When the Donor Steps Back: Can Indonesia's HIV Response Stand on Its Own? An interactive microsimulation and agent-based model of HIV prevention for MSM and transgender women in Jakarta and Bali, 2027–2029

This record contains an interactive, self-contained HTML presentation that embeds an individual-based simulation model of HIV prevention among men who have sex with men (MSM) and transgender women in Jakarta and Bali, Indonesia. It is designed to support programme planning as Indonesia's HIV response moves from Global Fund support in 2024–2026 to the 2027–2029 period, and towards greater domestic financing. Purpose. The model asks whether Indonesia's HIV prevention response for key populations can sustain its gains as donor financing declines. It explores five strategic questions from a health system perspective: The out-of-pocket transition. What happens when oral PrEP and condoms move to out-of-pocket purchase, and how large must a safety net be? Long-acting PrEP. What can long-acting injectable PrEP (lenacapavir) achieve given measured acceptability, and who should receive limited doses first? Reach versus retention. Does scaling up PrEP reach deliver impact when most users discontinue within three months? Case finding. Which testing intensity, network-based testing share and outreach caseload close the diagnosis gap? Transition risks. How robust is the response to partial domestic absorption of programme functions and to a hostile legal environment that disrupts outreach? Model. The deck runs a microsimulation with an agent-based network layer directly in the web browser: Population: 5,000 synthetic agents per run, simulated monthly from January 2026 to December 2032. Agent traits: age, gender identity, income, number of partners, sexual role, condom use, online partner seeking, PrEP history, lenacapavir preference and willingness to pay. These are drawn from aggregate distributions of the 2025–2026 network surveys in Jakarta and Bali. No individual respondent records are included. Partnerships: agents form and dissolve main, casual and one-off partnerships, and some have partners outside the city. HIV is transmitted act by act within partnerships, accounting for sexual role, condom use, viral suppression, acute infection and PrEP use. Health system layer: outreach workers link agents to HIV testing and PrEP, diagnosed agents refer partners for testing, and agents move through diagnosis, antiretroviral therapy and viral suppression. Calibration: done separately for Jakarta, Bali and both cities combined, so that 2026 HIV incidence is about 3.5 per 100 person-years and HIV status awareness stays near 67% under current support. Uncertainty: each scenario is run with several random seeds in parallel background workers. Scenarios are compared using common random numbers, and results are reported as medians with 5th–95th percentile ranges. Cross-check: results are compared with a deterministic compartmental model. Features. 17 slides in a storytelling structure (baseline, pivot, cliff, after the donor, decisions), with live sliders for every scenario parameter. City selector: Jakarta, Bali, or both combined. Results per 10,000 MSM and transgender women, or as absolute numbers when population size estimates are entered. Individual life-course trajectories of synthetic agents. Attribution of new infections by partnership type. Formula and legend panels on every modelling slide, plus a full parameter and source table. How to use. Download the HTML file and open it in a recent version of Chrome, Edge, Firefox or Safari. An internet connection is required to load the charting library (D3.js) and web fonts. Navigate with the arrow keys; press D for the contents. Most slides update within a few seconds on a typical laptop; the heaviest slides may take 5–10 seconds. Use the "Fast" precision setting for live demonstrations. Limitations. Results are illustrative projections under stated assumptions, not forecasts. They are intended to support, not replace, national estimation tools (such as Spectrum/AEM) and costed planning. Partnership formation uses random matching with simple age mixing, a simplified version of statistical network models. Act rates, partnership durations and per-act transmission probabilities come from the published literature. Survey respondents were recruited through snowball and service-based channels, so survey prevalence and PrEP use levels are not used directly. Lenacapavir uptake is based on stated preference, which typically overestimates actual uptake. Effects of outreach caseload, legal shocks and domestic absorption are assumption-driven and should be read as sensitivity analyses. Differences smaller than about 20 infections per 10,000 are within simulation noise. Data and ethics. The underlying survey data are not included in this record. The model embeds only aggregate distributions.