Biomedical Engineering Reference
In-Depth Information
They found that patients were more likely to choose drugs that allowed them to
participate in daily activities and sports, with minimal symptoms, fewer side effects,
and at a less cost. Based on the model estimates, simulations can be conducted to
quantify and compare the impact of various treatment attributes. For example, the
authors found that doubling the cost of the hypothetical drug from $50 to $100
would lead to a 2.6 % increase in demand for the current drug.
The above studies have modifi ed the experiment design (e.g., crossover in trials)
to some extent that is quite uncommon in industry practice. In a standard clinical
trials setup, researchers have used information on participant compliance together
with treatment outcomes to evaluate participants' preferences for treatment effec-
tiveness vs. less severe side effects. Substantial attrition or noncompliance (i.e.,
drop out from the trial before the end of the study) from participants has been docu-
mented as a common problem that plagued many clinical trials (e.g., Efron and
Feldman
1991
). Clinical trial literature has traditionally treated attrition as a sample
selection bias issue that can be addressed by various statistical methods (e.g.,
Frangakis and Rubin
1999
). However, given that it is a choice made by participants,
attrition may refl ect their evaluation of treatments (Lamiraud and Geoffard
2007
).
For example, if we fi nd that the attrition rate among participants in a treatment that
is more effective in reducing symptom but also has more severe side effects is higher
than the other drugs, this may indicate that side effects are more important than
effectiveness in participants' evaluation. Therefore, attrition may reveal information
regarding participants' preferences.
Built on this assumption, Chan and Hamilton (
2006
) constructed a structural
economic model in which individual participants make utility maximizing decisions
concerning dropout/compliance in a randomized clinical trial. They specifi ed utility
as a function of both “publicly observed” outcomes (i.e., the measured health status
in the experiment) and side effects privately observed by participants. They also
assumed that participants in the experiment have uncertainty regarding the treat-
ment effectiveness and side effects. They would acquire information through their
participation in the experiment to learn about these attributes and update their
beliefs according to the Bayesian rule. The authors applied the model to analyze
data from the AIDS randomized clinical trial ACTG 175 (Hammer et al.
1996
).
ACTG 175 was a large scale randomized double-blind clinical trial designed to
evaluate the effectiveness of four alternative therapies in treating HIV patients with
CD4 cell counts of between 200 and 500 mm
3
, including 600 mg of zidovudine
(AZT), 400 mg of didanosine (ddI), 600 mg of zidovudine plus 400 mg of didano-
sine (AZT + ddI), and 600 mg of zidovudine plus 2.25 mg of zalcitabine (AZT + ddC).
The “publicly observed” treatment outcomes in the data were participants' CD4
counts
5
measured at the outset of the trial and then at weeks 8, 20, 32, …, 104 of the
trial. Substantial attrition was observed in the ACTG 175 trial, as roughly half of
participants dropped out from the trials by the end of the second year.
5
CD4 count is a marker for the status of an individual's immune system, which has been widely
used to measure the progression of AIDS in the patient.
Search WWH ::
Custom Search