Clinical impact and cost-effectiveness of rapid versus non-rapid antiretroviral therapy initiation in HIV-positive men who have sex with men in China using modelling and 96-week multicenter cohort data

Abstract

International treatment guidelines recommend rapid initiation of antiretroviral therapy (ART) for individuals newly diagnosed with HIV-1 infection; however, robust long-term real-world evidence remains limited in China. We aimed to evaluate the clinical outcomes and cost-effectiveness of efavirenz (400 mg) plus lamivudine and tenofovir disoproxil fumarate (EFV + 3TC + TDF) versus the coformulated bictegravir/emtricitabine/tenofovir alafenamide (BIC/FTC/TAF) in rapid (≤ 14 days from diagnosis) versus non-rapid (> 14 days) initiation among HIV-diagnosed men who have sex with men (MSM). A real-world cohort of 301 ART-naïve HIV-positive MSM was stratified into four groups: rapid BIC (G1, n = 74), rapid EFV (G2, n = 77), non-rapid BIC (G3, n = 84), non-rapid EFV (G4, n = 66). Primary endpoint was CD4 + T-cell count change from baseline at weeks 48 and 96; secondary endpoints included viral suppression (< 50 copies/mL) and treatment persistence. Confounding was mitigated via the cloning, censoring, and weighting (CCW) method to emulate a randomized controlled trial (RCT), with cost-effectiveness analyzed using a hybrid economic evaluation model referencing China’s 2024 per capita GDP. At week 96, rapid initiation was associated with superior CD4 + T-cell recovery, particularly in the EFV-based regimen (mean difference: 144 cells/µL, p < 0.001 for rapid vs. non-rapid EFV). BIC/FTC/TAF demonstrated significantly better immunological recovery (mean difference: 151 cells/µL, p < 0.001 for non-rapid BIC vs. EFV) and higher treatment persistence (75.68%-84.52% vs. 53.25%-81.82%). Economic evaluation indicated a favorable profile for BIC/FTC/TAF. Rapid ART initiation achieved excellent virological suppression, superior long-term immunological recovery, and cost-effectiveness. Compared with EFV+3TC + TDF, BIC/FTC/TAF exhibited advantages in efficacy, economic value, and treatment persistence.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-43075-w.

Introduction

The global human immunodeficiency virus (HIV) epidemic remains a critical public health issue decades after its emergence. In China, an estimated 1.05 million people were living with HIV by 2020, with the epidemic imposing a substantial and rising health-economic burden, evidenced by the surge in national funding for HIV/AIDS control from 160 million CNY in 2004 to 3.7 billion CNY in 2020^1,2.

The management of HIV has undergone a paradigm shift over the past decade, moving from deferred treatment to the universal and immediate initiation of ART for all people with HIV (PWH). This strategy, termed rapid ART initiation, was defined by the World Health Organization (WHO) as starting treatment within seven days of HIV diagnosis for all individuals, with same-day initiation encouraged for prepared patients³. Compelling evidence has demonstrated that rapid ART accelerated viral suppression, enhanced immunological recovery, improved treatment retention, and reduced HIV-related mortality and transmission^4–6. Consequently, it has become a cornerstone of global efforts to achieve the UNAIDS 95-95-95 targets and is strongly endorsed by major international guidelines from the WHO, the U.S. Department of Health and Human Services (DHHS)⁷, and the European AIDS Clinical Society (EACS)⁸.

Consistent with this global consensus, Chinese national guidelines have progressively incorporated rapid ART into clinical practice. The 2021 Chinese Guidelines for Diagnosis and Treatment of HIV/AIDS and the 2023 National Free ART Manual formally recommend ART initiation within 30 days of diagnosis, while actively encouraging a shorter interval (≤ 7 days) for eligible patients^9,10. Recommended first-line regimens for rapid initiation consist of two nucleoside reverse transcriptase inhibitors (NRTIs) plus a third agent, such as an integrase strand transfer inhibitor (INSTI) or a non-nucleoside reverse transcriptase inhibitor (NNRTI). Among these, the single-tablet regimen bictegravir/emtricitabine/tenofovir alafenamide (BIC/FTC/TAF) and the multi-tablet regimen efavirenz/lamivudine/tenofovir disoproxil fumarate (EFV+3TC + TDF) are widely used in China, however, their comparative long-term effectiveness and economic value within rapid initiation frameworks remained insufficiently established^11,12.

This evidence gap was particularly salient for key populations, such as MSM, bearing a disproportionate burden of the HIV epidemic in China. While prior studies like the nationwide cohort analysis by Zhao et al.¹³ and the randomized trial by Wang et al.¹² have demonstrated the short-term efficacy and safety of rapid ART in Chinese MSM, critical questions persisted. Specifically, there remained a scarcity of robust, long-term real-world evidence comparing the clinical and economic outcomes of rapid versus non-rapid initiation strategies, especially when contrasting modern INSTI-based regimens (e.g., BIC/FTC/TAF) with traditional NNRTI-based regimens (e.g., EFV+3TC + TDF)^14,15.

To fill these gaps, we conducted a 96-week real-world cohort study to evaluate the comparative clinical effectiveness and cost-effectiveness of rapid versus non-rapid ART initiation, focusing on the BIC or EFV-based regimens among HIV-positive MSM in China. Our findings aimed to inform clinical practice and health policy by providing much-needed evidence on optimizing ART initiation strategies in real-world settings.

Methods

Study design

We conducted a retrospective cohort study using a prospectively maintained clinical database of PWH at Beijing Youan Hospital, Capital Medical University. Since March 2021, this database has systematically collected standardized clinical information from all visiting patients.

Patients were retrospectively identified from this database between March 2021 and March 2023, with follow-up lasting until March 2025 (96 weeks). Eligible participants were adult patients (age ≥ 18 years) diagnosed with HIV-1 infection between March 2021 and March 2023 who initiated ART with either an EFV-based regimen (efavirenz 400 mg + lamivudine + tenofovir disoproxil fumarate) or a BIC-based regimen (bictegravir/emtricitabine/tenofovir alafenamide).The 400 mg EFV regimen has demonstrated non‑inferior virological efficacy and improved tolerability compared with the 600 mg dose in randomized trials (ENCORE1) and is included as an alternative option in WHO policy recommendations. ART initiation was classified as “rapid” (≤ 14 days from diagnosis) or “non-rapid” (> 14 days from diagnosis). Based on the initiation timing and treatment regimen, patients were stratified into four mutually exclusive groups: G1: rapid initiation BIC group; G2: rapid initiation EFV group; G3: non-rapid initiation BIC group; G4: non-rapid initiation EFV group.

This study aimed to evaluate the comparative clinical effectiveness of BIC/FTC/TAF versus EFV+3TC + TDF under rapid and non-rapid ART initiation in HIV-1-infected adult MSM in China. Additionally, a comprehensive cost-effectiveness evaluation was performed to compare the long-term economic outcomes across groups. Uncertainty analyses were conducted to assess result robustness and identify key influential parameters, thereby generating evidence to support clinical policy decision-making regarding rapid ART initiation in China.

Participant recruitment and eligibility

Eligible participants were required to meet the following inclusion criteria: (1) Initiating ART for the first time (ART-naïve); (2) Availability of complete baseline demographic information and clinical laboratory data; (3) Availability of at least baseline and follow-up data at either week 48 or week 96. Exclusion criteria included: (1) Pregnancy or lactation; (2) Co-existing other severe immunodeficiency diseases or ongoing immunosuppressive therapy; (3) Significant missing baseline data.

Ultimately, 301 patients met the criteria and were included in the final analysis. The study was conducted in collaboration with Beijing Youan Hospital, with ethical approval obtained (Approval No: LL-2020-127-K).The requirement for individual patient consent was waived by the ethics committee due to the retrospective nature of the study and the use of de-identified clinical data. The data were collected between March 2021 and March 2025.

Variables and data sources

Patient follow-up visits were conducted at baseline, and at weeks 4, 8, 12, 24, 36, 48, and 96. At each visit, measurements included body weight, sleep quality assessment, and evaluation of adverse reactions, along with tests for total cholesterol, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol (LDL-C), triglycerides, alanine aminotransferase, aspartate aminotransferase, serum creatinine, and eGFR, Plasma HIV-1 RNA viral load, CD4 + T-cell count, CD8 count and the CD4/CD8 ratio. The clinical efficacy of EFV+3TC + TDF or BIC/FTC/TAF was assessed based on viral suppression and changes in CD4 + T-cell count from baseline to week 48 and 96, under both rapid and non-rapid ART initiation strategies.

The primary study endpoint was the proportion of participants achieving viral suppression (plasma HIV-1 RNA < 50 copies/mL) at weeks 48 or 96. Secondary endpoints included treatment safety, virological efficacy within predefined subgroups, analyzed according to changes from baseline in CD4 + T-cell count and viral suppression rate at weeks 48 and 96.

Data processing

Data cleaning addressed missing values, outliers, and inconsistencies. Incomplete or erroneous records were either corrected or excluded to ensure data quality. Records failing to meet follow-up requirements (e.g., incomplete baseline characteristics or missing CD4 + T-cell count change data) were excluded. Baseline patient characteristics were summarized using descriptive statistics, with variables defined based on research objectives, including baseline characteristics (e.g., CD4 + T-cell count, viral load), treatment regimen, and treatment outcomes (e.g., viral suppression rate).

To address potential baseline confounding and ensure group comparability, we employed the cloning, censoring, and weighting (CCW) approach, which was a causal inference method that emulates a target randomized trial using observational data. This framework mitigates selection bias and time-varying confounding often encountered in conventional treatment comparisons¹⁶.

Each participant was artificially cloned into four copies at baseline, with each copy assigned to one of the treatment strategies of interest (G1-G4). This cloning procedure aligns with the intention-to-treat principle by allowing all individuals to be eligible for each strategy, thereby breaking the link between baseline characteristics and treatment assignment. Thereafter, clones were dynamically censored once their observed treatment pathway deviated from their assigned strategy. For example, a patient assigned to rapid initiation with BIC but who switched to EFV by week 12 would have the corresponding clone censored at that time point. To account for selection bias introduced by censoring and to balance baseline covariates across strategy groups, we applied longitudinal inverse probability of censoring weights (IPCW). This weighting creates a pseudopopulation in which the assignment of treatment strategies is independent of baseline covariates, approximating the conditions of a randomized trial.

Two primary analysis datasets were defined. The primary efficacy estimates were derived from the intention-to-treat (ITT) dataset, which included all cloned participants according to their randomly assigned treatment strategies, irrespective of subsequent treatment deviations or discontinuations. This approach preserved the baseline randomization mimicry achieved by cloning and provided an unbiased estimate of the strategy’s effectiveness in a real-world clinical setting. To evaluate the robustness of our findings and to estimate efficacy under ideal conditions, a secondary per-protocol (PP) analysis was conducted. The PP dataset included only the clones who adhered to their assigned treatment strategy throughout the entire follow-up period. Results from this scenario analysis were presented as supplementary to the primary ITT analysis.

A comprehensive analysis of clinical characteristics was conducted within the MSM population, detailing baseline characteristics, disease progression, treatment responses, and adverse events. The effects of the two treatment regimens (BIC and EFV) across different initiation timings (rapid vs. non-rapid) were systematically evaluated in both the ITT and PP datasets. Key endpoints (CD4 + T-cell count changes, viral suppression rate, discontinuation rate, adverse event incidence) were incorporated into an infectious disease dynamic model for cost-effectiveness evaluation.

Statistical analysis

Continuous variables were presented as median and interquartile range (IQR) and compared using the Mann-Whitney U test. Categorical data were summarized as frequency and percentage, and comparisons were performed using the X² test or Fisher’s exact test, as appropriate. A two-sided P value < 0.05 was considered statistically significant. Additionally, 95% confidence intervals (CIs) were calculated for the difference in viral suppression rates and changes in CD4 + T-cell counts. All statistical analyses were performed using R (The R Project for Statistical Computing, version 4.3.3).

Economic evaluation model

Model design

This study integrated a decision tree-Markov model with an infectious disease dynamic model to simulate, respectively, the clinical progression and the population-level transmission of HIV among MSM in China and to evaluate the cost-effectiveness of treatment groups under different initiation conditions, model frameworks and datasets.

The Markov model adopted a lifetime horizon with monthly cycles, whereas the infectious disease dynamic model projected outcomes over a 30-year time horizon using annual cycles¹⁷. Health states were defined based on treatment line (first-line, second-line regimen or treatment failure), viral suppression status (suppressed or unsuppressed), and immunological stage according to the WHO classification for HIV/AIDS (CD4 + T-cell count: >500, 350–500, 200–350, 100–200, and ≤ 100 cells/µL)^18,19.

Detailed algorithms for state transitions, model explanations, and parameters were provided in the Appendix. The overall model structure was summarized in Fig. 1.

Fig. 1.

Economic Evaluation Model.

Parameter Explanations in Markov Model: No_P₁₋₅: Natural disease progression rates in ineffective states; ART_P_i1−5: ART effectiveness rates (i means different treatment); d_n: Natural mortality rate; d_1 ~ 5: Additional mortality rate due to HIV/AIDS in different CD4 + T-cell count health states; ade_1 ~ 5: Incidence rates of AIDS-defining events in different CD4 + T-cell count health states; ae_i: Incidence rates of adverse events in different treatment; First-line_vd: Discontinuation rate of first-line treatment due to drug resistance; First-line_nd: Discontinuation rate of first-line treatment due to non-resistance reasons; Second-line_vd: Discontinuation rate of second-line treatment due to drug resistance; Second-line_nd: Discontinuation rate of second-line treatment due to non-resistance reasons.

Parameter Explanations in Infectious Disease Dynamic Model: α: Number of individuals entering the model; γ: Natural mortality rate; δ: Additional mortality rate due to HIV/AIDS; ε₁₋₄: Natural disease progression rates in ineffective states; ζ₁₋₄: First-line treatment effectiveness rates; η₁₋₄: Second-line treatment effectiveness rates; λ: Diagnosis and treatment rate; µ: Viral suppression rate (effectiveness of first-line treatment); ξ: Viral suppression rate (effectiveness of second-line treatment); θ: Discontinuation rate of first-line treatment; κ: Discontinuation rate of second-line treatment; ω: Infection rate.

Cost-effectiveness analysis and uncertainty analysis

A comparative evaluation of the two ART regimens was performed by calculating the incremental cost-effectiveness ratio (ICER) under rapid and non-rapid initiation strategies and across different datasets. Cost-effectiveness thresholds were defined based on China’s 2024 per capita GDP. ($13,445)

To assess parameter uncertainty, both one-way sensitivity analysis (OWSA) and probabilistic sensitivity analysis (PSA) were conducted. In the OWSA, cost and utility parameters were varied by ± 20% from their baseline values, while clinical parameters were evaluated using their 95% CI. For the PSA, appropriate probability distributions were assigned to probabilities and costs. A cost-effectiveness acceptability curve (CEAC) was subsequently generated through Monte Carlo simulations.

Results

Clinical outcomes analysis

Patient enrollment

Figure 2 outlined the patient enrollment and follow-up flowchart. Between March 2021 and March 2022, a total of 340 newly diagnosed MSM HIV patients who met the inclusion criteria were enrolled. They were divided based on the interval from diagnosis to ART initiation.

Fig. 2.

Patient enrollment flowchart.

Among the 149 patients with rapid ART initiation (≤ 14 days), 73 were treated with BIC/FTC/TAF and 76 with EFV+3TC + TDF. In this group, 13 patients were excluded: 4 due to withdrawal (loss to follow-up or transfer out) and 9 due to treatment failure. Subsequently, among the 73 BIC/FTC/TAF-treated patients who completed follow-up, 59 remained on the original regimen and 14 discontinued for proactive reasons (e.g., improving compliance). Among the 63 EFV+3TC + TDF-treated patients who completed follow-up, 38 remained on the original regimen, 16 discontinued due to passive reasons, and 9 discontinued for proactive reasons.

Among the 191 patients without rapid ART initiation (> 14 days), 84 received BIC/FTC/TAF and 66 received EFV+3TC + TDF. In the BIC/FTC/TAF subgroup, 4 patients were excluded: 2 due to withdrawal and 2 due to treatment failure. In the EFV+3TC + TDF subgroup, 3 patients were excluded: 1 due to withdrawal and 2 due to treatment failure. Thereafter, among the 80 BIC/FTC/TAF-treated patients who completed follow-up, 71 remained on the original regimen, 6 discontinued due to passive reasons (adverse events, suspected resistance/virologic failure, or intolerance), and 3 discontinued for proactive reasons (planned optimization such as simplification, patient preference, or adherence support). Among the 63 EFV+3TC + TDF-treated patients who completed follow-up, 54 remained on the original regimen, 8 discontinued due to passive reasons, and 1 discontinued for proactive reasons.

Patient baseline characteristics

Baseline demographic and laboratory indicators were obtained from the prospectively maintained clinical database at ART initiation (baseline) and summarized in Table 1. Unmatched results described the crude observed cohort, whereas the matched results refer to the CCW-weighted ITT pseudopopulation used for primary causal inference. The median age across all groups was approximately 31 years, with no significant between-group differences in either the unmatched (p = 0.2503) or matched (ITT dataset; p = 0.659) analyses. Regarding biochemical profiles, high-density lipoprotein cholesterol (HDL-C) levels differed significantly among unmatched groups (p = 0.006), with the highest median values in G3 (1.02 mmol/L; IQR: 0.86–1.14) and G4 (1.02 mmol/L; IQR: 0.82–1.17), and the lowest in G1 and G2 (both 0.91 mmol/L; IQR, 0.80–1.09 and 0.80–1.01, respectively). Similarly, triglycerides showed significant variation (p = 0.0413), with the highest median in G4 (1.38 mmol/L; IQR: 1.06–1.93) and the lowest in G2 (1.12 mmol/L; IQR: 0.86–1.64). Fasting blood glucose levels also exhibited substantial differences (p < 0.001), increasing progressively from G1 (4.99 mmol/L; IQR: 4.69–5.47) to G3 (5.45 mmol/L; IQR: 5.18–5.96). No other biochemical markers—including AST, ALT, total cholesterol, LDL-C, and creatinine—showed statistically significant differences in the unmatched analysis. After matching, all biochemical indicators, including HDL-C, triglycerides, and blood glucose, were balanced and showed no significant variation among groups (all p > 0.05). For HIV-specific indicators at baseline, CD4 + T-cell counts differed significantly in the unmatched analysis (p = 0.0391), with the highest median in G2 (374.60 cells/µL; IQR: 262.98–507.00) and the lowest in G3 (289.50 cells/µL; IQR: 167.75–442.25). In contrast, CD4/CD8 ratio and plasma HIV-1 RNA levels were comparable across all unmatched groups (p = 0.362 and p = 0.7127, respectively). Following matching, all HIV-related baseline characteristics, including CD4 + T-cell count, CD4/CD8 ratio, and HIV-1 RNA were well-balanced and showed no significant differences (all p > 0.05), indicating that the matching procedure effectively minimized baseline confounding between groups.

Table 1.

Patient Baseline Characteristics.

(A) Unmatched
Characteristic, median (IQR)		G1 (n = 74)	G2 (n = 77)	G3 (n = 84)	G4 (n = 66)	P Value
Initiation time, day		3.00 (2.00–7.00)	4.00 (2.00–8.00)	26.00 (20.00–51.25.00.25)	24.00 (18.00–34.50.00.50)	< 0.001***
Age, year		30.19 (26.16–36.65)	30.48 (25.42–37.42)	30.59 (27.31–35.99)	35.48 (25.58–46.80)	0.2503
Biochemical indicators	AST, U/L	24.00 (21.00–28.75.00.75)	23.00 (19.00–28.00)	25.00 (21.00–35.00)	23.75 (21.25–27.75)	0.1753
	ALT, n	22.50 (17.00–32.75.00.75)	20.00 (15.00–29.00)	25.00 (18.00–35.25.00.25)	20.00 (15.25–29.00.25.00)	0.0848
	Cholesterol, mmol/L	4.25 (3.76–4.80)	4.20 (3.77–4.77)	4.45 (3.87–4.91)	4.20 (3.83–4.78)	0.5794
	HDL-C, mmol/L	0.91 (0.80–1.01)	0.91 (0.80–1.09)	1.02 (0.86–1.14)	1.02 (0.82–1.17)	0.006
	LDL-C, mmol/L	2.68 (2.24–3.16)	2.59 (2.09–3.09)	2.83 (2.39–3.17)	2.59 (2.32–3.04)	0.3555
	Triglycerides, mmol/L	1.32 (1.02–1.83)	1.12 (0.86–1.64)	1.38 (0.98–1.84)	1.38 (1.06–1.93)	0.0413
	Blood Glucose, mmol/L	4.99 (4.69–5.47)	5.11 (4.69–5.38)	5.45 (5.18–5.96)	5.25 (4.95–5.70)	< 0.001
	Creatinine, µmol/L	70.50 (63.00–75.00)	69.00 (64.00–76.00)	67.00 (62.75–75.25)	68.00 (61.00–75.75.00.75)	0.4694
HIV-related indicators	CD4 + T-cell count at baseline, cells/µL	350.25 (242.46–459.64.46.64)	374.60 (262.98–507.00)	289.50 (167.75–442.25.75.25)	333.50 (239.25–435.75.25.75)	0.0391
	CD4/CD8 ratio at baseline	0.32 (0.20–0.43)	0.37 (0.23–0.61)	0.32 (0.22–0.48)	0.34 (0.24–0.48)	0.362
	Plasma HIV-1 RNA at baseline, copies/mL	19,555.00 (5,396.50–85,940.00)	23,239.00 (6,739.00–67,267.00)	25,370.50 (5,361.75-90.75,344.50)	13,972.00 (5,710.00–59,303.50)	0.7127

(B) Matched (ITT dataset)
Characteristic, median (IQR)		G1 (n = 72.27)	G2 (n = 73.81)	G3 (n = 80.19)	G4 (n = 65.59)	P Value
Initiation time, day		3.00 (2.00–7.00)	4.00 (2.00–8.00)	24.00 (18.00–33.00)	26.00 (20.00–51.00)	< 0.001***
Age, year		30.08 (26.13–37.95)	30.47 (25.42–37.42)	35.34 (25.42–46.92)	30.48 (26.44–36.35)	0.659
Biochemical indicators	AST, U/L	24.00 (20.00–28.00)	23.00 (19.00–28.00)	24.00 (21.00–28.00)	24.00 (21.00–34.00)	0.311
	ALT, n	22.00 (17.00–33.00)	20.00 (15.00–29.00)	20.00 (16.00–29.00)	25.00 (18.00–34.00)	0.687
	Cholesterol, mmol/L	4.25 (3.75–4.80)	4.19 (3.76–4.77)	4.20 (3.83–4.78)	4.45 (3.89–4.95)	0.892
	HDL-C, mmol/L	0.93 (0.80–1.01)	0.90 (0.80–1.09)	1.02 (0.81–1.17)	1.02 (0.86–1.15)	0.492
	LDL-C, mmol/L	2.68 (2.23–3.18)	2.59 (2.07–3.09)	2.60 (2.31–3.04)	2.84 (2.42–3.19)	0.441
	Triglycerides, mmol/L	1.30 (1.02–1.83)	1.12 (0.86–1.64)	1.38 (1.05–1.95)	1.39 (0.97–1.84)	0.533
	Blood Glucose, mmol/L	4.99 (4.69–5.48)	5.07 (4.69–5.38)	5.26 (4.99–5.71)	5.45 (5.19–5.96)	0.856
	Creatinine, µmol/L	70.00 (63.00–75.00)	69.00 (63.00–76.00)	68.00 (61.00–76.00)	67.00 (63.00–75.00)	0.792
HIV-related indicators	CD4 + T-cell count at baseline, cells/µL	350.51 (246.83–463.00)	334.00 (236.50–465.39.50.39)	340.00 (249.00–473.00.00.00)	328.00 (203.00–509.00.00.00)	1
	CD4/CD8 ratio at baseline	0.36 (0.21–0.47)	0.31 (0.22–0.57)	0.34 (0.24–0.47)	0.32 (0.24–0.48)	1
	Plasma HIV-1 RNA at baseline, copies/mL	19,489.00 (4,643.00–90,594.00)	24,625.00 (7,060.00–70,056.00)	14,320.00 (5,694.00–62,024.00)	24,595.00 (4,067.00–77,321.00)	0.418

Efficacy analysis

Treatment persistence and immunological recovery were evaluated across the four study groups. As summarized in Table 2, a significantly higher proportion of participants in the non-rapid initiation groups remained on their initial regimen at the end of follow-up (G3: 84.52%; G4: 81.82%), compared to the rapid initiation groups (G1: 75.68%; G2: 53.25%; χ² = 28.685, df = 6, p < 0.001). Notably, the G2 group exhibited the highest rate of discontinuation due to treatment failure (11.69%) and the highest proportion of switches for reactive reasons such as drug resistance (65.22%). In contrast, the G1 group demonstrated a markedly higher rate of proactive regimen switches (93.76%), suggesting better tolerability or adherence management.

Table 2.

Treatment Regimen Persistence and Discontinuation.

Variable, n(%)	G1 (n = 74)	G2 (n = 77)	G3 (n = 84)	G4 (n = 66)	Statistical Test Results
Remained on Initial Regimen (first-line treatment)	56 (75.68%)	41 (53.25%)	71 (84.52%)	54 (81.82%)	χ²(6) = 28.685 p < 0.001
Second-line Regimen	16 (21.62%)	23 (29.87%)	9 (10.72%)	9 (13.63%)
Switched for Reactive Reasons (e.g., drug resistance)	1 (6.24%)	15 (65.22%)	6 (66.67%)	8 (88.86%)
Switched for Proactive Reasons (e.g., improving adherence)	15 (93.76%)	8 (34.78%)	3 (33.33%)	1 (11.14%)
Discontinuation due to Treatment Failure	0 (0.00%)	9 (11.69%)	2 (2.38%)	2 (3.03%)
Discontinued Study Participation	2 (2.70%)	4 (5.19%)	2 (2.38%)	1 (1.52%)

Immunological outcomes, as depicted in Fig. 3 and corresponding longitudinal data, revealed consistent CD4 + T-cell count recovery across all groups over 96 weeks. At baseline, no significant differences in median CD4 + T-cell counts were observed between rapid and non-rapid initiation groups within each regimen (all p > 0.05). By week 48, the rapid initiation groups showed numerically greater increases in CD4 + T-cell counts compared to non-rapid initiators, particularly in the EFV group (difference: 85.5 cells/µL, p = 0.122). By week 96, this difference became statistically significant in the EFV group (144 cells/µL, p < 0.001), while the BIC groups maintained stable and comparable counts between rapid and non-rapid arms (3 cells/µL difference, p = 0.613). Importantly, between-regimen comparisons at week 96 revealed a significant advantage of BIC over EFV in the non-rapid initiation group (151 cells/µL, p < 0.001).

Fig. 3.

Immunological outcomes from Baseline to 48 and 96 wks of Follow-up.

Economic outcomes analysis

Base case analyses

The base-case analysis results were summarized in Table 3. Panels A and B present the findings from the Markov model over a lifetime horizon. When comparing the BIC group to the EFV group (Panel A), rapid initiation was associated with substantially higher incremental costs ($43,970.58) for a minimal gain in effectiveness (0.01 QALYs), resulting in an ICER of $5,637,253.31 per QALY gained (419.28 times the per-capita GDP). In contrast, non-rapid initiation of BIC versus EFV yielded an ICER of $4,330.04 per QALY gained (0.32 times GDP), indicating cost-effectiveness. When comparing rapid versus non-rapid initiation strategies (Panel B), the BIC group led to an ICER of $89,491.68 per QALY gained (6.66 times GDP), while the EFV group resulted in a highly favorable ICER of $395.88 per QALY gained (0.03 times GDP).

Table 3.

Cost-Effectiveness Results: Cost, Effectiveness and ICER.

(A) Markov model-BIC group vs. EFV group
ART regimens	IC, $	IE, QALY	ICER	ICER/GDP
Rapid Initiation	43,970.58	0.01	5,637,253.31	419.28
Non-Rapid Initiation	11,814.96	2.73	4,330.04	0.32

(B) Markov model-Rapid Initiation vs. Non-Rapid Initiation
ART regimens	IC, $	IE, QALY	ICER	ICER/GDP
BIC group	33,380.40	0.37	89,491.68	6.66
EFV group	1,224.78	3.09	395.88	0.03

(C) Infectious disease dynamic model-BIC group vs. EFV group
ART regimens	IC, $	IE, QALY	ICER	ICER/GDP
Rapid Initiation	80.81	0.04	2,202.20	0.16
Non-Rapid Initiation	265.05	0.03	8,824.12	0.66

(D) Infectious disease dynamic model-Rapid Initiation vs. Non-Rapid Initiation
ART regimens	IC, $	IE, QALY	ICER	ICER/GDP
BIC group	3.38	0.01	419.41	0.03
EFV group	187.62	1.39E-03	134,509.38	10.00

IC incremental cost, IE incremental effectiveness.

Panels C and D displayed the results from the infectious disease dynamic model over a 30-year horizon. The comparison of BIC versus EFV group (Panel C) showed that under both rapid and non-rapid initiation, the ICERs were markedly lower than in the Markov model, at $2,202.20 and $8,824.12 per QALY gained, respectively (both below 1 times GDP). The comparison of initiation strategies (Panel D) revealed that for the BIC group, rapid initiation was dominant (less costly and more effective) with an ICER of $419.41 per QALY. For the EFV group, rapid initiation was associated with an ICER of $134,509.38 per QALY gained (10.00 times GDP).

Sensitivity analyses

In OWSA, conducted on the intention-to-treat dataset within the Markov model framework for the non-rapid BIC versus EFV group comparison, identified the drug cost of the BIC group as the primary driver of uncertainty. Discontinuation rate and viral suppression rate also emerged as influential parameters in other dataset or model scenarios, with the complete results visualized in Supplementary Appendix Figs. 6, 7, 8 and 9, panels A and B. Throughout the analysis, all outcome variations consistently remained within the pre-specified acceptable range defined by the willingness-to-pay (WTP) threshold.

PSA reflected substantial robustness of the model outcomes. The distribution of 10,000 ICERs calculated via PSA corroborated the stability of the base-case results, indicating consistent model performance. For example, the cost-effectiveness acceptability curve (CEAC) showed that at this WTP threshold (China’s 2024 per capita GDP), the probability ofthe non-rapid BIC group being cost-effective compared to the non-rapid EFV group was 100% in the ITT dataset based on the Markov model (Appendix Figs. 6, 7, 8 and 9, panels C-F).

Discussion

This 96-week real-world cohort study provided comprehensive evidence on the comparative clinical and cost-effectiveness outcomes of rapid (≤ 14 days) versus non-rapid ART initiation with two widely prescribed first-line regimens (BIC/FTC/TAF and EFV+3TC + TDF), among HIV-positive MSM in China. Our findings demonstrated that rapid ART initiation was associated with superior long-term immunological recovery. Furthermore, BIC/FTC/TAF consistently showed advantages over EFV+3TC + TDF in immunologic efficacy, treatment persistence, and cost-effectiveness, which is attributed to its higher genetic barrier to resistance, better tolerability, and single-tablet formulation that improves adherence.

A key methodological strength of economic evaluation was the adoption of a hybrid modeling approach that integrated a decision tree-Markov model with an infectious disease dynamic transmission model. This dual-model framework captured both individual-level clinical progression and population-level HIV transmission dynamics, addressing a critical limitation of traditional Markov models, which often overlook the secondary transmission benefits of viral suppression. The decision tree-Markov model, structured according to WHO immunological staging and treatment lines, was well-suited to estimate lifetime costs and QALYs from a healthcare system perspective, reflecting the natural history of HIV and treatment pathways at the patient level^18,20. However, traditional Markov models often fail to account for the secondary transmission benefits of viral suppression, potentially underestimating the full value of effective ART²¹. To quantify these population-level effects, we integrated a HIV-applicable Susceptible–Infectious–Removed (SIR) dynamic model, which projected the HIV epidemic among Chinese MSM over a 30-year horizon and incorporated behavioral parameters specific to this population, such as condom use rates and sexual contact patterns (Supplementary Appendix Table 7)^17,22. Importantly, the two components were run in parallel (i.e., simulated separately). The transmission model adopted a 30-year horizon to approximate the sexually active period of the Chinese MSM population, during which most secondary transmissions attributable to differences in time-to-suppression are expected to occur. Annual cycles were used because key epidemiologic and behavioral inputs are typically available on an annual basis, and finer cycles would require additional assumptions about within-year dynamics that are not well supported by routine data. Both components were parameterized with consistent clinical inputs from the same cohort (e.g., viral suppression, discontinuation, and mortality). We reported results from both perspectives to demonstrate how accounting for transmission can change the estimated value of rapid initiation; the transmission model outputs were not used to modify individual-level Markov transitions, and thus the models do not constitute a single fully coupled simulation^23,24.

The superior CD4 + T-cell count recovery observed in the rapid initiation groups, especially in G2, aligned with prior studies highlighting the benefits of early ART in preserving immune function and reducing time to viral suppression^4,5,12. Notably, the 144 cells/µL difference in CD4 + T-cell count at week 96 between rapid and non-rapid EFV groups underscored the potential of rapid initiation to mitigate early immune decline, even with a less contemporary regimen. However, the fact that BIC/FTC/TAF achieved comparable CD4 + T-cell count gains regardless of initiation timing suggests that regimens with higher genetic barriers to resistance and better tolerability may be less dependent on initiation speed for optimal immunological outcomes^11,25.

Regimen choice and the reasons for regimen changes are critical determinants of long-term success. In this study, we classified regimen changes as proactive (planned optimization) or reactive (passive) based on the clinical intent documented at the time of switch. Proactive changes referred to pre-planned optimization while maintaining virological control, such as simplification to a single-tablet regimen, patient preference, improved convenience, cost/access considerations, or adherence support. Reactive changes referred to switches prompted by clinical problems, including adverse events or intolerance (e.g., neuropsychiatric or metabolic symptoms with EFV), adherence-related discontinuation, and suspected resistance/virologic failure. The higher proportion of reactive switches attributed to resistance/virologic failure in EFV strategies should be interpreted in light of our long 96-week follow-up and the operational definition used in routine care: during follow-up, a plasma HIV-1 RNA level > 50 copies/mL at a scheduled visit triggered evaluation and was recorded as virologic failure/suspected resistance, which increases the likelihood of detecting and classifying such events over time, particularly for regimens with a lower genetic barrier to resistance. In contrast, the high rate of proactive switches observed in the rapid BIC group primarily reflected planned optimizations rather than treatment failures. Overall, the higher treatment persistence in the BIC groups (75.68%−84.52%) compared with the rapid EFV group (53.25%−81.82%) is consistent with the favorable tolerability and safety profile of INSTI-based regimens^11,26, while EFV discontinuations align with prior reports of neuropsychiatric and metabolic adverse effects^27,28.

Our economic evaluation further supported the preferential use of BIC/FTC/TAF, particularly in non-rapid initiation settings where it was highly cost-effective compared to EFV+3TC + TDF. The Markov model indicated that rapid BIC initiation was associated with a favorable ICER relative to China’s per-capita GDP, whereas the EFV-based rapid strategy, while clinically beneficial, was less economically attractive in the long term^29,30. The infectious disease dynamic model further highlighted the population-level benefits of rapid BIC initiation, which was both cost-saving and more effective over a 30-year horizon—a finding that underscores the public health value of combining modern regimens with rapid initiation protocols^17,31.

The robustness of our findings was confirmed through extensive sensitivity analyses, including one-way and probabilistic sensitivity analyses, which showed consistent results across a range of assumptions. The use of cloning, censoring, and weighting methods to emulate a target trial also strengthens the causal interpretation of our observational data, reducing biases inherent in conventional cohort studies³².

Several limitations should be acknowledged. First, despite rigorous statistical adjustment, unmeasured confounding might persist in this real-world study³³. Second, the single-center design might limit generalizability, although the study population was representative of a key demographic in China’s HIV epidemic³⁴. Third, we did not capture patient-reported outcomes or qualitative data on barriers to rapid initiation, which could inform implementation strategies³⁵. Finally, while the 96-week follow-up was longer than many real-world studies, even longer-term outcomes beyond two years warranted further investigation³⁶.

In conclusion, this study supported the adoption of rapid ART initiation among MSM in China, particularly with BIC/FTC/TAF, which offered a favorable balance of efficacy, tolerability, and cost-effectiveness. These findings aligned with global trends toward INSTI-based regimens and same-day or rapid-start models, and they provided locally relevant evidence to inform updates to Chinese national HIV treatment guidelines^3,9,10.

Conclusion

Rapid initiation of ART led to excellent rates of virological suppression, superior long-term immunological recovery, and was shown to be cost-effective among HIV-positive MSM in China. Compared with the EFV+3TC + TDF, BIC/FTC/TAF demonstrated superior efficacy, economic benefit, and treatment durability. These findings support the prioritization of rapid ART initiation in clinical practice and national HIV treatment guidelines for this key population.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

HIV

Human Immunodeficiency Virus

AIDS

Acquired Immunodeficiency Syndrome

ART

Antiretroviral Therapy

MSM

Men who have sex with men

CCW

Cloning, Censoring, and Weighting

QALY

Quality-adjusted Life Year

ICER

Incremental Cost Effectiveness Ratio

PWH

People with HIV

NRTI

Nucleoside Reverse Transcriptase Inhibitor

NNRTI

Non-nucleoside Reverse Transcriptase Inhibitor

INSTI

Integrase Strand Transfer Inhibitor

IPCW

Inverse Probability of Censoring Weights

ITT

Intention-to-Treat

PP

Per-Protocol

CI

Confidence Interval

IQR

Interquartile range

OWSA

One-way Sensitivity Analysis

PSA

Probabilistic Sensitivity Analysis

WHO

World Health Organization

DHHS

U.S. Department of Health and Human Services

UNAIDS

The Joint United Nations Programme on HIV/AIDS

WTP

Willingness-to-Pay

GDP

Gross Domestic Product

CPI

Consumer Price Index

AE

Adverse Event

ADE

AIDS-Defining Event

Author contributions

KZ analyzed and interpreted the data, and was a major contributor in drafting and revising the manuscript. XW coordinated across multiple institutions and contributed to the study’s conception and critical manuscript revisions. DZ contributed to data acquisition, visualization, and manuscript drafting. KZ, XW and DZ contributed to data acquisition, visualization, and manuscript revision. WT, and LD provided financial support, contributed to the study’s conception, and critically revised the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by the National Natural Science Foundation of China, International Cooperative Research Project (2023YFVA1002).

Data availability

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

Ethical standards of the relevant national and institutional committees on human experimentation and the Helsinki Declaration of 1975, as revised in 2008 were followed in all procedures. Ethical approval was obtained from the Ethics Committee of Capital Medical University-affiliated Beijing Youan Hospital (Batch number: LL-2020-127-K). Participants also signed informed consent forms before data collection began.