Is modelling complexity always needed? Insights from modelling PrEP introduction in South Africa

Mathematical models can be powerful policymaking tools. Simple, static models are user-friendly for policymakers. More complex, dynamic models account for time-dependent changes but are complicated to understand and produce. Under which conditions are static models adequate? We compare static and dynamic model predictions of whether behavioural disinhibition could undermine the impact of HIV pre-exposure prophylaxis (PrEP) provision to female sex workers in South Africa.

A static model of HIV risk was developed and adapted into a dynamic model. Both models were used to estimate the possible reduction in condom use, following PrEP introduction, without increasing HIV risk. The results were compared over a 20-year time horizon, in two contexts: at epidemic equilibrium and during an increasing epidemic.

Over time horizons of up to 5 years, the models are consistent. Over longer timeframes, the static model overstates the tolerated reduction in condom use where initial condom use is reasonably high ($\ge$50%) and/or PrEP effectiveness is low ($\le$45%), especially during an increasing epidemic.

Static models can provide useful deductions to guide policymaking around the introduction of a new HIV intervention over short-medium time horizons of up to 5 years. Over longer timeframes, static models may not sufficiently emphasise situations of programmatic importance, especially where underlying epidemics are still increasing.