Biomedical Engineering Reference
In-Depth Information
Recently, panel model at the physician level has become a standard analytical
practice for most pharmaceutical brands where physician level sales data are avail-
able. Pharmaceutical companies started to adopt this approach in late 1990s, just a
few years after physician level data (such as IMS Xponent) became available. Every
large pharmaceutical company by now has developed an enterprise level modeling
system that can produce promotion response analysis at brand and segment level on
a regular basis to support call planning, physician targeting, and resource allocation.
Typically, physicians' NWRx, TWRx, or share measures from IMS or NDC are
used as a dependent or response variable, and independent variables include number
of calls, samples left to physicians, meetings and events, and other promotion and
marketing variables such as DTC. The modeling methods are drawn from econo-
metrics and marketing science literature, including dynamic panel models, general-
ized linear models, mixed models (random effect models), count models (Poisson
and NBD), and hierarchic Bayesian models.
Most of the panel models as described above utilize internal promotion data and
prescription records from data vendor such as IMS. There are some well-known
limitations of using this internal promotion data. First, the promotion information is
subject to self-reporting errors (which may be caused by the company's incentive
structure). Second, no competitive promotion measures exist at the physician level.
Thus, the estimated promotion effectiveness can be signifi cantly biased especially in
a competitive market. Third, sample drop measure to any doctor may not be an accu-
rate measure of sample promotion due to sample sharing within a group practice,
sample signature practice, and the existence of other sample distribution channels.
In the past 5 years, panel modeling approach using patient transactional level
data, instead of monthly physician level data, has become a popular way to evaluate
the effectiveness of detailing, message, and promotion tactics. There are several
advantages of such an approach. First, it models stimuli and response relationship at
a more granular level; second, it considers timing between promotion and response
explicitly; third, it reduces multi-collinearity and data aggregation; and lastly, more
transactional level factors can be considered in the model. As nonpersonal or alter-
native channel promotion targeting of patients has gained a more important status in
recent years, many pharmaceutical companies also used this platform to evaluate
the effectiveness of nonpersonal promotion such as direct mail, voucher, and co-pay
card.
17.5.3
Sample Allocation Model
The ultimate question for every sales team is how many samples should be distrib-
uted to a targeted physician so that paid prescriptions can be maximized in the long
run. One approach adopted in the industry is the method of “one-period inventory
problem,” also called the “newsvendor problem.” The objective of this approach is
to fi nd the optimal number of samples to distribute that will maximize the expected
profi t. The expected profi t is the sum of expected profi t under the uncertainty of
Search WWH ::
Custom Search