen type at the start, Regimen substitution, Year of ART initiation, Registered phone number, OI prophylaxis (INH, CPT), Distance from the Hospitals), Disclosure status, Partner HIV status, Caregiver, Adherence level) were included.

Operational definitions

This study intended to determine the incidence of LTFU using seven-year data retrieved from medical record charts. LTFU is defined as not taking an ART refill for 3 consecutive months or longer after the last scheduled visit (from the last attendance for refill) not yet categorized as dead or transferred out [19]. The competing events were death and transferred out which is defined as patients recorded as dead on the patient’s exit form or whose outcome is recorded as death on the follow-up chart and patients formally transferred to another health facility respectively. A patient was classified as censored if he/she had still on follow-up at these hospitals at the end of the study period. Operationally the functional status was defined based on the ART guideline; Working FS: able to perform usual work inside or outside the home, Ambulatory FS: able to perform an activity of daily living, and bedridden FS: not able to perform an activity of daily living [45]. ART medication adherence is defined as the percentage of ART drug dosage calculated from a monthly total dose, and classified as good, fair, or poor. Therefore, good adherence was reported if equal to or greater than 95% or ≤ 3 doses missing per month, fair if 85–94% or 4–8 doses missing per month, or poor if less than 85% or ≥ 9 doses missing per month [46]. Disclosure in this study is defined as disclosure of the status that is being HIV positive to at least one individual. The caregiver is also defined as anyone who can support or assist an individual with HIV.

Data collection tool and procedure

ART patient data is stored in smart care as both a hard copy (chart) and a database. The ART follow-up forms served as the foundation for the data extraction checklist. The data extraction tool was carefully prepared from ART intake follow-up forms to ensure data quality. By randomly selecting and completing 42 sample chart reviews, we demonstrated consistency between data recording systems and the prepared checklist, resulting in minor modification to the data extraction checklist. Seven clinical nurses and health officers were recruited, as well as a data collector and three supervisors for each of the three hospitals.

Data processing and analysis

The Data was entered using Epi-data version 3.1 and then exported to Stata 14 software for analysis. Descriptive statistics including proportions, tables, and charts were done to describe the characteristics of the study participants. Non-parametric estimation of CIF was performed both graphically and using Gray’s test. After fitting the model, the proportional sub-distribution hazard assumption was also verified by using the plot of log (- log (1-CIF)) versus the log of time to failure for each covariate, by interacting each covariate with time (tvc) and using Schoenfeld residual test. Univariable competing risk regression analysis was fitted to identify factors associated with LTFU. Variables with a p-value of ≤ 0.2 in the univariable analysis were fit to the multivariable competing risk regression analysis once more. To express the strength of the association, the crude and adjusted sub-distribution hazard ratios were calculated, along with the corresponding 95% CI. Variables with a P-value of ≤ 0.05 were considered statistically significant in multivariable-analysis.

Ethical consideration

We obtained ethical consent from the Institutional Review Board of the University of Gondar, Institute of Public Health. Since the study used an analysis of secondary data from patient charts; we received a waiver for informed consent. To keep privacy, names and other personal identifiers were not included in the data collection tool.

Baseline socio-demographic characteristics

A total of 842 clients enrolled in ART care were included in the final analysis. Nearly half, 466 (53.34%) of the study participants were males. The median age of the participants was 35 (IQR=29-43) years. Of the total, 428 (50.83%) were married, and 270 (32.07%) participants had a primary level of education. The majority of subjects 759 (90.25%) had a caregiver and 673 (80.02%) study participants disclosed their HIV status (Table 1).

Clinical and treatment-related characteristics

About 240 (42.93%) participants were linked to care with a baseline CD4 count between 201-350 cells/ml. The median CD4 count was 292 (IQR=161-421 cells/ml), and 369 (43.82%) of the participants were WHO stage II, followed by WHO stage III & IV 244 (28.98%). Around 89.96% have been started with CPT prophylaxis and the majority (98.8%) of patients were screened for TB, of which 18.67% were positive (Table 2).

Incidence of LTFU                                                              

Study subjects were followed for different periods of time with a median time of 28.7 months (IQR: 13.9-50.43) months. The minimum and maximum follow-up time was one month and 85.06 months respectively with a total observation time to be 2324.3 person-year. Among the 842 patients enrolled, 91 (10.81%) were dead, 122 (14.49%) had been LTFU, 114 (13.54%) were transferred out, furthermore 515 (61.16%) on treatment until December31, 2021 (Figure 2). The overall incidence of LTFU was 5.25 per 100 person-year (95% CI: 2.1–7.2). LTFU was highest in the second 12 months of ART follow-up, 7.4 per 100 person-year (95% CI: 5.4, 10.05).

Non-parametric estimation of CIF of LTFU

CIFs across groups were checked statistically using Gray’s test (which is equivalent to the log-rank test of classical survival analysis) and graphically by plotting each predictor variables against failure time (Non parametrical checking).

Based on the result of the modified X² test (Gray’s test), there was a significant difference in CIF among categories of age, marital status, occupation, education, IPT, CPT, Functional status, disclosure status, baseline WHO stage, caregiver, and type of regiment at start. Graphically, belonging to age 15-24, poor adherence level (baseline), non-disclosure status and WHO stage III &IV, male gender, and bedridden functional status were all risk factors for LTFU (Figure 3).

Modeling bivariable & multivariable competing risk regression model

After fitting a bivariable competing risk regression model almost all the predictor variables except educational status, occupation, opportunistic infection & INH prophylaxis, were crudely associated with LTFU at 0.2 level of significance. The Schoenfeld residuals and The Cox-Snell residuals (together with their Nelson-Aalen cumulative incidence function) were done to check the goodness fit test or model fitness (Figure 4). The sub-hazard proportionality assumption with the Schoenfeld residual of the global test was 0.3479, indicating no evidence for violation of the proportional sub-hazard assumption. Finally, those variables like: Functional status, disclosure status, hemoglobin, recent adherence level, and CPT prophylaxis, were found to be significant predictors for LTFU at 5% level of significance.

In multivariable analysis disclosure status, adherence level (last known), CPT prophylaxis, functional status, hemoglobin level was statically significant for LTFU (p-value<0.05). The patients who did not disclose their HIV status can increase the sub-hazard of LTFU by 4.22 (asHR=4.22; 95%CI (2.11-8.47)) times compared with their counterparts. The sub-hazards of LTFU among the fair and poor level of recent adherence were 2.17(asHR=2.17; 95%CI (1.18-4.23)) and 2.27 (asHR=1.48; 95%CI (2.97-5.34)) times higher than those who have a good level of adherence respectively. The sub-hazard of LTFU is 2.47 higher among patients who do not take CPT prophylaxis 2.47 (asHR=2.47; 95%CI (2.99-6.15) compared with the counterparts. The sub-hazard of LTFU among patients with ambulatory functional status was 2.38 (asHR=2.38; 95%CI (1.21-4.68)) times higher compared with the patients in working functional status. The patient who had moderate and severe anemia increased the sub-hazard of LTFU by 2.22 (HR=2.22; 95%CI (1.09-4.52)) and 4.58 (asHR=4.58; 95% CI (1.28-11.39) times compared with that of those who had no anemia (Table 3).

In this study, we assessed the incidence and predictors of LTFU among adult HIV/AIDS patients under ART follow-up in Jigjiga Governmental Hospitals. A number of variables were used to explain the variation in the LTFU of HIV/AIDS patients using this competing risk regression model. In our study, we tried to assess socio-demographic, clinical, and treatment-related factors by using secondary data.

The estimated Incidence rate for LTFU was 5.25 per 100-person year observation. It is similar to the finding of a study conducted in different places in Ethiopia [18,21,23] Zimbabwe [14] and Zambia [47]. This might be due to the similarity in the quality of care given according to ART guidelines for HIV/AIDS patients across these different Hospitals since all study areas were in eastern African regions, have almost consistent guidelines prepared by WHO.

However, the current finding is higher than the findings in Debre Markos [24], Kenya [13], Republic of Democratic Congo [48]. This may be due to the study areas being located around the capital city of the region and the majority of the respondents were rural residents, so the flow of the population around the city is somehow low. In addition, the area is located at a small distance from the Ethio-somaliland border, and many of the long-track drivers and emigrants have joined the city most of the time. This also could be due to variations in study design, patient follow-ups, and definitions of LTFU is defined when a patient is lost for at least 90 days since some studies defined LTFU as 180 days [32,40].

The current finding is lower than the findings in Myanmar [37], Guinea-Bisaw [8], South- Africa [34], Cameroon [29], Nigeria [30], Uganda [42], Malawi [33], Ethiopia (mekelle, Jigjiga, Pawi, Gondar) [9,19,20,23]. The difference might be due to variations in study design and the operationalization of terms, as we know that our model is a competing risk regression model, majority of the preceding studies were classical survival analysis, so basically, it makes overestimate the incidence rate. Patients who were not taking CPT were more likely to be lost from ART care and this result is consistent with the study conducted in Myanmar [37], Gondar [16], and Tepi [43]. This might be due to the fact that CPT, given for the prevention of many opportunistic infections such as pneumocystis pneumonia, toxoplasmosis, bacterial infections, and diarrheal diseases [5] results from the patient feels better and intend to attend their schedule of ART follow-up accordingly. In contrast, patients who are not taking CPT, are more vulnerable to many opportunistic infections and finally, they may have such diseases and either they may develop drug toxicity due to drug-drug interaction or they may prefer to go to Traditional healers (Holly water) by discontinuing such burden of drugs and finally end up with LFTU or Death [10,31].

Patients who were in ambulatory functional status increased the risk of LTFU, this finding is in agreement with the study conducted in Indonesia [44], and different studies in Ethiopia [16,17,19,30]. This may be due to ambulatory patients being more likely to be LTFU could be due they become poor performance status at the initiation and due to the social, economic, and financial influences that are caused by their inability to work, this may lead to enabling to staying to the care [17]. The sub-hazard of LFTU was higher among patients with a fair and poor level of adherence, this is in agreement with studies in Gunie-Bisaw [8], Zimbabwe [14], Uganda [41], Ethiopia [17,18,31], this may be due to that the patients with the poor level of adherence for ART drug become interrupt the treatment schedule which leads to resisting to the virus. This treatment interruption make fail to suppress the viral load and become prone to poor treatment outcome like treatment failure, this leads the patient to lose hope for a good prognosis, and He/she prefers to lost from care after all the outcome turned to death [4,7].

The patients with severe and moderate anemia status were more likely to LFTU compared with their counterparts, this finding is in agreement with studies conducted in Mynamar [37], Togo [40], Mizan-Teferi [22], the possible reasons for this finding might be the azidothymidine (AZT) based regimen have an interruption on bone marrow function and it becomes resolved as time goes and can be treatable as easy, but the patients who develop anemia due to this AZT based regimen lack their trust about the treatment and they hesitate to continue the care [21]. The patient with severe anemia at baseline hope that immediate resolution by ART treatment and if the expectation is not soon, then they try other medication like traditional healer and they become lost from the care. Patients who did not disclose their HIV/AIDS status were 2.27 times more at the risk of lost from the treatment program as compared to their counter parts. This study is similar with studies conducted in South Africa [37], Tepi [43], Oromiya region [28]. This might be due to that patients may remain on treatment much more if they disclose their status and they had someone else to share their ideas and feelings about the treatment and the diseases condition. Since staying on treatment needs intensive support and care from different parts of the community as well [3]. This study has some limitations because of the retrospective nature of the study, which lacks completeness of some potentially important predictors (patient information) like substance use. Since the study uses baseline socio-demographic and clinical-related factors, there may be a change of these variables (change of exposure variable) after a time.

In this study, the incidence of LTFU was one person lost from twenty patients within a one-year follow-up time. For examining the predictor variables of LTFU, competing risk regression analysis was done considering death and transferred out as competing events. After all, patients on ART who disclosed their HIV status, not taking CPT prophylaxis, being ambulatory functional status, those have moderate to severe anemia at baseline, and have a poor level of adherence to ART were at higher risk for LTFU. Therefore, giving more consideration and close follow-up of these high-risk groups could reduce the rate of LTFU.

All the authors declared that they have no competing interest.

GSA: conception of the research idea, study design, data collection, analysis and interpretation, and manuscript write-up. KAG, AML, FMA, and YMC: analysis and interpretation, manuscript write-up, Writing – review & editing, visualization. All authors have read, validate, and approved the final manuscript.

I would like to express my deepest appreciation and special admiration for Jigjiga city Governmental Hospital administrative bodies and chart room experts for their cooperation as well as their permission to access this ART data. We are also thankful to the health professionals who work in the ART clinic and ART data clerk managers for giving relevant information. Finally, we would like to thank the data collectors and the supervisor for their tolerance and commitment to the data collection and research writing process.

Please contact the author for further dataset requests.

AIDS: Acquired Immune Deficiency Syndrome; cART: Combination Antiretroviral Therapy; CD4: Cluster Differentiation Four; PLWH: People Living with HIV; BMI: Body Mass Index; cSHR: Crude Sub Hazard Ratio; aSHR: Adjusted Sub Hazard Ratio; CIFs: Cumulative Incidence Functions; CPT: Cotrimoxazole Preventive Therapy; HIV: Human Immune Deficiency Virus; INH: Isoniazid Preventive Therapy; LTFU: Lost To Follow Up; OIs: Opportunistic Infections; FS: Functional Status; TB: Tuberculosis; WHO: World Health Organization; FMOH: Federal Ministry of Health; HAART: Highly Active Antiretroviral Therapy