Digital pathology can and does bring great benefits, but go in with eyes wide open. Walter H. Henricks, MD, surgical pathologist and director of the Center for Pathology Informatics at the Cleveland Clinic, tells you what it’s likely to mean for your clinical workflow, how to think about cost-per-slide calculations, how to cost-justify it, and what the accreditation requirements are. He presented these remarks at last fall’s conference on Advancing Practice, Instruction, and Innovation through Informatics, or APIII.
Walter H. Henricks, MD
My goal in this talk is to focus on the operational and workflow aspects that we and our vendor partners need to examine now that we’re starting to use images for clinical purposes. A number of things have to be accounted for in processes and procedures if digital pathology is to become anything close to a scalable workflow.
I’m going to define digital pathology as pathology practice based on digital representations of glass slides, and for this talk it’s mostly through whole-slide imaging. We all know it involves more than just scanning a slide and viewing it on a screen.
There’s a lot of talk about the merging of digital radiology and pathology and the similarities between the two. I appreciate and understand these discussions, but let’s take a small step back. Now and then I have encountered people outside of pathology who have said, ‘Look, it’s an image. You look at it, digitize it. Radiologists have done it. They’re out there making money doing this. Why can’t you?’ Well, guess what? Digital pathology is not digital radiology. And pathology isn’t radiology. Digitization is seamlessly integrated into radiology workflow and actually is the workflow. The images are created digitally, with no additional steps. They’re ready to go; they’re ready for interpretation.
We have that little inconvenience of, oh, by the way, actually having the tissue, having to look at the tissue before we have an image generated. We have to process the tissue, create a glass slide, stain it. All of the routine tissue procurement, gross exam, tissue processing, glass slide creation steps must still occur before digitization.
Therefore, digital pathology is extra work. It requires extra effort and time over and above all the steps that would have to take place anyway. Digital pathology means extra cost. We have to invest in image capture and digitization, transmission, displaying, viewing, and storage. In addition to the capital expense, there are labor and other expenses that come from having to incorporate digital pathology into any type of clinical workflow. These are all investments that have to be justified. The value proposition for digital pathology, then, hinges on benefits that outweigh the costs associated with the incremental work and required investments.
Let’s talk about the elements of work-flow that need to be considered, and this is outside of the histology laboratory. For the most part I’ll be talking about up-front case management and workflow distribution.
One of the things we have to fight, and will probably have to fight for a while, is that the workflow and data flow will be divided between the whole-slide imaging system and the anatomic pathology LIS. The first step, for example, is notification of case availability—that’s probably going to come from the whole-slide imaging system. The second step is accessioning into the AP LIS—that, of course, would be the AP LIS. Third is distribution of the case to the pathologist—in the digital world, that could be a shared function of the imaging system or AP LIS, or perhaps there’s an opportunity for integration. Fourth is slide review and interpretation, and that will, of course, be done in a whole-slide imaging system. The report is fifth and typically it will be generated by the AP LIS. Last, if additional studies are needed—additional sections, stains, other types of procedures—that typically will be done in the workflow engine of the laboratory, which is the AP LIS. Over time, and we’re beginning to see some implementations of this, the number of opportunities for integration between the AP LIS and the whole-slide imaging systems should increase. But we need to make sure those specifications match what we’re expecting.
I’ll start with notification of case availability. In routine pathology, the slides are created in the laboratory and we know they exist. For consult slides, a physical slide and, we hope, paperwork arrive in the lab and that’s the trigger that there’s a case that someone has to do something with. In the digital world, glass slides do not arrive. Paperwork may or may not arrive and, presumably, it won’t.
Depending on the type of case under consideration, such as internal cases or external consults, slides must be scanned at the originating site that’s sending the consult, or they must be scanned in your laboratory. And they must be scanned successfully. These scanners are not chemistry analyzers, and that will become clearer as pathologists implement digital pathology. Tremendous strides have been made, yes, but there’s still a necessary rescan rate. There needs to be a quality assurance process to look at the scanned slides because you don’t want to distribute slides that are not scanned correctly. Then there has to be some type of notification when electronic slides are available and someone needs to do something with them.
Cases have to be accessioned and given a case number. This is how we all work. Routine pathology is straightforward, if not always rapid, depending on whether it’s an outside or inside case. The slides are matched with accompanying documents to confirm that the patient ID and/or accession numbers and slide labels match. For slides that are sent in, we must be sure the correct slides have been sent, that all the slides expected have been received, that the slides have not been damaged, and that all necessary accompanying information is present. At that point, the case is accessioned.
In digital pathology, laboratory personnel must have access to this information that is typically all paper-based. When you receive a consult for an outside slide review, you almost don’t even have to think about it. You get the outside report and the insurance information. How are you going to get these things in a digital world if you’re accepting electronic consults? Is it going to be an electronic requisition? Will the information somehow be scanned into slide-viewing software? Will it be some type of secure e-mail system? For digital pathology, a mechanism is needed to get the following types of required case information to the accessioner and the pathologist: patient ID and demographic information, a description of the specimen and the number of glass slides submitted, clinical history and questions or requests for the pathologist, the pathology report if it’s a referral case, and insurance and billing information.
In routine pathology, the workflow is fairly well established. The slides, a working draft report from the LIS, and other paperwork are delivered to the pathologist. The pathologist knows cases are waiting because a stack of slides is in his or her box or on his or her desk. Or perhaps the pathologists will find out about their daily work by looking in the AP LIS. But typically, there’s the stimulus of, ‘There’s a stack of slides, you better get at it.’
How does that work in digital pathology? How will I know from my lab that I have slides to look up? I’ll need to take an active step. I’ll need to look at something occasionally to see if my work list has been updated, or I’ll need to get some type of presumably electronic notification—it’s that push-versus-pull concept. Do I need to keep checking the computer to see if I have received some type of notification? And where’s my working draft information from? Is there a gross description for internal cases? How will I have access to the paperwork I need if it’s an outside case? Am I looking at scanned documents? Is there a printout of an electronic requisition that was delivered to me as the pathologist at the time, and then that’s my stimulus to look up the cases in the digital pathology system? These steps have to be accounted for.
The pathologist then gets to look at the slides and make an interpretation. In routine pathology, the slide goes on the microscope. The pathologist navigates at his or her own pace, and the slides are moved around the screen pretty quickly in general. The pathologist in most cases arrives at a diagnosis based on review of the glass slides. In digital pathology, there are extra steps. Pathologists have to access the slides in the viewer system. They need to pull up the viewer to see the digital slides and view them on their monitor. And we’re assuming that the monitor and the ambient light are good and that everything is optimized, or at least good enough to make the diagnosis.
So pathologists need to know how to find the digital slides for the particular case. Maybe it’s presented within a work list through a nice interface with the whole-slide imaging system, but maybe it’s not, depending on the systems involved or how good the interface is or whether it’s tailored to do a pathologist’s work list. If they’re going into a viewer system and some type of directory structure, the pathologists need to know where to find the slides that are attached to that case, and that case has to be identified appropriately and without any question in the whole-slide imaging system.
When it comes to navigating the digital slides, that’s not going to be under the pathologist’s entire control. Any pathologists who have used these systems know that pushing a glass slide around the stage yourself is typically faster. It’s not only because of system latency, like network or server response time, but also just the act of using a mouse or a trackball or clicking on the button. It’s different. Over time that may change, and there are tradeoffs, of course, when you get a whole-slide view. But the additional time required of the pathologist per case, especially in the adoption phase, has to be accounted for. How do you apportion workload, and how do you assess capacity, when you consider that it is probably going to slow down the pathologist?
When using digital slides, there may be a need to defer to the glass slide. Image capture errors and failures—be it focus or tissue finding or other things that have happened with these images—may result in unreadable digital slides. How are those accounted for? Who was scanning the slides? Who rescanned the slides? Who digs out the slides to rescan? How do they find what slides need to be rescanned when the pathologist calls and says half the slide is missing or half the tissue is missing?
There are those who will argue this point, but many pathologists would say that the high-resolution microscope optics are better than the best digital image. Perhaps that’s open to interpretation. In terms of comfort and experience, pathologists at least in the near term, when it’s new in their environment, are going to be more comfortable with a microscope. Some cases are going to require additional sections or studies, such as special stains, and those need to be accounted for in your digital workflow.
Generating the pathology report is the last step. In routine pathology, this is usually accomplished through the LIS, and through dictation with transcription, or perhaps direct entry or speech-to-text. It’s directly into the LIS as the report engine, maybe out through an interface. In digital pathology, the options are probably the same. Most of the time, we’re not going to be sending our reports out of the digital pathology system, at least not this year or next. If part of the reason for doing digital pathology is for a pathologist at a distant location to access slides far from where they originated, then to generate a report the pathologist needs remote access to the tools used to generate the report.
Experience will be required for pathologists to acquire skills and to gain comfort in making diagnoses from digital slides. And most pathologists have little to no training or experience in digital pathology. Interest is growing, but many still have little to no interest. And expertise in microscopy does not automatically or immediately translate into proficiency in using digital pathology tools, and that needs to be accounted for, certainly in the adoption phase.
We’re all used to histological artifacts that we’ve learned to overlook over the years. Those still exist in digital pathology, but digital pathology is likely to introduce artifacts of its own, perhaps certain areas being slightly out of focus or out of synch. This will improve and probably become less important over time
What are the capital and operational costs of digital pathology? If purchased, the whole-slide imaging scanning system is typically a large part of the capital expense. There are lease options now. There’s digital storage and server capacity, network bandwidth, and high-resolution viewing monitors. Are you going to have one or a couple of telepathology rooms, or are you going to expect pathologists to do this in their offices? If they do it in their offices, you have to account for the number of offices you have, and the costs add up. What about at home? If you expect pathologists to do this remotely, or if that’s part of the value proposition, do you want them doing it on their laptops or do you need to provide the monitors because it’s a clinical application? There may be dedicated Web servers depending on the applications’ interfaces. Those are going to cost money when they’re available, and some of them are available now.
Who has time in your laboratories right now to scan the slides if you do digital pathology at scale? Probably no one, so you must account for FTE time. You’ll need FTEs for scanning and quality assurance and for IT technical support.
Cost per slide is a metric getting some attention. Whole-slide imaging systems represent very high fixed costs, and scanning more slides can reduce that fixed cost per slide. But you have to balance that with the variable costs of data storage, server performance, and network bandwidth upgrade requirements that will increase with the volume of scanned slides.
Lease or cost-per-slide-scanned agreements are being offered to mitigate these high fixed costs and, for that matter, some of the variable technical costs. How do you do the comparison? Is x number of dollars per slide a good or a bad deal? Of course, that gets into what your volume projections are. What are your applications? What are your workflow considerations? Part of the cost-per-slide analysis relates to what you’re scanning throughput is because that’s going to relate to your capacity. And slide scanning throughput in the real world is different from extrapolations of scan time per slide. Say you can scan a slide in four minutes. That doesn’t mean you’re going to get 15 slides an hour every hour for eight hours. No, these aren’t chemistry analyzers, and you may not even get the quoted throughput from chemistry analyzers. Slide scan time, which is typically quoted in minutes per slide, should not be extrapolated into throughput over extended periods. Actual throughput will be less than the fastest per-slide scan time quoted. And you need to factor in a percentage factor—that’s percentage of full capacity at which the scanners will be used; that can affect your cost-per-slide calculations.
What affects or degrades slide scanning throughput? For one, the scanning resolutions. Most of the studies have scanned slides at 20× and most of the quoted scan throughput times are based on 20× resolution. We’re still learning what’s appropriate to do at 20× and what’s appropriate to do at 40×. The amount of tissue on the slide will affect the scan rate. It’s faster to scan a biopsy than a chunk of colon, for example. Inefficiencies related to batch mode processing is another factor. This is not continuous flow random access. Therefore, time to load and off-load the slides, rescanning slides that did not scan appropriately, finding the slides that have to be rescanned—all will affect scan rate. Then there is the time to QA the scanned slides, whether it’s after they come off the scanner and before they’re distributed or at some point retrospectively. And there are other human factors because humans run these machines and we’re not 100 percent efficient.
Multiple variables go into cost-per-slide calculations. And it’s not just cost-per-slide calculations; it’s the overall capacity and expense planning. Are you going to be able to do 50 percent of a maximum rate? How many hours a day do you think you’ll run this scanner?—that will affect how many slides you can push through. That comes back to your workflow. You might say you’re going to run it a couple of shifts a day. Well, when do the slides come in that have to be scanned relative to when they’re going to be expected to be needed? Are you going to scan third shift, first shift? Cost of storage is another variable. You’re going to store more if you scan at 40× versus 20×. What’s your file size per slide? What’s your compression?
What will you allocate in your cost per slide for FTE time and cost? Does it take half of an FTE to run a scanner? If you put in four scanners, do you need two FTE or one FTE? How will you allocate your scanner costs if you purchased the scanner over time and you can depreciate that over four years? We don’t know. How long will these machines last? Can we depreciate them over three years, five years? What are the annual maintenance fees? They’re not trivial by any means.
We did calculations at the Cleveland Clinic using Excel and changing the variables. If we want to do the daily surgical pathology slide output based on what we thought was a reasonable model, which was two shifts every working day, it would require using 18 scanners, 260 terabytes a year, and nine FTEs dedicated to slide scanning, and it translated to a $4.6 million first-year cost and $1.4 million for years two, three, and four based on a four-year depreciation. That doesn’t factor in network upgrades and server upgrades because we were unable to come up with what we felt was a reasonable price to attach to those. So we’re not going to be digitizing our entire department tomorrow.
Let’s look at the cost-benefit justifications for digital pathology. Most of the benefits are intangible and do not have incremental revenue specifically attached to them. In contrast, when radiology departments went digital, it became apparent that they could go out and assume the radiology practice for multiple imaging centers that were in place and/or provide reads on a regional service model, and it translated immediately into a substantial incremental revenue opportunity.
We’re not there with pathology; we might be someday. But again, we have that ‘minor inconvenience’ of having to process the tissue and create glass slides. So how to justify digital pathology? To start, there’s improved quality and efficiency of patient care, which comes in part from physician and patient having access to subspecialty pathologist expertise when none would be available otherwise. Also, facilitated communications with clinical services. It’s easier to show clinicians slides instead of making them walk down to the department. Another justification: filling of gaps in pathology service coverage. This is related in part to the availability of subspecialty expertise but also to coverage of remote frozen sections. And digital pathology can support institutional telemedicine programs and be part of a package of services that’s offered. But, in short, the clinical applications that can justify digital pathology are, eventually, primary diagnosis; second opinion consultation—formal, external and informal, intradepartmental; frozen section coverage at remote or low-volume sites; and quality assurance—for example, second review of all new malignancies.
I would argue that the revenue from digital pathology activities is not likely to cover the costs, at least in the near term but possibly not ever. Still, there are metrics that can help tell the story before or after it’s been purchased, including just the sheer volume of slides scanned, regardless of what they’re used for. This might be a metric worth tracking because most sane people don’t spend a lot of time doing pointless activities. That is, if someone’s scanning a lot of slides, someone’s finding value in that activity. It might not be revenue-generating activity, but it is of some value. Maybe you set up 20 educational conferences in the past year, or maybe it’s now possible to hold conferences across a health system—activities that would not be feasible without digital pathology.
It might be worthwhile to track the number of cases in which digital pathology enabled access to subspecialty expert consultation and for which it would not have been available previously, or, if available, it would have been cumbersome. That would be an argument in favor of an intangible benefit, but the number at least puts some quantitation to it. And you can talk about pathologist efficiency—you can calculate improvements in efficiency if you’re using telepathology or remote diagnosis to eliminate pathologist travel or courier activity in terms of hours, miles, and other related costs.
Reduced time and costs associated with packaging and sending glass slides is another potential metric. That still remains to be proved, but it’s something we’re going to look at.
Perhaps digital pathology in your environment has improved turnaround time. Of course, it’s only going to improve turnaround time if the pathologist knows the slide’s there when it arrives and it’s not sitting in a digital queue for six hours because the pathologist doesn’t know there are digital slides to look at.
You could argue, and there are models on the market for this now, that your turnaround time is more rapid on technical component studies—for example, immunohistochemistry.
If you think you have improved clinician satisfaction by using digital pathology, you can try to measure that with a survey and use that as a metric.
Implementing digital pathology into routine practice is truly more than just what we see on the computer screen. Viewing the slides is only one part of it and just one of the many considerations that go into implementing a digital pathology program. Each step in the workflow has to be examined in your environment to understand and address what the requirements will be for digital pathology. It’s a considerable investment of not only capital but also time, people, and effort. And the value depends on benefits that may be difficult to monetize.