Background Pakistan‘s HIV epidemics among people who inject drugs (PWID) vary widely across different cities and has increased dramatically over time. To help guide future HIV programming, we used statistical and mathematical modelling to identify possible causes for these differing HIV epidemics.

Methods Cross-sectional integrated biological and behavioural surveillance (IBBS) data describing high-risk behaviours and HIV status among PWID were collected by the Canada-Pakistan HIV/AIDS Surveillance Project (HASP) over five rounds from 2005–2017. We used regression analyses to identify city-level associations between the prevalence of HIV infection and different risk behaviours. We then developed a dynamic HIV transmission model to reflect these associations. The model was calibrated to the IBBS dataset to estimate relative risks of HIV transmission due to each risk factor, and their respective population-attributable fractions (PAFs) over 10-years. Lastly, we investigated the prevention benefit that could be achieved if these behaviours were reduced across different settings to the lowest observed values.

Results Multivariable statistical analyses identified professional injecting use at last injection (‘ProfInj’) and heroin use within the past month (‘HeroinUse’) as key high-risk behaviours associated with city-level HIV infection, with HIV prevalence increasing by 4% and 2% for every 10% increase in the prevalence of each respective behaviour. Modelling projections estimated that ProfInj and HeroinUse increased the relative risk of HIV transmission by 7.9 (95%CrI 4.6–14.0) and 2.2 (1.1–3.7) times, respectively, with 10-year PAFs across all cities being 52% (37–61%) and 21% (3–34%), separately, and 61% (51–66%) combined. Lowering each behaviour’s prevalence across all cities to the lowest observed prevalence (11% ProfInj, 0% HeroinUse) reduced overall relative HIV incidence by 43.9% (36.9–47.7%) and 44.1% (11.0–60.5%) over 10-years, respectively.

Conclusion Pakistan’s HIV epidemic is heterogeneous, with professional injecting and heroin use likely large contributors to the differing epidemics. Interventions focussed on these behaviours, especially professional injecting, could substantially reduce HIV incidence.

Disclosure No significant relationships.