Virological impact of the programmatic dolutegravir transition at Mildmay Hospital Uganda: a before-and-after cohort study with subgroup and trajectory analysis, 2016–2023

Abstract

Background

Uganda began transitioning to Dolutegravir (DTG)-based antiretroviral therapy (ART) in 2018. However, patient-level longitudinal evidence quantifying the virological impact of this transition from Ugandan facilities is limited. We evaluated whether the programmatic transition to DTG improved viral load (VL) suppression at Mildmay Hospital Uganda and assessed whether benefits were uniform across age groups and sexes.

Methods

We conducted a retrospective cohort study using 114,406 VL measurements from 18,756 patients collected between 2016 and 2023. We defined VL suppression as ≤ 200 copies/mL in accordance with Uganda national ART guidelines. Regimen timelines reconstructed from 25,095 regimen-change events enabled time-varying classification of each VL test by the most recently recorded regimen before VL testing. The primary analysis compared VL suppression in a pre-switch window (3 to 18 months before the switch) with a 12-month post-switch window (9 to 15 months after) in 1,928 eligible paired patients using McNemar’s test. Secondary analyses included an interrupted time series (ITS) of monthly suppression rates, subgroup heterogeneity testing, and virological trajectory classification by regimen group.

Results

Overall VL suppression was 91.6%, rising from 89.2% in 2016 to 94.5% in 2023. 2.7% of tests recorded low-level viraemia (201–999 copies/mL) and 5.7% were unsuppressed. Among eligible switchers, suppression improved from 87.2% before the switch to 94.0% at 12 months after the switch (absolute difference + 6.8% points; 95% Confidence Interval (CI): 5.2–8.5; McNemar χ² = 63.09; p < 0.001), sustained at 94.7% at 24 months (p < 0.001). The adjusted odds ratio (aOR) for DTG exposure was 1.79 (95% CI: 1.63–1.97) and increased to 2.44 (95% CI: 2.13–2.80) by 12–24 months post-switch. All age groups and both sexes showed statistically significant improvements after the switch. Paediatric patients had the largest absolute gain of 12.2% points compared with 4.9% points for adults though post-switch suppression (89.8%) remained below 95%. DTG-direct initiators had the highest consistently suppressed trajectory rate at 82.0%, compared with 64.5% among documented non-DTG-only patients. Among 3,002 switchers with sufficient longitudinal data, 12.9% transitioned from non-suppressed to consistently suppressed trajectories after the switch.

Conclusions

The programmatic non-DTG-to-DTG transition was associated with a statistically significant and clinically meaningful improvement in VL suppression sustained through 24 months. Despite these gains, the cohort did not reach the Joint United Nations Programme on HIV/AIDS (UNAIDS) 95% VL suppression target under Uganda’s stricter ≤ 200 copies/mL definition. Moreover, paediatric and adolescent patients remain priority groups for targeted adherence support.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12879-026-13976-2.

Introduction

In 2022, an estimated 39 million people were living with Human Immunodeficiency Virus (HIV) globally, with sub-Saharan Africa bearing approximately two thirds of this burden [1]. Uganda’s HIV burden remains substantial, although national estimates indicate a gradual decline in adult HIV prevalence. Uganda Population-based HIV Impact Assessment (UPHIA) 2020/2021 estimated prevalence at 5.5% among persons aged 15–49 years. Uganda Aids Commission (UAC) Epi-data/Spectrum estimates reported 5.1% prevalence and 1,433,337 people living with HIV in 2022. The 2025 UAC HIV and AIDS Factsheet based on data ending 31 December 2024 estimated prevalence at 4.9% and 1,527,238 people living with HIV [2–4]. The Joint United Nations Programme on HIV/AIDS (UNAIDS) 95-95-95 framework calls for 95% of people on antiretroviral therapy (ART) to achieve Viral Load (VL) suppression by 2025 [1], making VL suppression the single most important metric for programme performance. At the population level, undetectable viral loads eliminate the risk of onward sexual transmission, a principle embodied in the Undetectable = Untransmittable scientific consensus [5].

Uganda’s HIV response underwent two transformative policy shifts during the study period. In 2016, Uganda expanded the 2014 “test & treat” policy, removing cluster of differentiation 4 (CD4) count thresholds for ART initiation [6]. Beginning in 2018, Uganda initiated ART optimisation by transitioning eligible patients from Non-nucleoside reverse transcriptase inhibitor (NNRTI)-based first-line regimens such as efavirenz (EFV)-based regimens, to dolutegravir (DTG)-based combinations. This was consistent with the World Health Organization (WHO) interim guidance recommending DTG-based regimens as preferred first-line ART and Uganda national guidance recommending DTG-based regimens as preferred first-line therapy for eligible people living with HIV [7–9]. The clinical trial evidence for this shift is robust. The ADVANCE trial in South Africa demonstrated DTG’s non-inferiority to EFV in a large, predominantly female, African population, with fewer treatment discontinuations and no integrase resistance detected in the DTG arms [10]. The NAMSAL ANRS 12,313 trial in Cameroon confirmed the non-inferior efficacy of DTG compared with low-dose EFV in a setting characterised by high HIV-1 genetic diversity, with faster viral suppression, and fewer virological failures [11, 12]. The NADIA trial demonstrated that DTG-based second-line ART was non-inferior to ritonavir-boosted darunavir-based therapy across East African sites [13]. A New England Journal of Medicine editorial accompanying these trial reports highlighted the importance of DTG-based therapy for global HIV treatment [14].

In Lesotho, the DO-REAL study reported high short-term viral suppression after programmatic NNRTI-to-DTG transition, with 98% suppressed to < 100 copies/mL at follow-up, while the larger VICONEL cohort showed sustained suppression after transition, reaching 93.9% at 12 months and 95.4% at 24 months using a < 50 copies/mL threshold [15, 16]. The KIULARCO cohort in rural Tanzania found higher 12-month viral suppression among treatment-naive patients initiating DTG-based ART than among those initiating NNRTI-based ART, at 85% versus 73%, respectively, in a programmatic setting [17]. A study from Mbeya, Tanzania reported comparable suppression before and after EFV-to-DTG switching in 250 patients [18]. The multinational AFRICOS cohort documented improved maintenance of viral suppression after programmatic transition to tenofovir disoproxil fumarate/lamivudine/dolutegravir TLD compared with remaining on non-DTG regimens [19]. Using the multi-country IeDEA consortium, Romo et al. showed that pre-switch VL status predicted subsequent outcomes after transition to DTG. Patients switching with VL ≥ 1,000 copies/mL had more than sixfold higher hazard of subsequent VL ≥ 1,000 copies/mL compared with those switching with VL < 200 copies/mL [20].

Within Mildmay Hospital Uganda itself, recent complementary studies have examined specific subpopulations. Ankunda et al. evaluated ART optimisation among 1,036 children and adolescents (≤ 19 years) in a descriptive cross-sectional design. They reported an overall improvement in suppression from 67.9% to 81.7% (p < 0.001) and documented adverse event profiles [21]. A separate study at Mildmay examined VL suppression across the maternal timeline among 1,291 pregnant and breastfeeding women. It found that DTG-based regimens maintained suppression rates above 94% throughout pregnancy [22]. Ankunda et al. also documented drug resistance mutation prevalence (59.6%) among 235 patients with virologic failure at Mildmay. They identified NNRTI mutations in 52.8% and NRTI mutations in 50.2% of tested patients [23]. Together, these studies provide important subpopulation-specific evidence, but there remains limited Ugandan facility-level evidence combining an all-age patient cohort, within-patient before-and-after assessment of the non-DTG-to-DTG transition, interrupted time-series analysis, subgroup heterogeneity testing, and longitudinal trajectory classification. Mildmay Hospital Uganda is located in Lweza, Kampala. It is one of Uganda’s largest dedicated HIV care centres. It is a faith-based, not-for-profit referral centre for comprehensive HIV and AIDS care in Lweza, Kampala district, Uganda. The hospital provides integrated HIV services including ART initiation and management, routine laboratory monitoring, paediatric HIV care, and psychosocial support. Between 2016 and 2023, the facility collected 114,406 VL measurements from 18,756 patients and recorded 25,095 regimen-change events across 19,070 patients. This dataset enables reconstruction of individual regimen timelines across the entire DTG transition period. This work reports the impact of the non-DTG-to-DTG transition on virological status, characterises the subgroup heterogeneity of the benefit, and analyses longitudinal treatment trajectories by regimen group.

Methods

Study design and setting

This was a retrospective cohort study conducted at Mildmay Hospital Uganda. It was designed to evaluate the virological impact of Uganda’s programmatic transition from non-DTG-based to DTG-based ART by linking serial HIV-1 ribonucleic acid (RNA) viral load results to individual-level ART regimen histories. HIV-1 RNA was quantified by automated real-time Polymerase Chain Reaction (PCR) with a limit of detection of 20–40 copies/mL.

Study period and population

The study period was 1 January 2016 to 31 December 2023. All patients of any age with at least one HIV-1 RNA PCR VL result during this period were eligible. There were no exclusion criteria based on ART regimen, treatment duration, or clinical stage. A small proportion of records (1.4% of VL entries) could not be linked to demographic data due to non-standard patient identifiers and were excluded from analyses requiring demographic information. January 2024 records present in the raw extraction (n = 1,617 tests) were excluded before analysis to maintain the prespecified 2016–2023 study window.

Data sources and linkage

The primary VL dataset contained patient identifiers, test dates, and quantitative results. The demographic register provided date of birth, sex, and clinical status while the CD4 count register provided baseline immunological data. These datasets were linked using unique M-format (MXXXXXX with X an integer) patient identifiers, achieving a 98.6% linkage rate. ART data were obtained from two HIV treatment database tables with different structures and purposes. The ART Regimen Table was a longitudinal event table and recorded 25,095 regimen-change events across 19,070 patients. It included regimen start date, drug combination, regimen line, switch reason, and WHO clinical stage at the time of the regimen event. This table was used to reconstruct regimen timelines and identify non-DTG-to-DTG switchers. In contrast, the Baseline ART Enrollment table recorded ART initiation regimens and baseline characteristics for 8,913 patients and was used only to classify documented ART initiators. Baseline enrolment records were therefore available for 8,913 of the 18,756 VL cohort patients (47.5%). Because switchers and initiators were identified from different source tables, these groups are not mutually exclusive and do not sum to the total cohort.

Regimen classification and switcher identification

Each ART regimen event was classified using the drug-combination field. Regimens were classified as DTG-based if the combination string contained “DTG” in any case format, and as non-DTG otherwise. This rule was verified against the complete list of 57 unique regimen strings (Table S3) in the ART Regimen table and identified 20 DTG-containing regimen strings. These included standard first-line combinations such as TDF/3TC/DTG, ABC/3TC/DTG, and AZT/3TC/DTG, as well as complex second- and third-line combinations containing DTG with boosted protease inhibitors (PIs). In regimen strings, TDF denotes tenofovir disoproxil fumarate, 3TC lamivudine, ABC abacavir, and AZT zidovudine. Switchers were defined as patients with at least one non-DTG regimen event followed chronologically by at least one DTG-containing event, with the switch date defined as the start date of the first DTG-containing regimen. Each VL test was assigned the most recently recorded regimen by matching the test date to the latest regimen start date on or before the VL test date. The pre-switch regimen of each switcher was further classified as efavirenz-based, nevirapine-based, protease inhibitor-based, or other non-DTG for regimen-class subgroup analysis.

Outcome and variable definitions

VL suppression was defined as HIV-1 RNA ≤ 200 copies/mL, in accordance with Uganda national ART guidelines [24]. A three-tier virological classification was applied: suppressed (≤ 200 copies/mL), low-level viraemia (201–999 copies/mL), and unsuppressed (≥ 1,000 copies/mL). The unsuppressed threshold of ≥ 1,000 copies/mL is consistent the Uganda national definition of virologic failure [8, 22, 25]. Results reported as undetectable were assigned a value of 0 copies/mL. Invalid or failed results were excluded. Throughout, the term viral suppression denotes HIV-1 RNA ≤ 200 copies/mL under the Uganda national threshold, the term undetectable is reserved for results below the assay limit of detection, and comparisons with the UNAIDS reporting convention apply the < 1,000 copies/mL threshold and are labelled as such. Age at each VL test was calculated from date of birth and categorised as paediatric (< 15 years), adolescent (15–24 years), adult (25–49 years), and older adult (≥ 50 years), in line with Uganda Ministry of Health stratifications [8]. Baseline CD4 count was defined as the measurement taken closest in time to each patient’s first recorded VL test.

Statistical analyses

Annual VL suppression rates were calculated with 95% Wilson score confidence intervals. Temporal trends were assessed using multivariable logistic regression with VL suppression as the binary outcome and calendar year, centred at 2016, as the main exposure. Models adjusted for sex and age group and included a year-by-sex interaction term. Logistic regression models were fitted in Python using SciPy. Sex was recorded as biological sex in the demographic register.

The primary analysis applied McNemar’s test with Edwards’ continuity correction to paired proportions among switchers. Eligibility required at least one VL test 3 to 18 months before the switch date and at least one VL test 9 to 15 months after. The most recent pre-switch and first eligible post-switch VL were selected for each patient. A parallel 24-month analysis used a 21 to 27 month post-switch window. The post-switch windows were centred on the 12- and 24-month anchors with a three-month tolerance on each side to match Uganda’s routine annual VL monitoring schedule. A multivariable logistic regression was fitted to all VL observations from the switcher cohort, with a time-varying binary post-switch indicator as the primary exposure, adjusted for sex, age group, and switch year centred at 2019. A Mundlak correction was applied by including the patient-level mean of the post-switch indicator, approximating a within-patient effect in the pooled framework [26]. Period-specific adjusted odds ratios were estimated separately for the 0 to 12 month and 12 to 24 month post-switch periods.

Monthly VL suppression rates, Sup(t), were calculated as the proportion of valid VL tests classified as suppressed in month t. These monthly rates were modelled using segmented linear regression: Sup(t) = β₀ + β₁t + β₂D_t + β₃(t − T₀)D_t + ε_t, where D_t was coded 0 before January 2019 and 1 from January 2019 onward, and T₀ denoted the January 2019 intervention month. β₂ estimated the immediate level change at the intervention point, while β₃ estimated the change in monthly trend after the intervention.

The 12-month McNemar analysis was repeated within strata of age group, sex, WHO clinical stage at switch where recorded, and switch era, defined as early adoption in 2018–2019 versus mass rollout in 2020–2021. Interaction terms between the post-switch indicator and age group and sex were tested using Wald tests. Patients with two or more VL tests and documented regimen histories were classified into the following five longitudinal trajectory categories: consistently suppressed, if all recorded VLs were ≤ 200 copies/mL; achieved suppression, if the first VL was > 200 copies/mL and the last VL was ≤ 200 copies/mL; virologic failure, if the first VL was ≤ 200 copies/mL and the last VL was > 200 copies/mL; fluctuating, if suppression status varied without fitting these directional patterns; and consistently unsuppressed, if all recorded VLs were > 200 copies/mL. Trajectory classification was performed separately by regimen group, and a pre-to-post switch transition matrix was computed for switchers with at least two VL tests before and at least two VL tests after the switch. For analytical adequacy, an approximate McNemar paired-proportion calculation assuming pre-switch suppression of 87%, post-switch suppression of 94%, two-sided α = 0.05, and 90% power indicated a required sample size of approximately 250 paired patients, well below the 1,928 pairs available. All analyses were performed in Python 3.12 using pandas, NumPy, SciPy, matplotlib, and seaborn.

Results

Cohort characteristics

After linkage and cleaning, 114,406 VL results from 18,756 patients were included (Table 1; Fig. 1). The majority were female (12,123; 64.6%), with a mean age at first VL test of 34.7 ± 13.6 years (median 34.8; Interquartile range (IQR): 26.0–43.9). Adults aged 25–49 years formed the largest group (12,107; 64.6%), followed by adolescents (2,580; 13.8%), older adults (2,390; 12.7%), and paediatric patients (1,679; 9.0%). A total of 1,706 patients (9.1%) had only a single VL test, and the remaining 17,050 (90.9%) had two or more tests. Among 7,125 patients with a baseline CD4 count from the CD4 register, the median was 444 cells/µL (IQR: 220–711).

Table 1.

Baseline and regimen-source characteristics of the study cohort. Mildmay Hospital Uganda, 2016–2023. N = 18,756 patients

Characteristic	Full cohort	EFV/NVP init.	DTG init.	Non-DTG-to-DTG switchers
N patients	18,756	3,971	4,492	4,022
VL tests, median (IQR)	7 (4–8)	—	—	—
Baseline CD4, median (IQR)	444 (220–711)	—	—	—

CD4 data were available for 7,125 of 18,756 patients (38.0%). EFV/NVP init. = patients whose baseline enrolment record indicated ART initiation on an efavirenz- or nevirapine-based regimen; DTG init. = patients whose baseline enrolment record indicated ART initiation directly on a DTG-based regimen; non-DTG-to-DTG switchers = patients with a documented non-DTG regimen followed by a DTG-containing regimen in the ART Regimen Table. Initiator groups were classified from baseline enrolment records, while switchers were classified from longitudinal ART regimen histories. Groups are not mutually exclusive and do not sum to the cohort total

Fig. 1.

Flow diagram of data sources and analysis cohorts. The diagram shows source datasets, linkage, exclusions, and analytic cohorts for the 2016–2023 Mildmay Hospital Uganda DTG transition analysis

ART regimen characteristics and the DTG transition

From 25,095 regimen-change events across 19,070 patients, 4,022 non-DTG-to-DTG switchers were identified. The transition accelerated rapidly with 112 switches in 2018, 366 in 2019, and 2,123 in 2020, which represented 52.8% of all documented switches. The recorded switch reasons were overwhelmingly programmatic, with “new drug available” accounting for 1,782 switches (44.3%), while toxicity and virologic failure each accounted for fewer than 2% of switches. Among patients with baseline enrolment data, 4,492 (50.4%) started ART directly on DTG and 3,971 (44.6%) started on EFV or NVP. By the end of the study period, 17,463 patients (91.6% of all patients in ART Regimen table) had been prescribed a DTG-containing regimen at least once. The pre-switch regimen was efavirenz-based in 2,594 switchers (64.5%), protease inhibitor-based in 1,244 (30.9%), and nevirapine-based in 182 (4.5%); we therefore refer to them collectively as non-DTG-to-DTG switchers. The most common pre-switch regimen was TDF/3TC/EFV (n = 2,424) and the most common post-switch regimen was TDF/3TC/DTG (n = 3,245); 3,148 switches were from first-line, 835 from second-line, and 39 from third-line regimens (Table S2). Participant flow is summarised in Fig. 1.

Overall VL suppression and annual trends

Using the Uganda national guideline definition of suppression (≤ 200 copies/mL), overall suppression across all 114,406 measurements was 91.6% (95% CI: 91.4–91.8%). An additional 2.7% of tests recorded low-level viraemia (201–999 copies/mL) and 5.7% were unsuppressed (≥ 1,000 copies/mL). Annual suppression rates are presented in Table 2; Fig. 2. The annual distribution of the suppression categories is shown in Fig. 3. Suppression rose from 89.2% in 2016 (95% CI: 88.4–90.0%) to 94.5% in 2023 (95% CI: 94.2–94.9%), with the largest single-year gain of + 3.1% points occurring between 2019 and 2020. This coincided with peak DTG rollout and the introduction of multi-month dispensing during the coronavirus disease 2019 (COVID-19) pandemic. The cohort did not reach the UNAIDS 95% VL suppression target under this definition in any study year, though 2022 (94.9%) and 2023 (94.5%) came closest. Under the WHO and UNAIDS < 1,000 copies/mL convention, suppression exceeded the 95% threshold from 2021 onward (95.8% in 2021 and 96.8% in 2023), as shown in Table 2. In multivariable logistic regression with year centred at 2016, the adjusted odds of suppression increased by 9.1% per calendar year (aOR 1.091; 95% CI: 1.074–1.109; p < 0.001), with suppression improving faster among women than men over the study period (year-by-female interaction aOR 1.046; 95% CI: 1.025–1.067; p < 0.001).

Table 2.

Annual viral load categories. Mildmay Hospital Uganda, 2016–2023

Year	N tests	Suppressed ≤ 200 (%)	95% CI	Low-level 201–999 (%)	Unsuppressed ≥ 1,000 (%)	Suppressed < 1000 (%)
2016	5,932	89.2	88.4–90.0	2.3	8.4	91.6%
2017	14,573	89.4	88.9–89.9	2.7	8.0	92.0%
2018	15,352	89.5	89.0–90.0	2.7	7.8	92.2%
2019	16,029	88.7	88.2–89.2	3.8	7.5	92.5%
2020	14,633	91.8	91.3–92.2	3.0	5.2	94.8%
2021	14,404	92.9	92.4–93.3	2.9	4.2	95.8%
2022	16,363	94.9	94.6–95.2	1.8	3.3	96.7%
2023	17,120	94.5	94.2–94.9	2.3	3.2	96.8%

CI = Wilson score 95% confidence interval. Suppressed = ≤ 200 copies/mL; low-level viraemia = 201–999 copies/mL; unsuppressed = ≥ 1,000 copies/mL, consistent with Uganda national ART guidelines. The < 1,000 copies/mL column represents the WHO/UNAIDS reporting convention for Suppressed VL (for sensitivity analysis)

Using the primary Uganda ≤ 200 copies/mL definition, annual test-level suppression approached but did not reach the UNAIDS 95% target, peaking at 94.9% in 2022 and reaching 94.5% in 2023. Under the < 1,000 copies/mL convention used in UNAIDS reporting, suppression exceeded 95% from 2021 onward

Fig. 2.

Annual HIV viral load suppression rate. Viral load suppression is defined as ≤ 200 copies/mL. Wilson score 95% confidence intervals are shown. The dashed red line denotes the UNAIDS 95% target. Mildmay Hospital Uganda, 2016–2023

Fig. 3.

Annual distribution of viral load categories. Categories use Uganda national guideline thresholds: suppressed (≤ 200 copies/mL), low-level viraemia (201–999 copies/mL), and unsuppressed (≥ 1,000 copies/mL). Mildmay Hospital Uganda, 2016–2023

Interrupted time series analysis

The Interruped time series (ITS) model of 91 monthly observations and fitted trends are shown in Fig. 4. It identified a pre-DTG secular trend of + 0.11% points per month (β₁ = 0.106; p = 0.067). The level change at January 2019 was non-significant (β₂ = − 1.268; 95% CI: −3.76 to 1.23; p = 0.32), as was the slope change (β₃ = 0.013; p = 0.83). The improvement in suppression therefore did not manifest as a discrete step-change at the DTG rollout date but rather as a continuation of a pre-existing upward trend, which is consistent with the gradual phased nature of the rollout that began with only 112 switches in 2018 before reaching its peak of 2,123 switches in 2020.

Fig. 4.

Interrupted time series of monthly viral load suppression. Suppression was defined as ≤ 200 copies/mL. The vertical dashed line marks January 2019, the onset of programmatic DTG rollout. Solid lines show fitted pre-DTG and post-DTG linear trends with 95% confidence bands. The dashed red extension shows the counterfactual trajectory. Grey dots are observed monthly rates. Mildmay Hospital Uganda, 2016–2023

Before-and-after analysis of non-DTG-to-DTG switchers

Of 4,022 identified switchers, 1,928 met eligibility criteria for the 12-month paired analysis. Among these patients, VL suppression improved from 87.2% before the switch to 94.0% at 12 months after the switch, an absolute improvement of + 6.8% points (95% CI: 5.2–8.5; McNemar χ² = 63.09; p < 0.001; Table 3). This improvement was sustained at 24 months: among 1,743 eligible patients, post-switch suppression was 94.7%, representing an absolute improvement of + 5.5% points (p < 0.001). In the Mundlak-corrected logistic regression fitted to 25,706 VL observations from the switcher cohort, the adjusted odds ratio for DTG exposure was 1.79 (95% CI: 1.63–1.97; p < 0.001). When stratified by post-switch duration, the aOR was 2.07 (95% CI: 1.82–2.35) for the first 12 months and 2.44 (95% CI: 2.13–2.80) for months 12 to 24. This shows that the benefit was not only sustained but continued to increase over the second year post-switch. Compared with the 2,094 excluded switchers, the 1,928 included patients were younger, more frequently monitored, and had higher baseline CD4 (Table S1). By pre-switch class, suppression rose by 2.5% points for EFV-to-DTG (93.4% to 95.9%; p = 0.007), 7.8 for NVP-to-DTG (90.9% to 98.7%; p = 0.04), and 14.0 for PI-to-DTG (76.6% to 90.6%; p < 0.001; Table S2).

Table 3.

Before-and-after analysis of VL suppression (≤ 200 copies/mL) among switchers (non-DTG-to-DTG), Mildmay Hospital Uganda, 2016–2023 (n = 1,928)

Parameter	Value
Total documented switchers	4,022
Eligible for 12-month paired analysis	1,928
Pre-switch suppression (%)	87.2
12-month post-switch suppression (%)	94.0
Absolute difference (95% CI)	+ 6.8 (5.2–8.5)
McNemar χ2	63.09
McNemar p-value	< 0.001
Eligible for 24-month paired analysis	1,743
24-month post-switch suppression (%)	94.7
24-month absolute difference	+ 5.5
24-month McNemar p-value	< 0.001
aOR for DTG (Mundlak model, overall)a	1.79 (1.63–1.97)
aOR for DTG (0–12 months post-switch)	2.07 (1.82–2.35)
aOR for DTG (12–24 months post-switch)	2.44 (2.13–2.80)

a Adjusted for sex, age group, switch year centred at 2019, and patient-level mean of the post-switch indicator (Mundlak correction). N = 25,706 VL observations from 4,022 switcher patients. Outcome: VL suppression ≤ 200 copies/mL

In a threshold sensitivity analysis among paired switchers (Table 2), suppression improved from 87.2% to 94.0% using the ≤ 200 copies/mL definition and from 92.4% to 96.2% using the < 1,000 copies/mL convention.

Subgroup heterogeneity of DTG benefit

Stratified McNemar analyses revealed substantial heterogeneity in the magnitude of the DTG benefit across age groups, with the pre-switch and 12-month post-switch rates shown in Fig. 5 and the corresponding odds ratios for improvement in Fig. 6. Paediatric patients (n = 393) showed the largest absolute improvement, gaining 12.2% points from 77.6% to 89.8% (p < 0.001). Adolescents (n = 354) gained 7.6% points from 84.5% to 92.1% (p = 1.5 × 10⁻³). Adults (n = 1,079) improved by + 4.9% points from 91.0% to 95.9% (p < 0.001). Older adults (n = 102) showed a + 3.9% point change from a high baseline of 93.1%, but this did not reach statistical significance (p = 0.29), consistent with their high baseline suppression and the small number of patients at risk of remaining unsuppressed.

Fig. 5.

Pre-switch and 12-month post-switch viral load suppression. Suppression rates are shown overall and by age group among non-DTG-to-DTG switchers. The dashed red line denotes the UNAIDS 95% target. Mildmay Hospital Uganda, 2016–2023. N = 1,928 switchers

Fig. 6.

Forest plot of subgroup heterogeneity in the DTG transition effect. Each row shows the odds ratio (OR) for improvement from the discordant pair ratio and 95% confidence interval. The vertical dashed line indicates OR = 1.0. The x-axis is on a log scale. Mildmay Hospital Uganda, 2016–2023 (n = 1,928 switchers)

By sex, females improved by + 6.7% points (p < 0.001) and males by + 7.2% points (p < 0.001). The mass rollout cohort (2020–2021 switchers; n = 1,624) improved by + 6.8% points (p < 0.001), whereas early adopters (2018–2019; n = 175) showed no significant improvement (+ 0.6% points; p = 1.0). Formal interaction tests found no statistically significant heterogeneity in the DTG benefit by age group or sex. The adolescent interaction aOR was 0.860 (p = 0.166), the paediatric interaction was 1.137 (p = 0.274), and the sex interaction was 0.983 (p = 0.859).

Suppression by sex and age group

Female patients had higher crude VL suppression than male patients (92.4% versus 90.0%; crude OR 1.364; 95% CI: 1.307–1.423; p < 0.001). In multivariable regression, the main effect of female sex remained statistically significant after adjustment for age group and calendar year (aOR 1.098; 95% CI: 1.013–1.191; p = 0.024), and suppression improved faster among women over time (year-by-female interaction aOR 1.046; p < 0.001). Adolescents had the lowest adjusted suppression (aOR 0.400; 95% CI: 0.379–0.422 versus adults; p < 0.001), while older adults had the highest (aOR 1.454; 95% CI: 1.351–1.565; p < 0.001). All groups improved over the study period (Fig. 7), though paediatric and adolescent patients remained furthest below the 95% target throughout. Figure 8 is a heat map of VL suppression rates by age group and calendar year while Fig. 9 is a forest plot of adjusted odds ratios from the full cohort logistic regression as detailed in Table 4.

Fig. 7.

Annual viral load suppression by sex. Suppression was defined as ≤ 200 copies/mL. Shaded bands represent 95% Wilson score confidence intervals. The dashed red line denotes the UNAIDS 95% target. Mildmay Hospital Uganda, 2016–2023

Fig. 8.

Viral load suppression heatmap by age group and year. Suppression was defined as ≤ 200 copies/mL. Colour scale ranges from red (lower suppression) to green (higher suppression). Mildmay Hospital Uganda, 2016–2023

Fig. 9.

Adjusted predictors of viral load suppression. The figure shows adjusted odds ratios from the full cohort logistic regression in Table 4. The vertical dashed line indicates aOR = 1.0. Asterisks denote p < 0.05

Table 4.

Multivariable logistic regression: predictors of VL suppression (≤ 200 copies/mL), full cohort, Mildmay Hospital Uganda, 2016–2023 (N = 114,406 observations)

Predictor	aOR	95% CI	p-value
Year (per year from 2016)	1.091	1.074–1.109	< 0.001
Female (vs. Male)	1.098	1.013–1.191	0.024
Paediatric < 15 (vs. Adult)	0.481	0.453–0.510	< 0.001
Adolescent 15–24 (vs. Adult)	0.400	0.379–0.422	< 0.001
Older Adult ≥ 50 (vs. Adult)	1.454	1.351–1.565	< 0.001
Year × Female interaction	1.046	1.025–1.067	< 0.001

N = 114,406 observations; Akaike Information Criterion (AIC) = 63,315.4. Reference age group: Adult 25–49 years. Outcome: VL suppression ≤ 200 copies/mL

Virological trajectories by regimen group

Among 17,050 patients with two or more VL tests, 8,353 could be assigned to one of the three regimen-defined groups used for trajectory analysis. DTG-direct initiators achieved the highest consistently suppressed trajectory rate at 82.0%, compared with 63.7% among switchers and 64.5% among documented non-DTG-only patients (Fig. 10). Virologic failure rates differed markedly with 7.7% in the documented non-DTG-only group, 3.8% among DTG-direct initiators, and 2.0% among switchers. The consistently unsuppressed trajectory was rare in both DTG groups at 0.5–0.6%, compared with 10.3% in the documented non-DTG-only group.

Fig. 10.

Virological trajectory distribution by ART regimen group. Trajectory classification uses the Uganda suppression threshold of ≤ 200 copies/mL. Mildmay Hospital Uganda, 2016–2023. N = 8,353 multi-test patients assigned to one of the three displayed regimen groups. A further 8,697 multi-test patients could not be assigned to these regimen-defined groups and are not shown

In a separate pre-to-post transition-matrix analysis restricted to switchers with at least two VL tests before and at least two VL tests after the switch date, 386 of 3,002 patients (12.9%) moved from non-suppressed trajectory categories to consistently suppressed after the switch. Of the 2,047 patients who were consistently suppressed before switching, 1,862 (91.0%) remained consistently suppressed afterwards, while 90 (4.4%) developed fluctuating patterns and 38 (1.9%) experienced virologic failure. This retention rate is consistent with the IeDEA finding that pre-switch suppression strongly predicts post-switch outcomes [20].

Time to viral load suppression

Among 1,698 initially unsuppressed patients with at least one subsequent VL test, 1,459 (85.9%) ultimately achieved VL suppression during the study period. The Kaplan-Meier estimate of median time to first suppression was 12.6 months (Fig. 11). No statistically significant difference in time to suppression was observed between male and female patients (log-rank p = 0.059).

Fig. 11.

Time to viral load suppression by sex. Kaplan-Meier cumulative probability of achieving VL suppression (≤ 200 copies/mL) among initially unsuppressed patients, stratified by sex. Mildmay Hospital Uganda, 2016–2023. N = 1,698; log-rank p = 0.059

Discussion

This study addressed several linked questions, and the key findings map onto each. Our primary objective was to quantify the virological impact of the programmatic transition to DTG. In the overall paired cohort of 1,928 switchers of all ages, suppression improved from 87.2% before the switch to 94.0% at 12 months afterwards, a 6.8%-point gain sustained at 94.7% at 24 months, with an adjusted odds ratio for DTG exposure of 1.79 that rose to 2.44 in the second year. Our second objective was to assess whether the benefit was uniform across subgroups. All age groups and both sexes improved significantly, with the largest absolute gain in paediatric patients (12.2% points). Paediatric and adolescent patients remained below the 95% target after switching and no formal effect modification by age or sex was detected. DTG-direct initiators achieved the highest consistently suppressed trajectory (82.0% versus 64.5% of documented non-DTG-only patients). Finally, at the programme level, the interrupted time series showed a pre-existing upward trend without a discrete step-change at rollout. This showed that the patient-level benefit was distributed across the phased transition rather than concentrated at a single calendar date.

Taken together, these results provide the first patient-level, within-patient before-and-after evidence that Uganda’s programmatic non-DTG-to-DTG transition was associated with a significant and durable improvement in VL suppression. Using the Uganda national guideline definition of ≤ 200 copies/mL, the 6.8%-point absolute improvement at 12 months post-switch and the adjusted odds ratio of 1.79 represent a clinically meaningful effect, with the benefit increasing further to an aOR of 2.44 by the second year. Three complementary analytical approaches converge on this conclusion. The McNemar analysis controls for time-invariant between-patient confounding by comparing each switcher against their own pre-switch status. The Mundlak-corrected regression approximates a within-patient effect in the pooled longitudinal framework [26]. The ITS contextualises the improvement as gradual rather than abrupt, consistent with the phased programmatic rollout. We interpret these three analyses as mutually corroborating rather than as independent confirmations, and given the multiplicity of analyses, we emphasise the consistency of direction across methods rather than any single p-value. These findings are broadly consistent in direction with evidence from Lesotho, Tanzania, and the IeDEA consortium [15–18, 20], though direct comparison of absolute suppression rates is complicated by the more stringent ≤ 200 threshold used here.

Notably, the application of Uganda’s ≤ 200 copies/mL suppression threshold produces suppression rates that are materially lower than would be obtained using the WHO < 1,000 threshold. In 2022, for example, suppression under ≤ 200 was 94.9%, whereas under < 1,000 it exceeds 96%. This distinction matters clinically. The 2.7% of tests recording low-level viraemia (201–999 copies/mL) represents patients who would be classified as suppressed by some international benchmarks but who are at elevated risk for treatment failure, resistance development, and onward transmission [25]. These patients represent a distinct clinical population requiring enhanced monitoring.

The present study complements and extends recent work from Mildmay Hospital Uganda using the same facility’s data. Ankunda et al. [21] evaluated ART optimisation among 1,036 children and adolescents in a descriptive cross-sectional design. Their study contributed important paediatric-specific transition-pathway data, particularly for PI-to-DTG and ABC/AZT-to-DTG switches, and documented adverse event profiles classified by Medical Dictionary for Regulatory Activities (MedDRA) organ class. A separate longitudinal analysis at Mildmay examined VL suppression across the maternal timeline among 1,291 pregnant and breastfeeding women [22]. Ankunda et al. further documented drug resistance mutation prevalence among patients with virologic failure at Mildmay, identifying NNRTI mutations in 52.8% and substantial NRTI cross-resistance [23]. The present study differs from these prior analyses in five important respects. First, it includes all age groups, employs a within-patient paired before-and-after design rather than repeated cross-sections, applies Mundlak-corrected logistic regression to approximate random-effects estimation, incorporates an interrupted time series to contextualise the ecological trend, and introduces trajectory transition matrices to quantify the proportion of patients whose longitudinal suppression trajectory changes could be associated with the DTG switch. These analytical features collectively provide stronger control for within-patient confounding than descriptive cross-sectional approaches, although the design remains observational and cannot firmly establish causation. An important caveat is that the largest single-year gain between 2019 and 2020 coincided with both peak DTG rollout and the introduction of multi-month dispensing during the COVID-19 pandemic. Their relative contributions cannot be fully separated using retrospective observational data. The apparent tension between the strong within-patient improvement and the flat interrupted time series is reconciled by the timing structure of the two analyses. The paired analysis detects improvement timed to each patient’s own switch date, whereas the interrupted time series tests for a discrete change at a single shared calendar date. Because switches were distributed across several years, a sharp population-level step-change at January 2019 would not be expected even when the individual-level effect is real, and the gradual rollout, secular trend, and concurrent service-delivery changes further smooth the ecological signal.

The subgroup analyses show that all age and sex groups improved significantly after the switch, and formal interaction tests found no statistically significant heterogeneity in the DTG benefit by age group or sex. Paediatric patients showed the largest absolute gain of 12.2% points, reflecting the substantial pharmacological room for improvement from the EFV era, though their post-switch suppression of 89.8% still fell short of 95%. This points to challenges specific to paediatric ART delivery including caregiver dependence and formulation palatability. Adolescents similarly improved significantly but remained below 95% post-switch at 92.1%, pointing to persistent structural adherence challenges beyond the drug itself [27]. These findings are concordant with the paediatric data from Ankunda et al. [21], who reported that EFV-to-DTG transitions improved suppression from 81.3% to 88.0% and NVP-to-DTG from 80.6% to 88.8% among children and adolescents at the same facility. The minimal improvement among early adopters who switched in 2018–2019 reflects their already high baseline suppression of 95.4%, a ceiling effect rather than an absence of benefit.

The trajectory analysis provides a finding not previously reported from a sub-Saharan African facility at this scale: 12.9% of switchers moved from non-suppressed trajectory categories to consistently suppressed after the switch. This is consistent with the switch being associated with improved virological trajectory in a minority of patients. The 91.0% retention rate among pre-switch consistently suppressed patients supports the virological safety of programmatic switching in stable patients.

These findings are most generalisable to large, urban and peri-urban, specialised HIV treatment centres in Uganda and comparable East African settings that implemented the same WHO-aligned DTG transition over the same period. Mildmay Hospital Uganda serves a predominantly Kampala metropolitan population, so the cohort captures the programmatic reality of a high-volume referral centre but may not reflect rural or lower-resourced facilities with less frequent viral load monitoring, a different patient mix, or slower regimen rollout. Because the transition we studied followed national policy rather than clinician discretion, the exposure is representative of programmatic practice across Uganda, which supports the external validity of the direction of effect even where absolute suppression rates differ by setting. The age and sex patterns we observed, in particular the persistent adolescent and paediatric gaps, are consistent with regional evidence and are likely to generalise, whereas the precise magnitude of the suppression gains should be extrapolated cautiously given the single-centre design and the better-monitored profile of the paired analytic cohort.

Several limitations apply to this study. The observational design cannot establish causality, and contributions of DTG, multi-month dispensing, and secular adherence improvements cannot be fully disentangled. The paired analysis excluded 52% of switchers due to irregular monitoring schedules, potentially introducing selection bias. As shown in Table S1, included patients were younger, more frequently monitored, and had higher baseline CD4 than excluded patients, so the paired estimate most plausibly overstates the average benefit. The direction of effect is therefore more secure than its precise magnitude. The ITS intervention date of January 2019 is operationally defined and sensitivity analyses using alternative cut-points have not been performed. Individual VL measurements were used as the analytical unit without clustering correction for within-patient correlation, meaning standard errors may be marginally underestimated. Resistance genotyping data and pharmacy refill records were unavailable. This is a single-centre study at a specialised referral hospital, and results may not generalise to rural or lower-resourced settings.

Conclusions

The programmatic non-DTG-to-DTG transition at Mildmay Hospital Uganda was associated with a 6.8%-point absolute improvement in VL suppression at 12 months, sustained at + 5.5% points at 24 months, with an adjusted odds ratio of 1.79 that increased to 2.44 by the second year post-switch. Under Uganda’s national ≤ 200 copies/mL suppression definition, the cohort had not reached the UNAIDS 95% target by 2023, with suppression of 94.5%, though the trajectory of improvement is encouraging. Paediatric and adolescent patients remain priority groups for targeted adherence support beyond regimen optimisation. The 12.9% of switchers who moved from non-suppressed to consistently suppressed trajectories is consistent with a meaningful programme-level effect for a significant minority of patients. Moreover, the finding that DTG-direct initiators achieved the most durable suppression (82.0% versus 64.5% among documented non-DTG-only patients) supports prioritising immediate DTG initiation for all newly diagnosed patients, especially children and adolescents, and actively transitioning those still on older regimens.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

RAA: Conceptualization, Data curation, Formal analysis, Methodology, Software, Visualization, Writing (original draft), Project administration. CS: Data curation, Software, Validation, Investigation. CA: Investigation, Data curation, Writing (review and editing). SKN: Data curation, Validation, Writing (review and editing). JM: Methodology, Validation, Writing (review and editing). UMW: Resources, Investigation, Writing (review and editing). DM: Validation, Writing (review and editing). PK: Project administration, Resources, Writing (review and editing). RM: Supervision, Funding acquisition, Resources, Writing (review and editing). All authors read and approved the final manuscript.

Funding

This study received no external funding. The research was conducted as part of the routine research activities of the Mildmay Research Centre Uganda.

Data availability

The data underlying this analysis are held by Mildmay Hospital Uganda and are available from the corresponding author upon reasonable request and subject to institutional data governance approval.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki. Ethical clearance was obtained from the Mildmay Uganda Research Ethics Committee (MUREC-2025-808). A waiver of individual informed consent was granted because the data were pre-existing, non-identifiable, and collected in the course of standard clinical care.

Consent for publication

Not applicable. No individual-level data are presented.

Competing interests

The authors declare no competing interests.