Duke 2008 | ASHE | Equity and Efficiency in Health and Healthcare

In this paper, I present a structural model of learning under ambiguity in the context of clinical trials. Patients are concern with learning the treatment effect of the experimental drug, but face the ambiguity of random group assignment. The learning model uses difference in side effect intensity across groups as a source of information to infer group assignment. Side effects are model to impose both an explicit cost (disutility) and an implicit benefit generate from learning. A two dimensional Bayesian model of learning incorporates side effect information along with health signals to capture patients' beliefs on the treatment effect and group assignment. These beliefs are then used to predict patient attrition in clinical trials. The model is estimated using data from two clinical trials. The first trial studies the effect of AZT to treat HIV patients. The second trial studies the effect of Topamax to reduce alcohol consumption. Patient learning is demonstrated to be slower when taking into account group ambiguity. Further, side effects are found to provide an implicit benefit. The implicit benefit of side effects is found to decrease attrition by 2% in both experiments. In addition, the model corrects for attrition bias in the estimated treatment effect. The effect of Topamax on alcoholism is found to be 3 times larger in the structural model than using standard evaluation methods.