Sensible plans for health-care needs and determination of priorities for expenditure require regular assessment of trends in HIV incidences. In particular, trends in the relative HIV incidences of different risk categories are useful when assessing whether current control strategies are working equally well for all risk categories. Here five tests for such trends are proposed for the analysis of AIDS incidence data and their performances are compared by a simulation study, assuming a log-linear trend in the HIV incidences for two risk categories. A convenient test based on a log-linear model for AIDS incidences is found both effective and robust to the nature of the underlying trend. The maximum likelihood estimate of the trend parameter is found stable even though estimates of other HIV incidence parameters are unstable. Smoothing of estimates of the other HIV incidence parameters is recommended because this dramatically reduces the rate of convergence of the iterative methods used to obtain the estimates.
I N T R O D U C T I O N
An important part of monitoring the HIV/AIDS epidemic is to look for trends in incidences for different risk categories. For example, there might be concern about the likelihood of increased incidences in heterosexually acquired HIV infection.' ~ Any detected trends can have a major influence on how resources for the control of HIV are allocated. The natural place for trends to originate is in the HIV incidences, and they are the trends of particular interest when considering disease control. However, it is AIDS incidences that are observed, not HIV incidences, a n d we must bear in mind that a trend in the former is usually a natural consequence of an earlier trend in the latter.
A variety of trends in the incidences of HIV infection and AIDS diagnoses are of interest when considering transmission control and resource allocation. Trends in the absolute number of HIV infections over time are clearly of interest. However, it is also necessary to look a t HIV incidences for one risk group relative to others. A small increase in HIV incidences over time for a risk group might not be of concern until viewed relative to the decrease in HIV incidence in other groups. It is then useful for decision makers to know if the apparent trend in the relative incidence is real or not, Such a trend might indicate that current control measures are working better for one risk group than for another. It might even indicate that if it were not for the current control measures then there would be a substantial epidemic in this risk group.
Here we consider methods for studying the trend in relative HIV incidences for two risk categories when observations on their AIDS incidences are available. A convenient test based on a simple model, with trend, for the AIDS incidences is compared with tests based on modelling