Data analysis on drug data

[CathyVI]

Hello,

I am asked to do both summary analysis and exploratory analysis on a data. I am new to sas but so eager to learn. I need some help understanding how to create codes and analyze for the following variables.

Summary Analysis (Encounter Level)

• Number of encounters that have any medication data by:
  • Setting
  • Source
  • Hospital Identifier
  • Code Systems
  • Verified Code systems
• Identify encounters that are missing medication data by:
  • Setting
  • Source
  • Hospital Identifier

Exploratory Analysis (Drug Level)

• Number of medications with codes
• Number of medications with labels
• Number of medication with codes but no labels
• Number of medications with labels but no codes
• Number of medications with codes and labels
• Number of medication with neither codes and labels

[Reeza]

What's your timeline for this analysis? I would say take the first two days/first week and complete the free SAS e course or the Coursera course keeping these objectives in mind. All of these can be done with PROC FREQ and a data step. PROC FREQ is pretty quick to learn, the data step I'm still working on 15 years later!

[Reeza]

And we can't really answer your question without any context of your data. How the data is formatted and stored and the exact rules of 'what does no label' or 'what does no code' mean are very important. If you need code help, then you need to provide example data and show what you want as the output. You likely can't post real data, so make fake data as close to your real data as possible.

[CathyVI]

Thank you for your help, I think you gave me a good lead into where i should start.

As for your question on meaning of code and label, the label is the drug name while the drug code represent the event of activity taking place.

See the following deliverables

MEDICATION_­CODE: The code representing the event or activity taking place, such as the code for ibuprofen.

MEDICATION_­NAME: The text name of the medication or immunization involved in the encounter. The text is the standard label of the codes where a code is present or the text provided by a custom extract when a code has not been provided.

[Reeza]

I've worked with health data quite a bit 🙂

What I mean by no label, does that mean the text 'no label', is it coded as 'NA' or 'Missing' or is it blank? Various systems have their own idiosyncrasies and we won't know your specific data. I can make an assumption that no labels means the field is empty but I've seen systems where that's not true as well, so it's easier to ask or be very specific with questions/requirements. Coding is details.

[CathyVI]

Thank you Reeza,

You are right the 'no label' is blank in the data. This is a very large data and its my first time working on large real data. All your recommendations  are very helpful, please. Also, do you recommend subseting the data to only variables i will be working on to minimize error as well as processing-time for the sas machine.

[Reeza]

What's your definition of big data here?

[CathyVI]

This data have over four million records with almost 90 variables. 

[Reeza]

Ok, that's big but not unmanageable by any means. Standard approaches will work here. You should be able to handle this data without any sub setting at all. You can drop variables if you find you're running into speed issues but you really shouldn't be unless you have 2GB of RAM and a really under-powered CPU.

[CathyVI]

Okay your information is helpful. Moving forward, how would you interpret this task. It's one of the final steps i am asked to analyze but am having difficulty simplifying into sas or analytical term.

Find;

Sample Dataset (Medication Level/Including all available data fields)

• 50 RxNorm observations
• 50 observations representative of other sources

[Reeza]

Random sampling of data rather than working with whole data. Look at PROC SURVEYSELECT.