The Successful Launch and Diffusion of New Therapies - Innovation and Marketing in the Pharmaceutical Industry - page 191

category study acceptance at the level of an individual person (Table 7.1 ). The for-
mer models are therefore called aggregate-level models, whereas the latter are dis-
aggregate-level models.
Diffusion models capture and forecast the cumulative number of new adopters
(i.e., the cumulative number of physicians prescribing the new drug for the first
time), whereas sales (growth) models capture the amount of the new drug's biologi-
cally active ingredient that is sold in a given market or region. This distinction
between diffusion models and sales (growth) models—i.e., the distinction between
the types of data they rely on—is important. The estimation of diffusion models in
the tradition of Bass ( 1969 ) is known to create estimation bias when estimated on
sales data rather than cumulative adoption data (see Van den Bulte and Lilien 1997 ;
Van den Bulte and Stremersch 2004 ).
A different type of method aims to predict the behavior of an individual physi-
cian (towards an individual patient) regarding a new treatment. Unlike aggregate-
level models, models that are based on this approach rely on disaggregate-level
data, evaluating the acceptance process of new drugs from the perspective of the
individual physician or patient. We, here, focus on models that are estimated on
experimental or behavioral data, not attitudinal data as often gathered in surveys.
The use of such individual-level (disaggregate) models requires technical sophisti-
cation and programming skills, and they are mostly suitable for heterogeneous
social systems, for social systems with unusual network structures, and for products
involving complex adoption decisions (Muller et al. 2010 ). Accordingly, such mod-
els fit the complex and unique structure of the pharmaceutical industry very well.
Four types of models can be used to assess treatment potential on the basis of
disaggregate-level data: prescription count models, learning models, consideration
and choice models, and conjoint studies.
Prescription count models predict the number of new prescriptions or the total num-
ber of prescriptions dispensed for a drug. These models' predictions are typically based
on drug characteristics, past prescription levels, drug prescription levels of other physi-
cians, and (own and competitive) detailing levels, among others. Learning models pre-
dict the utility physicians will perceive in a treatment for a particular patient. These
models emphasize the dynamic nature of physicians' perceptions regarding the quality
of a new drug, and the important role of these dynamics in the choice process.
Physicians' perceptions are estimated according to the physicians' initial beliefs regard-
ing the drug's quality and their eventual prescription behavior. Consideration and choice
models use past observations of physicians' choice (i.e., prescription) behaviors to pre-
dict whether a physician will prescribe the new drug to a particular patient. Conjoint
analysis predicts the utility of a new drug to a physician for a particular patient and
derives the likelihood that the physician will prescribe the drug to that patient.

7.1.1 Diffusion Models

Typical models in the diffusion literature predict the dynamic process of new prod-
uct adoption. The Bass diffusion model ( 1969 ) has been used extensively to