Adjusting AIDS incidence for non-stationary reporting delays: a necessity for country comparisons.

In many industrialized countries, infection with the human immunodeficiency virus (HIV) is one of the leading causes of mortality in adult persons below age 45. The incidence of the acquired immunodeficiency syndrome (AIDS) from surveillance systems is the most common indicator to compare the situation of the HIV-epidemic in different geographic regions or countries. Due to reporting delays, AIDS diagnoses in recent years are incompletely reported and need to be estimated. In this study, we analyze reporting delays in Switzerland and Spain for the period from 1988 to mid-1995 and estimate the number of AIDS diagnoses per year. A descriptive analysis for Switzerland shows increasing reporting delays in recent years. Then, a Bayesian generalized linear model on reverse-time hazards is used to model time trends of the reporting delay distribution. The model shows that in recent years (i) for Switzerland reporting delays became longer and yearly AIDS incidence might continue to increase, and (ii) for Spain, reporting delays became considerably shorter resulting in too large estimates of yearly AIDS incidence if stationarity of reporting delays is assumed. Critical issues of modeling non-stationarity of the reporting system are discussed and it is emphasized that estimates of recent AIDS incidence can be biased significantly if time trends of reporting are ignored-as in the example of Switzerland and Spain, this may severely distort comparisons of the AIDS epidemic in different countries.