OHDSI Home | Forums | Wiki | Github

Identification of home medication from in-hospital medication administration records

Researchers
#1

Hi,

I am currently conducting a project to investigate the effect of two different classes of medication on a specific comorbidity. I am using the MIMIC-IV data, which I have converted to the OMOP CDM. I have a quick question regarding this.

Due to the nature of the dataset, the drug_era defined in the dataset represents the duration of medication administered during the hospital stay. As a result, the average length for drug_era_end_date minus drug_era_start_date is around 6.5 days. I am facing difficulty in identifying whether the medications are “home medications” or not.

Has anyone encountered a similar challenge in distinguishing patients’ home medications from in-hospital medication history? If so, could you share your approach?

One possible method I am considering is aligning drug_era based on patient_id and determining if a medication was consistently administered in-hospital, then setting an arbitrary standard to classify it as a home medication. However, this approach feels too arbitrary, and I would appreciate input from the community on better methods or best practices.

Please let me know if there is any additional information I can provide.

Thank you!

#2

Since MIMIC-IV represents ICU patients, I don’t see how (would be interested if possible) to even know what post discharge behavior regarding drug consumption. In a previous project where I had pharmacy data - root prescription and refills - it still isn’t 100% clear what the patient does at home regarding compliance. This is a tough question to ask.

#3

I see. Clearly, trying to infer post-discharge behaviors like consumption of home medication from MIMIC-IV does not seem plausible given the dataset’s focus on ICU patients and in-hospital medication administration. My bad for trying to stay in a framework I’m comfortable with instead of finding something new :sweat_smile::sweat_smile:. I will explore alternative approaches to address the problem. Thank you!

t