Biomedical Engineering Reference
In-Depth Information
Physician surveys are the most common method used by pharmaceutical firms to
identify opinion leaders. It is a simple and effective method. Usually the physicians
are asked questions about who influences them in a particular therapeutic area.
Sometimes they ask for more information about how they know the opinion leader
or how often and through which channel they get information from the opinion
leader. These channels typically could be by direct contact; through patient referral;
medical school, hospital or group practice colleague; medical journals; or meetings
and conferences. Survey data has the advantage of simplicity and the ability to
choose which physicians to survey. Using survey data has several disadvantages
too. The first of these is the bias of physicians due to self-reporting. Physicians tend
to be more comfortable in naming the national key opinion leaders and more promi-
nent and senior physicians than their peers. Also this kind of survey methodology
leads to an incomplete list of opinion leaders because only a few physicians fill out
the survey, and possibly nonresponse bias because the physicians filling out the
survey may have more time or may be more responsive to detailing.
Another promising avenue for data for mapping physician networks extensively
is the anonymous patient level data (APLD). There are various data vendors provid-
ing longitudinal de-identified patient level data (in order to conform to HIPAA
regulations). Surveillance Data Incorporated (SDI) is a leading provider in this area
with the largest stream of patient-centric data available, with information from more
than 11 million unique patients per year. The SDI Patient Parameters: Source of
Business series of products offers information on the patient prescriptions written
by individual physicians. SDI is currently in the process of being acquired by IMS
Health. There are several other vendors which provide APLD such as Wolters
Kluwer and Dendrite. This patient level data can be used to monitor patient move-
ment between physicians, which can be used in turn to infer physician networks as
in Bhatia and Wang ( 2011 ). There is an opportunity to use this massive/passive data
(Chritakis and Fowler 2011 ) to map the full network of physicians and to look at
more facets of the physician network rather than just indegree and outdegree.
15.3.3
Latest Models of Physician Contagion Effects
We will describe the model and main results in the three latest research papers
modeling contagion effects in physician networks in detail here. These research
papers are Nair et al. ( 2010 , henceforth NMB), Bhatia and Wang ( 2011 , henceforth
BW), and Iyenger et al. ( 2011 , henceforth IVV). We will drop most subscripts here
for ease of presentation. The reader is encouraged to read the respective research
papers for more detail.
These models differ in the data and methodology used to identify the opinion
leaders. NMB and IVV use self-reported survey data while BW use patient movement
data between physicians to identify opinion leaders. NMB use a two-stage fixed-
effects panel data linear instrumental variables regression to estimate their model.
IVV estimate a discrete-time hazard model with a logit link function estimated using
Search WWH ::




Custom Search