Social networks and forecasting the spread of HIV infection

Abstract

This study is an initial effort to use network data to forecast the spread of HIV in a large U.S. city. Data were collected from a sample of drug users and sociodemographically matched nonusers in low-income areas of Houston, Texas. Two sample-based HIV prevalence models and two sociological models were combined with three published biological models to yield forecasts of the growth of HIV seroprevalence. The forecasts predict a compounded annual growth in HIV of between 2.4% and 16.5% among low-income residents of Houston's inner city. These results suggest that forecasts are most sensitive to the nature of the sociological model used. A random mixing model showed about a threefold overestimate of 20-year projected seroprevalence compared with the empiric network data. Thus, the collection of additional social network data is probably the most important requirement for more accurate projections.