Dynamic Sequencing of Drug Treatments for ADHD patients with Medicaid coverage

Almost 10% of children aged 4-17 were diagnosed with attention deficit hyperactivity disorder (ADHD) in the U.S. in 2007. While many believe that ADHD drugs are overprescribed, little is known about the existing prescribing practices, physician learning processes, and relative efficacies of various ADHD treatment strategies.

The evidence suggests that children diagnosed with ADHD face significant uncertainty regarding efficacy and severity of adverse effects of ADHD medications. Almost half of these children switch therapies during the first six months of treatment. This suggests a considerable amount of experimentation by doctors. I extend Crawford and Shum (2005)'s model to explore the effect of treatment interruptions (drug holidays) in addition to the effects of various drug therapies. Using South Carolina Medicaid claims data for 2003-2012, I estimate a dynamic model of demand for ADHD drugs under uncertainty. Uncertainty comes from two sources: little evidence on newly introduced ADHD treatments and uncertainty about the response to treatment of a particular patient. In the model, highly heterogeneous patients learn about the efficacy of available treatments through experimenting.

The baseline model estimates for a subsample of patients without accounting for drug holidays show that 40% of patients have a relatively mild form of ADHD, but their recovery probability at the start of the therapy is still low. The parameter estimates of both symptomatic and curative match values suggest that there is substantial heterogeneity in match values across patients. I also find that available treatments are horizontally differentiated by curative but not symptomatic properties.

I will evaluate the effect of interruptions in treatment on the overall treatment cost and disease duration, accounting for patient heterogeneity in response to treatment for ADHD. I will explore the potential to develop better guidelines that can improve the quality of drug-patient matches and patients outcomes.