The Successful Launch and Diffusion of New Therapies - Innovation and Marketing in the Pharmaceutical Industry - page 197

drugs. As an example, given that Nexium was a clear follow-on drug of (Pri)Losec,
one could have used physician response parameters of time, detailing, etc. on histori-
cal data on (Pri)Losec to estimate physician response to Nexium.

7.1.4 Learning Models

Learning models in particular exploit the uncertainty physicians perceive regarding
the quality of a new pharmaceutical drug. Physicians reduce their uncertainty about
the quality of a new drug over time on the basis of feedback from patients as well as
the firm's marketing efforts. Several studies have specified models to capture physi-
cians' learning with regard to new pharmaceutical drugs as these drugs diffuse
into the market (Camacho et al. 2011 ; Coscelli and Shum 2004 ; Crawford and Shum
2005 ; Narayanan et al. 2005 ; Narayanan and Manchanda 2009 ). Coscelli and
Shum ( 2004 ) suggest that the slow diffusion time of a new pharmaceutical drug in
an existing product category is due to slow learning by risk-averse physicians. The
only source of information in their model is patient feedback. Narayanan et al.
( 2005 ) investigated how the role of marketing communication for new products
changes over time in the presence of learning. They specified a learning model in
which marketing communication by firms as well as physicians' accumulated usage
experience contribute to physicians' learning about a new drug. Narayanan et al.
( 2005 ) found that marketing efforts by pharmaceutical companies—i.e., detailing—
have a primarily indirect effect (i.e., learning) in the early stages of the new drug's
life cycle and a primarily direct (i.e., persuasive) effect at later stages. Narayanan
and Manchanda ( 2009 ) find significant heterogeneity across physicians in learning
rates and show that there are asymmetries in the evolution of physicians' respon-
siveness to detailing over time. Chintagunta et al. ( 2009 ) suggest that the informa-
tion physicians retrieve from patients who were prescribed a new drug is subsequently
used in the physicians' learning process to update their beliefs regarding both the
drug's overall quality and a patient's idiosyncratic match with the drug. Their results
suggest that physicians are influenced by many sources of information, including
patient satisfaction, Medline articles, reports in the mass media and direct-to-
consumer advertising (DTCA).
Camacho et al. ( 2011 ) developed a generalized quasi-Bayesian learning model
that allows for decision-making biases that occur in physician decision making. In
essence, they argue that physicians can retrieve some pieces of information from
memory more easily than they can retrieve others. They show that physicians' belief
updating, and thus the speed of their new drug adoption process, is strongly influ-
enced by the salience of patient feedback. They find that negative patient feed-
back—feedback from patients whom the physician needed to switch to a different
drug—receives 7-10 times more weight than positive feedback does in the physi-
cian's quality belief formation. The authors show that this effect greatly reduces the
speed of diffusion of the new drug.
Firms can use learning models to gain knowledge about patterns in physician
adoption of new drugs, and they can subsequently take such patterns into account