Pulling themes from a qualitative variable

[cbt2119]

Hi SAS Community,

I am working with a qualitative data question that asks about what a pediatric patient has been told about their diagnosis from their parent/caregiver. The question is open-ended, which means the responses differ greatly from patient-to-patient, but themes are present in the responses. I know there is software for analyzing this type of data and extracting themes, but I want to know if anyone knows a way to do it in SAS. 

For instance:

Variable: whysick

Question asked: What has the child been told about why he/she visits doctor?

Responses:

1000: "That he takes meds to be strong"  - Non-specific, with misinformation.

1001: "To help his blood cells fight sickness" - Non-specific, correct information.

1002: "For his *wrong diagnosis*" - Specific, misinformation.

1003: "For his *correct diagnosis*" - specific, correct information.

Any procedures or programs that would allow me sift through this type of information?

Thanks,

Cara

[Reeza]

Do you have access to SAS text analytics tool?

[cbt2119]

Hi,

Thank you for your comment. Apologies for not specifying earlier. I don't believe we have access to that software - only to the basic SAS toolkit 9.4, SAS Enterprise Guide, SAS IML Studio 12.3 software. Is there a way to do this on these packages?

[ballardw]

Without the Text analytice you will end up having to provide the rules yourself. How many records are you talking about? If it is few hundred then it may be just as well to have someone familiar with the topic read them and assign flags for topics.

create something like this:

Response coding sheet

respondent	answer	Specific (1=yes 0=No)	Information (1=correct/0=incorrect)	Non-responsive X to remove
1000	That he takes meds to be strong
1001	To help his blood cells fight sickness
1002	For his *wrong diagnosis*
1003	For his *correct diagnosis*

Then you have someone mark the forme and then bring the respondent and code values into SAS and match merge the results.

If this is an open-ended response to a question that may have categorical answers already available then add  column(s) to indicate those as well.

Yes it's tedious and time consuming but maybe how to extract data from the question should have been considered at the start.

[cbt2119]

Hi,

I suspected that there wasn't an intuitive way to sift without access to Text Analytics and I'd have to manually code it. But thank you all for your input! 

Cara