INTRODUCTION

The rollout of antiretroviral treatment (ART) programmes has curbed the HIV epidemic to a major extent [1, 2]. However, the UNAIDS goals of 90% of people living with HIV (PLHIV) knowing their HIV status, 90% of those diagnosed being on treatment, and 90% of those on treatment being virally suppressed were not achieved by the end of 2020 [3]. Major reasons for these failures are lack of testing and attrition from care in PLHIV [4-6]. Risk factors for attrition include living without a partner, non-disclosure of HIV status, poor drug adherence and advanced HIV disease [5, 7].

Currently, there is no gold standard to measure adherence to treatment in order to predict the possibility of being lost to follow-up (LTFU) or virological failure (VF) [8, 9]. The methods mostly used in the sub-Saharan African setting – namely, pill counts, electronic monitoring systems, Adults Aids Clinical Trial Group adherence questionnaire, plasma drug levels and patient self-report – all have strengths and weaknesses [10-12]. Bringing back the dispensed pharmaceutical packaged antiretroviral pill bottle (pillbox) on the next visit in order to count pills is requested in many care and treatment centres in sub-Saharan Africa [13, 14] and is also recommended by the WHO [15]. However, carrying the pillbox is often associated with fear of being identified as HIV-positive [16], with stigma being one of the major factors leading to poor adherence and disclosure [17-19]. The assessment of pillbox return could serve as a possible proxy of poor adherence and potentially as a predictor of becoming LTFU. To our knowledge, this easily accessible information on a clinical visit has not yet been analysed as a predictor for being LTFU. Our study aims to determine the association between failed pillbox return and being LTFU.

MATERIALS AND METHODS Study design and setting

This study was nested within the prospective Kilombero and Ulanga Antiretroviral Cohort (KIULARCO), which is among the first rural HIV cohorts in East Africa, established in 2005 at the Chronic Diseases Clinic of Ifakara (CDCI) – the HIV care and treatment centre of the Saint Francis Referral Hospital [20, 21]. The CDCI provides HIV care and treatment services to people residing in the Kilombero and Ulanga Districts in Morogoro region, south-western Tanzania. According to the last census in 2012, these two districts had a total population of 673 083 [22]. Since its establishment, KIULARCO has enrolled over 11 000 PLHIV, with approximately 4200 on active follow-up in 2020. KIULARCO study procedures are described elsewhere [20, 21]. In brief, PLHIV newly tested HIV-positive at our site or being referred from another centre are enrolled into care, started on ART and – if stable – are seen 3-monthly for drug refill and twice a year by a nurse and pharmacist or by a medical doctor. Once yearly laboratory parameters such as HIV viral load, CD4 count and safety laboratory are done. All data are directly collected in an open medical record system (openMRS) and stored on a local server.

Study population

We included PLHIV aged ≥ 18 years enrolled in KIULARCO between January 2013 and March 2019 with follow-up through January 2020, who initiated ART and had at least 6 months of follow-up. Participants were excluded if they were < 18 years old, were in transit (visited facility for drug pick-up only), had never been initiated on ART, or had < 6 months of follow-up (to ensure participants were stable on ART). As per routine care, PLHIV starting ART were provided with drugs for 14–30 days in a pharmacy-packaged envelope. From the second visit onwards, clinically stable patients were dispensed pillboxes containing 30 tablets each for the duration of 3 months – or, if medically indicated, for shorter time periods. During the study period, the first-line ART regimen consisted of efavirenz, lamivudine and tenofivir as a once-daily single pill. The second-line regimen – atazanavir combined with zidovudine/lamivudine – is a once-daily two-pill regimen. In specific situations such as kidney failure or interaction with anti-tuberculous drugs, the treatment was individualized (e.g. abacavir instead of tenofovir, or lopinavir double dose instead of atazanavir). These alternative regimens could consist of up to five pills per day. Viral load data were available in a subset of PLHIV from 2017 onwards when routine viral load monitoring was rolled out in Tanzania.

Study objectives

The primary objective of this study was to determine the association between failed pillbox return and being LTFU. Secondary objectives were to evaluate the association between failed pillbox return and a combined end-point of LTFU/death, predictors of failed pillbox return, the association between failed pillbox return and VF, comparison between failed pillbox return and self-reported adherence to ART, and reasons for missed pills in PLHIV on ART.

Definitions and covariates

Baseline was defined as the latest of ART initiation (i.e. for those who initiated ART within KIULARCO) or enrolment in KIULARCO (i.e. for those who had initiated ART in another clinic before enrolment in KIULARCO). Patients were instructed from the first visit to bring their pillbox back to their next clinic visit for pill count. Failed pillbox return was defined as a failure to present the pillbox at a clinical visit during follow-up and was treated as a time-dependent covariate, updated at each clinic visit. Of course, pillbox return could only be assessed from the first follow-up visit onwards.

Visits were scheduled every 3 months; additionally unscheduled visits were recorded. Being LTFU was defined as not coming to the clinic for > 60 days after the last scheduled appointment [7, 23]. While patients could experience multiple LTFU events after returning to care, we report on the first LTFU event only [7, 24]. For those not LTFU, other possible outcomes were death, transfer to another clinic, or censored at database closure on 16 January 2020 for those active in care. We defined VF as the first elevated viral load > 1000 copies/mL. Poor self-reported adherence was defined as a report of any missed intake of medication in the 4 weeks prior to the scheduled clinical visit. Tuberculosis was recorded if within 3 months from enrolment acid-fast bacilli or a positive Xpert MTB/RIF assay (Cepheid, Sunnyvale, CA, USA) from sputum or an extrapulmonary site were documented, or if anti-tuberculosis drugs with an International Classification of Diseases, 10th Revision (ICD-10) code or clinical signs suggestive of tuberculosis were present. Unlikely tuberculosis was defined as no prescription of anti-tuberculosis drugs and no diagnosis of tuberculosis by ICD-10. For other cases, an indeterminate tuberculosis status was stated and treated as missing data.

Baseline covariates were age, gender, marital status, disclosure of HIV status, partner HIV status, highest education level, distance in kilometres of residence from the clinic, body mass index (BMI), HIV WHO stage, CD4 cell count, tuberculosis status, as defined earlier, ART initiation status (initiated ART after or before enrolment in KIULARCO) and calendar year. The BMI, HIV WHO stage, CD4 cell count and tuberculosis status were measurements closest to baseline, at most 6 months before and up to 3 months afterwards.

Statistical methods

Medians, interquartile ranges (IQRs), frequencies and proportions were used to describe baseline characteristics. We estimated the probability of LTFU over time by pillbox return status using Kaplan–Meier methods. These curves differ from standard Kaplan–Meier survival curves because pillbox return is a time-dependent covariate, i.e. it can change for any given patient at subsequent clinic visits. The estimation procedure for these curves allows for time-dependent pillbox return status [25].

Cause-specific Cox models were used to determine the association between time-dependent failed pillbox return and being LTFU [26]. Participants contributed follow-up time from the first follow-up visit (when pillbox return could first be assessed) until the first LTFU event. For PLHIV who were not LTFU, their follow-up time was censored at the earliest time of death, transfer to another clinic, or the date of database closure. Models were adjusted for baseline covariates. For baseline covariates with missing values, we used missing indicators in order to include all participants in the models (i.e. participants with missing data for a given variable where classified in a separate ‘missing’ category). In sensitivity analyses, we repeated the analyses: (a) incorporating an interaction between ART initiation status (initiated ART in or before KIUARLCO) and failed pillbox return; (b) restricted to participants who initiated ART within KIULARCO, and (c) restricted to participants with complete baseline covariates (those with missing values for any baseline covariate were excluded). We repeated the analysis with the composite outcome of LTFU or death, because in a previous study we found that 40% of KIULARCO participants who were LTFU and traced had died [27].

Baseline covariates (listed earlier) were evaluated as potential predictors of failed pillbox return, using generalized estimating equations for a multivariable repeated-measures logistic regression model to account for the correlation from the patient (i.e. multiple clinic visits for which pillbox return status was recorded). We used an exchangeable correlation structure.

The analysis for the association between failed pillbox return and VF was similar to that for the LTFU outcome described earlier, but restricted to the subset of patients with a viral load test done after implementation of routine viral load testing started in 2017. Patients contributed follow-up time from the first follow-up visit in 2017 to the first VF event, with delayed entry for those patients whose baseline was before 2017. For patients who had no VF, their follow-up time was censored at the earliest time of death, when they became LTFU, transfer to another clinic, or the date of database closure.

Comparison of failed pillbox return and self-reported adherence, and reasons for missing ART doses at each visit were assessed descriptively. Analyses were performed using Stata version 15 [28].

Ethical considerations

The study was nested within KIULARCO, which has attained ethical approval from the Ifakara Health Institute Review Board (IHI/IRB/No:16–2006) and the National Health Research Committee of the National Institute for Medical Research of Tanzania (NIMR/HQ/R.8a/Vol. IX/620), which are both updated annually. Only participants who consented to enrolment into KIULARCO were included in the study.

RESULTS

In total, 4520 PLHIV were enrolled in KIULARCO between January 2013 and March 2019. Of these, 1968 were excluded from the analysis for the following reasons: 469 were below 18 years of age, 408 were not receiving care from our clinic (in transit), 308 had not initiated ART and 783 had a follow-up of < 6 months (Figure 1).

Study population flow chart. Flow chart of patients enrolled in the Kilombero and Ulanga Antiretroviral Cohort (KIULARCO) and included in this analysis. PLHIV, people living with HIV; ART, antiretroviral treatment. Transit are patients who came for drug pickup only

Patients’ baseline characteristics

Among 2552 PLHIV included, at baseline, the median age of patients was 38.4 years (IQR: 31.7–46.2); the majority were female (N = 1735, 68.0%), had normal BMI (N = 1577, 62.7%), were married or cohabiting (N = 1565, 61.3%), and had disclosed their HIV status (N = 1943, 76.1%) mostly to a family member (N = 1227, 49.5%), %) (Table 1). Over half of patients (N = 1417, 59.7%) were at HIV WHO stage I/II and a third (N = 750, 33.5%) had a CD4 count ≥ 350 cells/μL. Most of the patients had a primary school level of education (N = 2108, 82.6%) and lived < 1 km from the clinic (N = 1104, 44.4%). Tuberculosis was diagnosed in 322 (13.0%) patients. The majority of patients were initiated on ART after enrolment in KIULARCO (N = 2118, 83.0%).

TABLE 1. Patients’ characteristics at baseline Patient characteristics All patients (n = 2552) Socio-demographics Age (years) [median (IQR)] 38.4 (31.7–46.2) Age (years) [n (%)] 18–24 197 (7.7%) 25–34 734 (28.8%) 35–44 898 (35.2%) ≥45 723 (28.3%) Gender, female [n (%)] 1735 (68.0%) Marital status [n (%)] Married/cohabiting 1565 (61.3%) Never married 206 (8.1%) Separated/divorced/widowed 781 (30.6%) Disclosed HIV status [n (%)] No 535 (21.6%) Yes 1943 (76.1%) Missing 74 (2.9%) Person to whom disclosed HIV status [n (%)]b Family member 1227 (49.5%) Partner 886 (35.8%) Non-family member 37 (1.5%) Partner HIV status [n (%)] Positive 543 (21.9%) Negative 300 (12.1%) Not tested/unknown 719 (29.0%) No partner 916 (37.0%) Missing 74 (2.9%) Education [n (%)] None 236 (9.3%) Primary 2108 (82.6%) Secondary and above 208 (8.2%) Distance of residence to clinic [n (%)] ≤ 1 km 1104 (44.4%) 2 to < 50 km 825 (33.2%) ≥ 50 km 557 (22.4%) Missing 66 (2.6%) Clinical Body mass indexc,d, c,d (kg/m2) [n (%)] Underweight, < 18.5 370 (14.7%) Normal, 18.5 to < 25 1577 (62.7%) Overweight, ≥ 25 567 (22.6%) Missing 38 (1.5%) WHO stagec [n (%)] I 1033 (43.5%) II 384 (16.2%) III 709 (29.8%) IV 250 (10.5%) Missing 176 (6.9%) CD4 count (cells/µL)c [n (%)] < 100 473 (21.1%) 100–349 1014 (45.3%) ≥ 350 750 (33.5%) Missing 315 (12.3%) Tuberculosis status (TB)c [n (%)] Unlikely 2155 (87.0%) Yes 322 (13.0%) Missing 75 (2.9%) ART initiation status [n (%)] Had already initiated ART in another clinic before enrolment in KIULARCO 434 (17.0%) Initiated ART within 1 month of enrolment in KIULARCO 1835 (71.9%) Initiated ART more than 1 month after enrolment in KIULARCO 283 (11.1%) Calendar year [n (%)] 2013–2014 767 (30.0%) 2015–2016 926 (36.3%) 2017–2019 859 (33.7%) Note Results are number and column % of those with non-missing data; missing data rows are number and column %. Abbreviations: ART, antiretroviral therapy; IQR, interquartile range; KIULARCO, Kilombero and Ulanga Antiretroviral Cohort. Follow-up visits

The total number of visits among 2552 patients was 21 420, with a median of six visits per patient (IQR: 4–12). Of these visits, patients did not bring back their pillbox in 7438 (34.7%), while the pillbox was returned every time in 431 (16.9%). Of the 7438 visits where patients did not bring back their pillbox, in 1956 (26.3%) PLHIV later became LTFU and in 5482 (73.7%) PLHIV remained in care. During a median follow-up of 33.1 months (IQR: 17.5–52.4), 909 (35.6%) patients were LTFU, 43 (1.7%) died and 194 (7.6%) were transferred to another clinic.

Failed pillbox return and LTFU, and LTFU/death

At 24 months, the probability of being LTFU was higher among patients with a failed pillbox return than among those who returned their pillbox (log-rank test, p < 0.001), at 30.0% (95% CI: 26.8–33.2%) versus 19.4% (95% CI: 17.4–21.6%), respectively (Figure 2). This was confirmed by an adjusted hazard ratio (HR) of 1.67 (95% CI: 1.46–1.90, p < 0.001). We also found an association between failed pillbox return and the combined endpoint of LTFU/death (HR = 1.73, 95% CI: 1.52–1.97, p < 0.001) (Table 2). The results were broadly similar in all sensitivity analyses. In particular, there was no evidence of a difference in the association between failed pillbox return and being LTFU or LTFU/death according to whether participants initiated ART in or before KIULARCO (interaction p = 0.69 and p = 0.55, respectively, with broadly similar effect estimates). Other factors associated with LTFU were younger age, living far from the clinic and advanced HIV WHO stage (Table S1).

Probability of being lost to follow-up (LTFU) over time by pillbox return status. Kaplan–Meier estimation of being LTFU for patients with and without pillbox return. Patients who failed to bring back the pillbox are shown in red and those who brought it back are shown in blue

TABLE 2. Association between failed pillbox return and being lost to follow-up (LTFU), and LTFU/death Characteristics Cox model: LTFU Cox model: LTFU/death Univariable Multivariable Univariable Multivariable HR (95% CI)b HR (95% CI)b,c, b,c HR (95% CI)b HR (95% CI)b,c, b,c Failed pillbox return (N = 2552)c No Reference Reference Reference Reference Yes 1.64 (1.44–1.87) 1.67 (1.46–1.90) 1.69 (1.49–1.92) 1.73 (1.52–1.97) Failed pillbox return, among those who initiated ART in KIULARCO (N = 2118)c,d, c,d No Reference Reference Reference Reference Yes 1.67 (1.44–1.92) 1.69 (1.46–1.95) 1.72 (1.50–1.98) 1.76 (1.52–2.03) Failed pillbox return among those who initiated ART before enrolment in KIULARCO (N = 434)c,d, c,d No Reference Reference Reference Reference Yes 1.53 (1.12–2.10) 1.57 (1.14–2.15) 1.52 (1.13–2.09) 1.58 (1.16–2.16) Failed pillbox return (restricted to patients who initiated ART in KIULARCO; N = 2118)c No Reference Reference Reference Reference Yes 1.67 (1.45–1.93) 1.69 (1.46–1.96) 1.73 (1.50–1.98) 1.77 (1.53–2.04) Failed pillbox return (restricted to patients with no missing baseline covariate values; N = 2050) No Reference Reference Reference Reference Yes 1.71 (1.47–1.98) 1.77 (1.52–2.06) 1.77 (1.53–2.04) 1.84 (1.59–2.13) Predictors of failed pillbox return

Factors associated with a higher probability of failed pillbox return were being male, younger age, living closer to the clinic and having a less advanced HIV WHO stage, a higher baseline CD4 count, and later baseline calendar year (Table 3).

TABLE 3. Predictors of failed pillbox return Characteristics Multivariable Multivariable OR (95% CI)d OR (95% CI)d N = 2552 (Missing indicator used for missing covariates) N = 2050 (No missing baseline covariate values) Age (years) 18–24 1.95 (1.60–2.37) 1.86 (1.49–2.31) 25–34 1.65 (1.46–1.87) 1.71 (1.49–1.97) 35–44 1.13 (1.01–1.27) 1.18 (1.04–1.35) ≥ 45 Reference Reference Gender Male Reference Reference Female 0.61 (0.55–0.68) 0.62 (0.55–0.70) Marital status Never married Reference Reference Married/cohabiting 1.06 (0.87–1.29) 1.03 (0.83–1.28) Separated/divorced/widowed 1.01 (0.83–1.24) 1.00 (0.81–1.25) Disclosed HIV status No Reference Reference Yes 1.01 (0.90–1.13) 1.02 (0.90–1.16) Missing 0.74 (0.53–1.05) Partner HIV status Positive Reference Reference Negative 1.04 (0.88–1.22) 1.12 (0.94–1.35) Not tested/unknown 0.98 (0.86–1.11) 1.02 (0.88–1.17) No partner 0.96 (0.83–1.11) 0.97 (0.82–1.14) Missing 1.01 (0.85–1.26) Education None Reference Reference Primary 1.09 (0.94–1.27) 1.10 (0.92–1.30) Secondary and above 1.18 (0.94–1.47) 1.11 (0.87–1.42) Distance from residence to clinic ≤ 1 km Reference Reference 2 to < 50 km 0.86 (0.77–0.95) 0.77 (0.85–0.97) ≥ 50 km 0.75 (0.66–0.84) 0.75 (0.66–0.86) Missing 0.78 (0.56–1.09) Tuberclosis status No Reference Reference Yes 1.12 (0.96–1.30) 1.08 (0.93–1.26) Missing 1.17 (0.95–1.1.51) Body mass index (kg/m2) Underweight, < 18.5 0.94 (0.82–1.08) 0.92 (0.79–1.07) Normal, 18.5 to < 25 Reference Reference Overweight, ≥ 25 1.08 (0.96–1.21) 1.03(0.90–1.17) Missing 0.88 (0.61–1.27) WHO stage I Reference Reference II 0.92 (0.80–1.06) 0.94 (0.81–1.09) III 0.83 (0.73–0.94) 0.81 (0.71–0.93) V 0.80 (0.68–0.97) 0.82 (0.67–0.99) Missing 1.06 (0.83–1.34) CD4 count (cells/µL) < 100 Reference Reference 100–349 1.00 (0.87–1.14) 1.04 (0.90–1.19) ≥ 350 1.25 (1.08–1.44) 1.30 (1.11–1.52) Missing 1.16 (0.96–1.40) Initiated ART in KIULARCO No Reference Reference Yes 0.95 (0.83–1.10) 0.95 (0.81–1.11) Calendar year 2013–2014 Reference Reference 2015–2016 1.42 (1.28–1.59) 1.39 (1.24–1.57) 2017–2020 2.05 (1.82–2.32) 2.02 (1.76– 2.31)