Subsequent Prescribing of New Drugs: METHODS

6 May

Study Sample

To answer the study question, we randomly selected 2,187 physicians who were involved with at least one phase 3 clinical study as principal investigators at 2,897 clinical sites from January 1, 1995, to December 31, 1996. (Some physicians conducted studies for more than one indication.) For physician selection, we used a database that tracked data on clinical study investigator grant payments and protocol design information from more than 40 pharmaceutical companies that were conducting phase 3 clinical studies in the U.S. The 2,187 physicians were randomly selected within each indication group and represented about one third of the physicians in each of these groups. We could not use a complete census of physicians in each indication group because of the expense involved in matching individual prescribing level data from IMS Health, Inc.

We matched several characteristics of each investigator-physician in the IMS Health database of practicing physicians with those of a physician who had not been involved with the conduct of a clinical study. Although a larger control group (e.g., four times the number of cases) would have been an optimal design, we had to use a research design with a one-to-one matching ratio because of the financial constraints of obtaining these additional physician data points. 

The matching variables for the two physician groups were (1) the physicians’ geographic location; (2) their specialty, as classified by the American Medical Association (AMA); and (3) their prescribing behavior three months before their participation in the clinical study, including:

  • the total number of prescriptions written.
    • the total number of sponsored company prescriptions as a percentage of all prescriptions written by the physician.
    • the number of prescriptions written by the physician for all of the study sponsor’s drugs.
    • the number of prescriptions written by the physician in the relevant drug class.

We matched geographic areas by using U.S. Post Office Zip Codes. There were no statistically significant differences between the two sets of physicians on the matched variables.

We confirmed the clinical investigators from information that had been supplied to the Food and Drug Administration (FDA). Before an investigational new drug is allowed to be used with enrolled patients at a site, the FDA’s 1572 Form must be completed, identifying the principal investigators and the sub-investigators. All of the investigators’ names appeared in the database. None of the control physicians’ names appeared in that FDA database from 1988 to 1997.

Additional descriptive information was available from another database about each investigator and control physician, such as age, sex, board certification, and practice type.
canadian cialis online

These data came from the AMA’s Annual Survey of Practicing Physicians. These four variables were not used to match the case (investigator) and control groups.

Investigator and control physicians differed statistically on three of these additional descriptive variables—namely, sex, board certification, and practice type. Because the literature had pointed to a possible relationship between age, sex, board certification, and new drug adoption, we included these three variables, along with practice type, in subsequent multivariate analyses.