The Effects of Detailing on Prescribing Decisions under Quality Uncertainty

1
The Effects of Detailing on Prescribing Decisions
under Quality Uncertainty

• Andrew Ching
• Masakazu Ishihara
• Rotman School of Management
• University of Toronto

2
Advertising in the Pharmaceutical Industry

• In 2000, detailing costs 4.9 billion dollars
  journal advertising costs 0.25 billion dollars
  direct-to-consumer (DTC) advertising costs 2.5
  billion dollars.
• Roles of detailing informative and persuasive.
• Uncertainty about product attributes in the early
  stage of drug lifecycle (Lasser et al. (2002)).
• Understanding the relative importance of
  informative and persuasive advertising is
  important from both managerial and public policy
  viewpoint.

3
Observations

• Slow diffusion of new drugs suggests learning
  is important.
• Detailing expenditures are higher than revenues
  during the initial stage of the drug lifecycle
  suggests detailing may have long-lived effect.
• Relative detailing declines as the demand for a
  product decreases consistent with advertising
  is informative.

4
Objectives of Research

• Develop an empirical dynamic structural demand
  model that incorporates learning, persuasive and
  informative advertising, and its long-lived
  effect.
• Measure the importance of learning.
• Disentangle the effect of persuasive advertising
  and informative advertising.

5
Literature Review

• Consumer Learning Erdem and Keane (1996),
  Crawford and Shum (2005), Ching (2003).
• Persuasive vs Informative Advertising Leffler
  (1981), Hurwitz and Caves (1988), Berndt, Bui,
  Lucking-Reiley and Urban (1997), Ackerberg (2001)
  (2002).
• New Informative Advertising Model Goeree (2005).
• Pharmaceutical Marketing Mukherji (2003),
  Narayanan, Manchanda, Chintagunta (2005).

6
Model

• Agents physicians, firms, and a public agency.
• Each firm has one product.
• There is one outside alternative (0).
• The public agency uses patients experiences with
  the drug to update the public information set
  about the quality of each drug (in a bayesian
  manner).
• Firms have access to the public information set.
• But not every physician accesses the public
  information set.

7

• If a physician is influenced by firm js
  detailing, he has the most current information
  about drug j (informative effect). Otherwise,
  his information set will simply be the initial
  prior.
• Detailing could also change physicians
  preferences (persuasive effect).
• Physicians decide which drug to prescribe based
  on his information set.
• After patients have taken the prescribed drugs,
  some of them reveal their noisy experience
  signals.
• There is a continuum of physicians with measure
  one. Each physician sees m patients randomly.
• Assume there are only two products.

8
Bayesian updating of the public information set

• Experience variability qijt qj dijt,
• where dijt N(0, s2d).
• Initial prior for qj N(qj, s2).
• Expected quality
• EqjI(t1) EqjI(t) ?j(t)(qjt
  EqjI(t)),
• where qjt is the sample mean of experiences
  signals revealed for product j in period t.
• Perception variance
• s2j(t1) 1 / (1/s2j(t) ?njt/s2d),
• where njt is the quantity sold for drug j in
  time t
• 0lt?lt1, is a scaling factor.
• Ij(t) (qj?,njt), ?1,,t I(t) (I1(t),
  I2(t)).

9
Long-lived Effect of Detailing

• Let Mj(t) be the measure of physicians who have
  the most updated information about product j at
  time t (i.e., Ij(t)).
• Let asjt be the advertising stock, and ?M be the
  depreciation rate.
• asjt (1- ?M)asjt-1 Ajt.
• Mjt f(asjt, as-jt).
• E.g., let D ß0 ß1as1ß2as1as2ß3as2,
• M1 exp(D)/(1exp(D)).

10
Consumer heterogeneity with endogenous weights

• Measure of physicians with current information
  about both products M1M2 (I1(t), I2(t)).
• Measure of physicians with current information
  about only one product Mj(1-Mk), for j?k
  (Ij(t), Ik(0)).
• Measure of physicians who do not have current
  information at all (1-M1)(1-M2) (I1(0),
  I2(0)).
• Four types of physicians.

11
Physicians Choice

• Patient is utility of consuming drug j
• uijt aj - exp(-rqijt) - pppjt eijt.
• If physician h is influenced by detailing of drug
  j, his utility of choosing drug j for patient i
  will be
• EUhijIh(t) EuijtIh(t) ?asjtG
• aj - exp(-rEqjI(t)-1/2r2(s2ds2(t)) - pppjt
  ?asjtG eijt,
• where ? captures the effect of persuasive
  advertising.
• If physician h does not have current information
  about drug j,
• EUhijIh(t) EuijtI(0)
• aj - exp(-rqj-1/2r2(s2d s2) - pppjt eijt.

12
Identification

• The stock of well-informed physicians generates
  more flexible diffusion path.
• The rate of increase in empirical variances of
  the consumption path helps identifying the rate
  of building up Mj(t) (i.e., f(asj, as-j ß)).

13
Estimation Strategy

• Follow Ching (2003)
• Endogeneity problem Ajt and (EqjI(t))j1, 2
  may be correlated.
• Let sjt (EqjI(t), sj(t), Mj(t)).
• Ajt aj(sjt, s-jt)?jt, where ?jt is the
  prediction error.
• log(Ajt) log(aj(sjt, s-jt)) log(?jt).
• Use a polynomial series estimator to approximate
  log(aj(.)).
• Jointly estimate this pseudo-advertising policy
  function with the demand model.
• Need to integrate out the unobserved state
  variables simulated maximum likelihood.

14
Data

• Monthly Canadian data on detail advertising,
  revenue and number of prescriptions from March 93
  to Feb 99 for ACE-inhibitor with diuretic.
• Why Canada?
• Subject to price regulation Patented Medicine
  Prices Review Board.
• Why ACE-inhibitor with diuretic?
• No Direct-to-consumer advertising.
• Only two dominant drugs (Vaseretic and
  Zestoretic).
• Treat high-blood pressure patients/physicians
  are likely to be risk averse.

15
Estimated Preference Parameters
16
Pseudo-advertising policy functions
17
(No Transcript)
18
(No Transcript)
19
(No Transcript)
20
(No Transcript)
21
(No Transcript)
22
Conclusion

• Introduce a new way to model informative
  advertising.
• Advertising is mainly informative.
• Persuasive effect of advertising is positive and
  statistically significant, but very small.
• Managerial implications If you know your drugs
  are not as good as your rivals, save your
  detailing efforts.
• What else can this model address? Some policy
  experiments.
• Completely remove advertising.
• Solve for the socially optimal advertising policy
  function.

23
(No Transcript)
24
(No Transcript)
25
(No Transcript)
26
Actual functional form used

• Mjt exp(ß0 ß1as1)/(1exp(ß0 ß1as1)).
• log Ajt ?j0 ?j1(1-Mjt)(Eqjt-Eq-jt)
• ?j2(1-Mjt)(Eqjt-Eqj0) ?j3 IVjt
  ?jt.