Biomedical Engineering Reference
In-Depth Information
prescriptions of influenced physicians. NMB calculate a social multiplier for each
opinion leader by multiplying the additional revenue from the influenced physician(s)
as a result of the increase in opinion leader prescriptions (marginal effect of detailing
*
δ
* revenue generated by one prescription) by the number of physicians the opinion
leader influences in their sample. Their social multipliers range from 1.04 to 1.35.
IVV study the adoption behavior of a new drug designed to treat a chronic viral
infection. They collected physician self-reported data on the referral and discussion
networks and self-reported assessment of leadership in three cities. This enables IVV
to distinguish between self-reported leadership and referral leadership (or sociometric
leadership). IVV model the adoption of the new drug at time
t
as a hazard function.
The discrete-time hazard of adoption is modeled as below. The probability that the
new drug is adopted at time
t
given it was not adopted at time
t
− 1 is a standard normal
function of covariates
x
and parameters
β
to be estimated.
(
)
=
Py
=
1
|
y
=
0
F x
(
b
)
(15.3)
it
it
−
1
it
The covariates used by IVV are the indegree (number of physicians who nominate
a particular physician), self-reported leadership construct, social contagion measures,
detailing, and control variables (physician characteristics, category level prescrip-
tion volume, outdegree (number of nominations given), city and time dummies).
IVV find that physicians with high indegree adopt earlier. These are the opinion
leaders identified by surveys based on their peer-to-peer connections. IVV test three
types of social contagion measures based on adoption (did the opinion leader adopt
or not), current prescription (did the opinion leader prescribe in the time period
t
− 1), and prescription volume (how much did the opinion leader prescribe). They
find only volume contagion out of the three measures of social contagion to be
present. This is an interesting finding showing that adoption by a physician is depen-
dent on the prescription volume of the opinion leader, not just the fact that she has
adopted the drug and has prescribed it in the last time period. Volume contagion is
moderated by self-reported leadership and not be indegree. The correlation between
heavy users and influential users (high indegree physicians) is only moderate,
suggesting that “just focusing on heavy users will fail to leverage all potential
influential seeding points (IVV p. 196).”
Chritakis and Fowler (
2011
) argue for new techniques to identify the connections
among entire networks of physicians. BW set out to do just that by using patient
movement data between physicians. Since patient movements can be generated by
patients or by physicians, and only the latter will contain information about physician
networks, BW impose a simple framework dividing patient movements into three
groups: (1) primary care physician (PCP) to specialist and back, (2) specialist to
specialist, and (3) PCP to PCP. They suggest that the PCP to PCP movements
are purely patient-generated, PCP to specialist movements are mostly physician-
generated, and specialist to specialist movements are a mix of patient- and physician-
generated movements based on a physician survey.
BW model the contagion between pairs of physicians
i
and
j
where there is at
least one patient movement from physician
i
to physician
j
. Similar to NMB's
Search WWH ::
Custom Search