In this mixed-methods study of a sequential sample of adults presenting to initiate or re-initiate HIV treatment in South Africa and self-reporting no prior ART use within the past 90 days, 55% had at least one indicator suggesting previous ART use. Roughly one fifth were found to have ARV metabolites in blood samples indicating recent (within 90-day) use. An additional third had records of prior viral load tests or ART care, evidence that they were likely restarting ART after a short or long interruption. The qualitative interviews of 11 respondents with discordant results suggest a variety of reasons for both interrupting treatment and non-disclosure of prior experience, including deliberately presenting as new to avoid potential stigma from clinic staff and burdensome transfer processes.

The primary strength of this study was the use of multiple data sources to find evidence of prior ART exposure in the same sample. We found a high level of non-concordance among the sources, with only 10% of those who appear to have prior ART experience showing positive indicators across all sources. This is not entirely surprising, as some indicators—self report and EMR and lab records—represent lifetime experience, while others—metabolite testing (and baseline viral loads, to the extent captured)—reflect only recent use. It does make clear that none of the available indicators alone is sufficient for an accurate estimate. In our study, laboratory records and EMR records were the most accurate, while self-report was the least.

Metabolite testing and baseline viral load tests, both of which can accurately discern recent ART use, also both pose a number of problems for routine use. Metabolite testing is expensive, can only be carried out in specialized laboratories and, in our analysis, did not perform better than self-report. Baseline viral load tests, which are being introduced into routine care in some settings[11] but are not recommended as part of routine care in South African treatment guidelines, are less expensive but, in the absence of rapid point-of-care technologies, provide information only days or weeks after the initiation visit, by which time many patients have already disengaged from care. Both of these approaches, moreover, only capture ART exposure within the past 1-3 months, before viral rebound or metabolite clearance. In view of this, active review of patients’ clinic records (EMR) and/or laboratory records may be the most practical approach to identifying re-initiators, particularly where such records are available to clinicians in real time at point of care. In South Africa, for example, a counselor or clerk may be able to look up patients’ clinic and lab histories while a patient waits for services and convey any evidence of prior treatment to the nurse before the treatment initiation consultation begins.

This study follows on a systematic review we conducted that found limited data on prior treatment exposure[12]. We found that there was no consistency in defining populations where prior treatment exposure was being reported and varied estimates in prior exposure measured in similar ways. For example, self-reported estimates in the literature range from 2-21%[13–18], which places our estimate of 18% on the high end. When using EMR data, estimates have ranged from 22% in a population that had ART naivete in the inclusion criteria[19] to 69% in a dataset that considered all initiations and allowed for individuals to be tracked across multiple facilities[3]. Metabolite testing also ranged from 19% when testing only blood specimens for tenofovir and emtricitabine among participants with undetectable viral loads at initiation to 53% when using both blood and hair samples to test for tenofovir, efavirenz and emtricitabine among a study population who all reported being ART naive[19,20]. Baseline viral load suppression is another measure reported in the literature with relatively high yield of previous exposure (30-52%)[21,22]. Among the small number of participants in our study who had a viral load performed as part of their initiation laboratory tests, however, none were suppressed.

A further strength of this study is that in addition to these measures, we also conducted qualitative interviews with participants who had tested positive for ART metabolites after reporting being treatment naïve or not using ART in the preceding 90 days. One interviewee reported using PrEP, which commonly includes tenofovir and can remain detectable in dried blood specimens for up to 3 months[9]. As PrEP use expands, testing for metabolites of drugs common to PrEP and ART may become less useful. It is perplexing and concerning that three of the 11 participants we interviewed continued to deny any known prior exposure to ART before the day of enrollment into the study. While it is possible that participants did not remember prior use or were inadvertently exposed through recreational drugs, neither of these explanations seems likely. We speculate that it is more likely that these three individuals simply represent more adamant examples of the seven clients who deliberately presented to the facilities as naive and determined that it was in their interest not to reveal prior exposure even to researchers.

For those who did reveal to our interviewer that they had deliberately concealed their prior ART use, the quotations included here offer a range of reasons that are consistent with prior research[23,24] and appear eminently rational from the perspective of the patients. Despite national guidelines that advise clinic staff to be welcoming and encouraging to re-engagers[25–27], there is abundant evidence that facilities and providers deter patients from revealing prior disengagement[23,24,28]. Previous research suggests that provider disposition, in particular, is an important driver of patient behavior [29]. Re-engaging patients report being chastised (or worse), sent to the back of the queue, or told to return on another day with an official transfer letter from another facility. All of these actions on the part of providers would lead patients to hide their previous treatment, particularly when initiation procedures for naive patients have become simpler and faster in recent years [30,31]. The time required to locate existing files is also an obstacle, compared to opening a new file. It is striking that one respondent reported that clinic staff advised him to present as a new patient, suggesting that staff, too, regard the re-engagement process as burdensome.

Our study had a number of limitations, leading us to regard it as exploratory rather than definitive. First, our inclusion criteria required participants to conform that they were ART naïve or had a treatment interruption of 90 days or more and were informed that we would conduct metabolite testing and search medical records for prior ART exposure. This process may have led some potential participants to self-select out of the study at the time of screening because they were aware of having undisclosed, prior treatment experience, leading us to underestimate the proportion of initiators with positive measures of prior exposure. Second, we could not access EMR records from other clinics participants may have attended previously, making our EMR-based result for prior exposure almost certainly an under-estimate. Third, we did not have baseline viral load test results for most participants. We did conduct viral load tests on the same dried blood specimens used for metabolite testing for all participants, but results were sufficiently inconsistent with the routinely reported viral load tests conducted by clinics for the 10 clients mentioned above that we discarded the study DBS findings as unreliable. This may have been due to a long delay between sample collection and ultimate viral load testing of DBS or the unreliability of DBS samples for viral load estimates. Fourth, while we are not aware of other studies using multiple data sources in the same sample to compare the sensitivity of different indicators, our sample was small and limits our ability to comment on predictors of prior treatment exposure with confidence. Fifth, our small number of facilities in just three districts makes generalizability to other locations uncertain, such that our results should be regarded as exploratory only. Sixth, there may be a difference between self-reported prior exposure reported to study interviews and that revealed to clinic staff. Finally, our qualitative analysis might have been subject to survey bias and we are uncertain if we reached saturation or predictability, though we did see repeated themes that elicited important nuance. Despite these limitations, our results are consistent with prior studies using single indicators.