Prediction and validation in the public health modelling of HIV/AIDS

Mathematical models are an integral part of long-range scientific research and are broadly equivalent to the hypotheses to be tested. Validation consists: (1) in checking whether theoretical expectations are sufficiently close to observed values; and (2) in showing that theoretical constructions that pass the first test can also make verifiable predictions of future events. When modelling is used in operational situations to assist practical decision-making, as in the public health surveillance, prediction and control of infectious diseases, especially HIV/AIDS, it is easy to use the first criterion, but not so simple to implement the second. The paper discusses various methods of improving the validation of a specific classical compartmental model of HIV/AIDS geared to good serial public health data on AIDS incidence. These methods include model fitting to existing data, cross-checking findings with independent research results, general circumstantial support, and the possibility in special situations of the quasi-prediction of present or recent data using models fitted only to sufficiently distant past data.