An exclusive interview with Paul Egerman, eScription Chairman and CEO

Paul Egerman spent time at MEDITECH in the early days, co-founded and was COO of IDX, and founded medical transcription technology company eScription, where he remains CEO and Chairman. I confess to having had limited knowledge and interest in transcription before this interview, but as is often the case, talking to someone who knows it well and has a passion for it changed my tune. Paul is a nice guy, not only to participate in this fifth HIStalk CEO interview, but because he seems to genuinely care for people, both in the industry and outside it.

Someone proclaims every year “The Year of Speech Recognition,” but it always seems to be still another year or two away. Why is that?

Your question is absolutely correct - there’s an incredible consistency going back nearly 20 years that people have been saying that speech recognition is just a year or two or three away. They don’t really understand the problem or the technology. You wouldn’t say “Database technology is a year away” or “handheld devices are a year away.” The real issue is: what is the problem you’re trying to solve and how is technology being used to solve that problem?

The problem with speech recognition is that people have this science fiction image of it, that it’s somehow like Star Trek where you can talk as fast as you can talk and the system will just understand it, no big deal. That’s not a year away, that’s not three years away. Maybe it’s 10 or 20 years away, but certainly it’s not in the foreseeable future.

Most people think it’s a problem that will be solved at some specific moment or that cheaper hardware will provide the answer.

When you talk about real-time recognition, that won’t happen. It isn’t a one-size-fits-all solution. It’s just not going to work. It’s a very hard problem to solve, and it's being solved statistically, not with hardware power. Speech recognition can work now, but only if you have a clearly defined problem you’re trying to solve and an environment in which 100% accuracy is not a requirement. If you have those factors, speech recognition is an interesting tool, but to simply say it will be there in a couple of years and will solve lots of problems is ridiculous.

Is it unusual that eScription developed its own speech recognition engine?

It is a little unusual, but not as unusual as you would think. We used an open source engine as a starting point, which we got from Carnegie Mellon University. Every speech recognition engine that’s available commercially comes from the same source. There’s an algorithm called the Hidden Markov Model, which is always a good phrase to use during a sales presentation because it sounds like you know what you’re talking about [laughs.] That model and that technology was developed by the United States Government and the defense industry. That’s the source of most of the engines.

We really just did what everyone else has done by taking the stuff that’s in the public domain and altering it for our application. What is unusual about our approach is that it’s a background model. It’s not intended to work in real time. Being in the background, we have some flexibility to do some things that others wouldn’t think to do, like processing dictation more than once to change what you did in the first pass based on information you get later in the dictation.

How do you position Computer Aided Transcription in the marketplace?

We have very demonstrable and dramatic cost savings. This is a solution that requires no change at all on the part of the physician. That’s quite a bit different from what people in healthcare IT are talking about these days, where the physician has to be trained or has to do something. We lead with the ROI analysis, but can show an improvement in turnaround time in getting the document into the medical record. I suspect as organizations go more and more toward electronic medical records, the turnaround time argument will prove to be more and more important.

What are the challenges in selling transcription solutions to hospitals?

The number one challenge is inertia. Most hospital IT departments have a long list of applications and tasks they’re trying to accomplish. So, you have these very fancy things like CPOE that involve a lot of work to roll out. IT departments often don’t have the bandwidth to deal with our application, so we must convince them that it's important and not a lot of work. Second, people within the HIM department may have a expensive system, but it works, so there will be resistance to change. And, people are skeptical that the system will really work, although I think we’ve gotten past that challenge with over 50 customers and lots of data.

How do you convince hospitals that transcription has an impact on CPOE and clinical systems?

We say that our transcription solution is consistent with whatever their healthcare strategy is for automating the medical record. The information in dictation and transcription is really an important part of the medical record document. In terms of quantity, it’s probably the biggest. IT departments are more used to viewing transaction-oriented systems, which are easier to sell. This is a slightly different concept, but a critical piece of the medical record.

Are there dimensions beyond service levels and cost that would be compelling?

Service levels and costs are important. We generally will reduce cost by 25% to a third. For a good-sized institution, that can be a savings of over $1 million a year. That’s a dramatic savings that’s justifiable. Service levels are important, and we’re very proud that we came in first in KLAS in that area. Turnaround time and accessibility of the information will also be important. Physicians are less concerned about the hospital’s ROI, but they need the information.

Transcription delays can impact length of stay. Should transcription documents be routed based on their priority?

That’s what we do now. We’re known mostly for background speech recognition, but the workflow capabilities of our system or any transcription system are critically important. We route documents based on priority. For example, if a patient in an acute care facility needs to be transferred to an extended care facility and is being held up for a discharge summary, we can expedite that process. A radiology interpretation may be needed in the ED very fast, perhaps within 30 minutes of dictation, and we can expedite that.

Gartner predicted a dramatic drop in the need for transcriptionists a few years ago. Did that prediction play out?

No. My view of the transcription industry is that technology, such as background speech recognition, will cause that group to be transformed. Instead of typists, the transcriptionists will become the editors and quality assurance people. Transcriptionists have a huge amount of knowledge, such as medical terminology and an understanding of the institution and physicians they’re transcribing for. They will always be necessary. In fact, some people think that the amount of dictation will go down with data entry, but I think the amount of dictation, at least n the short term, will increase.

What effect did the offshore transcriptionist who threatened to disclose PHI have on the industry?

Without a doubt, that one incident had a very chilling effect on the industry. It caused us to do a lot of work to proactively make sure that nothing like that would ever happen to our customers. That occurred with someone supposedly located in Pakistan, but could have just as easily occurred with someone in the United States. Ultimately, protection of PHI can be achieved when people work offsite. It just requires better technology that wasn’t used in that case.

What was different in that case was that person was outside US jurisdication. Wouldn’t that be a problem only with offshore transcriptionists?

I think that’s probably the case, but we have a clever solution for that. We can allow offshore people to do editing or transcription on our system in such a way that we remove the patient identification first. All they see is the body of the document and the audio. We do other things, such as keeping things in separate files. In our experience, the United States isn’t the only country wrestling with this. We actually lost a possible transcription sale to a group in Canada because the Canadian health system didn’t want their medical records to enter the United States. They felt they wouldn’t have jurisdiction over American citizens if they were to gain access.

Is the future of eScription to get better at core business or to expand to other areas?

We are focused entirely on computer aided medical transcription. We’re very excited by it and we think there’s a lot of potential in that area. I believe there is also interesting potential in integrating voice at the same time date entry occurs, so ultimately we will get involved in other data entry vehicles and expand to other functions in the HIM department. Right now, we’re doing very well with computer aided medical transcription.

Who are the big dogs in the speech recognition marketplace?

The largest is Dictaphone, which has a number of products. There used to be a company called Talk Technology that did a lot of work with radiology and was acquired by Agfa. They’re definitely a player. SoftMed is a good company and a player. We like to think we’re one of the big dogs. We have over 50 customers and over 300 million lines of transcription a year, which puts us right up there.

Did the Lernout & Hauspie bankruptcy that Dictaphone was wrapped up in set the industry back?

It certainly set back speech recognition. It was outrageous in terms of the financial games that occurred. I can’t imagine that was good for anybody. It was a very bad thing.

Where does transcription fit in the discussion about interoperability?

It’s another system that has to exist in accordance with whatever industry standards are in place. Interoperability is a nice word, but it’s critically important. The transcription industry has historically been behind in that area. Vendors have tried to sell systems based on selling hardware and devices, like when Arnold Palmer was on television trying to sell handheld devices. That’s really not the future for healthcare or the transcription industry. I suspect most vendors are behind the curve on interoperability compared to other healthcare vendors.

Is transcribed speech a step back from decomposed, structured data, the equivalent of free text that just happens to have been spoken instead of keyed?

You’ll still need to do dictation. One size does not fit all. You can look at any situation and you might be able to do a check-off or answer objective questions about what happened in a physician-patient interaction, but inevitably there’s something that doesn’t fit into any of those buckets that has to be typed, handwritten, or dictated.

Dictated data often starts with discrete data, such as in a template, but the applications turn it into free text. It seems like it's the opposite of data entry.

That’s a reasonable observation. There will be a marriage between these two, where a clinician may place an order, perhaps answers some questions, and then get to a section that doesn’t fit, like a social description, and then dictate that section. No one’s thought a lot about inconsistencies, where the dictation note is not in the structured document or vice-versa. I think you can fix that with workflow.

Do we need better natural language text search tools to mine the data in transcriptions?

That’s part of our formula for success. We couple our speech recognition with our own natural language processing approach. As a result, we can do some things that most people can’t. That will be part of how these systems will work in the future. This isn’t a replacement for structured data entry, but it can help considerably.

The Hidden Markov model uses proximity, frequency, and context to interpret speech. Couldn’t you build a keyword table at the same time for searching the document later?

We’re doing something very close to that. We take the unstructured dictation and turn it into a normalized document. No matter how someone dictates, all documents of the same type look the same. We’ll find in the dictation where the person said "pre-op diagnosis" or "admitting diagnosis," no matter what they call it, and we’ll put it in the right section of the transcription. We’re already restructuring the dictation in that way.

You need the transcription people to QA the final result. Speech recognition on its own is pretty unimpressive. You need some natural language or knowledge based system that will structure the text to make it useful.

The assignment of a content section, that should allow more refined searches and more consistent presentation as we move to shared data.

It’s helpful to have every op note and discharge summary in exactly the same format. The whole concept of the electronic medical record is that other people can read it and make decisions, so that consistency is critically important. We also are consistent with how we write and express all the terminology. We spent six months watching transcriptionists and reading dictation when we started the business. There must be 600 ways to write the word amoxicillin. We standardize that, which makes the document easier to read and to lay the groundwork for the kind of keyword search you asked about.

For interoperability, transcription may be the most useful information available. Standardizing the information is important. It’s a people challenge. There are many changes, especially in university settings where residents change every year.

You worked at MEDITECH a long time ago. What did you learn there?

I learned a lot. There’s one thing that Neil Pappalardo said that impacted me early on. He said, “When we make a mistake, we admit it and we fix it.” I’ve carried that commitment to telling the truth with me everywhere. It’s certainly true about MEDITECH. Another thing that MEDITECH is very good at is selling the systems they have, not the systems they want to have. That’s what I’ve tried to do.

You were involved with the founding of IDX.

I was one of the original founders of IDX, although the company had two foundings. There was the original company, Burlington Data Processing, which was formed in 1969 by Rich Tarrant and Bob Hoehl in Burlington, Vermont. I started my business with a partner, Terry Ragan, in 1974 in Boston in the basement of my apartment building. We were the classic undercapitalized technology company who had good ideas and were working hard, but we really didn’t know what we were doing, although I’m not sure we would have admitted that at the time. We got together with Rich and Bob in maybe 1976. They saw the technology and invested in us, and we eventually merged our two companies together.

People have this image of companies that it’s like a rocket in the sky, it shoots up overnight, but it was a lot of work and one step at a time. I was always interested in the administrative aspects of healthcare, those aspects that other people aren’t paying a lot of attention to. At IDX, that was physician billing systems. Now it’s speech recognition. There’s a lot of exciting stuff you can do with electronic medical records and CPOE and I think I’m as smart as anyone else, but I figured there were a lot of smart people in those areas and I’d rather be in an area where I didn’t see as many smart people. Back in the 1970s, that was billing systems.

Where do you want eScription to end up?

It’s not like I have an end game or a final goal. We want to be the best at what we do, to constantly improve, to constantly change, and to see where the market and our customers take us. That’s it. We work hard, listen to the customers, tell the truth, and see where it all goes.

You’re a big financial supporter of Democratic political causes and candidates. Aren’t CEOs supposed to be Republican?

A lot of CEOs are Republicans, but my own personal views came from my upbringing. I grew up in a single family home. We didn’t have a lot of money. In fact, it was before there was welfare, and I got my healthcare at a county hospital. Our family got by as a result of a lot of help from a lot of people, and I’m very fortunate that I’m to be a member of what I call the Winner’s Circle right now. It was a wonderful ride and I’m very fortunate that I’m able to do well.

The reason that I have "blue state" political beliefs is that I personally know that I couldn’t have made it without the help of the government and lot of people. I think the Winner’s Circle should be expanded and other people should have that same opportunity that I had.

I respect other people’s opinions and I keep politics separate from my business, but my involvement in politics is only because I’m interested in good government. I'm also interested in a number of community efforts. I’m vice chairman of the Museum of Science here in Boston. I’m very proud of that. I think our business can succeed based on the availability of highly trained technical people, and a science museum is a critical component of getting people interested in science. I suspect almost all healthcare vendors are in cities where there’s a good educational system and a good university presence. I feel responsibility to support the community.

Where is the government going in healthcare IT involvement and where does it need to go?

They’re doing much better than I would have expected. There’s clearly a lot more discussion on interoperability issues, medical records, and errors in healthcare than existed in years before. That’s positive, for the most part. I’m a little bit concerned a what is happening with electronic medical records, where people are taking an element of faith that the EMR will just be there when we need it. I don’t view it that way at all.

I don’t know where the government should be going. I would hope that they continue to focus on interoperability, standards, and benefits of automation. What I’m fearful of is that people have a picture that this magic bullet will reduce the cost of healthcare in the United States and I’m pretty skeptical of that.

You’re a business owner as well as a healthcare IT vendor. Do you think healthcare IT will reduce your healthcare costs?

IT has the potential to save a fair amount of money, but it’s not the solution to the exploding cost of healthcare. It’s not enough by itself. If everyone had an EMR tomorrow, it’s not like healthcare would cost 30-40% less.

Is the US healthcare model sustainable?

No. There was a great presentation at HIMSS by Stuart Altman. He gave a terrific assessment of the last 20-30 years of efforts to contain healthcare costs in the United States. He said you can do lots of things that would work -- managed care, government allocation of resources, free enterprise – but the basic problem is that there has to be pain sooner or later if you want to reduce costs. Someone has to lose. Historically, when faced with the possibility of pain or losers, public policy makers have tried something else. Our current system needs change and there’s probably six different things you could do to it that would be successful. The question is whether our leaders really have he courage to make the change they need to make.

What healthcare IT companies are best positioned?

MEDITECH is extremely well-positioned, Epic is well-positioned, and despite their bad press recently, Eclipsys has huge potential and is well-positioned. There’s a tendency to throw out the names of industry leaders, but you could look at companies like eScription and small vendors who have found a niche and are doing very good work. In some sense, there’s a history of the very big vendors not succeeding. The companies you might choose today wouldn’t necessarily be the same ones from five years ago. That says how hard it is to be a big player and stay on top of the game.

You didn’t mention GE, Cerner, Siemens, or McKesson. Do you think they will stumble?

I just don’t know them as well. Siemens and McKesson I question because they’re healthcare divisions within a bigger company, and I’m a bit skeptical of that environment historically. Maybe it's the lack of a sense of industry history or their ability to react. Cerner certainly is a major player and must be doing something right. I personally do not understand the GE strategy or the conglomeration of products they have. I’ve been told that they’re the only company on the top list from 100 years ago who’s still on it, so they must know what they’re doing.

What did you like and dislike about the HIMSS conference?

My first reaction is to remember how tired I was at the end of it. I always like the opportunity to interact with customers and people I know. The attendees are terrific people and I really enjoy that. This is a big opportunity and we were successful. Of what I didn’t like, the thing sometimes feels way too big to me. It’s too hard to get around the floor. It's like trying to drink from a fire hydrant.

I’m surprised that it’s in San Diego next year. We wanted to have a bigger booth, but they don’t have enough space. I wonder if they shouldn’t let the smaller vendors expand and the bigger ones contract. Another thing I don’t like about HIMSS is that I worry that the money spent on marketing and entertainment by the large vendors is excessive. I’m sympathetic, though, because if they don’t spend the money, customers think they’re cheap or that something is wrong. When we’re talking about the cost of healthcare, though, the nature of some of the booths and the entertainment seems excessive.

Did you see any companies or products on the show floor that caught your eye?

No. There was not any technology or anything different enough to catch my eye.

What would you be doing if you weren’t running eScription?

I’m fortunate that when I left IDX I could take time off, travel around the world, get involved with politics, and get involved with philanthropic institutions. I spent 4-5 years basically not working. It started to drive me crazy.  I couldn’t stand waking up in the morning with nothing to do. I found it stressful to not have any stress. If I weren’t doing sScription, I would do some other healthcare software company. I’m not doing this because I need to pay the bills. I genuinely enjoy it.

What does a guy with a lot of money do differently that a guy without?

It’s not a lot different. Compared to when we started IDX, I’m using the same management style and working as hard as I ever did. This time around, however, there’s less pressure on me to perform. Before, it was all or nothing,. That was my whole future. Not having that pressure helps me make better business decisions.

Who do you admire in the industry?

Number one would be Neil Pappalardo. Not only for his business capabilities, but because he does so much in our community. He’s very quiet. There’s nothing about him that is self-promoting. I hate to pick two people from the same company, but Howard Messing doesn’t get anywhere near the credit he deserves. He’s just an excellent CEO. Another person who I like personally and admire is Harvey Wilson, founder of Eclipsys. He’s a really terrific guy. I don’t know Judy Faulkner, but she seems like an interesting person. I have a huge amount of respect for her. I also really like the current CEO of IDX, Jim Crook. Jim used to work for me. A lot of people don’t know him, but I think he’s got exciting potential and is a very good person.