The Chronic Diseases Clinic at St. Francis Referral Hospital, Ifakara, Tanzania enrolls HIV-positive patients in a prospective cohort (Kilombero and Ulanga Antiretroviral Cohort (KIULARCO)). Written informed consent was obtained from the patient or, if younger than 18 years, the caregiver. Since its conception in 2005, KIULARCO enrolled more than 10,000 HIV-infected patients. Demographic, clinical, and treatment information is collected 4 times per year. Plasma is sampled twice yearly with storage in an onsite biobank. The cohort has been described in detail in other publications [11, 12]. For this study, we included all patients, enrolled into KIULARCO from 2005–2016, who were started on bPI-based ART, and who had a stored plasma sample taken 6–12 months after the start of treatment. We also used data from those newly enrolled on bPI treatment with a plasma sample taken at 6–12 months after enrolment. No routine viral monitoring was in place during the study period; however, the treating physician upon suspected immunologic or clinical failure could order viral load testing.

Data on demographics, clinical progression and ART was extracted from the KIULARCO electronic medical records. We recorded risk factors for treatment failure at the start of bPI treatment; namely sex, age, VF on first-line ART defined as viral load (VL) ≥1,000 copies/mL, advanced HIV disease (CD4 cells <200/μL and/or WHO stage III/IV), BMI (only in adults), and self-reported non-adherence [4, 13]. VL was tested on samples stored 6–12 months after bPI start (‘6–12 months sample’). If available, VL was additionally measured on samples taken within 6 months prior to start of bPI (‘pre-bPI sample’). Furthermore, we determined patients’ VL in a sample from >12 months after initiation of bPI, and at the time of clinical or immunologic treatment failure on bPI treatment (‘failure sample’). Resistance testing was carried out in samples with a VL ≥1,000 copies/mL, if amplification was successful.

Cell-free plasma was collected after centrifugation of EDTA blood at 1,900 rcf for 5 minutes and stored at -20°C until tested for VL, DRM or, after 2 weeks of being stored at -80°C. HIV RNA was extracted from 200 μL of plasma using the Pure Link Viral RNA/DNA Mini Kit (Thermofisher Scientific, Allschwil, Switzerland), according to the manufacturer’s instructions. Viral RNA quantification was performed with a validated in-house protocol with the Taqman qPCR Master Mix (Thermofisher) using the StepOne^TM System (Thermofisher), with a detection limit of 100 viral RNA copies/mL; details have been reported elsewhere (dx.doi.org/10.17504/protocols.io.749hqz6) [14]. For all VLs above 1000 copies/mL, genotyping of HIV-1 DRM was performed with a validated in-house protocol by population sequencing on an ABI 3130 capillary genetic analyzer, using six primers covering protease and reverse transcriptase [14]. HIV-1 drug resistance was interpreted according to the Stanford HIV Drug Resistance Database algorithm, version 8.4 (http://hivdb.stanford.edu). All DRM conferring high-level resistance were considered. CD4 cell counts were determined by flow cytometry (FACS Calibur, BD Company, Franklin Lakes, New Jersey, USA or Partec Cyflow counter, Partec GmbH, Germany).

The primary outcome was VF, defined as HIV-1 RNA levels ≥1000 copies/mL. The secondary outcome was the prevalence of drug resistance in patients with VF, as per the Major HIV-1 DRM list [15].

Baseline was defined as the time point of starting a bPI ART or, if already on bPI ART, at inclusion into the cohort; variables and co-variates for risk factor analyses were considered up to 6 months before this date. Non-adherence was defined as missed at least one pill at any time during the past 6 months. VL results between 100–1000 copies/mL were defined as low-level viremia (LLV), whereas levels ≥1000 copies/mL where defined as VF. Advanced HIV disease was defined as CD4 cells <200/μL and/or a clinical WHO stage III/IV.

The study was conducted according the current version of the World Medical Association Declaration of Helsinki. The KIULARCO study was approved by the Ifakara Health Institute institutional review board (IHI/IRB/No 16–2006), the Health Research Ethics Review Committee of the National Institute for Medical Research of Tanzania (NIMR/HQ/R.8a/Vol. IX/620), and the Ethical Review Board of the Canton of Basel, Switzerland.

Demographic factors were summarized using frequencies and percentages for categorical data and medians and IQR for continuous data. Univariable and multivariable logistic regression models were used to estimate the association between risk factors and VF at 6–12 months. Non-linearity of the association with continuous variables was tested and modeled accordingly. Results are presented as odds ratios with 95% confidence intervals. Statistical analyses were performed using STATA 14 (StataCorpLP, College Station, TX, USA).

Between 2005–2016, 584 participants of the KIULARCO cohort, received a bPI-based treatment (Fig 1). Of these, 332 were not eligible for the following reasons: being on bPI for a duration of less than 6 months, being switched back to a non-bPI regimen before 6 months, or no sample available 6–12 months after bPI start. Characteristics at the start of bPI treatment of the eligible 252 patients (199 adults and 53 children) included in the final analysis are summarized in Table 1. Out of 199 adults, 121 (60.8%) were female (2 were pregnant), median age and BMI were 41 years (IQR 33–49), and 21.5 (IQR 19.1–23.5) respectively. An advanced WHO stage (III/IV) was recorded in 105 (52.8%) patients. The median CD4 cell count was 163/μL (IQR 71–276), and median time on ART before start of a bPI was 3.6 years (IQR 1.5–5.8). Out of 53 children, 20 (37.7%) were female, median age was 8 years (IQR 5–10), 39 (73.6%) were in WHO stage III/IV, and median CD4 cell count was 590/μL (IQR 250.0–940.5) with a median time on first-line ART of 3.3 years (IQR 2–5.2).

The pre-bPI ART consisted of one NNRTI, mostly combined with zidovudine (69.9%), stavudine (42.9%) or tenofovir (32.9%). Due to reasons such as phasing out of certain regimens, adverse reactions, non-adherence, patient decision or stock-outs, many patients experienced several changes in their first-line treatment (median 2 (IQR 1–4)) before the start of a bPI-based treatment.

Eleven out of 199 (5.5%) adults and 7/53 (13.2%) children had already been prescribed a bPI treatment at the time of enrolment into our clinic. In adults, this was due to previously suspected treatment failure. In contrast, children were started directly on a bPI regimen, which was based on new guidelines [3]. At the start of the bPI-based regimen, tenofovir was the most commonly prescribed backbone drug (204 (81.0%)).

Reasons for prescribing a bPI-based regimen were immunologic failure in 127 patients (50.4%), clinical failure in 20 patients (7.9%), or as VL testing was not routine in Tanzania at the time of the study, suspected VF (107 patients (42.5%)).

In our study population (85 available sequences from 75 patients) subtype C was most prevalent (35 isolates, 46.7%), followed by subtype A (23 isolates, 30.7%) and subtype D (9 isolates, 12%). This correlates well with the previously reported subtype distribution for the region [14]. Recombinant HIV-1 forms were identified in 8 samples, 7 of subtypes C and D, including CRF10_CD, and 1 of B/D.

Overall, we found bPI DRM only in 2 adult patients. Both had been on a long-term bPI regimen (1027 and 1643 days, respectively) and both had accumulated multiple major bPI DRM. The first of these patients had the following mutations: M46I, L76V, V82A and I84IV. The second patient had L24I, M46I, I50L, I54A, and V82A.

The most common NNRTI mutation was K103NS, detected in 29 (51.8%) adults and 7 (36.8%) children. Additionally, Y181CIV was found in 15 (26.8%) and 4 (21.1%), E138KAQG in 10 (17.9%) and 4 (21.1%) and G190ASEQ in 12 (21.4%) and 5 (26.3%) adults and children, respectively.

For the NRTI, the most common non-TAM DRM was M184VI, which was found in 35 (62.5%) adults and 14 (73.7%) children. TAMs were found in 18 (32.1%) adults and 11 (57.9%) in children (Table 3).

Prior to starting bPI, 27/199 (13.5%) adults and 15/53 (28.3%) children had reported episodes of non-adherence to ART intake. After 6–12 months on bPI, 19/199 (9.5%) adults and 9/53 (17.0%) children still reported non-adherence (S1 Table). Most common reasons for non-adherence were ‘having lost’ or ‘run out of’ medication, reported in 9/42 (21.4%) cases on first-line treatment and 11/28 (39.3%) on second-line treatment. One patient indicated depression as a reason for not taking pills during first-line treatment (1/42; 2.4%) and one on second-line (1/28; 3.6%). Reasons for poor adherence reported on first-line treatment only were ‘being too ill or too healthy’ (3/42; 7.1%), ‘being concerned about stigma’ (2/42; 4.8%) and ‘problems with delivery or stock out’ (1/42; 2.4%).

For adults, the presence of advanced HIV disease (WHO stages III/IV or CD4<200) was the only significant factor associated with a positive VL at 6–12 months on bPI (univariate analysis Odds ratio 5.04; 95% confidence interval 1.15–22.15, p = 0.03). For children, none of the investigated factors was significantly associated with a positive VL (Table 4).