As Eprescribing Grows, Medication History as-a-Service Plays Crucial Role

New services are filling in where HIEs and national EHR exchange data fall short.

Eprescribing continues to grow, and with it, healthcare IT organizations face new challenges and opportunities to share the information associated with it.

Eprescribing provider DrFirst recently announced that its iPrescribe service continues to grow steadily. In 2021, nearly 6.5 million prescriptions were processed through iPrescribe, representing more than 100% growth year-over-year. Its user base grew to more than 26,000 prescribers, an 80% jump over the previous year.

Two years ago, DrFirst hired Colin Banas, MD, as its chief medical officer. Banas had previously served as the chief medical information officer of VCU Health from 2010 to 2019. HealthLeaders spoke to Dr. Banas about the growth of eprescribing and related issues. This interview has been lightly edited for clarity and brevity.

HealthLeaders: This notion of medication history as-a-service sounds like it's something that may or may not be closely tied to the rest of the electronic health record.

Colin Banas, MD: A lot of traditional EMRs give their partners the opportunity to make a connection to an external medication history feed, and it's usually at a cost. With the big EMRs like Cerner, Epic, Allscripts, and Meditech, you can subscribe to a medication history feed. If your facility has it, and you've trained your users, the patient presents to the ER, and the clinician can click a button, which queries a database and brings back data for what the patient has gotten filled in the last three months, six months, up to a year.

The traditional medication history feed is from a company called Surescripts. And a lot of their data is adjudicated claims data. They're connected to pharmacy benefit managers. They know that I used insurance to go pick up my Lisinopril or whatnot. And they have connections to some of the big-box retail pharmacies. DrFirst has partnered with Surescripts, so we have that data feed, but also we've been an eprescribing company for 22 years. Over the course of two decades, we've made a lot of relationships, whether it's with health information exchanges (HIE), our own eprescribing data, and then, we also have connections to a lot of the smaller pharmacies, because we've made connectivity to their dispensing software systems. So right there, you've got more data coming in. 

My former role was as a hospitalist and CMIO at an academic health system, and I remember how great that med history button was, because before that we didn't have anything. It was the interview or bust, or the patient brings in a giant bag of pills and you sort it out. And while that button was great, we learned quickly that it wasn't perfect in two ways.

One, some medications would not be on the list. Second, the instruction information (i.e., take one pill by mouth daily) can often be lacking in the data feed as well. The reason is, if it's coming from a PBM, they only care about the [national drug code (NDC)], or the drug, and the number of pills that went out the door, so the instruction data gets stripped, or never gets into the feed. So there's another source of potential error, where if you don't know the instructions, you don't know how the patient was supposed to be taking it. It can get dicey in terms of clinical care.

HL: How is medication history data updated? How do you correct errors?

Banas: We have that core feed that I described before, plus the local pharmacies that we're able to make connections with. We also have some connections, depending on geography, to health information exchanges.

We have some ability to incorporate patient-reported medications on one of our mobile applications. So if you curate a med list on the mobile app, it'll get into the feed as well. A lot of local pharmacy data is getting batch uploaded at midnight, every night. One of the first things we have to do is clean it up. We have to remove the duplicates.

Then the real game changer, and my aha moment as a CMIO and then joining DrFirst as the CMO, was this technology. Anywhere in the DrFirst lexicon you see the prefix "Smart," it means that our patented technology, the AI, is cleaning up the data, de-duplicating it, but also turning any free text back into structure, and inferring any of the missing pieces. That's how you get a more complete and safer medication history service.

HL: Do interoperability platforms such as CommonWell, Carequality, and eHealth Exchange deliver medication history as-a-service or not?

Banas: The interoperability landscape is tricky. They don't necessarily serve up med history as-a-service per se as much as they serve as a repository for those [continuity of care documents (CCD)] that get automatically transmitted after certain care events. A lot of that data can come in and allow the user to import. But again, that's only as good as what went out. It doesn't necessarily reflect what got picked up at the pharmacy or what didn't get picked up. It has the potential to be a corollary, but I wouldn't depend on it as a core medication history service. There are too many holes.

HL: Are you doing anything to promote improvement on that front? Is there something the industry as a whole can do to make that better?

Banas: One of the problems with interoperability is this notion of semantic interoperability. We're good at moving chunks of data from A to B. But we're not so good at the receiving system consum[ing] it without a lot of manual intervention. There's a couple of problems in the industry.

One, there's five different drug compendia out there. There's First Databank. There's Multum. And so the sending system might be using FDB, but the receiving system is expecting Multum. For the most part, it matches. And when it doesn't, it becomes free text. You lose all the interaction, checking on drug-drug, drug-allergy interactions. And a human has to fix it. Smart can sit in between the compendia and make sure that the NDCs are going to match no matter what.

The second thing Smart does is, we know how you've set up your nomenclature. I'll give you an example. There's a paper out of University of Michigan. They studied the most common sig [instruction] in the industry: take one tablet by mouth daily. Makes sense. They found 835 permutations of that instruction set. Some people have "tablets." Some people have "PO." Some people have "oral." We learn how you set up all of your various nomenclatures. We intercept the message, we restructure how the system expects it, and then we land it.

You're saving massive clicks and keystrokes. You're making the pharm tech, the nurse, and the clinician much more efficient. But more importantly, it's just safer. We have two papers in peer review right now showing the impact of our technology on reduction in adverse drug events and events reaching patients. But you can imagine putting that technology in front of an HIE and improving that portion of interoperability as well.

HL: According to the NIH, more than 40% of medication errors are believed to result from inadequate reconciliation in handoffs during admission, transfer, and discharge of patients. How does your platform help reduce those errors?

Banas: Part of the software also shows you the gaps that we're filling in monthly in terms of patients you were previously missing, or drugs that you were previously missing; we were able to add to the dataset. One of the reports is about dangerous medications that we were able to fill in. I sat on medication safety committees in my former role. A good example is a patient came in for a bypass surgery. It was revascularisation of the leg. The seizure medication was missed on the med history. Therefore, it didn't get reconciled and the patient never got started on it. The patient did fine.

On post-op day five, they're getting ready to get discharged. They have a seizure. Because they seized, they aspirated. They ended up on a ventilator [with a] two-week stay in the ICU. That's entirely preventable, and it's exactly what [the] NIH study alluded to, which is missing medications up front has a cascading effect throughout the continuum of care. Having a more complete data set is a huge piece of it.

The second piece is reducing the human intervention, the human tendency to introduce error into a process. Anytime I'm asking a human to re-transcribe something, or re-input something, I'm risking them making a mistake. The one becomes an 11, or the twice a day becomes four times a day because I went too fast on the scroll wheel. When you study medication errors, it's all about the Swiss cheese model, where it slips through all of the holes in the Swiss cheese, and reaches the patient. Anywhere that you can introduce automation and technology to augment the clinician experience, you're reducing error.

HL: A 2014 report by the Institute for Safe Medication Practices found that electronically transmitted medication history information, used for medication reconciliation, has potential for inaccuracy. To this day, physicians report continuing concern for some of these transmissions to introduce errors due to missing special characters, such as a decimal point, a forward slash, or a percentage in some records. How prevalent a problem are they, and what amount of patient harm or physician and pharmacist rework do problems like this introduce?

Banas: [Here] is a real-world example. There was a mistake on some of the data feeds where [the instructions say] to take a quarter tablet every day. Well, [in] 1/4, the dash would fall off, or the system would interpret it wrong. And it would be interpreted as 14 [instead]. Take 14 tablets every day. Some of the core data feeds turned off the instructions, because they knew that that error was out there.

The Smart technology has statistical and clinical guardrails around it. It's about a decade in the making, and it's processing 15 million transactions per day. There's a team of clinicians that oversees it. And we're only making these restructuring translations or inferences when it matches up with what's statistically normal and safe. In the example of one quarter becoming 14, we would know in our data that 14 tablets per day for that drug was never normal.

The AI is trained to only [bring over data] when it is statistically safe and clinically normal. And we do it with about 93% accuracy. And the 7% that we don't do is intentional. We want a human to look at it because something's off. We don't want to get this wrong.

HL: How confident are you when you remove a duplicate that it is truly a duplicate and not something that should not be removed?

Banas: 100% confident. I have 10 different data sources, but it all represents one fill event. One of them came from the PBM. One of them came from the prescribing software. One of them came because I'm connected to the local pharmacy, but it's still you picking up that Lisinopril, so I know that all 10 of those have the same timestamp. And I know I can mush them together and [know] that's one fill event, and here's the most complete data set for that fill event.

HL: What needs to happen next?

Banas: If your doctor writes you a prescription or a set of prescriptions, about 15 minutes later, in automatic fashion, we're going to ping you with a text message and deliver up an app-like experience. We're not going to make you go get an app. Literature shows that people respond to SMS, or at least look at them. And the app serves up education, short videos, coupon cards … or copay assist. And we've been able to show a 10% reduction in abandonment through patient engagement. My mission as CMO is, where can we insert our technology in the lifecycle of the patient journey, and specifically, the patient medication journey to make it safer, and achieve the quadruple aim?

Editor's note: This story was updated on March 18, 2022.