Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

Conjoint design and estimation for Pharma research study

I am helping my client in a pharma research study focussed on patients with Diabetes - the objective of the study is to understand what role do the patient profiles play in prescribing them a given drug and what attributes of patient profile are most important in driving prescription of a certain drug.  I have 3-4 type of drugs, and 5 attributes in patient profile (Age, presence of Cardiovascular disease and 3 more) with 3-4 levels each.  At the end of conjoint analysis I want to be able to identify that for a given patient profile, what is the distribution of prescription across those 3-4 drugs (sum to 100%).   Could you please shed some light on how should I create the design - and do the utility estimation?  Thanks in advance.
asked Jul 24, 2017 by kshitijkumarsingh (160 points)

1 Answer

0 votes
I know I've already replied to you privately, but for purposes of the forum I'll reply here as well.  

This isn't so much a conjoint experiment (where we have respondents make choices among experimentally designed profiles) as it is what I'd call a Situational Choice Experiment (SCE).  In SCE we have an experimentally designed situation (patient in your case) and the choice is among a set of fixed alternatives (in your case drug therapies).  

You can design the experiment in any software that you could use to make a single set of conjoint profiles - so you could use our CVA designer, our CBC designer where you request a single concept vs a "none," or you could use SAS, SPSS or Ngene to make your design.  

Analysis uses a different logit model than traditional choice models:  rather than the conditional logit model it uses an unconditional logit model.  You an run this model very easily in our MBC software or you can do a bit of manual manipulation to your design file and then trick our standard logit software (CBC/HB or Latent Class) to estimate utilities for you.  You can also run the unconditional model in SAS, SPSS, SYSTAT or NLOGIT.

You simulate choices straightforwardly using the logit choice rule.  

The model produces a separate vector of utilities for each alternative respondents might choose, so some folks are confused by this non-typical output.  We have a chapter in our upcoming book "Becoming an Expert in Choice Modeling" devoted to this model and we also cover it in our "Becoming an Expert" tutorial, which we'll teach next in Barcelona in September.
answered Jul 24, 2017 by Keith Chrzan Platinum Sawtooth Software, Inc. (95,675 points)