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

There is a substantial literature demonstrating that patients receive a very different quantity or type of medical care depending on where they live and the physician they choose. Most studies conjecture that the variation in treatment styles is due to differences in physicians' perceptions regarding the optimal way to treat patients. However, little is known regarding how physicians form their treatment styles and how these styles change when a new treatment method becomes available. The objective of this research project is to use detailed patient-level data from the SEER-Medicare data set to examine how oncologists learn about the efficacy and side effects of new biotech and pharmaceutical drugs, and to help explain why those drugs are adopted at different rates in different communities.

Specifically, we will use regression analysis to estimate how important the following factors are in determining whether a patient with a specific type of cancer (e.g., breast) receives a particular drug or drug regimen: 1) the attributes of the drug (e.g., median months of survival among patients in the phase 3 trial, percentage of patients in phase 3 trials who experience a grade 3 or grade 4 side effect) relative to other drugs approved for the patient's type of cancer; 2) patient characteristics (e.g., age, gender, stage of cancer at diagnosis); 3) physician characteristics (e.g., gender, specialty, number of years of post-residency experience); 4) the health outcomes of a physician's patients who have recently received the same drug; and 5) the health outcomes of patients in a physician's peer group, such as physicians who admit patients to the same hospital or who attended the same residency program. The relative importance of each of these channels will be assessed through the estimation of a dynamic programming model of physician learning, and counterfactual simulations of the outcomes of interest in the absence of a particular channel for learning.

This analysis will allow us to see the extent to which physicians base treatment decisions on the results of randomized controlled phase 3 trials, early clinical results among their own patients, and clinical results from a broader practice setting. For example, the first factor above is important whereas the fourth and fifth factors are not, this would indicate that physicians learn largely from the randomized controlled phase 3 trials. If the first, fourth, and fifth factors are all important, this would indicate that a physician's own experience supplements the randomized trials and might explain why certain communities adopt new drugs at different rates. By looking at interactions of the above categories of variables, we will be able to examine which type of physicians switch all/most of their patients to new drugs, which type of physicians target new drugs to specific types of patients, and which type of physicians avoid the new product all together (perhaps after experimenting initially).