choice-metrics.com

Dear Ngene users,

I am going to design a CE to study preferences for treatment. In my design, I have 3 attributes related to patient characteristics (age, quality of life and life expectancy), and 2 attributes related to treatment characteristics (effect of treatment and risk).

I am going to reduce the choice task complexity for respondents and provide some overlap between attributes. I have read the Ngene manual about explicit partial profiles, but could find find a solution to this question:

Can I have a design where in some tasks the the patient characteristics are having overlap and attributes in treatment characteristics are changing and in some other tasks it is treatment characteristics which have overlap?

I appreciate your help.

Partial profile designs create overlap across multiple characteristics. For example, in choice task 1 it creates overlap across treatment attributes, and in choice task 2 it creates overlap across patient characteristics. This overlap needs to vary from choice task to choice task as otherwise it is not possible to estimate all parameters. For example, of treatment attributes are always overlapping, then their associated parameters cannot be estimated. A partial profile design optimises the design in such a way that all parameters can be efficiently estimated by creating different overlap situations.

When you create an external candidate set for the partial profile design as explained in the manual, you can create overlap across attributes of your choice. So you can include choice tasks that have overlapping treatment characteristics and you can have choice tasks with overlapping patient characteristics. Ngene will then select the best choice tasks from this candidate set to include in the design. It is not possible to ask Ngene to include exactly X choice tasks where treatment attributes are overlapping and exactly Y choice tasks where patient attributes are overlapping.

Does this answer your question?

One comment. You mention that you want to include patient as an attribute, characterised by age, QoL and life expectancy. I can imagine a choice task like this:

Consider a patient of age 70. Which treatment would you recommend:
Option A: Effect of treatment = ..., Risk = ..., QoL = ..., Life expectancy = ...
Option B: Effect of treatment = ..., Risk = ..., QoL = ..., Life expectancy = ...

The age of the patient above is a scenario variable. It is not appropriate to put the age of a patient inside the options since this is constant across both options. When including scenario variables in your utility function, you will need to create interaction effects if you are using unlabelled alternatives as above, or you can add age as a main effect in J-1 labelled alternatives, where J is the number of alternatives. For exampe, if you have an optout alternative, you could add age in both Option A and B and omit it in the optout. The Ngene manual explains how to use scenario variables, but essentially you would use something like b1 * age[age] in your utility functions, which ensures that age is the same across the alternatives.

Michiel

Dear Michiel,

Thanks for your help. I will try creating an external candidate set for the partial profile design.

With regard to your comment, age and the other two patient characteristics (QoL if un-treated and Life expectancy if un-treated) are not scenario variables. These are attributes that explaine patient characteristics and are changing in each choice task.

Best regards,
Nasrin

Happy to hear that age is not a scenario variable, then all should be fine.

Michiel