CAP TODAY Pathology/Laboratory Medicine/Laboratory Management
Dr. Simpson
Ross Simpson, what are your thoughts as you listen to this conversation?
Ross Simpson, MD, head of pathology informatics, Methodist Hospital, HealthPartners, St. Louis Park, Minn.: I like the integration of LIS and pathology data. It comes down to one thing for me—structured data. We’ve done it in cancer synoptics, but it needs to be expanded in pathology. Pathology has all this data, this narrative, which, from a computer-readable standpoint, is kind of dead, and we need actual data. Pathologists initially object to that kind of structure in their reporting, but after using the electronic cancer protocols they would not go back. There are many parts, starting with cancer reporting and going down to benign and preliminary findings, that are standard across the United States but currently exist in a narrative format. As a result, we can’t analyze that data as we’d like to.
We’re able to gather all our data on breast biopsies. We’ve been using those templates for a couple of years. It’s interesting to look at the different tumor types and margin positivity and how those patients do, and to look at our older populations and decide whether we need to send lymph nodes. We changed the way we obtain our lumpectomy specimens. We used a different marking system and then looked at that data and at margin positivity impact. Those kinds of changes are easy when you have discrete data in pathology.
The power of all the data we have in anatomic pathology is becoming more clear as we computerize everything. We expect it to be readable and able to link with our digital images so we have discrete data with a digital image that a pathologist has looked at and verified.
Years ago we thought we had to produce a PDF report for every pathology report. That’s an area where I’d like to see AP and CP come together. I don’t think that’s necessary. I’ve gone through the federal regulations and people have spoken to CLIA personnel several times, and they said you don’t need to have a PDF permanently. A pathology report is no different than any other lab test—if you don’t need it for a lab test, you don’t need it for AP. Yet almost all the systems generate a PDF of a static report and store it with the images. It’s a beautiful report, but you can produce that in real time as needed, just like you can for CP.
I also think of George Birdsong’s integrated disease report. Cancer is a journey; it’s not a single point. Once the data is discrete, you have the same data sitting on both sides, and now you’re pulling it forward in a way that you have the current status. What’s the current mutation load on this tumor? Did it develop a mutation that now makes it resistant to the drug the patient was on?
What about the practices that are serving vastly different sets of physicians, patients, and institutions and are feeling an excessive burden to supply this data in the midst of a day on which they’re busy trying to turn out cases? Alexis Carter, can you comment on that?
Dr. Carter (Children’s Healthcare of Atlanta): Most of the surgical pathologists I work with are busy signing out cases and can’t worry about how their data is being made discrete or lumped into a report—to the point where I’ve seen practices find themselves in trouble when they put in an LIS and the pathologists were peripheral to the project. All of a sudden they can’t get any of their cases signed out.
While discrete data is useful, I would caution the group that it also enables that data to be taken extremely out of context. I’m chairing a group for the Association for Molecular Pathology that is looking at getting genomic data into electronic health records. The reason people want to see data discretely is that they have an inherent desire to integrate data from multiple pathology, genetic, flow, or other reports into a single overall report, sometimes referred to as an integrated report in clinical informatics circles. But you have to be careful when some of those data and diagnoses get divorced from the pathologist’s interpretation of that data or from the specimen type it came from or the date the specimen was taken, all of which are included in the original report. For example, a molecular report in which molecular lab A subtracts out the germline variants and another report in which molecular lab B includes the somatic and germline variants in its report will give very different variant allele fractions, or VAFs.
How we use discrete data is critically important. I work with very smart clinician colleagues, but the way they want to handle these types of data is not currently where it needs to be. They think you can split a pathology, molecular, or flow report apart and then reconstruct that data into a single, integrated cancer report on a patient, where now you have lost significant pieces, including information a pathologist might have put into a comment.
The HL7 Clinical Genomics Work Group has an interface standard for genomic data that is technically and syntactically interoperable, but it loses the meaning and semantics along the way. For example, we’ve seen instances in which people are implementing certain genomics modules and issue a report that shows the CFTR [cystic fibrosis transmembrane conductance regulator] delta F508 variant at 50 percent VAF and is described as “pathogenic.” People who don’t understand what variant allele fraction means will look at that—and you can’t see the interpretation or you have to hover or click on it to get it—and won’t understand that it means the patient is a carrier. They may instead misinterpret that the patient is affected. This is the danger of divorcing the technical details, or for example a cancer synoptic report, from the specimen you did it on.
Michelle Stram, do you agree with those caveats?
Dr. Stram (NYU Grossman School of Medicine): Yes. Our grand rounds this week is on how the Bureau of Vital Statistics uses the information we put in death certificates. We have the same concerns when we see how the data we generate is used downstream. They’re not getting our final diagnostic list, the details of the information; they’re just getting the information from the death certificate and there’s a failure to understand the caveats that come with this limited information.
For example, let’s say there’s an overdose death due to a mixed drug intoxication and the overdose includes fentanyl and fentanyl analogs, such as acetylfentanyl. In addition, our death certificate may say there is morphine present. Conceptually, we understand that the presence of morphine is not likely to mean morphine in a hospital context; it likely means that this overdose included heroin, but the metabolite that allows us to specify it as heroin, 6-monoacetylmorphine, is absent because it has metabolized. Consequently, we’re only able to say, “Morphine.” The downstream user, with their interpretation of morphine, can’t discern this difference. There are failures of communication when they’re left with just the discrete data they received, and it is, in essence, out of context. There hasn’t been a great way for us to bridge that divide.
We do a synoptic report for sudden unexpected infant deaths, or SUID, which was once called SIDS. There is a form we complete that is used nationally to have a data registry so the factors that go into sudden unexpected infant deaths can be evaluated by the CDC and other groups. The way in which we suffer the same fate as many other pathologists is that this is a form, a synoptic, that we complete on top of all the other paperwork we do for these cases. It is burdensome, but the data generated from it is more useful than the death certificate data I just mentioned because this is a form created for tracking data points that the CDC and other organizations have said are most important for being able to learn more about these types of infant deaths and ultimately, ideally, prevent them. We’re capturing the data they want but in the most time-consuming, duplicative manner. In an ideal world, we would be able to accomplish the task everybody wants from a public health perspective but without the burden. So we’re in the same boat as the rest of pathology in that we’re double working or getting suboptimal results when data is taken out of context.
Eder
Many physicians who get pathology or clinical laboratory reports don’t fully understand what it is they’re getting, which is why there is a line that says, This is what it is and this is what you need to do. As we add complexity from other modalities of testing, et cetera, this problem will only grow. Beth Eder, how do you solve it?
Beth Eder (Orchard): Each pathology lab has its own unique challenges so it must be taken case by case. It’s going to be an investigation, a partnership with the laboratory we’re working with, to be able to get down to what the need is and what the deliverable is to whoever it is, attorneys or patients.
Suren Avunjian, this sounds like customized reporting for the customer that we’ve heard about in anatomic pathology. What are your thoughts on this?
Suren Avunjian (LigoLab): It’s essential to provide tailor-made reports that cater to the needs and preferences of health care providers. This goes beyond delivering data; it’s about presenting information in a way that is most actionable and relevant to clinicians’ diagnostic and treatment decisions. To achieve this, the LIS must support a high degree of flexibility and adaptability, enabling clear and concise reports that are augmented with educational resources and training for ordering providers. By incorporating decision support tools within the LIS, we can guide providers through the process of ordering appropriate tests and interpreting complex results.
Advanced algorithms, AI, and machine learning promise to enhance visualization, refine differential diagnoses, and provide alerts for critical values or trends that might otherwise be overlooked. They can also facilitate predictive analytics.
We envision a future where LIS customization will enable a more patient-centric approach, in which patients can access understandable and actionable health information.