CAP TODAY Pathology/Laboratory Medicine/Laboratory Management
That’s part of the point I’m trying to get at here.
Dr. de Baca (MDPath and Pacific Pathology Partners): Yes, but I’m not sure what our in-between option is. One thing we know about fax transmission is that if you have a fax machine that is a known number and you send to that, you can be pretty sure the report will arrive at the place it was intended to arrive and that the patients’ health information is not shared elsewhere. If a larger practice has many clients and is trying to share things over secured email, that’s always a possibility, but it’s not yet the preferred way of transmitting reports.
Dr. Prichard, how do you deal with it when the clinician says, “I wish you’d just fax this to me”?
Dr. Prichard (Geisinger): We fax it to them. I’m surprised at how much faxing still goes on, and it’s not something we’ve been able to get away from. It’s the culture of some of the clinicians and their offices. But the fax does have a certain kind of security to it—that is one of the benefits that keeps it hanging on—but it would be my goal in the future to get away from faxing if we can, especially the paper part of it because it’s a degradation in the report as they print the paper out and then scan it back into their EHR. If we could have them receive things in digital format in the fax and then not have to print it but then store it into their EHR and skip that paper step, we’d be going in the right direction. I haven’t been able to champion that through.
Rick, how many physician systems do you interface with at NovoPath? And do you have an observation to make about this question of faxing?
Rick Callahan (NovoPath): We have one lab that interfaces to 150 clinicians through the same EMR vendor. That could be different EMR versions, but it is the same vendor’s EMR. And with that we’ve set up an interface engine, which the lab manages, so we’ll send the information of the reports to the interface engine, and then the lab will distribute the reports to the appropriate clinician in the EMR.
In my world, faxing is not as well received or used as much as clinicians accessing a web portal and pulling their reports through the lab’s web portal into their EMR. That seems to be one of the more popular means of receiving reports. The other is through the HL7 interface to the EMR. Dr. Prichard had mentioned some enterprisewide vendors being unable to receive PDFs, and there are several of them. What Geisinger has done is embed a link into the HL7 where the clinician who does have access to the EMR clicks and then pulls the report into their computer through the HL7 interface. This functionality is available in NovoPath, and it enables our clients to provide reports in a PDF format through an EMR that is unable to accept PDFs.
Regarding what Dr. de Baca said about textual transmission, we’re finding that many NovoPath hospitals are moving away from textual transmission of pathology reports. If you’re not aware of the ability to install a link into the HL7, then the worst possible scenario is to send the report in a textual format.
Dr. Liberman, can you comment on this question of getting reports into physicians’ hands and what your experience is across the spectrum of customers?
Dr. Liberman (Computer Trust): There are two levels of question there: how and what. The question of how goes to faxing; it’s secure, point-to-point. It’s grainy and black and white; that’s not great if you’re showing images or color. If they then rescan that to load it, as Dr. Prichard said, it gets even grainier. If they can accept an EMR with a link or directly with uuencoding or 64-bit encoding of a PDF into an HL7 and then load it into theirs—clients that have EMRs that will accept such a report prefer to do it that way. We have some that don’t have that, and they prefer—if it’s a big enough volume client—to put a remote printer in the office and have our system send directly to that remote printer.
Now for the what. Your original question was how do you make sure you’re matching up what you’re communicating from the pathology lab to what the clinician needs clinically. I don’t think most EMRs have adopted this, but there’s one fairly widely used EMR in dermatopathology that has a suggested diagnosis feature, and that matches up with our diagnosis category. The lab customers who use that love it because it helps communicate to the client in a format they really get in terms of their clinical need.
Dr. de Baca (MDPath and Pacific Pathology Partners): We’ve spoken about the how and the what. It would be interesting to talk about the “wouldn’t it be nice if. . . .” I used to be a clinician, and I know from experience how I as a surgical ophthalmologist read the pathology reports that came to me. It was different from the way most pathologists think a person reads the reports. If we think about cancer reporting, for example, and we think about synoptic reports, which have thankfully become the standard, we still have a host of clinician clients who are reading these reports, each one with specific interests and needs. We also have patients who are reading these reports, and then there are the insurers that are reading the reports and the people from the cancer registries, and each one of these clients consumes our reports in a different way. When our life was paper, it was necessary for all the information to be on every report. But now that most of our reports are being consumed digitally, one could foresee the moment in which, as a consumer of a specific type of report, I could choose to see specific information. For instance, if I’m the surgeon, perhaps I just want to know if the margins were clear and what the diagnosis was, and I’m not interested in seeing the gross description or the billing information. If I’m the patient, perhaps I need to see not only the report as written by the pathologist but also with natural language, if you will, or colloquial language annotations so that I can see what was written but I can also see the comments that say “this means something to this effect” in real-life language. I’m envisioning the pathology report more as a Rubik’s cube, and every consumer of that report would be able to compose the face of the cube they need to see for their specialty or their level of specialization.
Chad Meyers (Sunquest): The future Dr. de Baca envisions is within our reach given some of the work being done by the Integrating the Healthcare Enterprise PaLM group on the diagnostic report template and FHIR [Fast Healthcare Interoperability Resources] for clinical document architecture usage. We’ve made good strides in getting some of the report body content into synoptic or discrete reporting but need to get to the point where the whole report is done that way. To be able to allow that would be powerful.
On the previous topic of faxing—the beauty of the fax is it’s a series of numbers to identify and set up a new connection, and every fax machine works the same way so there’s no testing required and no interface setup on both sides; that’s part of what makes it frictionless. And we’ve continued to try to do that on the HL7 front, but there’s still allowance for customization there. As we move toward that full report being in more of a consumable format that separates the data from the presentation, it’ll allow greater consistency in how that’s consumed across both the consumers and downstream systems.
Dr. Prichard, what would be on your wish list as you reflect on what we’ve been talking about?
Dr. Prichard (Geisinger): The next step for the AP LIS is integration with whole slide imaging and the whole slide imaging workflow and making it as efficient and easy to adopt as we can because there’s going to be resistance from some of our pathologists. That’s an obvious next step.
The one I’m facing now and trying to find a solution for has to do more with molecular testing and communicating and coordinating what is an algorithmic step-by-step if-then-else process of what test comes next and to make sure you’re moving through that algorithm efficiently, so that by the time the patient gets into the office, you have the answer to all the testing questions. Molecular testing is so complex and the guidelines are changing so often that knowing the next test you’re supposed to order has been confusing. I would like help coordinating these molecular laboratory testing algorithms.
I’d like a brief comment or two about artificial intelligence. It’s brooded about a great deal, and it’s become a buzzword. Dr. Liberman, what two or three things should I know or would you like to share about your view on artificial intelligence and anatomic pathology?
Dr. Liberman (Computer Trust): I’m going to use an advertising analogy. With advertising there’s the content, but you also need white space in an ad because without it people can’t read what you have to say. The LIS should largely be, in my view, the white space, and laboratorians should be able to do what they do best—accession, make and read the slides, deliver reports. The intelligence evolves as humans figure out what they need, and they need to be able to express that and codify that into the LIS, and it needs to be able to just do it. It can do it repetitively and check for a zillion exceptions at rapid speed much better than a human can focus on all these different things.
Not too long ago we saw multiple Post-its on transcriptionists’ screens. If it’s this doctor, do this; if it’s such-and-such type of case, do that—as if they’re supposed to be checking every Post-it on every case and thinking about each one. That’s hard for a human to do. Artificial intelligence should be focused on building and codifying those things.
Chad, I’d like a brief comment on artificial intelligence either from your perspective or that of Sunquest. Is it ready for prime time? Is it overhyped?
Chad Meyers (Sunquest): It’s ready for prime time, but it makes sense to test it out and validate its applications. An example the group was talking about earlier is the ordering rules and trying to implement decisions and flows. Artificial intelligence could learn the patterns there based on historical ordering and try to optimize it. If you wanted to have a standard protocol, then you might need to use that instead. There are more applications of AI if we extend it to the various types of rules that are traditionally built in LISs and other solution applications—for example, the progress being made with image analysis and what’s being done with quantitative scoring on whole slide imaging. With all of these application possibilities, we need to make sure we’re proving them out as an industry and prioritizing the right use cases for our clients based on what helps them the most.
I’m excited about the possibilities, and we continue to monitor artificial intelligence progress to identify and perfect the right cases in which to apply the technology.
Dr. de Baca, what are your views on AI in pathology and pathology systems?
Dr. de Baca (MDPath and Pacific Pathology Partners): We’re already experiencing and are comfortable with a lot of machine learning tools. Prognostic scoring and simple risk calculators have been used for quite a while, and artificial intelligence has also been used for a handful of years in immunohistochemistry for calculating percent positives in nuclear staining with Ki-67 or ER or PR staining, for example. There’s a huge possibility for AI and clinical decision support, be that on the level of the patient-facing clinicians or as pathologists, and of course the applications with whole slide imaging abound.
One of the questions is: What are the right tools to apply? There will be a lot of people with hammers who think every possible thing a pathologist does is a nail where AI could be implemented. Another question we need to wrap our heads around is: If we have artificial intelligence algorithms that are constantly learning, how do we make sure they’re adequately validated? And once initially validated, how do we continually validate them?
We’re stepping into a new galaxy, if you will, of questions about the assuredness with which we are implementing some of these tools. There are scientific, computational, and ethical questions that we need to be tackling as more of these tools come to the market and as we determine which of them are algorithms that will help us save time and improve or maintain quality and won’t add risk.
Rick, is AI a frequent topic of conversation among your customers and potential customers?
Rick Callahan (NovoPath): No, it’s not. You asked if we are ready for prime time or if it is hype, and my own opinion is, after any necessary government approvals, we’re ready for prime time, but we don’t have the early acceptance or the early adopters yet that would turn it into an acceptable tool to use in a general laboratory. Once we have early adopters that have shown the benefits of computer-aided diagnosis, we’ll have more of a hockey stick adoption by other labs.
Dr. Pritchard
It reminds me of the early reluctance toward digital pathology. It seems AI is tracking along the same lines, part of which is concern about professional security. Dr. Prichard, do you see the concerns about AI and those about digital pathology in the past being similar?
Dr. Prichard (Geisinger): My experience of it is pretty similar, and I think we’re going to have to go through a process with AI like we did with whole slide imaging and teaching the FDA what it is they’re approving. Once we got to the point where the FDA was willing to grant approvals for whole slide imaging, adoption started to pick up. We’re ready for AI, but I don’t think the regulatory environment is ready to take on the challenge of approving what would be actively learning systems. We’re going to be restricted to having some static algorithm that they can approve, and then we’ll improve on that and we’ll have to submit it again. It’s going to be the FDA learning how to regulate in a learning system. That’s what will hold us up, much like it did whole slide imaging.n