A multistate approach for estimating the incidence of human immunodeficiency virus by using HIV and AIDS French surveillance data.

Before the introduction of highly active antiretroviral therapy (HAART) in 1996, monitoring the epidemics of HIV infection was extensively based on back-calculating the number of new HIV infections from AIDS surveillance data and knowledge of the AIDS incubation period distribution. However, the increase in the AIDS incubation period induced by HAART has complicated the use of AIDS diagnosis data only, and made it necessary to supplement these data with HIV diagnosis data. We explore the advantage of combining HIV and AIDS surveillance data in the context where the HIV diagnosis data are available only for the most recent years. Extending the earlier work of Aalen et al. (Statist. Med. 1997; 16(19):2191-2210) based on a discrete-time Markov model that describes simultaneously disease progression, HIV diagnosis and treatment intake, we propose a penalized likelihood approach to estimate smooth HIV incidence, together with HIV diagnosis rates. The smoothing parameter is chosen using an approximated cross-validation criterion. In a simulation study, we show that incorporation of HIV test information improves the precision of the estimation of the incidence of infections in recent periods. The method is illustrated using HIV and AIDS surveillance data collected by the Institut de Veille Sanitaire, France. (c) 2009 John Wiley & Sons, Ltd.