HIV programmatic outcomes following implementation of the ‘Treat‐All’ policy in a public sector setting in Eswatini: a prospective cohort study

Abstract Introduction The Treat‐All policy – antiretroviral therapy (ART) initiation irrespective of CD4 cell criteria – increases access to treatment. Many ART programmes, however, reported increasing attrition and viral failure during treatment expansion, questioning the programmatic feasibility of Treat‐All in resource‐limited settings. We aimed to describe and compare programmatic outcomes between Treat‐All and standard of care (SOC) in the public sectors of Eswatini. Methods This is a prospective cohort study of ≥16‐year‐old HIV‐positive patients initiated on first‐line ART under Treat‐All and SOC in 18 health facilities of the Shiselweni region, from October 2014 to March 2016. SOC followed the CD4 350 and 500 cells/mm3 treatment eligibility thresholds. Kaplan‐Meier estimates were used to describe crude programmatic outcomes. Multivariate flexible parametric survival models were built to assess associations of time from ART initiation with the composite unfavourable outcome of all‐cause attrition and viral failure. Results Of the 3170 patients, 1888 (59.6%) initiated ART under Treat‐All at a median CD4 cell count of 329 (IQR 168 to 488) cells/mm3 compared with 292 (IQR 161 to 430) (p < 0.001) under SOC. Although crude programme retention at 36 months tended to be lower under Treat‐All (71%) than SOC (75%) (p = 0.002), it was similar in covariate‐adjusted analysis (adjusted hazard ratio [aHR] 1.06, 95% CI 0.91 to 1.23). The hazard of viral suppression was higher for Treat‐All (aHR 1.12, 95% CI 1.01 to 1.23), while the hazard of viral failure was comparable (Treat‐All: aHR 0.89, 95% CI 0.53 to 1.49). Among patients with advanced HIV disease (n = 1080), those under Treat‐All (aHR 1.13, 95% CI 0.88 to 1.44) had a similar risk of an composite unfavourable outcome to SOC. Factors increasing the risk of the composite unfavourable outcome under both interventions were aged 16 to 24 years, being unmarried, anaemia, ART initiation on the same day as HIV care enrolment and CD4 ≤ 100 cells/mm3. Under Treat‐All only, the risk of the unfavourable outcome was higher for pregnant women, WHO III/IV clinical stage and elevated creatinine. Conclusions Compared to SOC, Treat‐All resulted in comparable retention, improved viral suppression and comparable composite outcomes of retention without viral failure.

The impact of changing ART eligibility criteria in RLS is poorly understood because of the lack of recent programme data [28], the gap between supporting health policies and efficient operationalization [29,30] and inconclusive treatment outcome data from ongoing Treat-All trials [31][32][33][34]. Eswatini (formerly Swaziland) is one of the few countries that piloted the Treat-All policy before it became a WHO recommendation in 2016 [1]. While Treat-All has increased timely ART initiation in routine settings [35][36][37], studies on longer-term outcomes are scarce. To inform scale-up of Treat-All in RLS, we aimed to assess varying patterns of associations with treatment outcomes under Treat-All and under the concurrent national standard of care (SOC) at the time, and to compare programmatic outcomes between both interventions.

| Study design
This is a prospective cohort study of ≥16-year-old HIV-positive patients initiated on first-line ART under the Treat-All programmatic approach and under SOC in 18 public sector health facilities of the Shiselweni region (Eswatini), from 20 October 2014 to 31 March 2016.

| Study setting
The setting has been described previously [38]. The predominantly rural Shiselweni region has a population of~210,000 [39] and HIV prevalence is 31% in 18-to 49-year-olds [40,41]. The study was conducted in two neighbouring health zones, each comprising eight HIV/TB care integrated primary care facilities and one HIV/TB care collocated secondary care outpatient department. The Treat-All health zone offered prompt facility-based ART initiation irrespective of CD4 and clinical criteria for all newly diagnosed patients and those already enrolled in pre-ART care. The neighbouring SOC health zone followed national treatment guidelines with ART initiation at CD4 ≤ 350 (October 2014 to October 2015) and ≤500 cell/mm 3 (November 2015 onwards), WHO III/IV clinical staging and the prevention of mother-to-child transmission programmatic approach option B+ (PMTCTB+).
Trained lay counsellors conducted HIV testing, and pretreatment and treatment adherence counselling. ART initiation and follow-up care were performed by nurses in primary care clinics and supported by onsite medical doctors in secondary care outpatient departments. Patients usually had a baseline CD4 cell count and laboratory test (haemoglobin, alanine aminotransferase (ALT), creatinine). The CD4 result was not a requirement for ART initiation under Treat-All. Routine viral load (VL) monitoring was available (using the Biocentric platform [42,43]) with VL testing recommended at six and twelve months after ART initiation, and annually thereafter [43]. Enhanced adherence counselling was provided for patients with a VL ≥ 1000 copies/mL, with treatment switching in case of viral failure (two consecutive VLs ≥ 1000 copies/mL). Telephonic and physical defaulter tracing was recommended for patients missing clinical appointments.

| Analyses, outcomes and definitions
Several analyses were conducted ( Figure 1). First, baseline factors were described separately for both interventions. Laboratory measures were recorded at the time of ART initiation and a TB case was defined as a patient receiving TB treatment between six months before and three months after ART initiation. Calendar time was divided into time period-1 and time period-2, corresponding to the WHO 2010 (October 2014 to October 2015) and WHO 2013 (November 2015 onwards) treatment guideline implementation periods followed under SOC. Same-day ART initiation was defined as patients starting ART on the same day as HIV care enrolment (the date of opening a patient file at the health facility).
Second, we describe crude and covariate adjusted programmatic indicators. Retention was defined as patients in ART care at different time points (without the outcome of death or LTFU). We chose this end point because the vital status was not actively ascertained in both interventions. LTFU was defined as six months without a clinic visit measured from the last clinic visit. Follow-up time was censored at database closure (31 October 2017) or date of transfer out of the facility. Then we describe VL testing uptake (the probability of receiving at least one VL test) and viral suppression, defined as the proportion of VLs < 1000 copies/mL among patients with a first VL measurement recorded. Finally, viral failure was compared, defined as two consecutive VLs ≥ 1000 copies/mL measured at least five months after ART initiation and performed ≥1.5 months apart or treatment switching to a protease inhibitor based regimen with two new drugs in the absence of documented viral failure. All viral load outcomes were measured from five months after ART initiation.
Third, Treat-All aims to retain patients on virally suppressed ART to improve patient level outcomes and reduce transmission of HIV. Accordingly, we established a composite primary endpoint of death, LTFU and viral failure. To assess whether the unfavourable outcome was more likely to occur in the Treat-All zone, we compared the interventions directly, first for the entire cohort and then restricted to patients presenting with advanced HIV disease defined as CD4 < 200 cells/ mm 3 and/or WHO III/IV clinical staging.
Finally, we conducted separate analyses of the composite primary endpoint for both models of care to assess possible varying patterns of associations with Treat-All and SOC during the implementation of different treatment guidelines.
All data were collected by trained data clerks from individual-level clinic records and paper registers, and entered into EpiData software. VL and TB data were complemented with data from separate electronic databases used for routine programme monitoring.

| Statistics
Baseline characteristics and crude programmatic outcomes were described with frequencies and proportions, and compared using the Pearson's chi-squared test for categorical variables and Wilcoxon's rank-sum test for continuous variables. Kaplan-Meier estimates were used to describe retention, VL testing uptake and viral failure.
Variables for inclusion in multivariate analyses were selected a priori based on clinical relevance and literature review. We used multiple imputation by chained equations [44] to impute missing covariate data using 20 datasets (Table S1). Multiple imputation diagnostics were satisfied according to trace plots and Kernel density plots ( Figure S1 and S2). Covariate adjusted parametric survival models (Royston-Parmar models) [45] were built to describe associations with time to the composite outcome. We used Akaike's information criteria to determine the number and location of internal knots for the baseline spline function. Covariates violating the proportional hazards assumption (assessed with Schoenfeld residual statistics) were included in the models as time-varying effects. All analyses were performed with Stata 14.1 (StataCorp, College Station, Texas, USA).

| Ethics
This study was approved by the Scientific and Ethics Committee of Eswatini, the Research Ethics Committees of MSF and the University of Cape Town, South Africa. Informed written consent was obtained before ART initiation from patients in Treat-All who were ineligible for ART according to the national SOC.

| Baseline characteristics
3.1.1 | Treat-All and SOC Figure 1 shows the study flow. Thirty-five patients were removed from the analysis because study eligibility was unclear. Of the remaining 3170 patients (  Figure 1. Study flow and analyses performed. 1 The analysis directly compares Treat-All with SOC irrespective of CD4 and WHO clinical staging criteria. 2 The analysis directly compares Treat-All with SOC restricted to patients with advanced HIV disease (CD<200 cells/mm 3 and/or WHO III/IV). 3 The treat-All and SOC interventions were analysed separately. ART, antiretroviral therapy; n, number; IQR, interquartile range; LTFU, loss to follow-up; SOC, standard of care; TFO, transferred out.  (Table S2).

| Programmatic outcomes
The frequency of the outcomes is presented in Figure 1.

Treat-all versus SOC
Comparing both health zones, crude 6-and 36-month retention were 84% and 71% under Treat-All compared with 89% and 75% under SOC (p = 0.005) (Figure 2d). Retention tended to be lower for pregnant women than non-pregnant  Figure 3). In both analyses, the hazard was higher for BMI < 18.5 kg/ m 2 , haemoglobin ≤9 g/dL, creatinine ≥121 µmol/L and being unmarried. The hazard was higher for the entire cohort only for the CD4 strata ≤200 cells/mm 3 (vs. CD4 201 to 350), WHO clinical stage III/IV (vs. WHO stage I), age 16 to 24 years (vs. 25 to 49 years) and shorter time since HIV care enrolment. The hazard of the unfavourable outcome varied over time for TB, with lower hazard during the first nine months after ART initiation and similar hazard thereafter ( Figure S4).

| Predictors of the composite unfavourable outcome (Treat-All vs. SOC)
A breakdown of crude outcomes is shown in Figure 1 and predictors in Table 4. In both health zones, the hazard of the unfavourable outcome was higher for young adults aged 16 to 24 years (vs. 25 to 49 years), unmarried patients, haemoglobin ≤9 g/dL, ART initiation on the same day or within three months of HIV care enrolment (vs. ART initiation after three months) and CD4 cell count ≤100 cells/mm 3 . The effect of same-day ART initiation varied over time under Treat-All, with higher hazards during the first 1.1 years after ART initiation while the difference in hazard ceased thereafter (Figure 4a,c). The effect of baseline CD4 varied over time under SOC, with the highest hazard during the first year of treatment for CD4 ≤ 100 cells/mm 3 (vs. CD4 201 to 350) (Figure 4b,d). Although the hazard difference decreased thereafter, it remained higher for almost the entire observation period. Other factors did not show any strong associations.  Under SOC only, the hazard was higher for low BMI < 18.5 kg/m 2 (aHR 2.21, 95% CI 1.52 to 3.21) and lower for the later implementation period (aHR 0.71, 95% CI 0.54 to 0.92). Under Treat-All only, the hazard was higher for pregnant women (aHR 1.37, 95% CI 1.10 to 1.71), WHO staging III/IV (aHR 1.41, 95% CI 1.05 to 1.90) (vs. WHO stage I) and elevated creatinine ≥121 µmol/L (aHR 1.73, 95% CI 1.00 to 2.99).
In supplementary analysis, including health facility as a covariate instead of primary versus secondary care level, two primary care facilities under Treat-All and one primary care facility under SOC showed an increased hazard of an unfavourable outcome when compared with the secondary care facility at each health zone ( Figure S5).

| DISCUSSION
This study assessed programmatic and patient outcomes of universal ART provision (Treat-All) in a predominantly rural   public sector setting in Eswatini. Compared to SOC, Treat-All resulted in comparable retention, improved viral suppression and comparable composite outcomes of retention without viral failure, after adjusting for differences between patients accessing each service. Although crudely a higher proportion of patients were lost to care under Treat-All, this service also enrolled more patients with higher CD4 counts [36], suggesting improved coverage of the overall HIV population.

| Explanation of findings
Similarly to other settings [46], low CD4 cell count was associated with an adverse outcome. The association was more pronounced for patients presenting with advanced HIV disease, and the effect of CD4 cell count varied with time under SOC, with higher hazard early during treatment. Notably, under Treat-All, outcomes for patients with high CD4 cell counts were similar to those for patients with 201 to 350 cells/mm 3 , a pattern confirmed by two other Treat-All trials [31,32] but in contrast to findings from the Western Cape, South Africa, where attrition was increased for CD4 >500 cells/mm 3 [33].
Overall, the median CD4 cell count at ART initiation was only slightly higher under Treat-All (37 cells/mm 3 ), possibly explained by concurrent expansion of treatment eligibility criteria under SOC during the study period. Pregnant and lactating women were already eligible for prompt ART (PMTCTB+) under SOC and treatment eligibility for non-pregnant adults was expanded from ≤350 to ≤500 cell/mm 3 . Restricting analysis to non-pregnant adults under period-1 (WHO 2010 treatment guideline implementation), the difference in median CD4 cell count increased to 73.5 cells/mm 3 .
ART initiation on the same day as HIV care enrolment was associated with an adverse outcome, possibly stronger during the first year of treatment. Data on same-day ART initiation remain conflicting, with randomized controlled trials showing benefits and observational studies indicating no benefit or increased risk of unfavourable treatment outcomes [46][47][48][49][50]. A reason could be that observational studies may not be able to sufficiently adjust for timedependent confounding. For instance, patients with higher CD4 cell count may be less likely to initiate ART the same day and more likely to have a favourable outcome while  immunocompromised patients may be more likely to start treatment on the same day and to have an unfavourable outcome. Patients initiating ART are a subset of those diagnosed, linked and enrolled into care, thus our findings not comparable with studies with follow-up starting at the time of HIV diagnosis. Further studies are needed to understand same-day ART and its impact on HIV programmes implementing Treat-All, specifically because many RLS already apply rapid treatment initiation [3]. Nevertheless, attention is needed to identify patients not ready for same-day ART and to provide adequate adherence support after same-day ART initiation [51]. Men and non-pregnant women had the same risk of an unfavourable outcome. Although men in general show worse HIV care outcomes [46,52,53], findings from Eswatini remain inconsistent, with increased and similar risk for men [54][55][56]. Adverse treatment outcomes were high for pregnant women under Treat-All, which is in line with findings from PMTCT B+ and general ART programmes [46,48,57]. Specific interventions supporting pregnant women under Treat-All may be needed to achieve the full benefits of universal ART expansion for this group.
Similar to other studies [37,48,[58][59][60], younger age, being unmarried and clinical factors (BMI, haemoglobin, creatinine) increased the risk of adverse outcomes in both health zones and irrespective of disease progression. In contrast to another setting [58], the level of education did not show associations. While this setting showed significant variations for ART initiation across facilities [36], the variations with respect to programmatic outcomes were minor.
The later WHO 2013 guideline implementation period (time period-2) showed a lower risk of an unfavourable outcome under SOC. Temporal trends have also been reported from other settings [16][17][18][19][20][21][22]. In our case, quality of follow-up care may be one explanation. This time period coincided with the expansion of differentiated communitycentred ART care models for patients stable on ART and was more pronounced under SOC, which may have supported long-term adherence and decongested busy facilities [61].
WHO emphasizes that patients in greatest need of ART should not be de-prioritized during treatment scale-up [1]. Although about one third of patients in both health zones presented with advanced HIV disease and had an increased likelihood of adverse outcome if CD4 was ≤100 cells/mm 3 , covariate-adjusted analysis indicated that the risk was similar under Treat-All and SOC. In addition, TB co-infection emerged as a protective factor during early treatment. As per national guideline recommendations, co-infected patients may have received more attention by health workers given their high risk of mortality, resulting in less loss to care. Although median CD4 cell count increased during ART programme expansion internationally [62], the challenge of advanced HIV disease is likely to persist [63]. Scale-up of optimized packages of care for patients with advanced HIV (e.g. better diagnostics and effective prophylactic treatment) is essential to further reduce mortality and morbidity [63,64]. Our findings are encouraging in that patients with advanced HIV disease were probably not de-prioritized under Treat-All compared with SOC.

| Findings in context
Overall retention was comparable to ART programmes in low-and middle-income countries [65] and two Treat-All trials in Southern Africa [32,33]. However, point estimates of retention tended to be lower than under SOC, than previous retention estimates from this setting before the introduction of Treat-All [38] and than in a streamlined combination intervention trial in Eastern Africa [31]. Similarly to another Treat-All trial in South Africa [32], 6% of patients never returned for a clinic visit after ART initiation (vs. 3% under SOC). A broad range of supportive interventions may improve retention (e.g. community-based adherence support, health technology interventions) [66][67][68][69], potentially also under Treat-All [27].
While VL testing uptake was low and delayed in both health zones, crude viral suppression tended to be slightly higher under Treat-All and comparable to other settings [70]. Overall, viral failure seemed lower than in other ART programmes [71,72], possibly explained by variability of definitions, underestimation of true viral failure because of suboptimal VL testing coverage and record keeping, and high viral re-suppression rates (~60%) in patients with single elevated VLs [43,73].

| Limitations and strengths
First, our estimates of ART retention are conservative. Previous studies showed that transient treatment interruptions and movements between clinics are common and many patients recorded with LTFU are retained [26,74,75]. Because of limitation in routine monitoring and limited tracking of patients lost to follow-up, this study was not able to adjust for silent transfer between treatment sites and silent return to care. In addition, ART retention in clinic was measured rather than retention in care or national-level retention, likely biasing estimation of retention downwards [76,77]. We also did not report on overall HIV care retention of patients entering care as done in other routine Treat-All settings [37], thus possibly not detecting a higher care retention benefit of Treat-All when compared with SOC. Finally, not accounting for transient treatment interruptions possibly introduced a spurious trend of increased LTFU in our cohort, which had a relatively short follow-up time (analysis bias) compared with other cohorts [78]. Second, given the observational study design and comparison of two different health zones, we may not have been able to adjust for all unobserved variables (e.g. exposure to differentiated service delivery model for patients stable on ART). Third, assessing ART coverage and population-level viral suppression due to Treat-All was beyond the scope of this analysis. Nevertheless, ART initiation rates measured from the time of facility-based HIV care enrolment was higher under Treat-All (91%) than SOC (74%; p < 0.001) in this setting [36]. This possible additional ART coverage under Treat-All may have an increased overall effect on viral suppression of the entire population living with HIV despite lower retention in crude analysis. In addition, ART has been progressively expanded in this setting since 2006 [38], achieving 82.7% population-level ART coverage and 79.1% population-level VL suppression among people living with HIV in 2016/17 [79].
Despite the wide-scale adoption of Treat-All in RLS [5,80], studies accounting for this policy change under routine conditions are lacking. This study began two years before publication of the WHO Treat-All guidelines, and thus has the potential to inform implementation of this policy in similar rural settings. We adjusted for a wide range of covariates, which likely enabled us to show a comprehensive picture of Treat-All. In addition, we encountered risk factors that have not been widely described previously (e.g. same-day ART initiation) but that may affect programmatic outcomes of large HIV programmes. Finally, we assessed the programmatic impact of treatment expansion on patients with advanced HIV disease.

| CONCLUSIONS
Compared to SOC, Treat-All resulted in comparable retention, improved viral suppression and comparable composite outcomes of retention without viral failure. Patients with advanced HIV disease were possibly not de-prioritized and predictors of unfavourable outcomes were comparable between Treat-All and SOC. This study contributes to evidence that treatment expansion through the Treat-All programmatic approach may be feasible in RLS without increasing unfavourable outcomes, and as such is likely to have public health benefits. advised on final analyses. BK, MS, KJ, SMK, RT, EM, SMH, BR, IC and AB interpreted the data, contributed to the writing of the manuscript and approved the final version.

A B B R E V I A T I O N S
aHR, Adjusted hazard ratio; ALT, Alanine aminotransferase; ART, Antiretroviral therapy/treatment; BMI, Body mass index; CI, Confidence interval; EFV, Efavirenz; IQR, Interquartile range; LTFU, Loss to follow-up; PMTCT B+; Prevention of mother-to-child transmission option B+; RLS, Resource-limited setting; SOC, Standard of care; TB, Tuberculosis; TDF, Tenofovir disoproxil fumarate; VL, Viral load; WHO, World Health Organization.

A C K N O W L E D G E M E N T S
We thank all the patients and healthcare workers who were involved in piloting the Treat-All approach in the Shiselweni region, and specifically the patients in Nhlangano health zone. In addition, we thank all the MSF teams involved in data collection and data cleaning.

SUPPORTING INFORMATION
Additional information may be found under the Supporting Information tab for this article. Table S1. Complete and missing values for covariate and imputation procedures. Table S2. Distribution of CD4 cell count and WHO clinical staging for patients with advanced IV disease under Treat-All and SOC (n = 1060). Table S3. Kaplan-Meier estimates of retention under Treat-All for selected variables. Table S4. Predictors of the unfavourable outcome for the entire cohort (Treat-All and SOC combined) initiated on firstline ART (n = 3170). Figure S1. Trace plots of imputed data for all covariates with missing values. Figure S2. Kernel density plots for imputed haemoglobin for all imputed datasets as an example using the midiagplots command in Stata. Figure S3. Kaplan-Meier graphs of retention under Treat-All for selected variables. Figure S4. Absolute difference in hazard of an unfavourable outcome by TB status for the entire cohort (Treat-All and SOC combined). Figure S5. Variations in adjusted hazard ratios of the composite unfavourable outcome comparing primary care facilities with the secondary care facility under Treat-All (facility 1) and under standard of care (facility 10).