A fuzzy discrete event system with self-learning capability for HIV/AIDS treatment regimen selection

Based on the fuzzy discrete event system theory, we originally created, we recently reported the development of an innovative regimen selection system for the first round of highly active antiretroviral therapy of HIV/AIDS patients. The core of the system consisted of fuzzy finite state machine models for treatment regimens and a genetic-algorithm-based optimizer. In the present paper, we studied the inherent self-learning capability of the system. We focused on four historically popular treatment regimens with 32 different associated treatment objectives involving the four most important regimen factors (potency, adherence, adverse effects, and future drug options). Depending on what is to be learned, the highest self-learning accuracy was 100% and the lowest 81% with the average and standard deviation being 93% and 6.3%, respectively. These results establish our approach as a novel supervised learning mechanism. One major advantage of it over the popular neural network learning is that a reasoning chain between input and output of the system is always readily available for humans to understand its decisions. Our approach proves it to be feasible to quantitatively estimate clinical utility of a regimen and compare it with other regimens even before it is available, all with minimal involvement of AIDS experts.