In hematology, making the most of automated solutions

Fernando Chaves, can you comment on the issue of reference ranges? It seems that without well-thought-out and well-established reference ranges for patient populations, much of the work we do interpretively or in AI may be misleading. In other words, if we’re going to feed the AI, we need to do so with the best, most pristine reference ranges for the patient.
Dr. Chaves (Siemens Healthineers): All the AI algorithms and solutions that go in that space open a vast array of opportunity for going to personalized health care. Then you can cut the data any way you want. Because we have the technology now that enables us to capture and analyze the data much more efficiently than in the past, we can understand what is normal for different groups and what is abnormal. Reference ranges are an essential part of enabling all the solutions we have been envisioning here.

Bob Roopra, how do you control for reference ranges with your application? I’m assuming you have a vast number of users from different populations and places.
Bob Roopra (Sight Dx): We have eight age- and sex-related default ranges within the device, all of which can be selected. For reference range validation, local health care providers are advised to conduct studies on their own patient populations.

The biggest challenge we’ve had in the past 30 or 40 years is the way the populations have changed not just with growth within a population but also because of migratory effects. In hematology there should be best practices around setting and understanding what’s normal in your own backyard. Maybe with AI there’s a tool that can be used rapidly in the lab to define those ranges for each place.

The most interesting samples, for that matter, are the ones with poor results that get referenced and end up requiring film review, which you can see in patients with severe clinical conditions. They’re the most interesting because they start to bend the rates of what we call normal, and normal is limited to the technology. So we’re pushing the boundaries on that because we’re looking at it differently. It’s an interesting space to look at and should be taken seriously

Ihab Zidan, have you found an improvement in the percentage of manual differentials your laboratory customers are performing by virtue of their better understanding and better use of the automated results?
Ihab Zidan (Abbott): We are seeing more dependence on the technology these analyzers offer. Review rate is a major determining factor for labs when considering new analyzers. Looking deeper into scatterplots and scattergrams, which are often overlooked, can provide a better picture for the user and can impact review rates. Because certain parameters do not appear the same way or are not measured the same way across analyzers, that can create confusion among users, especially if you’re not looking at the same system or the same technology within the health system. Standardized technologies coupled with the increased adoption of digital morphology, AI, and clinical decision support can help reduce the percentage of manual differentials.

Susan Behnke, does Horiba set a goal or an ideal of a percentage of manual differentials that you would expect your users to have to perform?
Susan Behnke (Horiba): Not specifically because it depends on the patient population of a practice. The manual differential rate for an analyzer in an oncology practice will be different than the rate in a general practice. But with the technology analyzers utilize to perform the differential, including mobile thresholds, cell differential is excellent. The cells are placed into the correct location on the matrix. Most practices can keep it below 10 percent.

Jonathan Galeotti, do you have a benchmark in your mind for your patient population for manual differentials? I realize it’s an easy and yet impossible question.
Dr. Galeotti (UNC School of Medicine): We are a large academic hospital with many affiliates that send us things to review, so our patient population is varied and we see more from our oncology patients. There are also challenges with autoverifying differentials from our ICU patients; we review a lot of those. Our goal is to minimize the ones that don’t need a manual review. There are many that a pathologist must look at, but it would be ideal to eliminate the ones that could be autoverified or not get normal ones that have been flagged for incorrect reasons or should not have been flagged.

Jill Crist, what should our readers know as of 2023 about manual differential rates?
Jill Crist (Sysmex): The key is trusting the technology of the instruments. The smaller community hospitals may be less trusting of an automated differential. The more advanced places tend to be more accepting of the automated differential because they understand the technology and have abnormal populations they can learn from. With the exclusion of commercial reference labs and oncology and pediatric populations, the overall autovalidation rate is around 85 percent. With staff shortages and more and more sick patients, people need to embrace their automation and technology.

Tim Skelton, you’re in a large, diverse system—if you had a policy statement on the manual differential, what would it open with?
Dr. Skelton (Beth Israel Lahey Health): Recently the biggest gains we’ve had in reducing manual differentials are from using the CBC and absolute neutrophil count instead of the CBC and diff. There are fewer instrument flags that require manual review on an absolute neutrophil count than there are on the manual diff. By working with the clinicians and saying maybe you just need a CBC and absolute neutrophil count here, we’ve seen an uptake in the use of that test order rather than the CBC and diff. For our very sick inpatients, our bone marrow and liver transplant patients, patients on chemotherapy, and also for screening for sepsis, we use the CBC and absolute neutrophil count instead of the CBC and diff. And like Jill said, you need to understand the technology and push it to its limits. The time to test what the instrument is capable of and optimize the auto-reporting of the differential is when you bring in a new instrument.

Dr. Galeotti (UNC School of Medicine): We did something similar where we limited the number of manual differentials for individual patients. If a patient had a manual diff within 24 hours, we wouldn’t do another one unless there was an urgent need to repeat it.

Jill Crist (Sysmex): I’d like to ask Drs. Skelton and Galeotti—pediatricians and neonatologists tend not to want to go with automated differentials and instead prefer manual diffs. Do you also see that and, if so, do you see it changing? How can we get these physicians to embrace and accept the automated diff? Or will it always be a challenge?

Dr. Skelton (Beth Israel Lahey Health): The issue is the band count. Automated instruments cannot tell the difference between a band and a segmented neutrophil but technologists can, or believe they can. There’s a huge variation. The skill of the medical technologist will affect the band count. Also, it’s subjective: What’s a band and what’s a segmented neutrophil? The neonatologists have not given it up; they want a band count. And the neonatologists’ professional literature supports that.

From a pathology point of view, the band count is not precise or robust enough. And the automated instruments can’t do it, so if I had my way I would lump the bands and the segmented neutrophils together. But we don’t get buy-in from the neonatologists on that.

Dr. Galeotti (UNC School of Medicine): Similar answer—they’re not going to give it up. Most of those currently do not make it to the pathologist; they are reviewed by our technologists. I do worry about that as we lose our senior technologists. It’s going to be a bigger problem.

Dr. Chaves (Siemens Healthineers): As far as in the industry, we see two sides of the story. There is a pressure to lower the manual counts because of staffing challenges. At the same time, if it is easier to process manual slide reviews through digital morphology and automated solutions that facilitate the interpretation of the images, maybe the conversation will shift from the percentage of differentials to how difficult it is to process and obtain reliable information from those novel tools. If it’s easy and automated to review a differential, we could even have higher diff counts and that would not be much of an issue.

Scott Dunbar, where do you think the CellaVision application will be in the next few years and how will it continue to enhance the field? What’s in your pipeline?
Scott Dunbar (CellaVision): It’s always important to add more assays, more offerings to tedious tests in the laboratory, whether for bone marrows, which are coming out shortly, or for more esoteric things like Kleihauer testing, which is tedious because when you say something is positive, it might be 1.4 percent. For a remote laboratory that doesn’t have a person who is good at Gram stains, it would be wonderful to put Gram stains and other assays on that could be read by a microbiologist at a mothership.

Band counts and standardization are hard. I’m in laboratories every day and medical technologists will not agree on a cell. I can show five cells on a screen and everyone will call them something different, and if you look at Rumke’s table of variability, everyone is right or wrong. Someone once asked me, How do you become a better morphologist? You could easily say, Do more differentials. But if you have a bias, you become a more consistent morphologist, not a better one. The only way to become better is from coaching, having someone give you input to help you become a better morphologist—not by doing more differentials.

Bob Roopra (Sight Dx): I’d like to make a point about the effect on manual diffs. In oncology we have had an impact on how many samples get sent back to the lab, fundamentally because the way our CBC solution works is essentially as an automated smear with AI. By nature AI is consistent and not prone to a specific morphologist interpretation, as the system is trained on extensive amounts of blood samples and getting better and better through R&D improvements, and it has affected how many patient results end up requiring manual differentials. So eventually, AI has the potential to make the answers better and better.

Nitsan Maayan-Rabinowich, tell us what’s ahead in the next two years for Sight.
Nitsan Maayan-Rabinowich (Sight Dx): At Sight we skipped the stage of going from manual review to digital review and went directly to AI-based CBC digital imaging processing, which will eventually reduce the need for further blood smear reviews of any kind. On top of that, we see a trend going from hematology to cell morphology, and this is exactly what our platform is about. Our name, Sight Diagnostics, means we aim to diagnose everything you can see in a blood sample. We started with malaria, continued to CBC, and now we’re working on additional applications with the same core technology. These applications will solve big challenges for clinicians and patients.

Jill Crist, what are one or two things in Sysmex’s pipeline in the next couple of years?
Jill Crist (Sysmex): We have a lot of big data and we’re working on things with AI and continuing to bring automation to the laboratory. We recently released three modules that have helped with that, and there will be others. We’re going to move with the industry.

Susan Behnke, same question.
Susan Behnke (Horiba): We have a next-generation hematology analyzer, which is available outside the United States, that includes flagging for infectious diseases such as malaria and dengue.

Tim Skelton, what one thing could industry do to make your life a little easier?
Dr. Skelton (Beth Israel Lahey Health): Critical value calls. With the advances in IT, a lot of the providers already have the information before the laboratory calls. In the OR the anesthesiologists have a screen, and the inpatient nursing area has a track board that pops up the critical values and they have already acted on it by the time they get called. We’ve reviewed and cut back a lot on the critical value calls in hematology. We only call things like the platelet count, white count, and the absolute neutrophil count the first time it’s critical, per encounter. The real-time push delivery of results has enabled that.

Jonathan Galeotti, what is the one thing for you?
Dr. Galeotti (UNC School of Medicine): The ability to review remotely, full slide imaging, all the things that streamline workflows and make it easier for pathologists to get their hands on slides and maximize the efficiency of things that are going to happen regardless of the technology we’re using.