HIStalk Interviews William Cavanaugh, CEO, Concord Technologies

William Cavanaugh, MBA is CEO of Concord Technologies.

Tell me about yourself and the company.

I’ve been in technology for over 30 years and health tech for 20. I’ve worn just about every hat there is to wear in a healthcare technology company, from making the coffee, developing the software, taking out the trash, closing the deals, and writing the business.

The high-level mission of Concord Technologies is to advance healthcare through universal exchange and intelligent processing of data. We leverage advanced AI to drive a smarter, faster, and more connected healthcare ecosystem.

What kinds of documents do health systems receive and what challenges do they experience in processing them?

The big challenge in healthcare is the exchange of data between disparate healthcare entities. You have to look at the volume of data. There are 2.3 zettabytes of data generated every year across healthcare. What’s a zettabyte? I can tell you that it’s a billion gigabytes, but that doesn’t really represent the challenge.

If you look at one hospital to paint the picture, one hospital creates 50 petabytes every year. Again, that is difficult to comprehend. Picture yourself in an NFL stadium, in the upper bowl. If you printed the physical equivalent of the annual data from just one average hospital, it would fill 750 NFL stadiums to the brim, and it is growing at 36% a year.

Now you need to share that data. You can’t email it to a doctor because it will go to junk or spam. You need a secure, ubiquitous way of sharing that data. Everyone thinks that the big EHR vendors are going to solve the problem, but there are 500 EHR vendors. They are also not the only player when you add in radiology information, PACS, payers, and pharma. There are thousands of different systems.

On the entity side, the US has 6,000 hospitals, but the number blooms over 200,000 disparate entities and growing when you add in post-acute, outpatient, private practices, urgent cares, specialty practices, et cetera. The problem that we are solving spans 200,000 disparate entities, 1,000 software vendors, 2.3 zettabytes of data growing at 36% a year, and you need to share data.

The space that we play in is documents. Think about documents between your payer, pharmacist, EHR, specialty, and primary. Our very large customers do big volumes. We do about 22 million pages a day through our network. Our big customers do over 50 million documents a month. One of our big EHRs does 90 million a month. We bring that data through an exchange protocol, universal protocol, and then we like to say that we bring it to life. We classify the document, extract key pieces of information, and then insert it into the systems that we’re on.

People might think of interoperability as a FHIR-based data exchange. How does that approach coexist with how documents are managed?

I always say that we’re not in the fax business. But at 10,000 feet, we are a fax company, even though we don’t use paper and fax machines. We use the digital fax protocol to exchange these documents.

FHIR has been around for a long time, plus HL7, integration engines, QHINs, and HIEs are trying to create the structured data exchange. We keep it simple. You have a phone number, and from any EHR, you click “send document.” If MD Anderson wants to send a document to Debbie’s Dermatology in Rice Lake, Minnesota from the EHR, they click “send document” and Debbie’s Dermatology, if she has a fax number, receives a document. Then it automatically sends a response back to the referring physician at MD Anderson that the document was received.

That’s what we do very simply, but we don’t stop there. Your big dermatology clinic gets 5,000 documents in a month. What is this document? We classify it. Then a dermatology clinic is looking for different pieces of information in that 50-page chart that just came across and that a urology clinic would be looking at. We extract the pieces of information, leveraging AI, that are relevant to the receiver of the document. That’s where we bring it to life.

Fax gets a bad rep in the market. I almost didn’t take this job as CEO because I heard we were a fax company, but we’re in the digital exchange business, using a universal protocol.

You asked about FHIR, though. There are instances where FHIR comes into play. We use FHIR to do a lookup to find that patient in Debbie’s Dermatology to match it so we can insert into the system a record. Then we use HL7, which has been around for 15 to 20-plus years as well.

The mental picture of faxing is someone watching thermal paper spool off a fax machine that is covered by taped-on “send” numbers. Is healthcare the only industry where faxing is still a viable way to exchange information?

When you say fax, you think of the curly paper, and if you’re as old as I am, the dial tone. That’s not the business we’re in.

We had a third party do some market research and I’m still surprised by the number of fax machines and paper faxing that is still done in healthcare. Anywhere from at least 10% to 15% of the documents still go through that old-fashioned, corded phone protocol.

Other entities also use fax, both digital and old-fashioned fax. Legal still uses it to fax documents. Payers, the FBI, and the IRS still use it. Other big government entities and institutions, along with mortgage companies, use old-fashioned fax. They’re also migrating to digital fax.

There is still that need when you want a secure ubiquitous protocol to send and receive documents where email doesn’t work, and that fax protocol is still used outside of healthcare. But I would say that around 70% of the digital document exchange via that fax protocol is within healthcare.

How does the process change in moving to digital fax, and what technology criticisms does that eliminate?

The biggest criticism of digital fax is that it’s not structured. By structured, I mean that you are mapping specific data fields from one system to the next. Fax comes in as an unstructured document, such as a PDF, Word document, or chart. It’s not broken down into its discrete fields. 

When that document is received, whether it’s a two-page prior authorization or a 500-page patient chart, it’s just a big PDF. What am I going to do with that big, unstructured document? If you stop just with the digital transmission, even through a cloud-based digital fax protocol, that’s the knock on fax. It doesn’t get me to where I need to be. I still need to scan through the document or read it to figure out what it entails.

With the introduction of large language models, which is the generative AI that is permeating all parts of society, I see the ability to grab unstructured data, pieces of information, from a 500-page patient chart through a large language model that can understand the context as well, which large language models are really good at. They extract the key pieces of information that are needed for the recipient. That will transform how digital fax will have higher quality, lower cost, and better efficiencies for healthcare than try to use things that have been around for a long time. I get to be too geeky, but it’s called CCDAs to structure all these fields in HL7 and FHIR to map all these discrete fields from one system to the other.

Why don’t we just do this mapping and do all this structured data exchange? Again, you just have to look at the volume. Epic has anywhere from 50,000 to 150,000 discrete data elements, based on the configuration, and every configuration of Epic alone is different. Doing that mapping isn’t rocket science, but it takes a lot of one-time work and ongoing effort to keep that up versus just sending the whole document through a secure, ubiquitous protocol that everybody has. You don’t need FHIR, HL7, a QHIN, or HIE. You have a phone number, so you can leverage the telecommunication backbone and security that is already there. Now let technology do the work to bring that unstructured document to life.

That’s relatively new even for our company, and within the overall digital fax industry. But it’s a way to transform interoperability within healthcare.

How much of the information in those documents needs to be integrated into the EHR and other systems?

The unstructured document that comes into the hospital, usually through digital fax protocol, is still probably at least 80% of the transmissions in healthcare. We’re seeing Direct Secure Messaging, and think of that as secure email. Maybe it’s about 10% of the transmissions right now. When you do it through a Direct Secure Message, it comes in through structured, but the challenge is that it doesn’t represent all of the data.

You can’t put an image in there, obviously. You’re not going to structure clinical notes. You still have to provide some unstructured data, which gives context to the recipient, the physician who needs to review the patient who was just imaged at a facility or gone to an emergency room, to get the whole context of the patient.

You call your AI approach “Practical AI.” What does that mean?

We call it Practical AI because it’s exactly what it is. A lot of AI doesn’t add much value. Ours is practical because it’s pretty straightforward and we’re focused on solving real, practical problems. So with 10,000 documents coming into a payer, hospital, or pharmacy, is it a purchase order that goes to finance? Is it a prior authorization with high priority that needs to be responded to within the next 30 minutes because there’s a patient in an ER waiting for that prior authorization? Or is it a claim that needs to be processed in the next 30 days? The first part of our Practical AI is that we’re going to look at this document that just came in and identify its type. 

The other part of the practical side is that in healthcare, nine times out of 10, there’s a patient associated with it, and probably a provider and a record number. We have to extract the patient and identify them by date of birth and address so we can find that patient in the recipient system. That’s a practical use of AI to classify, extract, and then decide what the system needs out of this 50-page document. Sometimes 20 pages and sometimes only three fields. We will make it practical in terms of what’s needed for this incoming transmission for that hospital provider or payer.

How does AI fit into the hype cycle and your company’s business strategy?

It is definitely advancing along the hype cycle and finding some real practical uses. We who use ChatGPT or any of the tools see its ability to digest information in human speech, synthesize information, and create really nice clinical summaries. If the meeting you’re in has three action items, you don’t have to take notes, because it’s going to find it for you. That’s the practical side of how AI is being used.

In our world, we’ve been doing machine learning for over 10 years. It requires a lot of training and use. It gets more challenging and specific with the introduction of large language models. Now you can throw large pieces of information at a large language model, especially when it’s been fine tuned with customized prompts for healthcare, to add real advantages of efficiency, accuracy, and clinical efficacy in the delivery of care.