HIV Prevention Barrier Modeling for PWID

⚠️ SUPERSEDED: This record has been replaced by https://doi.org/10.5281/zenodo.18746065, which is linked to the dedicated HIV_Prevention_PWID repository. This archive is retained for historical reference only. HIV Prevention Barrier Modeling for PWID Code repository for: Demidont AC. Structural barriers drive near-zero population-level effectiveness of Long Acting Injectable HIV prevention (LAI-PrEP) among people who inject drugs: A Computational Modeling Study. The Lancet HIV (2025). [Manuscript ID: thelancethiv-D-25-00576, Submitted Dec 29, 2025] Overview This repository contains the computational models, simulation code, and visualization scripts supporting the analysis of structural barriers to HIV prevention among people who inject drugs (PWID). The model demonstrates how policy, stigma, and infrastructure barriers compound across an 8-step prevention cascade to produce near-zero population-level effectiveness despite highly efficacious pharmacological interventions. Key Findings Current policy achieves ~0.003% sustained HIV prevention among PWID vs. 16.3% for MSM receiving identical interventions Barrier multiplication effect: Even moderate barriers at each cascade step compound to near-total prevention failure Policy-determined outcomes: Results are structural, not individual choice-based 63% outbreak probability within 5 years under current conditions Repository Structure ├── architectural_barrier_model.py # Main simulation model ├── visualize_md_results.py # Publication figure generation ├── architectural_barrier_results.json # Monte Carlo simulation outputs ├── pwid_simulation_results.json # PWID-specific analyses ├── manuscript/ # LaTeX manuscript files │ └── lancet_hiv_manuscript_acd_2025.tex └── figures/ # Publication-ready figures ├── Fig1_CascadeComparison.png ├── Fig3_PolicyScenarios.png └── Fig5_SNR_LOOCV.png Model Architecture Three-Layer Barrier Framework Layer 1: Pathogen Biology HIV establishes irreversible R₀ > 0 within hours of infection No intervention can restore R₀ = 0 post-exposure Layer 2: Testing Barriers Acute infection detection gaps (RNA testing not standard of care) 4th gen Ag/Ab tests miss first 2-4 weeks Layer 3: Architectural Barriers Policy (criminalization, incarceration) Stigma (healthcare discrimination, disclosure barriers) Infrastructure (MSM-centric cascade design) Research exclusion (LOOCV framework) Machine learning (algorithmic deprioritization) Eight-Step Prevention Cascade Awareness of LAI-PrEP availability Willingness to seek care despite system visibility Healthcare access Disclosure of injection drug use Provider willingness to prescribe Adequate HIV testing First injection received Sustained engagement Policy Scenarios Scenario P(R₀=0) Current Policy 0.003% Decriminalization Only 0.20% Decrim + Stigma Reduction 0.45% SSP-Integrated Delivery 5.0% Full Harm Reduction 9.5% Full HR + PURPOSE-4 Data 11.9% Full HR + Algorithmic Debiasing 18.6% Theoretical Maximum 19.7% MSM (Comparison) 16.3% Installation Requirements Python 3.9+ NumPy Matplotlib SciPy (optional, for advanced statistics) Setup git clone https://github.com/Nyx-Dynamics/hiv-prevention-master.git cd hiv-prevention-master pip install -r requirements.txt Usage 1. Main Simulation (architectural_barrier_model.py) Run the core Monte Carlo simulation to evaluate cascade completion across multiple policy scenarios. python architectural_barrier_model.py --output-dir ./results --n-individuals 100000 --output-dir: Directory to save .json and .csv results (default: .) --n-individuals: Number of individuals to simulate per scenario (default: 100000) --n-sa-sims: Number of stochastic avoidance simulations (default: 10000) 2. Cascade Sensitivity (cascade_sensitivity_analysis.py) Analyze how parameter uncertainty and barrier removal affect prevention probability. python cascade_sensitivity_analysis.py --output-dir ./outputs --n-samples 1000 --output-dir: Directory for figures and results (default: outputs) --n-samples: Number of Probabilistic Sensitivity Analysis (PSA) samples (default: 1000) 3. Enhanced Stochastic Avoidance (stochastic_avoidance_enhanced.py) Forecast regional outbreak probabilities and methamphetamine prevalence trajectories. python stochastic_avoidance_enhanced.py --output-dir ./outputs --n-sims 2000 --n-psa 500 --output-dir: Directory for figures and results (default: outputs) --n-sims: Number of simulations for national forecast (default: 2000) --n-psa: Number of PSA samples (default: 500) 4. Publication Figures (visualize_md_results.py) Generate the primary figures used in the Lancet HIV manuscript. python visualize_md_results.py --input architectural_barrier_results.json --output-dir ./figures --input: Path to the simulation results JSON file --output-dir: Directory where figures (Fig 1-5) will be saved 5. Quick Visualization (visualize_results.py) Quickly generate a comparison bar chart from simulation results. python visualize_results.py --input architectural_barrier_results.json --output ./figures/scenario_comparison.png Outputs architectural_barrier_results.json/csv: Core simulation data and barrier decomposition. cascade_sensitivity_results.json/csv: PSA summary and step importance rankings. stochastic_avoidance_sensitivity_results.json/csv: Regional outbreak forecasts and tornado analyses. outputs/ or figures/: Publication-quality PNG and PDF figures. Data Sources All model parameters are derived from peer-reviewed literature: Parameter Source PWID epidemiology Degenhardt et al. (2017), Lancet Criminalization effects DeBeck et al. (2017), Lancet HIV PrEP cascade Mistler et al. (2021), AIDS Behav Incarceration impact Altice et al. (2016), Lancet LAI-PrEP efficacy PURPOSE-1/2 trials, Bekker et al. (2024), NEJM Healthcare stigma Biancarelli et al. (2019), Social Science & Medicine Reproducibility All analyses are fully reproducible: Random seed: 42 Monte Carlo iterations: 100,000 Bootstrap replicates: 10,000 Citation If you use this code, please cite: @article{demidont2025structural, title={Structural barriers drive near-zero population-level effectiveness of Long Acting Injectable HIV prevention (LAI-PrEP) among people who inject drugs: A Computational Modeling Study}, author={Demidont, AC}, journal={The Lancet HIV}, year={2025}, note={Manuscript ID: thelancethiv-D-25-00576, Submitted} } License This project is licensed under the CC 4.0 International and MIT Licenses - see the LICENSE file for details. Contact AC Demidont, DONyx Dynamics LLCEmail: [e-mail]: 0000-0002-9216-8569 Acknowledgments HIV prevention research community for published parameters PWID community advocates for barrier characterization AI tools (Anthropic Claude 4.5), Jetbrains Junie, Zotero AI, Numpy, Matplotlib, Scipy, Arviz, OpenAI GPT 5.2 used as assistive technology for literature search, reference management and code refinement. Author maintainces responsibility for all aspects of project and responsbility for use and software use. This work honors the 44.1 M lives lost to HIV since the beginning of the epidemic and strives to use Mathematics and Machines to end the epidemic.