I was really hoping the presence of an E/M code would be the key to identifying patient-provider interactions in our EHR encounter data. However, I found some ambiguity in using this method. If anyone can shed some light on how I can distinguish patient-provider interactions from the transactional EHR encounters, I would love to talk to you and/or host you on the Healthcare Systems call.
I compared the presence of an E/M code in the combined clinical event table, a massive, staging table where all source clinical events are combined before moving to their home in the CDM, with the encounters in our Visit Occurrence & Visit Detail tables. A few things to note about our source data. First, we are currently bringing in all source encounters/visits unless the visit is in the future, was canceled or never occurred (no show, left before being seen, etc.). Second, our OMOP CDM consists of data from the UCHealth system and the CU Medicine physicians group among other sources. The other sources aren’t relevant for this topic. Third, our EHR data comes from Epic’s Caboodle data warehouse.
The following are my findings:
-
Some visits were billed/coded by the physician’s group. The CU Medicine physicians group represents some, but not all of the physicians who see patients in our healthcare system. So, we might be missing E/M codes billed by other physicians.
-
Not all inpatient hospitalization records have an associated CPT4 E/M code in our source data. A quick check of these IP records show the person was in the hospital for days, weeks, and even months. So, they definitely interacted with a clinician.
-
I dug into Long Covid Clinic OP visits. I found some Telehealth (video visit and telephone visit) visits didn’t have E/M codes. Upon further examination, I took 3 random visit records from this group and all 3 had progress notes including care plan indicating the provider saw the patient.