Introduction

As HIV programmes have scaled up and access to antiretroviral therapy (ART) improved, sustaining viral suppression and improving outcomes among individuals with viremia has become a critical priority [1]. Existing evidence suggests that across Africa, 50%–70% of individuals with treatment failure (i.e., viral load [VL] ≥1,000 copies/ml) have documented re-suppression after receiving enhanced adherence counseling (EAC), a structured counseling intervention to identify and address barriers to medication adherence for 3–6 months [2], though re-suppression rates are likely higher among those on integrase-based regimens [3–7]. Furthermore, with ongoing shifts from tenofovir disoproxil fumarate (TDF)/lamivudine or emtricitabine (XTC)/efavirenz (EFV) (TLE) to TDF/XTC/dolutegravir (DTG) (TLD) as first-line ART regimens, it remains important to understand how these recent transitions impact fidelity to treatment failure algorithms, rates of second-line switch, and overall re-suppression rates [8,9].

Identifying care gaps and their contribution to suboptimal clinical outcomes is essential for developing effective strategies to optimize care delivery after treatment failure [10,11]. Rates of documented re-suppression reflect a complex interplay of client-level challenges, health system constraints, implementation issues, and biomedical factors (i.e., drug resistance). On the client side, inadequate adherence and drug resistance remain concerns, while delays in returning to care for EAC and poor retention that also delay VL testing may also be important drivers of treatment failure [12]. On the health system side, challenges include inadequate outreach to reengage individuals after identifying treatment failure, poor quality EAC that could exacerbate engagement challenges, delays in completing EAC, failure to order repeat VL monitoring, and missed opportunities to switch to second-line therapy when appropriate [12]. By examining transitions through key steps in the treatment failure cascade, including repeat VL monitoring, initiating EAC, and switching treatment regimens, we can gain insights into health system and client factors that drive outcomes. This approach also helps assess whether the transition from TLE to TLD has influenced care patterns, provider behaviors, and outcomes.

In this manuscript, we employ multistate analytic approaches to comprehensively examine care journeys of people living with HIV who experience treatment failure during routine care in Zambia. Multistate analytic approaches account for client transitions in between different care cascade states over time and can help to better characterize care patterns and identify the magnitude of care gaps at different stages [13–15]. Using data from 24 public HIV health facilities in Lusaka, Zambia, we examine the longitudinal dynamics of clients’ care engagement after treatment failure with respect to return clinic visits (as a proxy for EAC initiation) and retention in care, repeat VL monitoring, switch to second-line therapy, rates of re-suppression, and how these differ on TLE versus TLD [13,16]. We hypothesize that delays in key steps of the treatment failure cascade may significantly contribute to suboptimal VL suppression rates among individuals with treatment failure in Zambia.

Methods

Statistical analyses

VL monitoring outcomes after initial treatment failure (i.e., VL ≥ 1,000 copies/mL).

We used multistate analytic methods to describe the longitudinal care and VL monitoring of clients after treatment failure (i.e., their first elevated VL test ≥1,000 copies/ml). We tracked transitions between the various care states that occur over time up until a subsequent VL measure was repeated per guideline recommendations (e.g., return visits without repeat VL test, treatment interruption [>60 days late for scheduled visit], repeat VL test suppressed versus unsuppressed) [13]. Time zero (i.e., cohort entry) was the date of first elevated VL test, and clients were followed until censoring occurred at the time of a repeat VL measure, transfer, death, or administratively at the end of the observation period of 10 January 2022. In the first stage analysis, individuals were categorized into 1 of 9 mutually exclusive and exhaustive states based on their clinical status at each time point after cohort entry (i.e., first elevated VL test) (Fig 1A): (1) in care but no return visit since initial elevated VL (treatment failure), (2) 1 return visit with no repeat VL measure, (3) 2 return visits with no repeat VL measure, (4) 3+ return visits with no repeat VL measure, (5) treatment interruption, (6) repeat VL not suppressed (≥1,000 copies/ml), (7) repeat VL suppressed (<1,000 copies/ml), (8) documented transfer, and (9) death. We then applied nonparametric multistate analysis based on the Aalen–Johansen method that accounted for client movements between care states over time [19]. Individuals could transition between nonabsorbable states an unlimited number of times based on the Markov assumption (i.e., after entering a state, prior states do not impact subsequent outcomes) until they transitioned into an absorbable state (i.e., repeat VL measure, transfer, death) or were [24] censored (Fig 1) [19]. We estimated the probability of a client being in any particular state over time, and generated estimates for composite states such as any history of return visit, VL test rechecked, ever >60 days late. We also calculated median times with interquartile range (IQR) and instantaneous hazards of transition between states. We stratified analyses based on whether individuals were on TLE or TLD at the time of initial treatment failure to examine differences in care delivery and outcomes across these groups. We also conducted sensitivity analyses where we stratified by secular time periods relative to the COVID-19 pandemic. We used bootstrapping (1,000 iterations) to obtain 95% confidence intervals (CI).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. State transition flowchart for multistate analysis.

This figure depicts all possible transitions between care states over time. A. At each time point after cohort entry, clients were categorized into 1 of 9 mutually exclusive and exhaustive states. For the first stage analysis, those with one elevated VL ≥1,000 copies/ml could be in states: (1) treatment failure, with no return visit and not returned to care, (2) 1 attended visit for EAC without a repeat VL, (3) 2 attended visits for EAC without a repeat VL, (4) 3+ attended visits for EAC without a repeat VL, (5) treatment interruption (i.e., >60 days late for visit after an elevated VL), (6) repeat VL not suppressed (≥1,000 copies/ml), (7) repeat VL suppressed (<1,000 copies/ml), (8) transfer, and (9) death. B. For our second stage analysis, those with a second elevated VL ≥1,000 copies/ml (6) were then recategorized as (10) now due for second-line switch and were categorized into another 1 of 10 mutually exclusive and exhaustive states at each time point after the second elevated VL: (10) treatment failure, due for second-line switch and with no switch or return visit, (11) 1 attended visit for EAC and no switch made, (12) 2 attended visits for EAC and no switch made, (13) 3+ attended visits for EAC and no switch made, (14) treatment interruption (i.e., >60 days late for visit after an elevated VL), (15) repeat VL not suppressed, (16) repeat VL suppressed (<1,000 copies/ml), (17) ART regimen switch, (18) transfer, and (19) death. This figure depicts all the possible transitions clients could make from each state. Note: Clients attended visits presumably for EAC, though this is not confirmed. Abbreviations: EAC, enhanced adherence counseling; VL, viral load. *If a client had a prior unsuppressed repeat VL and had a visit after treatment interruption, they were classified as unsuppressed at that visit.

https://doi.org/10.1371/journal.pmed.1004720.g001

Guideline concordant switch to second-line ART regimens (i.e., after two elevated VL ≥1,000 copies/mL).

Among individuals with two consecutive elevated VL tests, we repeated multistate analyses to assess the longitudinal experience of switching to second-line therapy (State 6 from first multistate analysis) in a second-stage analysis. To ensure regimen changes were due to treatment failure rather than programmatic roll out of TLD, this analysis was restricted to periods after individuals had two consecutive elevated VL tests. For the second stage analysis, time zero was the date of second consecutive elevated VL test, and clients were censored at the time of a suppressed repeat VL test (after already having at least two consecutive elevated VL tests), regimen switch to second-line, transfer, death, or administratively censored. Individuals were again categorized in mutually exclusive and exhaustive states over time (Fig 1B): (10) treatment failure, due for second-line switch with no return visit, (11) 1 attended visit with no switch made, (12) 2 attended visits with no switch made, or (13) 3+ attended visits with no switch made, (14) treatment interruption, (15) repeated VL test unsuppressed (≥1,000 copies/mL), (16) repeated VL test suppressed (<1,000 copies/mL), (17) switched to second-line regimen, (18) documented transfer, and (19) death. For this analysis, absorbable states included a suppressed repeat VL test (after already having at least two consecutive elevated VL tests), regimen switch to second-line, transfer, and death. A repeat unsuppressed VL test was considered nonabsorbent as individuals were still indicated for a regimen switch (Fig 1B). Analyses were stratified based on whether individuals were on TLE or TLD at the time of initial treatment failure and time periods. We used bootstrapping (1,000 iterations) to obtain 95% confidence intervals.

Among individuals who switched to second-line regimens after two consecutive elevated VL tests, an alluvial diagram was used to depict history of ART regimen changes and assess concordance of these changes with the national guidelines at the time of initial treatment failure, subsequent VL test, and at switch to second-line. ART regimens were categorized based on their main drug class (i.e., EFV, DTG, or LPVr) and NRTIn (nucleoside/nucleotide reverse transcriptase inhibitors) backbone (TDF/XTC compared to AZT/XTC). This analysis was restricted to clients who had an ART regimen recorded in the electronic health record.

Results

Client characteristics

Between 1 August 2019 and 31 November 2021, 179,885 clients accessed care at one of the 24 study facilities after being on ART for at least 180 days, and 143,857 (79.9%) had a VL test on record. Among those with a VL test, 10,758 (7.5%) individuals had a VL test ≥1,000 copies/ml and of those, 7,916 (73.6%) individuals had their first-line drug regimen documented at the time of treatment failure and were included in this analysis (Fig 2). Of these, 4,138 (52.3%) were female, median age was 36.7 years (IQR 29.9–43.6), and median time on ART at treatment failure was 3.3 years (IQR 1.7–6.6) (Table 1). Of those who had their first-line drug regimen documented, 4,323 (54.6%) were on TLD and 3,593 (45.4%) on TLE, at the time of the first elevated VL test (Table 1, S1 Fig). Those on TLD were less likely to be female and more likely to have been on ART for less than 2 years, but other baseline characteristics were similar compared to those on TLE.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Baseline characteristics of clients with an elevated viral load.

https://doi.org/10.1371/journal.pmed.1004720.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Flowchart of client inclusion and outcomes.

The flowchart shows the selection of patients with an elevated VL from an initial cohort of 179,855 patients on ART for ≥6 months with visits between 12 Aug 2019 and 12 Nov 2021, outcomes following. We excluded 35,998 (20.0%) as they had no VL test, and 133,099 (92.5%) had no elevated VL in the electronic health record. We excluded 2,842 (2.0%) who had an elevated VL but no described drug regimen. We included a total of 7,916 (5.5%) in our analysis cohort with an initial elevated VL test and drug regimen described at failure. Of these, 48.2% achieved suppression, 19.5% had a second elevated VL test, 30.4% had no follow-up VL test, and 2% had either transferred or died. For those with two consecutive elevated VLs (n = 1,540), outcomes included subsequent suppression without switch (18.4%), regimen switch (56.9%), transfer or death (0.8%), or no further action (23.9%). Abbreviations: ART, antiretroviral therapy; VL, viral load.

https://doi.org/10.1371/journal.pmed.1004720.g002

Outcomes after first elevated viral load among clients due for EAC and repeat viral load in 3–6 months

Thirty-seven percent (37.0%, CI 36.2, 37.8%) and 72.2% (CI 71.3, 73.0%) had returned to clinic for at least 1 visit (including visits with and without a VL test) by 3- and 6-months (Figs 3A and S2 and S1 and S2 Tables), respectively, after the initial elevated VL test (median time to return 92 days IQR 42–168 among returners). At 6- and 12-months, 36.4% (CI 35.5, 37.2%) and 70.1% (CI 69.3, 70.9%) in the overall cohort had a repeat VL test (Figs 3A and S2 and S2 Table). Among those who had returned for at least 1 visit, 49.9% (CI 48.9, 51.0%) and 78.4% (CI 77.6, 79.3%) had a VL test repeated at 6- and 12-months, respectively. Of those with a repeat VL test, median overall time from initial VL test to repeat VL test was 180 days (IQR 117–254), and median time from returning to clinic to having VL test repeated was 63 days (IQR 0–132). At 12 months, 16.6% (CI 15.9, 17.2%) of individuals were currently experiencing a treatment interruption, 32.5% (CI 31.8, 33.3%) had ever experienced a treatment interruption since treatment failure, and 11.4% (CI 10.3, 12.4%) had return visits but no repeat VL test. Additionally, 59.3% (CI 57.5, 61.0%) of those who did not have a VL rechecked by 1 year were in a treatment interruption (Figs 3A and S2 and S2 Table). Results were similar when stratified by COVID-19 period (S3 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Longitudinal outcomes after first VL ≥1,000 copies/ml.

This figure shows the proportion of clients estimated to be in each state at any given time after an individual’s first elevated VL, accounting for the transitions made between different states over time. (A) Outcomes following a first VL ≥1,000 copies/ml for all regimens (n = 7,916); (B) Outcomes following a first VL ≥1,000 copies/ml, regimen at cohort entry TDF/XTC/DTG (TLD) (n = 4,323); (C) Outcomes following a first VL ≥1,000 copies/ml, regimen at cohort entry TDF/XTC/EFV (TLE) (n = 3,593). Abbreviations: TLD, tenofovir disoproxil fumarate/lamivudine or emtricitabine/dolutegravir TDF/XTC/DTG; TLE, tenofovir disoproxil fumarate/lamivudine or emtricitabine/efavirenz TDF/XTC/EFV; VL, viral load.

https://doi.org/10.1371/journal.pmed.1004720.g003

In the overall cohort, 50.1% (CI 48.9, 51.3%) of individuals had documented re-suppression at 12 months. At 6 months, 29.2% (CI 28.1, 30.3%) of individuals on TLD and 21.0% (CI 20.0, 22.0%) on TLE had documented re-suppression, increasing to 59.2% (CI 57.9, 60.4%) on TLD and 41.1% (CI 40.0, 42.2%) on TLE at 12 months. Among those who had a VL test repeated by 12 months, 85.0% (CI 83.9, 86.1%) on TLD resuppressed compared to 58.2% (CI 56.8, 59.8%) on TLE (risk difference 26.8% CI 25.0, 28.6%) (Figs 3B, 3C, and S2 and S2 Table).

Outcomes after two consecutive elevated viral loads among clients due for switch

A total of 1,540 individuals had a subsequent elevated VL test and were due for a switch to second-line regimen per national ART guidelines (889 [57.7%] female; median age 35.5 [IQR 28.6–42.1]), (390 [25.3%] on TLD and 1,150 [74.7%] on TLE) (S4 Table). After a second elevated VL test and being due for switch, 63.3% (CI 61.5, 65.2%) and 89.2% (CI 87.9, 90.6%) had returned to clinic for at least 1 visit by 3- and 6-months, respectively (median time to return 49 days IQR 29–92 among returners) (Figs 4A and S3 and S5 Table). Among those on TLD, at 6 months, 24.0% (CI 20.5, 27.5%) were switched and 33.2% (CI 29.4, 37.3%) had another VL test rechecked for a third time prior to switch (68.0% CI 61.6, 75.2% suppressed among those with repeated VL), going up to 27.9% (CI 24.1, 31.5%) switched and 52.4% (CI 47.2, 57.4%) had another VL test rechecked (70.0% CI 64.1, 76.6% suppressed) at 12-months (Fig 4B and 4C and S5 Table). In contrast, among those on TLE, 51.7% (CI 49.5,54.2%) were switched and 22.6% (CI 20.9, 24.6%) had a VL test rechecked prior to switch (40.7% CI 36.1, 45.4 suppressed) at 6-months, and that increased to 66.6% (CI 64.5, 68.9%) switched and 32.1% (CI 29.9, 34.1%) had a VL test rechecked (42.9% CI 39.1, 47.0% suppressed) by 12-months. Overall, at 12 months, 38.7% (CI 34.5, 43.6%) fewer individuals on TLD had been switched compared to those on TLE, but 24.1% (CI 19.1, 29.4%) more had resuppressed prior to switch (Figs 4A and S3 and S5 Table). Lastly, 13.2% (CI 11.7, 14.5%) of individuals were currently experiencing treatment interruption and 23.4% (CI 21.6, 24.8%) had ever had treatment interruptions prior to switch or re-suppression at 12 months in the overall cohort.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Longitudinal ART regimen switch outcomes after two consecutive elevated VL ≥1,000 copies/ml, TLD vs. TLE.

This figure shows the proportion of clients estimated to be in each state at any given time after an individual is due for a regimen switch (i.e., two consecutive elevated VLs), accounting for the transitions made between different states over time. (A) Outcomes following two VL ≥1,000 copies/ml (n = 1,540). (B) Outcomes following two VL ≥1,000 copies/ml, regimen at cohort entry TDF/XTC/DTG (TLD) (n = 390); (C) Outcomes following two VL ≥1,000 copies/ml, regimen at cohort entry TDF/XTC/EFV (TLE) (n = 1,150). Abbreviations: ART, antiretroviral therapy; TLD, tenofovir disoproxil fumarate/lamivudine or emtricitabine/dolutegravir TDF/XTC/DTG; TLE, tenofovir disoproxil fumarate/lamivudine or emtricitabine/efavirenz TDF/XTC/EFV; VL, viral load.

https://doi.org/10.1371/journal.pmed.1004720.g004

Trajectories of ART regimens

876 individuals were switched to second-line regimen after at least two consecutive elevated VL tests (581 [66.3%] female; median age 35.9 [IQR 29.2–42.1]), (96 [11.0%] on TLD and 780 [89.0%] on TLE) (Fig 5). Median time from initial treatment failure and from second elevated VL test to switch was 273 days (IQR 189–379) and 91 days (IQR 39–167), respectively. Among those on TLD at time of treatment failure, 46 (47.9%) switched to AZT/XTC/LPVr, 35 (36.5%) only had TDF changed to AZT (i.e., switched to AZT/XTC/DTG), and 8 (8.3%) were transitioned to an NNRTI-based regimen. Among those on TLE, 366 (46.9%) switched to AZT/XTC/DTG, 248 (31.8%) to TDF/XTC/DTG, and 139 (17.8%) to AZT/XTC/LPVr. Only 14 (1.8%) had a TDF switch to AZT. At 6-months after switch, 54.3% (CI 51.5, 57.2%) had a VL test rechecked overall and this increased to 80.7% (CI 78.3, 83.3%) by 1-year; 83.0% (CI 80.7, 85.3%) of those who had VL test rechecked by 1 year were suppressed (75.4% [CI 66.6, 84.8%] on TLD first-line versus 83.7% [CI 81.4, 86.0%] on TLE first-line).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Drug at regimen switch trajectories.

This figure illustrates the common regimen transitions among 876 patients from the time of the first elevated VL to the second elevated VL and eventual switch. Those on TDF/XTC/DTG most frequently switched to TDF/XTC/DTG and AZT/XTC/DTG. Those on TDF/XTC/EFV had diverse switches with most being switched to AZT/XTC/DTG and TDF/XTC/DTG, with a smaller proportion being switched to AZT/XTC/LPVr. Abbreviations: AZT, Azidothymidine; DTG, dolutegravir; EFV, efavirenz; LPVr, ritonavir-boosted lopinavir; TDF, tenofovir disoproxil fumarate; VL, viral load; XTC, lamivudine or emtricitabine.

https://doi.org/10.1371/journal.pmed.1004720.g005

Discussion

Among individuals with treatment failure, we identified substantial delays and missed opportunities in rechecking VL test status and switching to second-line therapy in routine care in Zambia. Only 36.4% of individuals had a VL test rechecked by 6 months after initial treatment failure, driven by delays in returning to clinic, treatment interruptions, and missed opportunities to do a VL test after return. At 1-year, only 50.1% had documented re-suppression. However, rates of re-suppression were high among those who had a VL test rechecked (70.1% overall), and even higher for those on TLD (85.0%). Similar patterns occurred after confirmation of failure based on two consecutive elevated VL tests, with fewer than 25% switching within 6 months of initial treatment failure (upper limit of guideline recommendations). Notably, individuals on TLD were far more likely to resuppress (85.0% versus 58.2% on TLE), despite similar care patterns and timing of VL test rechecks. Providers were also more reluctant to switch those on TLD to second-line regimens. These findings reflect real-world outcomes under routine programmatic settings and underscore the critical care gaps driven by both client and health system barriers that impede timely delivery of recommended care.

First, many clients did not return promptly, which likely delayed both the initiation of EAC and timely assessment of response. By 3 months, only 37% had returned to the clinic, and only 7% had a VL test rechecked. The length of the scheduled appointment interval at the time of elevated VL was highly associated with longer time to return, suggesting the health system did not adjust care delivery to account for the urgency of treatment failure. While shifting towards longer ART dispensations has helped reduce clinic visit burden on clients [21–24], it complicates timely response to viremia when clients do not return until their next refill is due [25]. Results are not immediately available and providers are not able to act in real time, and this ultimately requires responsive systems that can adapt once VL results are available. This includes formalized procedures for laboratory notification of providers, proactive outreach to clients, timely EAC initiation in manner that accommodates individual needs, followed by close follow-up and reassessing response to ART. Innovations like telehealth could help initiate prompt adherence counseling prior to individuals physically returning for their next clinical visit. Point-of-care VL testing has promise for enabling immediate action [26], but such approaches must still be complemented by systems that ensure continued follow-up of clients [27–29].

Second, treatment interruptions are common and represent a major barrier to rechecking VL tests and subsequent outcomes. In our study, approximately 32.5% of individuals were ≥60 days late for an appointment at least once after treatment failure, and 59.3% of those who did not have a VL rechecked by 1 year were in a treatment interruption. Similar gaps occurred after individuals were due for switch. As programmes mature, treatment interruptions remain one of the largest drivers of viremia in the community, but this major challenge often gets obscured behind metrics that only consider those in whom a VL test is documented [30]. While re-suppression was high among those who were retested, it was substantially lower when also accounting for those with treatment interruptions. A complex interplay of structural, clinic-based, and psychosocial factors undermines retention and access to medications [31]. Judgmental and antagonistic interactions with providers may be especially heightened for individuals experiencing adherence challenges [32–35]. Person-centered approaches that support providers in delivering nonjudgmental, tailored counseling should be considered central to strategies aimed at improving outcomes after treatment failure [36,37].

Third, when clients did return to care, implementation gaps persisted at the health system level. Providers often missed opportunities to recheck a VL test or switch to second-line therapy despite clients being due at these visits. For example, 40.8% of those who had not had a VL test rechecked by 1 year were still in care. Delays may be driven by rigid insistence on completing all three EAC sessions, provider perception that clients were not yet adherent, or simple lack of familiarity with treatment failure guidelines [32,38]. Addressing these issues requires system-level interventions to ensure that each health interaction is fully taken advantage of to support optimal care. This requires health system strategies that strengthen organizational culture, promote team-level accountability, ensure provider coherence [39], and clear communication on managing treatment failure. Flexible interpretation of guidelines, such as not delaying rechecking VL tests while waiting for all EAC sessions may help improve responsiveness [40].

Our findings also provide additional insights as HIV programmes shift from TLE to TLD. In this study, a high number of individuals resuppressed while maintaining TLD. Despite similar patterns of care, the prevalence of re-suppression was 25% higher in individuals on TLD compared to TLE among those who had a VL checked [3,5,41,42]. Assuming comparable adherence, this difference likely represents the gap in acquired resistance that frequently occurs on TLE but rarely on TLD. Providers were also much less likely to switch clients on TLD, often rechecking VLs multiple times instead even when due. Prior studies have demonstrated very low levels of DTG resistance after treatment failure on TLD [4,43,44] to date, compared to fairly high levels after treatment failure on TLE [45]. The effectiveness and forgiving nature of DTG-based regimens has shifted focus away from resistance concerns and switching to second-line regimens [4,8,41,42] and towards adherence counseling as a primary focus. Nonetheless, there are emerging concerns that DTG resistance will continue to grow, and an evidence-based approach to treatment failure on TLD is needed [46,47]. In our study, 45% of those who had switched regimens on TLD were likely switched to an inappropriate regimen (i.e., 37% only had TDF changed to AZT, 8% switched to NNRTI-based regimen), although it is also worth noting that a small proportion of individuals on TLD required regimen switches and that formal guidance on use of integrase-based second-line regimens only after TLE failure were not issued until after the study period in 2022. Results from NADIA and D2EFT suggest that maintaining TLD or switching to DTG plus PI may be appropriate strategies, although these studies were in individuals failing TLE and not TLD [48–50]. REVAMP demonstrated little benefit of genotypic assays to identify resistance at treatment failure on TLE despite 68% with detectable resistance, and any clinical benefit is likely even further diminished with low rates of resistance on TLD (although it may still be cost-effective given cost of PIs) [51,52]. Evidence on adapting approaches to treatment failure, including monitoring for suppression, when to consider switching regimens, and what to switch to, are needed with widespread roll out of TLD as first-line regimens.

There are several limitations to this study. First, Electronic Health Record (EHR) data did not include whether outreach activities were performed, whether EAC was delivered, or provider rationale (if any) to delay VL. Nevertheless, this analysis highlights the major delays in the timeliness and success of current approaches to treatment failure and identifies gaps that can be further characterized in future studies. Second, there may be incomplete EHR data capture, although prior analyses have demonstrated high-quality data [53]. Third, we could not link visits to individual providers, limiting assessment of provider-specific practices. Fourth, our analysis only included individuals with documented treatment failure, likely missing individuals with viremia that goes undocumented due to not being in care or missing VL measures [30]. Fifth, TLD was rolled out during the study period, and some individuals may have switched to TLD due to this programmatic scale-up rather than to manage treatment failure on TLE. Although this would likely improve re-suppression rates after TLE failure, we still noted marked differences in re-suppression on TLD versus TLE, and we would better capture current fidelity to guideline-based care algorithms by categorizing regimen switches only after two consecutive elevated VLs. Sixth, potential selection bias may exist, as individuals on TLD are less likely to experience treatment failure to begin with. Lastly, we could not verify whether clients with treatment interruptions had transferred to other sites, although prior studies show long delays in ART reinitiation among many of those transferring without official documentation [54].

In summary, there are substantial gaps along the treatment failure cascade precluding viral re-suppression, driven by delays in outreach, return for adherence counseling, and repeat VL monitoring to assess response; treatment interruptions; and missed opportunities in rechecking VL test status or switching to second-line therapy in routine care in Zambia. These gaps stem from both client-related and health system inefficiencies and increasing shifts to six-month ART distribution, underscoring the need for systemic health system approaches to reengage individuals with treatment failure promptly. Furthermore, re-suppression rates were significantly higher among individuals who failed TLD compared to TLE, and the high re-suppression rates on TLD suggest that an increasingly lower number of treatment failures are driven by resistance, and switching may not always be necessary. Although these findings are from Zambia, they remain broadly relevant for public sector HIV programs with similar implementation challenges. In the modern era of ART with TLD used as first-line regimens, it is critical to reimagine approaches to treatment failure and create formal and explicit guidance to prioritize person-centered strategies for timely engagement and support for client retention and adherence and likely less emphasis of resistance concerns and switches to second-line. Future efforts should prioritize understanding how to best advance approaches to treatment failure that align with the evolving needs of the TLD era and integrate person-centered approaches to target the particular care gaps that exist along the treatment failure cascade.

Supporting information

Acknowledgments

We thank the Zambian Ministry of Health, the US President’s Emergency Plan for AIDS Relief through the US Centers for Disease Control and Prevention/Zambia, and the Centre for Infectious Disease Research in Zambia for their leadership in ensuring that people in HIV care continue to receive life-saving HIV treatment.