October 2026—Differential reviews, reference ranges, longitudinal data, AI, and information silos were front and center when CAP TODAY publisher Bob McGonnagle led a discussion online in early September about hematology analyzers and the needs of laboratories and clinicians.
“As long as the future will help automate the predictable and allow people to focus on the exceptional or abnormal, we will have made the right progress,” said Julia Radchenko of Beckman Coulter.
With McGonnagle and Radchenko were representatives of Sysmex America, Siemens Healthineers, and Abbott, in addition to the chair and a member of the CAP’s Hematology/Clinical Microscopy Committee: Olga Pozdnyakova, MD, PhD, of the Hospital of the University of Pennsylvania, and Michelle Grant, DO, of Geisinger Medical Laboratories.
Click here for CAP TODAY’s interactive Hematology product guide.
Olga Pozdnyakova, in our hematology analyzers roundtable two years ago, you talked about leading an effort to standardize criteria for the manual differential review. Can you update us on your progress?
Olga Pozdnyakova, MD, PhD, director, Division of Hematopathology, and professor, pathology and laboratory medicine, Hospital of the University of Pennsylvania: We formed a working group in 2024 with representatives from the CAP, Society for Hematopathology, International Society for Laboratory Hematology, International Council for Standardization in Hematology, and others. We tackled the ISLH rules, which came out 21 years ago [Barnes PW, et al. Lab Hematol. 2005;11(2):83–90], and we published two papers.
The first paper summarized what has happened in the field of hematology and the manual differential review since the publication of the rules, how people have been using the rules, and what’s been helpful and not helpful. We did an extensive review of the literature and found that few institutions use the rules as they were; each customized the rules to their practice setting [Courville EL, et al. Int J Lab Hematol. 2026;48(5):1008–1022]. Many institutions use it as a first step to approach the manual peripheral blood differential.
The second publication focused on standardizing the peripheral smear review [Pozdnyakova O, et al. Int J Lab Hematol. 2026;48(5):937–950]. The more I was writing and thinking about it, the more I realized it’s difficult to standardize because we all practice in different settings and ask different questions about what we think is important in our patient population. So it turned out to be a set of what I would call common-sense rules—important items to consider when standardizing rules for your setting. For example, it is essential that you know how your instrumentation works. They all come with factory settings, but there’s a lot we can tweak to increase the sensitivity or specificity. There’s a discussion about deltas, because there are five different ways to measure deltas, and what’s important and what’s not. It’s not a standardized, one-size-fits-all approach but a set of points that explain why it is important to customize the rules for a particular hospital or hospital system.
Kevin Croghan, can you speak to how efforts to better understand how people are implementing thresholds for manual differentials will ultimately benefit the laboratory user and the patient populations they’re addressing?
Kevin Croghan, MBA, director, hematology strategy and solutions, IVD product marketing, Sysmex America: It is establishing guiding principles or a methodology to understand what adjustments need to be made on the instrument and software to help ensure that the only smears that get made are the ones that are required. We continue to reinforce with our customers and our staff that every site is going to be different. Each site has its own standards and patient population and how they’re affected, but there is still one goal of making smears only when you need to.
We all share that goal, and progress is slow and there isn’t a simple road map.
Kevin Croghan (Sysmex): Not at all. There are many variables, whether it’s the stain, the water, the smear itself, or the hematocrit. When customers talk to us about smears and stains, there’s a laundry list of items we have to troubleshoot and help eliminate. But ultimately, it’s the same goal.
Amjad Azizi, can you comment on trying to get a handle on the appropriate reference ranges for patients and the conditions hematopathologists are encountering in their practice?
Amjad Azizi, director, automation and informatics strategy and solutions, Sysmex America: The constant challenge is standardizing the data, because everyone gets their data in different data sources or data lakes, and there needs to be a standardized way to share it on a broader scale. Not everyone wants to share their data, and the people who are willing to share have data that is not as specific or well maintained as it needs to be for our use. We in informatics call it labeled data, which we can get more insights from by running algorithms or AI on it.
We are seeing bigger players moving forward now. For example, Epic announced in September its partnership with OpenAI and ChatGPT. This could mean every Epic customer, which is 60 to 70 percent of the U.S. market, has at least a standardized way to acquire data, run normalization on it, and work with it, and then it can be shared on a larger scale.
Julia Radchenko, talk to us about how high-quality reference ranges and standards can help save laboratories money because they’re no longer reporting what I’d call a false-positive, which demands a working up of the patient. Proper reference ranges put that patient in an area that indicates no further investigation is required.
Julia Radchenko, MBA, senior director, international commercial and hematology/urinalysis product management, Beckman Coulter: Our responsibility as manufacturers is to give laboratories data, evidence, and transparency so they can establish meaningful reference ranges for their patient population. A step forward in improving this position is to think about the longitudinal information—a patient’s result today may still fall within the reference range, but it could represent a meaningful change from that patient’s own baseline. If we establish an algorithm using population reference intervals, population-specific trends, and appropriate clinical decision limits, I think we will advance to the next level.
Stephen Smith, can you comment on what you’ve been hearing?
Stephen Smith, MSc, global head of hematology, Siemens Healthineers: Hematology has no shortage of data. Of all the diagnostic tests, it’s probably the most abundant in data. The real challenge and opportunity is how we turn that data into useful information. I wouldn’t suggest a change in reference ranges. I work with clinicians and ask what the best outcome is for them. Our ultimate goal is to turn that data into meaningful information that provides greater clinical context and supports patient care.
The analysis of hematology data has a richness that when properly understood will benefit patient care and help clinicians do a better job in taking care of their patients. Do you agree?
Stephen Smith (Siemens Healthineers): Yes. Hematology is one, but it goes across other modalities. By combining hematology data with hemostasis or coagulation data, you can get a better insight into the patient. That’s what we’re hearing and what we see as the potential future for the diagnostic industry.
Michelle Grant, can you comment on what we’ve been discussing?
Michelle Grant, DO, system director of coagulation, Geisinger Medical Laboratories, and assistant professor, Geisinger Commonwealth School of Medicine: As Stephen said, it’s more than just setting up a reference range based on high-quality data. Sometimes results fall outside the reference range but they’re not necessarily actionable. Determining when results should be acted on is important. As data becomes more accessible, there is a greater potential to refine these ranges to be more clinically actionable.
Dr. Pozdnyakova (Penn): There’s been a paradigm shift in reference ranges. The challenge I’m facing now is that as we create more and more hospital systems, hospitals or outpatient clinics within the systems are often using different hematology analyzers and instrumentation, and there is a push to combine and create one reference range. Maybe by using AI with longitudinal data, we can look at what’s happening not just within a generic reference range but on a patient-to-patient basis. What is normal and within the reference range for one person could be abnormal for another person because they somehow changed it from point A to point B over the course of a year. This is another challenge I’m seeing.
Lawrence Johnson, expert professionals are trying to get a grip on the questions of standardization and harmonization among analyzers from different vendors. How is this playing out for you at Abbott?
Lawrence R. Johnson, MD, systems director, hematology, Abbott: I’ve always challenged the notion that you can get an adequate reference range from 120 or 240 patients, if you think there’s a gender difference. I did an unpublished study on 1.5 million uniquely matched patients with data going back to 1998 using an enterprise master patient index, and after filtering it extensively, we found differences by gender and decade of age. You could never find that with 120 or 240 patients after filtering that many patients over that period of time. So I think we need to change our idea of what an adequate representation is when determining reference ranges or trying to standardize or harmonize.
I agree that studying longitudinal trends and understanding the velocity of change is valuable. For example, if someone starts out with a hemoglobin of 16.5 grams per deciliter and then drops to 13.5, the question is germane: Did it happen overnight or over 10 years? If it happened overnight, then they either lysed, bled, or had a transfusion, or it’s the wrong patient. We’re looking at the statistics in the wrong way. We’re looking at a sample size that’s wholly inadequate, but it’s what people have settled on. I think the reason they’ve stuck with 120 patients is it’s easy to work with statistically. I cannot find literature that justifies that sample size. We need to challenge this idea, and we need to look at much longer-term and bigger data sets and then filter them extensively.
At Abbott, we support the patient and clinician in the lab by having them validate their own population, which is critical. I say this having come from a laboratory where we worked with Native Americans who have different reference ranges; they also have unique hemoglobinopathies. And then there are the blood phenotypes that change other reference ranges.
We need to look at long-term effects and at the tempo of the changes in patients, in addition to absolute and relative changes, because we know from a Lewin Group survey that lab results drive 80 percent of all medical decisions and yet we account for less than five percent of total dollars expended. Consequently, it behooves us to get it right.
We are a big believer in artificial intelligence, and agentic AI in particular, and its ability to look at a patient’s longitudinal data and correlate it with other analytes and then draw meaningful inferences from that.
Amjad Azizi, do you have a comment to make?
Amjad Azizi (Sysmex): I want to clarify why I’m excited about the ChatGPT-Epic collaboration and what we are trying to achieve with the reference range. We’re not trying to standardize reference ranges so everyone would use the same reference range. It would be personalized; patient number one will not have the same reference range as patient number two, and those reference ranges would be based on many data points. We might in the future be able to change the way we classify patients. The decision will always be made by the physician—the algorithms and IT will be an aid in getting information and giving a full picture, and the physician can then make an informative decision about that patient.
AI and machine learning companies are trying to get those documents right now. They will get vitals from smartwatches, combine it with lab data, add other data, and then get a customized reference range that fits a specific patient. AI/machine learning is good at analyzing a large amount of data, but when it gets a large amount of data that’s not harmonized or identified, it cannot connect the data points.
Dr. Grant, how do laboratories and laboratory professionals communicate these results and their meaning to clinicians and to patients?
Dr. Grant (Geisinger): Clinical providers have less time to review results, and patients are more likely to see their results before their clinical provider does. Therefore, the most actionable results should be easy to recognize. Many laboratories change the color of all results that fall outside the reference range to a different color like red or flag them as abnormal. We’ve modified this approach and have chosen to highlight only a few of the out-of-range CBC parameters including WBC counts, hemoglobins, hematocrits, platelet counts, and WBC differentials. This focuses clinicians and patients on the most important values. It has sped up clinical teams’ review of results and decreased the number of calls to physicians from patients about nonactionable out-of-range results. In the future, it would be ideal to provide narratives with the highlighted results to further enhance result interpretation.
Julia Radchenko, it seems we’re putting a lot of burden on the people who supply the instrumentation and assays and on laboratories and their IT systems. In other words, you can have the most wonderful instrumentation and brains in the laboratory, but if your IT system is somehow inadequate, there’s a failure to connect everything for the best care. What would you say?
Julia Radchenko (Beckman Coulter): Two themes come to mind—future ecosystems and what the strongest solution looks like. We shouldn’t be operating in silos. IVD manufacturers, LIS and EHR vendors, AI and software developers—we should all be disciplined enough to integrate.
Technologies should be working together to come up with parameters that are ready for clinical use. As IVD manufacturers, we have a responsibility to make sure new parameters are analytically sound, clinically validated, and supported by the appropriate intended use. When you have appropriate intended-use claims, obviously requiring confirmatory clinical assessment, extensive clinical trials, then we can help answer the questions: Is this parameter reliable? Have the findings been repeated in other studies? Can it help change a clinical decision? Simply put, can I trust it, can I understand it, and can I act on it as a receiving clinician?
We continue to have staffing challenges. In particular, many senior pathologists and senior clinical laboratory scientists are retiring. It’s difficult to backfill years of experience and expertise. Stephen Smith, tell us about Siemens Healthineers’ help here for laboratories.
Stephen Smith (Siemens Healthineers): We look at how we can help with automation, but the measure of automation isn’t just how much technology we throw into the lab. We look at how much unnecessary work we can remove from the clinician side. For instance, remote review with our Scopio solution removes requirements for labor to be located in a certain area. You could have a centralized place where the clinician sits and the expert can review it, and the slides are processed somewhere else. The goal is to remove the unnecessary work and put the skilled labor where it needs to be. This is a focus for Siemens Healthineers and one that many clinicians are calling for.
Stephen hints at something I’m hearing increasingly in meetings and lectures, which is everyone in the laboratory wants everyone there to be working at the so-called top of their license. Kevin Croghan, can you comment on how Sysmex is involved in educating its customers and others to achieve this?
Kevin Croghan (Sysmex): We focus on providing education. There is not one portion of these topics that could not stand alone, whether it’s staffing, artificial intelligence, or longitudinal data. As manufacturers we want to continue to deliver what we’re good at—quality, accuracy, dependability, efficiency—so we can provide the best results possible.
Laboratorians reinforce when we talk to them that they’re not the ones who diagnose, and we don’t necessarily want the instruments to diagnose either. That’s the clinician’s responsibility. We want to give them all the information possible based off trends, not spot results. No one can diagnose or help interpret based on one result. How can we help not diagnose but raise a question for the clinician to ask or to ponder in determining what must happen next?
Dr. Johnson, give us your reaction to that comment, given the well-documented difficulties of clinicians having the time to deal in depth with patient results.
Dr. Johnson (Abbott): We need an integrated approach and to better connect all our instruments and data. That’s at least one luxury of Abbott having the chemistry and immunoassay instruments along with the hematology analyzer. I’ll give you an example of why that’s useful. It is too hard for clinicians to stay up on hematology, let alone other areas of laboratory medicine, so we need to give interpretive or narrative comments. Let’s say a patient comes in and you measure their CBC and they have marked microcytosis perhaps with a slight degree of anemia; for example, an MCV of 60 fL and an MCH of 22 pg. That’s one of two things in my book, which an agentic AI should be able to propose: You either have thalassemia or iron deficiency, or both, but you do not have anemia of chronic disease as the sole cause of that degree of microcytosis. If a system could dive back into previous laboratory results and even the patient’s parents’ medical history, because thalassemia is an inherited disorder, you could help them arrive quite comfortably at a diagnosis if you had the ability to tie that data together.
Genetic diseases are playing an increasingly central role as we discover more about them, and our ability to tie results together, not only the hematology result but also other analytes in chemistry, genetics, and so on, will be helpful. There is a tremendous spectrum of knowledge variation among clinicians in their ability to comfortably deal with hematologic issues. We need explanatory narrative comments that do much of the legwork for them. It would also help reduce strain on the laboratory because there will be fewer calls if the clinician gets the information that he or she needs in that final result.
The way to improve staffing issues is to have fewer touches, but also to take work off the bench and propose narrative comments that are helpful to clinicians. You do that by having your finger in the data of different instruments and analytes, and better reference ranges, as well as the patient’s longitudinal data.
Dr. Pozdnyakova, is there a hospitalist-type role that might work in systems to help laboratories and specialized clinicians connect? Or do you foresee that role being served by middleware solutions with a souped-up IT component?
Dr. Pozdnyakova (Penn): There is a role for both. We have increasingly been using what we call physician extenders. They might have been medical technologists and have accumulated vast knowledge but don’t have their medical degree. They’re more than capable of combining all the data in the LIS and EHR and presenting it to a pathologist or clinical team in a way that is digestible and understandable. The challenge is there is no clear path to defining these people and getting to that position.
I don’t know whether it’s a middleware, but there could and should be a system. We work in silos, so there is no communication between analyzers, divisions, departments, even between areas of the lab. We can call it middleware or something else; it compiles the data from all the analyzers and talks to electronic health records and helps produce the interpretive comment. That can be a lot more efficient than our opening the chart and trying to figure out why the patient’s WBC count jumped from one to 15 in a day, only to find out the patient has GCSF [granulocyte colony-stimulating factor]. It’s a simplistic example but we still get a delta, we still get phone calls, we still get smears for review. All it takes is our opening the chart and finding the word Neupogen and then you’re done. There is a big role for AI or Cloud or Dome that connects everything within labs and hospital systems.
Dr. Grant, what are your thoughts about communicating results as efficiently and accurately as possible to all who need to see and act on them?
Dr. Grant (Geisinger): Efficient and accurate communication requires strong laboratory and clinical team collaborations to determine what results are most actionable. One example of this is the blood smear review, a time-consuming process for technologists and pathologists. By having open discussions with technologists, pathologists, and clinical teams, we were able to standardize the criteria for blood smear reviews across hospitals and clinic sites, resulting in decreased turnaround times as well as a reduction in laboratory resources spent on blood smear review.
Using AI to enhance this communication has huge potential in the future, and I’m excited to see how technology evolves, particularly in interpreting multiparameter results. Many people use canned comments now when performing routine blood smear review. Imagine if an algorithm could recognize predictable CBC patterns and create a narrative comment based on the results, eliminating the need for a blood smear review altogether. This could free up pathologists to focus on those smears that were most challenging, further enhancing efficiencies.
Would anyone like to make a final comment?
Julia Radchenko (Beckman Coulter): Regarding the staffing shortage and AI, I hope we are not going to materialize yet another dashboard with which clinicians have to collaborate and that creates more cognitive burden rather than helps them focus on what matters. As long as the future will help automate the predictable and allow people to focus on the exceptional or abnormal, we will have made the right progress.
Amjad Azizi (Sysmex): The biggest challenge we have in the middleware is that we are missing the actual diagnosis. What I mean by that is we have the look-back tools. We have the history of the patient, if the patient stays in the same hospital or has been analyzed on the same entity. Then you create the best AI that gets you a personalized reference range based on that information. This information goes to the physician, who makes a diagnosis, and that goes into Epic, where the patient sees it in MyChart with limited comments because they still have to see the doctor and get the information. The lab has zero idea what’s happening after it reports the results. Making decisions after that or making sure the model in the lab is more iterative is still missing and, we hope, being addressed at the higher level, the EHR or Epic or Cerner, where both results live—the lab result and the physician diagnosis.
There are other parameters that live in the lab, but don’t live in the higher-level system, that can provide added value. AI could flag “This looks like . . . ,” and start to find the patterns. Once you find the right pattern—say, 25,000 patients AI reviewed last month all share a specific extra parameter from hematology, chemistry, or immunoassay—that’s where trends emerge. AI needs data that’s labeled and traceable. It’s not about which company’s software or middleware has to make the decision; it’s where the data lives. The more data we have, the better the outcome. That’s why I’m hopeful about the collaboration between Epic and ChatGPT, because something is going on on a bigger scale. We are not looking at silos anymore. It’s not IVD- or vendor-dependent. Everyone can collaborate.
Dr. Johnson (Abbott): An important development coming down the pike is the direct integration of the peripheral blood morphology into the CBC. Another is that modern AI does more than pop a flag or notification and starts being true agentic AI, helping think with you and parsing and analyzing large amounts of unsiloed data.
Finally, we integrate hematology into the realm of other systems. That would be a huge advance. We need a CBC that knows what it means. We need to move far past the five-part differential.
Dr. Pozdnyakova (Penn): Many laboratorians, whether they work at the bench or direct labs, may not be aware of the issues we have been discussing today. The more we expose them to what they need to learn about, the better we’re going to be because they will better understand why we’re talking about what we’re talking about.
There’s a shortage of staff everywhere, but especially in IT. Much of what we’ve discussed requires someone with that knowledge. It’s a challenge that can hinder development because there are not enough people who know how to implement the ideas we have.