There are a lot of exciting things going on these days at the Hospital of the University of Pennsylvania (HUP) and the health system of which it is the flagship facility, the University of Pennsylvania Health System (Penn Medicine). And among a constellation of forward-looking initiatives, one particularly fascinating one is in the imaging informatics area, where Woojin Kim, M.D., associate director of imaging informatics at the Hospital of the University of Pennsylvania, has led the development of a highly innovative radiology search engine called Montage, which first went live in 2009 at HUP.
Prior to Montage, Dr. Kim and colleagues, including Dr. Khan Siddiqui, Dr. Nabile Safdar, and Dr. William Boonn, were behind creation of Yottalook™, which was made publicly available to radiologists, other physicians, and medical researchers, now being accessed in over 170 countries. Like many good ideas, the concept for Yottalook emerged organically out of a set of problems to be solved. Its name, by the way, came out of International System of Units (SI) where, as Dr. Kim points out, “Yotta is the largest accepted SI.” It is also a play on the phrase, “you outghta look”.
Dr. Kim spoke recently with HCI Editor-in-Chief Mark Hagland and shared the story behind the development of Montage, and of his leadership in that development. Below are excerpts from that interview.
How did the idea for Montage emerge?
The thing that I worked on when I first came to Penn was that I had trouble finding my own radiology reports. In the world of radiology, our main product is radiology reports; everything hinges on that—the quality, the value we have for referring physicians, those are all based on radiology reports. So here’s an analogy: If I wanted to go out to a restaurant tonight and be within a certain geographical range and wanted a certain cuisine and look up some reviews, I could Google all that and find a restaurant that I want to go to, right? But if I wanted to do similar things with radiology reports where I’m looking at a case, and I say to myself, I had a case just like this, but I can’t remember the patient’s name, for example, what do I do? The ability to rapidly search radiology reports didn’t exist. So I created a Google-like search engine that can be used at Penn. We had an old rudimentary search tool, but it was very slow, limited, and the user interface was cumbersome to use. It wasn’t like Google, which works in less than a second with a very simple user interface. Now, if it took more than a second to do a Google search to get results, you’d think there was something wrong with your Internet service, right? Yes, the expectations have completely changed among search engine users. So we created an application that works like Google. For example, at Penn, we have over 13 million radiology reports digitized since the late 1980s. And that amount of data gives you a tremendous amount of possibilities. So that’s how it got started. And I said, why don’t I make this vendor-agnostic in terms of its backend, so that I can plug it into other specialties within medicine, like pathology?
So now I can search pathology reports as well as radiology reports within the same institution; and what’s more, I can also do pathology-radiology correlational searches. And that’s how we came up with the idea for Montage. Over a weekend, I created a prototype version on my own laptop and showed it to my IT guys, and they really liked it. So they said, go ahead, tap into the entire report database. They gave me my own server, and the very first version of this search tool was born. It was initially called PRESTO (Pathology-Radiology Enterprise Search Tool); Dr. Boonn [William Boonn, M.D., a radiologist and radiology informaticist at HUP] came up with that name. Now, we have over 300 users, which include radiologists, administrators, non-radiology physicians, and research coordinators.
Woojin Kim, M.D.
And it’s already in use by many different physicians and others, within the U.S. and even abroad. Now whenever I gave presentations at national meetings on search and data mining, after each presentation, people would come up and ask me, how can I get this? As a result, we ended up commercializing it, and created a company with Dr. Curtis Langlotz, Dr. Rajan Agarwal and Dr. William Boonn . The commercial product is called Montage. In radiology, everybody focuses on images, but the product really is the radiology reports. And I started incorporating other specialties within medicine, like pathology and cardiology; there is really no limit; we could go into surgical operative notes, discharge summaries, progress notes, etc. And that’s what folks are interested in at Penn. A lot of hospitals are building their own data warehouse or buying one so that they can federate their databases and mine them.
So this is a nice application that can sit on top of such system, where its easy to use interface allows for powerful searching capability across multiple different databases. Going a step further, radiology reports are a unique type of medical document. There are numerous nuances around the ways that radiologists dictate and write their reports, so we’ve created a tool that accounts for that. For example, let’s say that you’re searching for a report that discusses pneumothorax, which simply is a hole in the lung, causing presence of air between the lung and the chest wall. Now, many radiologists dictate absence of pneumothorax in chest radiography reports routinely. However, if you are doing a search for “pneumothorax”, you are likely looking for positive cases of pneumothorax and want to exclude all those reports that say, “there is no pneumothorax.”
Well, you can perform “negation searches” to remove all such cases so that you are left with only positive cases of pneuomothorax. Understanding such unique elements in any medical specialty are very important; this is why it is crucial to have domain expertise when creating search and data mining tools in medicine. That is also what makes Montage unique; it’s developed by physicians for physicians. We have also created various dashboards and analytics with quality assurance capabilities like error-checking within radiology reports to improve report quality and patient safety.
What lessons have you and your colleagues learned so far?
One thing is, when you give end-user physicians the ability to data mine and search their own reports, the ideas they come up with are absolutely fantastic—in particular, some really neat quality improvement projects have come out as a result. And you engage the younger guys; and by doing so you also promote academic research and quality improvement. So they’ll come up with things like, how many malpositioned catheterizations have there been in the past year? Or, how many times is the endotracheal tube being placed in too deeply? Or how many times a feeding tube is put into the airway instead of the esophagus? I’m sure the hospital has its own procedure and incident report database, but how often are radiologists identifying malpositioning?
How about how many times has a radiologist made a mistake and identified the wrong side of the body in his reports, such as a fracture of the left hand instead of the right hand? People really come up with creative ideas and solutions. Having this ability to search and data mine really empowers end-users. Speaking of the left and right issue, I have created a whole dashboard just on laterality errors—and when I first created this, the error rates were higher than I’d expected. But when I started sending out these error reports to physicians and when the physicians found out this was being monitored, the error rate dropped by almost 50 percent; purely the awareness that someone was watching dropped dictation error rates by more than half. Such behavior modification has been well documented so it’s no surprise, but it is nice to see such tools can be used to improve quality and patient safety.
And radiology is one of the specialties where a lot of people haven’t yet thought about quality improvement, right?
They have, but not enough. And this is just one specialty in medicine. But if you give this ability to do search, in a HIPAA-compliant fashion, the kind of information you can get is absolutely fantastic and mind-blowing.
What thoughts would you like to share with CIOs and other IT leaders?
People like to say in healthcare IT world we’re many years behind everyone else [outside healthcare] in terms of IT, such as financial/banking services. Now if you were a product in WalMart, WalMart knows exactly where you are and have been at all times. Yet how many times a day do we lose lab test results or have studies that go unread in healthcare? So if you look outside healthcare for potential paths forward, that will help. Take for example the whole concept of dashboards and business analytics; those tools have been used for decades outside healthcare. That’s why I like to look outside healthcare for ideas and implementation strategies. So the simple comment I would make to CIOs and CMIOs is, what do you do every day when you go online? Half of the time, you are searching for something. And imagine what you can do if you could do that inside your own hospital. And right now, if I took Google away from you, you’d feel pretty limited. We are living with just such limitations but don’t know just how much. The fact is that we don’t have Google-like search capabilities when it comes to patient care right now. But imagine what could happen if we really did have that ability.