Backcalculation models of age-specific HIV incidence rates.

This paper extends the use of backcalculation to estimate past incidence of HIV infection in different age groups from age-specific counts of AIDS incidence. The approach is flexible and allows the distribution of age at HIV infection to change over time. In this new approach, the incubation distributions used to backcalculate HIV infection rates depend on the age at HIV infection, because younger age is associated with slower progression. The effect of age on progression is estimated from a joint analysis of natural history data from several cohort studies of gay and bisexual men. As in previous applications, the incubation distributions also change over time to accommodate treatment effects and the 1987 revision of the surveillance definition. The method is applied to AIDS incidence data for the United States population. Estimated infection incidence for the entire population declined from peak levels seen in the mid-1980s. However, persons under age 25 years have accounted for an increasing proportion of new HIV infections as the epidemic has progressed. Quantitative estimates were sensitive to the assumed incubation distribution, but the trend toward younger age at HIV infection was apparent using several models of the incubation distribution. Although the models are computationally intensive, they provide useful information about seroincidence trends in different age groups.