My personal view is that this stuff is basically best modeled as a multi-state markov chain. (Or if you are a real bear for punishment a semi-markov chain instead). At its simplest, the model has a chain that is linear. Each node as a transition probability to the state on the right and on the left as well as a transition probability to itself and a transition probability to the rightmost stage of PML onset. At the begining of therapy, you are in the left most state. Depending on what therapy you get, you can trannsit to the right or transit yourself or transit to PML. Conditioned on the sort of therapy, the transition probabilities change. In normal folks with no treatment, the transition probability to the absorbing PML state is extremely small. Each state has an increasing probability of transition to PML and increasing probability of transition to the next state on the right as you go from left to right in the chain. Likewise, the transition probability from right to left goes down and the self transition probability goes down. As you increase the agression in your therapy, you skew the transition probabilities to favor the transition to the right and the transition to PML. Avonex by itself and from what we have seen of the limited Tysabri mono trials does not seem to skew the probabilities much. Add other supressive agents and you head off to the right in a bigger hurry. But even with full blown AIDS, you only have a 5% probability of doing the transition to that PML absorbing state. The reason I have a set of non-disease states is to allow for the effects of usage over time. The fact that avonex decreases Tysabri clearance means that the first dosage has a different effect than the nth dosage.
Most exposure to disease causing agents can be modeled in this manner, be they infectious or otherwise.