What AI Can't See: The Irreplaceable Role of Medical Laboratory Scientists

Deandre White: Where do we begin?

A needle is put into the test tube, a sample is withdrawn, and bingo bango at the other end. There are values that come out, and in most places today, those values get automatically posted to the patient's chart. So it's highly automated. And it's hugely more diverse. Maybe we had, oh, I don't know, I'm going to go on a limb and say maybe 50 to 75 tests back in the 1990s, and now we have thousands. So it's a hugely changing field. At one point, the knowledge base required of a medical laboratory science doubled every three years. So a lot of change that went along there. What has happened in that whole mess is that some person, sometimes it's a certified phlebotomist, sometimes it's a nurse, sometimes in a teaching hospital, it might be a medical student, or a clinical laboratory scientist will come into your room, have a brief conversation with you, draw test tubes worth of blood. You say as a patient, what are you doing? They say drawing blood. You say, what are you doing it for? And in many places, they will require that the person drawing the blood say, I don't know, which is kind of like not an increase in your own security.

There are some where you can't eat bananas for three days. So there are specific preparatory steps that a patient needs to have done. AI can't tell you that. AI says, this is a test tube that has a label on it for Deandre White. Yeah, okay, fine. But it can't tell you any of that beginning stuff. It cannot tell you, is this a test that is so important to your status that I have to stop the machine that I'm using and put your specimen first in line? If it comes from the OR, well, that's obvious. If it comes from the ER, well, that's obvious. Except the label that has been printed out with your name, if you are a patient in a hospital, is your bedroom number. So I'm looking at something that says, White, Deandre, room 335. It does not tell the AI that you are in the OR. I have to look that up. Now I have a piece of paper right by my side that tells me, oh, this is stat, this is from the ER, OR, I have to move these things around. AI can't do that.

The specimen can go through the instrument that I set up. AI can't set that thing up and make sure that it's working correctly. When I look at your answers, before they automatically go to your chart, I'm the one who has to say, this makes sense, this doesn't make sense, this was drawn improperly, this wasn't treated correctly. And I can do something before that laboratory result goes up to the floor, wherever you happen to be, so that your physician can see it. So that interpretive level of pre-analytical, as we call it, did your patient do what they were supposed to do to give me the best test they could possibly, specimen that they could possibly take, and setting the instrument up, that's the analytical part, and then that post-analytical, oh no, wait a minute, I did this test on you yesterday, and something has changed, something important has changed. I better track that down before it goes up to the physician because he or she needs to know in the OR what has changed. So maybe someday AI will get bigger and more complex and more nuanced, and I'm sure there are people out there that are trying to do that, but at the present, they don't do that. They don't.

Dr. Susan Leclair: Yeah, sure. One of the better ones, and we've mentioned, at least I've mentioned it a lot in whatever talk I'm doing, is the role of medications. The average patient in a hospital, and we'll just stick with that one for the moment, is on 13 medications at any given time. The average outpatient, let's see, if you're a cardiac patient, you're on aspirin, you're on a diuretic, you're on maybe a beta blocker, you're on an ACE inhibitor, maybe a calcium channel blocker, okay, they're counting, you know? And the rule is at the moment that if you're on five or more medications at any given time, we don't know how they interact with the individual person. So if AI could say, oh, White, Deandre is on the following medications, and they could cause trouble with this result, well, then I can look at that result, maybe do something, write a note, make a phone call, do something like that to communicate that concern. I could, in the case of the one test that involves the most common complication in cardiovascular surgery called heparin-induced thrombocytopenia, you don't need to know the name. The test that we've had for a number of years is really less than wonderful, has a lot of false negatives.

If I know that this test is being done on someone in the OR, then as a laboratorian, I should be able to call and say, is this person in surgery for cardiovascular disease? Because I got to tell you, this negative could be wrong. So trust your gut, trust what you're seeing in the patient clinically, and don't trust this answer. And that's something AI can't do.

Dr. Susan Leclair: Oh, sure. I'm a storyteller. Just accept it. I was in a lab that was a very... Laboratories reflect the personality in a sense of the administrative structure. I was in a laboratory that did not allow laboratorians to speak to physicians. They had to do, they had to go through word secretaries. It is my first day there. I'm just going to tell you more about me, I think, than anything else. I was doing a test called a CBC. And it had happened that when the specimen came in, the person who delivered the specimen said, oh, it's really too bad. This is a six-year-old boy. He just flew across the country to spend some time with his grandparents and now he fainted and they're all concerned, blah, blah. She leaves. I'm doing the test. I hear noise at the laboratory door and a pediatrician comes in and says, do you have this test done yet? And I said, not quite, but you don't need to worry about, and I gave a specific complication that occurs in children who have had some kind of antigenic shock, whether it's traveling east to west, now you've got a whole lot of antigens you're not used to, or you're just getting over a cold or other viral disease, something like that.

So I said to him, you don't have to worry. It isn't hemolytic uremia. And he said, how do you know? And I was doing the differential, which is a microscopic evaluation of the test, and I leaned back and said, you wanna see? Because they don't look like they would if the person, this child had hemolytic uremia. And he turned around and said to somebody that was next to him, cancel the IVP and cancel the kidney scan, which was at that time probably $5,000 worth of non-laboratory imaging testing that no one had to eat, the insurance companies didn't have to pay, the child didn't have to go through. It just wasn't those. But he was concerned because it's a very common complication. So he was following his line of thought, and I could tell him to stop. And when that happens, it makes you feel good for longer than the rest of the day.

Dr. Susan Leclair: You have to. There's no way for you to... People are not robots. They're not automatons. Every 6'3", 24-year-old male is not identical. Every six-year-old girl with a twirly dress is not identical. They are unique individuals, and you have to look at those numbers in that mindset that they are special. And when you look at the lives that we lead, the stressors that are out there, the things that can happen to people, those are unique to them. Somebody can have chickenpox, and it's from head to toe. The same age kid who got it from the kid who's got the smallpox from head to toe has got a little line here and maybe a small strip across their belly of rash, and that's it. Everybody's different. Every lab test, every imaging, everything that a physician or a nurse practitioner or a PA asks of you has to be filtered through the fact that you are unique.

Dr. Susan Leclair: Oh, welcome. In the 1960s, now in the 1950s, we didn't have many tests and we did them all manually. In the 1960s, automation revolutionized the laboratory. And it ended up we needed more of us. And we needed more of us at a higher level of education because the automation tests that were once six hours to do became a half an hour to do. Tests that once required three or four days could get done in 18 hours. And more tests became available. So we were doing not the old tests, those were automated, but we had newer tests and newer panels of tests to do together. What I see happening now is essentially that process. Is it comfortable? By the way, the answer is no. But I'm seeing maybe AI, when it gets good enough, will be able to handle the common, ordinary, frequently done, because that's how usually it works. You get the frequently done tests become automated or they will go with AI, but that will leave a different kind of clinical laboratory scientist to do the genetics, to do chromosomal analysis, to do variations on CRISPR, to see what could happen.

And I see 50 years ago, if you had told me that there would have been doctorate level trained personnel in a clinical laboratory, I probably would have said, why? Because that was the level of information we had back then. It was also the types of patients we had back then. But now when you move into the 2030s, let's say, moving five years out, we have patients that have chronic diseases, cardiovascular disease, number one in the United States, we'll use it. So I got a 55 year old patient, or we all have a 55 year old patient who has got type 2 diabetes, non-insulin dependent diabetes. They've got a little bit of coronary vascular disease, and now they have, they have colorectal cancer. This person is going to be on meds for the diabetes, the cardiovascular, and the cancer. Which drugs are best for this patient? Should they just take glipizide, or should they take Ozempic? That's the diabetes. Which drugs are they going to take for the cardiovascular disease? And there's legions of them now out there. Which oncology drugs are specific to that genetically derived colon cancer? So should there, could there, might there be tests in which we take a biopsy of your malignancy, grow those cells up a bit, and then test them against 10 or 12 different oncological agents so that in 24 hours I can report to you as a physician and say, this malignancy is sensitive to these four drugs. Don't bother with these six, it's not going to work?

Now wouldn't that be information that a health care provider would be thrilled to know? We currently don't have that test, but can you see how it could move along? We could do it. We can't do it at the same time we're doing blood glucoses. It's just so much time in the day. You either have to train up a whole bunch more of us, or we have to go in and do these tests at a level that requires that PhD, and then requires that PhD to look at the physician or the nurse practitioner or the PA and say, I am equal to you because of what I know in the laboratory. Pharmacology, I don't know this patient's family history, but I know this patient needs these meds. Wouldn't that be cool?

Dr. Susan Leclair: Well, I was going to say thinking. But I don't think that's... But that's hard to get across. AI doesn't, at the moment, does not think. AI has a huge, vast batch of knowledge and it runs through every single piece of knowledge to get to the right answer. If you say to it, what is the first letter of the alphabet? They will start A, B, C, D and run it all the way through to 26 and say A. And if you say to them, what's the third letter of the alphabet? They'll start at A and they'll run all the way through 26 and pick C because they're just very good at very fast sorting in that sense. I'm the one, my colleagues are the ones who sit there saying, "Oh, no, I know that they're there. They're just not there in that amount. So I can't call it this, but I can't..." We're the ones who are doing the educated nuances. And again, I don't think that AI is ready for that, and I don't think it will be ready for it for a while. And even when it is, someone is going to sit there and say, I don't think this AI is right. This just looks a little funny to me. And that level of being able to say, I think, will be the last thing that goes in the clinical laboratory.

Deandre White: Yes, it's very hard to manufacture intuition.

Dr. Susan Leclair: That's right.

Deandre White: I don't think they could ever do. But thank you, Dr. Leclair, for reminding us that lab science is not just a technical field, it's a human one. So while AI is transforming how we work, it can't replace what truly makes a difference. And that's the insight, the ethics, the pattern recognition, and the empathy that only trained professionals like you bring to the lab branch every single day. If you're a student thinking about a career in lab science or a patient who just wants to understand the people that are behind the test result, just know this. Your story deserves a human interpreter. Thanks for joining us and stay curious.

Diagnostic Equity Resources

Staring at your lab test report with lab test questions? Understanding your lab test results starts here.