|
Abstract:
|
To determine whether a drug is effective in treating alcoholism, a subject's drinking behavior must be monitored over a period of time. To capture complex features of a subject's drinking behavior, like the difference between a steady, moderate drinker and one who alternates between abstinence and binges, Markov chain models are useful. We extend Albert's Markov model for chronic diseases (Albert, 1994) to a hidden Markov model (HMM), whose hidden states represent "in-control" vs. "out-of-control" drinking. The HMM allows us to describe drinking behavior flexibly; instead of defining a subject's state at a given time strictly based on his or her observed data, the HMM essentially smoothes the observed data and lets us estimate a probability of being in each hidden state at a given time. We present simulation results, as well as an application to real data.
|
