Transcript of Web Conference
This information is for reference purposes only. It was current when produced and may now be outdated. Archive material is no longer maintained, and some links may not work. Persons with disabilities having difficulty accessing this information should contact us at: https://info.ahrq.gov. Let us know the nature of the problem, the Web address of what you want, and your contact information.
Please go to www.ahrq.gov for current information.
This Web assisted audioconference, broadcast on October 21, 2003, was the fourth in a series on bioterrorism conducted throughout 2003 via the World Wide Web and telephone. This Web conference focused on how information technologies and surveillance systems are being used in bioterrorism preparedness programs. The User Liaison Program (ULP) of the Agency for Healthcare Research and Quality (AHRQ) developed and sponsored the program.
Cindy DiBiasi: Good afternoon, welcome to "The Role of Information/Communication Technology and Monitoring and Surveillance Systems in Bioterrorism Preparedness." This is the fourth event in the series of Web-assisted audio conferences on Bioterrorism and Health System Preparedness. These events are designed for State and local health policymakers and health systems decisionmakers.
The series is sponsored by the U.S. Department of Health and Human Services Agency for Healthcare Research and Quality, often referred to by the acronym AHRQ.
My name is Cindy DiBiasi and I'll be your moderator for today's session. The context for these calls is clear and compelling. Bioterrorism represents a significant public health threat to the United States. Addressing this threat requires the rapid development of Federal, State and local capacity to respond to potential bioterrorism events. This means improving the abilities of both our public health and healthcare delivery systems to detect and respond to such threats. It also means ensuring that public health, health systems, facilities and providers communicate and coordinate effectively with one another, as well as with other related systems such as emergency preparedness and law enforcement.
The intended audience for these calls, State and local health policymakers, program administrators and health system decisionmakers all play an essential role in these efforts. Within their own jurisdictions, regions or delivery systems, they must develop capacity and coordinate efforts across public health, healthcare, law enforcement and related systems. It's therefore extremely important that they have information about emerging health systems research, promising approaches and available tools that assist in the development of readiness plans.
In addition to today's event, one more call will be conducted as part of this series. The final event is scheduled for Tuesday, December 16th. I'll tell you more about the December call later in the broadcast, but right now, let's turn to today's call. We'll be examining the role of information and communication technology and monitoring and surveillance systems in bioterrorism preparedness. Let me begin by introducing today's panelists. In the studio with me I have Dr. Michael Shannon, Chief of the Department of Emergency Services at Childrens' Hospital at Harvard University, and Dr. Michael Wagner, the Director of the Real-Time Outbreak and Disease Surveillance Laboratory, and Professor of Medicine and Intelligence Systems at the Center for Biomedical Informatics at the University of Pittsburgh. And joining us remotely, Dr. John Loonsk, the Associate Director for Informatics at the Federal Centers for Disease Control and Prevention.
Before we begin our discussion I'd like to tell the audience a bit more about the format of this audio conference. First, we'll talk with our three panelists, then open up the lines to take your questions. And we'll give instructions on how to send your questions to us later in the program. In the meantime, if you experience any Web-related technical difficulties during this event, please click the 'help' button in your window to troubleshoot your Web connection. If it appears that the slides are not advancing, you may need to restart your browser and log on again.
Well, now I think we're ready to discuss today's topic. And let's begin with Dr. Michael Shannon, again, the Chief of the Department of Emergency Services at Childrens' Hospital at Harvard University. Michael and his colleagues have been examining the role of information technology and improving syndromic surveillance and decision-making in the event of an act of bioterrorism, and Michael, for those of us who may not be familiar with the term, what exactly is syndromic surveillance?
Dr. Shannon: Good afternoon, Cindy. Sure, let me try to answer that question. Syndromic surveillance is a relatively new term in clinical medicine, and it basically describes a technique for the active monitoring of a disease cluster. Its goal simply is to provide early detection that something unusual is happening among your population. Now, it's not totally new. What we used to call syndromic surveillance was the astute clinician, or a group of public health officers who would investigate some unusual occurrence to find out if there truly has been an outbreak. But now, with information technology, we can do many of those things electronically.
Cindy: How exactly is information technology improving the surveillance?
Dr. Shannon: Well, information technology has offered quite a few benefits in improving syndromic surveillance. First, its electronic nature improves detectability quite a bit. If nothing else, it reduces the burden on clinicians to always play the role of the astute clinician that identifies when something unusual has occurred. The electronic nature permits you to examine something in real time. Meaning rather than have a human at the end of the day or the week look through a set of records and say, "Hmmm, something unusual has happened," you have a system that can monitor activities every minute, every second, and as soon as something unusual appears, it can report it. And then finally, information technology offers the advantage in permitting you to monitor data that we already collect routinely. We already collect prescription habits, absenteeism, the number of visits to a physician's office or an emergency department. And now, with electronic systems you can mine those data already being obtained, and improve your ability at detection.
Cindy: Let's talk a bit about automated decision support, what it is and how it has developed.
Dr. Shannon: What it is, simply speaking, automated decision support of the use of the computer to assist the clinician in making the diagnosis. Automated decision support tries to take that to a more sophisticated level, generating more useful and robust assistance to the clinician. I'd like to try to outline the steps of developing an automated decision support system. I think a useful example would be how you might try to create an automated decision support system to assist you in detecting anthrax, or smallpox.
The steps you would need to take are these: first you would want to assemble the signs and symptoms of anthrax or of smallpox. You would want to assemble a group of infectious disease experts, create a so-called Delphi panel, who would assist you in outlining those signs and symptoms and in what order they appear. From there, with the assistance of informaticians, or computer engineers, you can begin to construct a differential diagnosis, and even probabilities around signs and symptoms.
For example, it's well known that fever precedes the rash of smallpox. If you can teach the computer that that always occurs, and a clinician comes to the program and says—I have someone who has a rash but no fever. Then the program can not only generate a differential diagnosis, but the probability that that is the disease of interest. So, if the patient has a rash, a suspicious rash, but no fever, the bioagent diagnosis program would be able to say—Here's your differential, smallpox very unlikely, probability less than 5 percent based on that. So those are the steps and they can be created for potentially any disease of interest.
Cindy: Now, can you give us an overview of the types of information systems that your program has been developing?
Dr. Shannon: Our center for bio-preparedness at Childrens' Hospital in Boston has been working in this area for more than 3 years now, and over that course of time, we've either completed or are continuing to develop quite a few different models. Five of them are in the area of syndromic surveillance. They have various names such as ED Scope, AEGIS, the Boston Data Linkage Project, the PECARN Data Linkage Project, the School Monitoring Project. And then we have created an automated decision support prototype known as the bioagent diagnosis program.
I'll tell you a little bit about the ED Scope. ED Scope is a prototype for identifying—it's a prototype for syndromic surveillance for pediatric illnesses. In order to create ED Scope what we did was to examine visits to our emergency department at Childrens' Hospital, Boston, over ten years, which totaled 500,000 visits by children over the course of 10 years. And in doing so, we were able to find that visits by children to emergency departments are incredibly predictable. In fact, they are so predictable that in looking at 10 years of prior data, we are now able to predict within 5-10 percent accuracy the number of children we'll see in our emergency department and what their chief complaints will be.
And having that information, we are able to, as the ED Scope shows you here, identify for a set of chief complaints that forecasts numbers, the number we would have expected to see on a day of the week. We can set alarms, so if that threshold or number is exceeded, a red flag appears. So as you see here, we have red flags indicating that for a specific day we saw a few more children than we would suspect with rash, or with meningitis. So, it's purely a research prototype, but this is an example of what can be accomplished with syndromic surveillance. In the area of automated decision support we've created a program known as the Bioagent Diagnosis Program, which again, tries to accomplish, as I outlined, the purposes of automated decision support. And simply what we've done is created a number of—identified a number of bioweapons induced illnesses of interest. We met with a group of infectious disease experts. They let us know what the signs and symptoms and order of appearance of these signs and symptoms would be, and then working with collaborators at the MIT artificial intelligence lab, we've been able to construct some probabilities around each of those signs and symptoms.
So that now, in theory, if we post this program on the Web, any clinician taking care of any patient, can go to this site, can enter the signs and symptoms of their patient, provide some temporality, that is how long these symptoms have been present. And then the output is a list of potential bioweapons induced illnesses, and the probability that that is the agent.
What's most exciting about this is it brings us what's so important in this whole area of preparedness, which is so-called dual use functionality. Meaning that we've created this solely for the purpose of assisting the clinician in identifying, for example, a victim of anthrax or smallpox. But it can potentially be used to identify emerging infections such as West Nile Virus, or SARS through the very same process of assembling the Delphi group, identifying the signs and symptoms, establishing probabilities around them, and then working with collaborators and artificial intelligence generating a differential diagnosis along with probabilities.
Cindy: Do any of these support systems have particular importance to public health preparedness around specific populations?
Dr. Shannon: Well, I'm a pediatrician, so now you're going to hear my bias! They have a lot of importance in making sure that children are taken care of. For example, syndromic surveillance as a tool could be extremely useful by the school nurse, who is going to, on a day-to-day basis, identify when absenteeism has a certain pattern, or when more children than expected come to the nurse's office with a chief complaint such as an asthma attack. There's an obvious place where syndromic surveillance can be used in an important environment such as the school. I think that the decision support tool can also be useful in thinking about all the issues that will be faced in thinking about children. For example, if the clinician has limited experience, or knowledge about childhood diseases, and for example doesn't really know the details about chicken pox versus smallpox, and can go to the bioagent diagnosis program for assistance there.
Cindy: Now you say you're a pediatrician, and obviously this area has received a lot of attention in the past year. Where are you focusing your attention—what areas are you looking at and what areas do you think are going to be requiring additional development and support?
Dr. Shannon: Well, first let me say that that attention is appropriate. First because I think as we thought so much about preparedness over the last several years that we've only begun to really embrace and think in greater detail about the issues that children would face. Despite the fact that we know that children will be disproportionately affected in multiple ways in an act of bioterrorism or any other type of disasters. In my mind there are several priorities that must be addressed and we think about the preparedness and where it should go from here. For example, there's nothing more important than thinking about school and the issue—because children spend more than 80 percent of their waking hours in school, and what exactly will happen if there is a disaster that involves children in school. The concept of surge capacity—I hope at least that every emergency department has an ability to take care of a group of children who are injured or involved in some type of disaster. But are they able to take care of several hundred children involved in a disaster?
So, the issues of surge capacity that we talk about so much do have a pediatric focus as well that has not received adequate attention. There's the entire issue of training first responders—fire, police, EMTs. Those that will go to the site of a disaster and will be responsible for the triage, the early care, the decontamination of children, and children of course are not small adults. If they are very young they can't speak, they can't give you a chief complaint or even tell you what hurts. So, there's a real challenge in being able to manage children and it's going to require much more training and education to first responders.
Nationally, for many, many years, we've been creating so-called disaster medical assistance teams, response teams that can go to the site of a disaster and provide assistance to local resources. But there are only two pediatric disaster response teams in existence in the entire U.S. at the moment. So there clearly needs to be greater work there.
And then finally, we know that children are more likely than adults to have enduring, long-term mental health and behavioral consequences after being witness to, or victims of, disaster. So there needs to be a well-conceived plan for identifying these children, and clinicians, nurses, parents—will need instruction on how you identify the child who is having any type of enduring behavioral consequences from being witness to violence or a disaster.
Cindy: Well, Michael, you've given us a lot of great things to think about. We will be back to talk to you during the question and answer portion. But first, I'd like to turn to Dr. Michael Wagner. Michael is the director of the Real-Time Outbreak and Disease Surveillance Laboratory also known as the RODS Lab. And he's investigating methods for the real-time detection and assessment of outbreaks of disease using information technology. This work has resulted in the development of two major systems: The National Retail Data Monitor and the Real-Time Outbreak and Disease Surveillance or RODS system. So, Michael, can you begin by telling us a bit about the RODS Lab?
Dr. Wagner: Sure, Cindy. The RODS Laboratory involves 20 computer scientists from the University of Pittsburgh and Carnegie Mellon. And the name of the laboratory really says it all. We're focusing on the use of information technology to accelerate the process of outbreak detection as much as possible. And the word 'real-time' or the phrase 'real-time' really conveys the goal. On the slide that's titled, 'What is the mission of the RODS Laboratory,' you see the basic public health investigation cycle that we're trying to accelerate with information technology. On the left we have the first hint of trouble. Which figures in further analysis and characterization of whatever data happened to be available at the time, and answers questions like, "Do we really have an emergency on our hands?" "Do we know enough about it to recommend preliminary measures like quarantine of the population?" "Should we activate the national pharmaceutical stockpile?" and, "What additional data collection should occur?" Either via shoe leather process, meaning public health calls. People on the phone whether it's emergency rooms, specialized rapid testing like microchip testing or something in the future—electronic decisions reported at the point of care that's modified dynamically based on changes in the belief about what's going on in the population. So for example, that additional testing is done on people with diarrhea. So we're interested in every aspect of the use of information technology to accelerate this so that the delay between steps is measured in seconds or nanoseconds rather than days at present. Now, if I can draw your attention to the length of this slide. You see that there are three ways at present that public health for the community may notice that there's something anomalous going on. The first being statistical analysis of data that are already being collected for other purposes.
And just like Michael Shannon was discussing earlier, is really a focus of syndromic surveillance. And in the second and third types I'm just including for completeness. Namely an astute clinician or a citizen reports something to the health department, or as in the case of Boca Raton back in October of 2001, a laboratory recovers—or somebody notices an instance of anthrax, or something that just never happens in the absence of terrorism.
Cindy: Let's talk a bit about the systems that your lab has developed. What is the National Retail Data Monitor, or the NRDM?
Dr. Wagner: The National Retail Data Monitor and also the other system that I'm going to talk about in a minute, both share two things in common. First, they leverage data that are already being collected on a massive scale by industry. And second, they try to get that data for every individual in the population. Not in an identifiable manner, but just so that we don't miss any of the sick individuals, or the behaviors that are showing up in routinely collected data. So, in the case of the National Retail Data Monitor, we have this enormous retail pharmacy industry and grocery industry that sells products that people use when they're sick. So, people wake up one morning with some infectious disease, then after some time delay like 24 hours, some fraction of these individuals will actually purchase cough syrup, anti-diarrheals, thermometers, pediatric electrolyte solution especially for the kids. And it turns out that the retail industry has built this enormous infrastructure, which I like to think of as an Egyptian pyramid that's just missing the very top stone. And all we did was add the top stone. So what has the retail industry built for us at a cost of tens of billions of dollars? Well, they designed a universal coding system that the printed on every package. They installed optical scanners in every pharmacy, in every grocery store. They didn't do this for our benefit, but they actually merged many, many, many individual stores and consolidated them into very large national chains. So at the present time, ten large national pharmacy companies, and mass companies, account for about 60 percent of every one of these products sold. And if you kick the number up to 20 chains, if you're able to get 20 chains to participate, you have 76 percent coverage of every product sold in the country at the present time.
But what did we do? What kind of capstone did we put on the pyramid? We simply asked for the data. And miraculously seven of those corporations eventually said 'Yes.' And so at the present time, these corporations transmit the data on a daily basis by 3:00 PM today for yesterday's sales data, to Pittsburgh. And then we redistribute the data nationally to 33 States in which there are user accounts, so that's eight entities which actually get subsets or entire copies of the raw data, including the Centers for Disease Control. And there's a slide that shows the geographic distribution at the current time of the system, and you can see that the entire country has—those are—by the way—counties that you're looking at. And the color coding refers to whether the level of sales are normal on the day that this picture was taken, or whether they're abnormal.
And I won't scare anybody by telling you what the exact mapping between colors and level of abnormality is. Or even what the product category was. But that gives you a sense of how the data are being monitored. And then to the right of that picture of the United States, there is a graph which shows the daily usage figures for about 20 States of the system. This slide is about a month old. So on the left we have the blue ribbon States that are looking at the system every single weekday. That what our 100 percent means, that they're looking every single weekday. And that's the burgundy color vertical bars. And the light blue colored bars indicate that they're actually looking at the system on weekends.
Cindy: What are the future plans for the system?
Dr. Wagner: This, Cindy, is definitely under development. Although it's been used for frontline protection of the public since last December, which was just two months after we started working on the project. At the present time we have 30 percent of market share for the country, and we're trying to get to 70 percent. So we want to enlist enough additional retail chains to get to the 70 percent level. Which means we have to get about 15 or 17 national companies. We would like to decrease the time latency below 24 hours. These products are being scanned in real-time at the optical scanners, but the retail industry has had no need in the past to transmit the data from the individual stores to the national warehouses any more frequently than at midnight. And they can speed that up, it's just a matter of making a consolidated 'ask' on behalf of the country to these companies to do something very patriotic.
Those are the main things. We would also like to add monitoring of antibiotics. Because that's an additional tip that something anomalous is going on. We think it should be done in our neighboring countries, like Canada. We could definitely use more automation of the map reading, so people don't have to log in everyday on weekends. And then we're looking for a stable home for this, and a long-term funding model for it.
Cindy: Before we move off of this, how can anybody interested in getting more information on the National Retail Data Monitor get that?
Dr. Wagner: Well, most importantly, any public health official can get a free account. And that's the second point on this slide. By sending E-mail to the address given. For more general information, there's a wealth of it, facts about the system at the URL given there, www.health.pitt.edu/RODS. And finally, there is a recent publication in the September issue of the Journal of the American Medical Association that's specifically about this system, and it goes into things in great detail.
Cindy: Let's move onto the other major system that your lab has developed. It's the RODS System. And Michael, if you could tell us briefly how this works?
Dr. Wagner: The RODS System is just like the National Retail Data Monitor in that it's leveraging an enormous infrastructure out there. But instead of leveraging the retail industry's infrastructure, it's leveraging the hospital industry's, or the healthcare industry's infrastructure. It turns out for the past 20 years the healthcare industry has built an enormous computing infrastructure. And in particular, any time a patient registers in an emergency room in almost any hospital in this country, their chief complaint, and a lot of other information related to their illness, is recorded. And this information is transmitted, and this is already done routinely by the health system, to a very central location, a machine called an integration engine, or an HL7 message router. And it then becomes available to be sent to public health if somebody adds a little capstone of information processing to that infrastructure. So, for those of you with an information technology bent, the next couple of slides are intended for you. The rest of you can sort of just turn your brains off for a moment.
What you're looking at here is an HL7 admission discharge and transfer message. So this is the infrastructure that the hospital industry has in place, HL7 messaging. And this particular message has the ZIP code, date of service, and chief complaint in anonymous form for one individual registering at an emergency department. So you can see 'Sore throat and cough' is what this individual was complaining of. Now, people complain when they have pain in their throat and they're making these coughing noises, with all sorts of different English phrases. And then the transcriptionist mistype them in all sorts of creative ways. So the next piece of technology that's needed, and this is part of the capstone, is something that's called Na´ve Based Text Classifier. So we've built this. And you'll see on the left of this slide, that there's one of these creative encodings of the chief complaint produced by a transcriptionist. N/V/D. Which stands for 'nausea, vomiting and diarrhea'. And the Na´ve Based Classifier is like a black box that takes this kind of input and attempts to classify it into a syndromic category, like respiratory disease or diarrheal disease or hemorrhagic disease. And in this case you can see that it's computed likelihood of 0.9 that this particular chief complaint is a GI chief complaint, and it puts out the syndrome, the GI syndrome in this case as it's classification for that patient based on the Na´ve Based black box. So, that's the secret to processing a chief complaint. And it differs from processing retail data, because retail data is already clearly coded using Universal Product Code, a UPC coding system. We do not have a similar thing as of yet in medicine that works as well as this.
Now the next slide just sort of closes the story, or the next two slides, on how we process this chief complaint information. The slide that's titled 'Prime Series' shows the eight different syndromic categories that are produced by the chief complaint classifier, the base classifier. But it doesn't show the results for a single patient, but it shows the results for the entire region being surveyed. In this case the region is Utah. And this was the system that was in place for the 2002 Olympics. And so you can see in the left upper corner, that little graph titled 'All Visits' shows the past seven days of emergency room visits in the entire State. And the number is about 2500. But right next to it is the subset of visits that were due to respiratory illness as detected from the chief complaints provided by the patient. And you can see it numbers about 150 or 200 per day. I can't even read it myself. Now the key point here is that these data are available in real-time, even before the physicians know. As we all know, when you go to the emergency room you talk to the triage person. Sometimes it's several hours, or even more, before you see the doctor.
Proceed to Next Section
Return to Audioconference Session