Barry Jacobs Talks to Jim Cimino at Columbia University
Barry Jacobs: What are you working on in terms of IT in health care?
Jim
Cimino: Right now, a study of clinician information needs that arise while
clinicians are using electronic medical record systems. Decisions are often
made while clinicians are using EMR, whether they're doing order entry or looking
at new lab results, that's the moment where they're getting new clinical information
and making diagnostic and therapeutic decisions. We can help respond to their
information needs by providing access to online information resources.
BJ: Why hasn't IT technology penetrated the health care industry?
JC: You have to look at the difference between using computers to run your business and using computers to do what it is you do. For instance, a farmer could use a computer to run his business — to handle his inventory, etc. — but it's not going to help him raise crops, unless he's got something that helps him plot climate or whatever. In health care, we have information systems used by managers for billing, inventory, and payroll. That's easier to justify since you can show an ROI. With patient care, patients are going to get better, whether you have the system or not. It's a lot harder to determine ROI. Maybe there'll be fewer mistakes, or maybe they'll get out of the hospital faster, but a lot of that is unknown. Another problem in health care is, institutions are working on the margin and don't have the money to invest upfront. In addition, there are no standards of practice in health care institutions; every hospital does things differently. It's hard for a vendor to come up with an application that they can sell in lots of institutions, because each institution wants it tweaked differently.
BJ: How can IT help with clinical decision support?
JC: First, we have to know a lot more about clinical decisions. We have to know what we are doing wrong. A physician tells me that a patient is in the hospital too long. They know that because they've calculated a DRG (Diagnosis-Related Group), and they know that the patient's DRG says they should be in 10 days and the patient's been in 20. But the problem is, they don't understand why the patient wasn't discharged. The patient might have a complication, or a social service issue. We can't solve those problems unless we start looking at the level of patient care, instead of at high-level statistics.
Second, we need a more evidence-based practice. We're constantly learning new things. We need to bring information systems into the loop in patient care, and the way people are talking about achieving that right now is physician order entry. I think that's a good short-term goal. Ultimately, computers are going to help us with managing patient care. If a patient comes in with pneumonia, we don't want the computer to just say, here's a rule about which drug to use. We want it to actually help us monitor the patient overall. If the patient's got pneumonia, the system might suggest another chest X-ray.
When some people talk about clinical decision support, they're talking about bringing very complex rules into play, but there's simple stuff with a big bang for our buck. Physician order entry with alerts and reminders is one. Alerts and reminders are great. They reduce costs; they shorten length of stay; they improve health care. But a computer can't make clinical decisions unless you have a way to translate some number into some term that's in a decision-support rule.
For example, a patient had her last immunization 29 days ago. She's not due for her next immunization until tomorrow. I decide to do it today. That's a fuzzy sort of approach — a human being wouldn't even think twice. They'd say, oh, 29 days. Give it to her. I'm not going to see her again for three months. The computer would say she's not due for it until tomorrow, make another appointment.
It's similar with guidelines. They're written in natural language for human beings to understand. Computers can't make heads or tails of these guidelines. We have to write guidelines that convert into something that a computer can understand.
BJ: What about things like a picture-archiving system over the Internet? Like a patient comes in and he's got a strange blue splotch, and you've got a handheld PDA, and you see that another doctor has seen a similar thing. And that guy knew that it was caused by spider bite. What about those types of IT — a connection to instantaneous information?
JC: You have to look at where the greatest benefit is going to be. You probably want to spend your IT dollars on something that, for example, reminds doctors to give a congestive-heart-failure patient a flu shot. There are a heck of a lot more people with congestive heart failure than there are people with mysterious skin lesions, and they live a lot longer if you give them flu shots.
When I was doing my fellowship at Harvard, I worked on a decision-support system that helped with medical diagnosis. You'd put in findings and it'd spit out a list of diagnoses. The problem was, usually you make medical diagnoses based on the probability of disease. It's like that thing about zebras; if you hear hoof beats, you think of horses, not zebras, because horses are a lot more common than zebras.
But the problem is that if it's a common disease, people aren't going to go to a decision-support database and ask, what are these hoof beats? They know it's a horse. It's when they've already figured out that it's not a horse that they ask the systems. So, if you've already eliminated 95 percent of the most typical cases, the best you could do is help with 5 percent of the cases. Is that worth the investment? Diagnosis isn't always the biggest problem; the problem is management. That's where those alerts and reminders are helpful, because computers don't get bored checking checklists over and over.
BJ: Why is IT important with health care as it relates to bio-terrorism?
JC: First thing that comes to everybody's mind is the issue of surveillance. The first time somebody gets smallpox, God forbid, we'll know that we have a bio-terrorism problem. But anthrax is a different story. Anthrax, as we've seen, looks like some very common diseases. People down in the trenches might not notice one particular case. Surveillance helps us notice if a bunch of cases emerge. We know we have a problem then.
Helping with recognition of uncommon conditions is also an area, because most of the potential threats are not common diseases. West Nile virus is an example. We had an occasional case of West Nile in New York; now we've got them all over the place, and we needed help recognizing it when it first came along.
The third area where IT can help is in response. This is true of any medical emergency at the public health level. For a coordinated response, information systems are the way to go. It's not going to help — if you've suddenly got cases of anthrax — to start calling around and asking if anybody has Cipro or empty beds. That has to be done by computer systems that are tracking that information in advance. The emergency response system can track where the resources are and help mobilize them.
BJ: You have a unique perspective on IT as it relates to genomics.
JC: Again, IT relates in several different ways. One is simply pattern recognition, identifying sequences. Figuring out what part of a genome a protein comes from isn't as straightforward as simply looking up a bunch of nucleotides in a genome and seeing that those nucleotides match this protein. Figuring out the sequence of a protein doesn't necessarily tell us what the shape of the protein is or what its activity is likely to be. So there's a lot of computational biology, that is, modeling that represents what's going on at the molecular level. It requires pattern matching, and that deals with huge amounts of data. If you find a gene and you want to know how it matches proteins, you need a big database of genes and a big database of proteins. It's not something you're going to be able to do manually.
Another role is function prediction: figuring out what a gene produces. The last area is mapping the genome and what some people like to call the phenome — what's really happening at the organism level — eye color, cholesterol level; all those things are tied to genetics, but we don't know how. The genome is a parts list, and it doesn't tell us how everything is assembled.
At the clinical level, by tying the genome database to clinical databases, we can see that while we don't know what this gene does, 95 percent of the people with this disease have this gene. They all get better on this drug. Let's give everybody who has this gene this drug. We can jump right over the understanding if we can map what we're learning in the laboratory with what we've collected in our clinical databases.
