Evaluating the Impact of Treatment Effectiveness and Side Effects in Prescribing Choices - Innovation and Marketing in the Pharmaceutical Industry - page 182

drug, physicians and patients may have large uncertainty in these treatment
outcomes. It is of great interest from the policy and managerial perspective to sepa-
rately identify how physicians and patients evaluate treatment effectiveness and side
effects of different drugs and how they resolve their uncertainties using information
from various marketing channels.
To achieve this objective, researchers have used observed treatment outcomes in
addition to prescription choice data. Crawford and Shum ( 2005 ) is a good example.
In the paper, the authors proposed a dynamic match model of demand under uncer-
tainty in which patients learn from experience about two distinctive treatment
effects of alternative drugs: the symptomatic effect and curative effect. A drug's
symptomatic effect impacts a patient's per-period utility directly via symptom
relief and/or side effect, while its curative effect impacts a patient's probability of
recovery. The authors used drug choices from patients to identify the parameters
related to symptomatic effect and used observed treatment lengths conditional on
different sequences of drug choices to identify the parameters related to curative
effect. They estimated the model using a patient-level panel data in the antiulcer
drug market. They found that existing drugs in the market are ranked differently
along these two dimensions, suggesting a trade-off has to be made when patients
decide on their treatments. They also found that learning from ones' own prescrip-
tion experience can help patients and their doctors overcome the costs of uncer-
tainty in this market: the uncertainty in each dimension was sharply reduced even
after a single prescription.
Treatment outcome data may be diffi cult to collect. Furthermore, for those cate-
gories that only relieve symptoms (e.g., antidepressants or erectile dysfunction (ED)
drugs), objective measures of treatment outcomes (e.g., blood or cholesterol count)
are not observed. A different identifi cation strategy, therefore, is required to sepa-
rately identify the impact of treatment effectiveness and side effects. In Chan et al.
( 2013 ), the authors rely on an additional data source, self-reported reasons for
switching drugs, as well as the observed treatment choice for each physician-patient
pair to achieve such research objective. While the overall quality evaluation of drugs
from physicians and patients can be inferred from treatment choices, self-reported
reasons for switching help to identify effectiveness and side effects as well as the
heterogeneity of their impacts. A drug that accounts for greater proportion of those
switching out due to treatment ineffectiveness (side effects) implies that, compared
with other drugs, more patients fi nd this drug less effective (with severe side effects)
than expected. Since the data also reports to which other drugs these patients switch,
further inference can be made that drug switched into is more effective (with less
severe side effects) for that particular patient. Hence the potential correlations of
both treatment effectiveness and side effects across drugs can be estimated from the
data. Self-reported consumer survey data was proposed by Manski ( 2004 ) to help
understand the extent of consumer uncertainty. Berry et al. ( 2004 ) used the data of
consumers' self-reported secondary choice in the automobile market to identify the
correlation of consumer preferences for product attributes. The approach used in
Chan et al. ( 2013 ) is similar to theirs. The authors developed a structural model of a
physician maximizing a patient-physician joint utility function under uncertainty