Trusting the lab over the patient is the new medical error [PODCAST]

Listen and followApple PodcastsSpotifyYouTubeMore apps · Browse all episodes

The lab says the potassium is critically low. You walk in, and the patient looks completely fine. Duane Penshorn is an emergency physician and author. This episode is based on his article “Automation bias is the new medical error,” published on KevinMD. He explains that automation bias got its name from studies of airline pilots, who had seen an alert be correct so many times that they stopped checking whether this one was. You will hear why point-of-care machines never answer “I don’t know,” and why most errors trace to how the sample was handled rather than to the machine. One clumped platelet count sent a child to the hospital for nothing. He is direct about the fix. If the number does not fit the patient in front of you, order it again, and say so out loud to the radiologist when the scan does not fit either. If you have ever scanned a result for the red flags and moved on, this conversation will make you look twice.

Partner with me on the KevinMD platform. With over three million monthly readers and half a million social media followers, I give you direct access to the doctors and patients who matter most. Whether you need a sponsored article, email campaign, video interview, or a spot right here on the podcast, I offer the trusted space your brand deserves to be heard. Let’s work together to tell your story.

PARTNER WITH KEVINMD → https://kevinmd.com/influencer

SUBSCRIBE TO THE PODCAST → https://www.kevinmd.com/podcast

Transcript

Kevin Pho: Hi, and welcome to the show. Subscribe at KevinMD.com/podcast. Today we welcome Duane Penshorn. He’s an emergency physician and author. Today’s KevinMD article is “Automation bias is the new medical error.” Duane, welcome to the show.

Duane Penshorn: Hey, thanks very much, Dr. Pho.

Kevin Pho: All right, let’s start by briefly sharing your story, and then we’ll jump right into your KevinMD article.

Duane Penshorn: My story? Well, let’s see. I’m twenty years out of residency. I started off working as a paramedic for a while, then went to nursing school, worked as an ER nurse, and then ended up in medical school. I have been in emergency medicine for the last twenty years.

Kevin Pho: And why did you decide to write this particular article on KevinMD?

Duane Penshorn: I’d run across a lot of lab errors where the laboratory results didn’t correlate with how the patient was presenting. In my article, I mentioned one in particular, where the lab reported a critically low potassium and calcium on a patient. It was a patient I was getting ready to pick up in a turnover at the end of another practitioner’s shift. We went and looked at the patient, and they didn’t look bad at all. When we looked at the labs closely, it was actually an error: The draw had been mixed with saline, so it had artifactually lowered those values.

Over time, especially with point-of-care testing becoming much more widespread, I’m seeing labs that don’t make sense. Say a renal failure patient has missed dialysis, and their potassium doesn’t come back elevated even though they’ve got EKG changes, things like that. When we rerun the labs, they come back as we expect them to. As I saw more and more of them, that’s when I learned about automation bias, where we become dependent on these laboratory studies, or even scans. We trust them to be true and honest, and the data to be correct, and a lot of times they’re not.

Kevin Pho: When you talk about automation bias, which of course can lead to lab error or diagnostic error, what exactly is the cause of that? How does error creep into automated workflows?

Duane Penshorn: Well, the term was actually coined in the ’90s, when they were studying airline pilots. They noticed that when they tested these airline pilots, they would give them an alert that was incorrect. The pilots had seen those alerts thousands and thousands of times when they were correct, and they just assumed it was correct. They didn’t really look at what was happening and determine whether that alert was actually true or not.

So over time, when we see lab results patient after patient, hundreds and hundreds of patients where they’re correct, when one shows up incorrect and doesn’t correlate with what we see clinically, we tend to believe the lab, or the scan, or whatever the results are, rather than our own eyes and what we see clinically.

Kevin Pho: From your experience in the emergency department, how often do you get these errors from automation?

Duane Penshorn: It’s not often, but the problem is that sometimes it’ll give an abnormal result when you’re expecting something normal, or a normal result when you’re expecting something abnormal. If you’re really not paying attention, something can slip through the cracks. It’s a really busy environment. You’re seeing twenty-five or thirty patients a shift, and you don’t have the time. You have the time to glance at the labs. You look for the critical high or the critical low, and you move on.

Those things were designed to make life easier or quicker for us, and we take those shortcuts when we can. As an ER physician, you’re making a thousand decisions a shift, and you want to look for shortcuts. That’s where you can get in trouble, because it can happen, and then you’re doing the wrong thing for the patient.

Kevin Pho: For those who just aren’t versed in how the laboratory works, are there any fail-safes for errors that are obviously abnormal, other than relying on a physician’s judgment? What kinds of fail-safes are there typically?

Duane Penshorn: Well, the lab will alert you if something abnormal shows up. Typically, you’ll have an alert show up that you have a critically elevated or critically low value for whatever it is you’re testing. Unfortunately, to streamline throughput and flow, a lot of times this alert will pop up, and the pop-up will say, “Hey, here’s what you can do to correct that potassium,” or whatever. And you don’t stop and think, “OK, is that thing really what it is?” If you’re in a hurry, and again, throughput and flow are everything, you can get in trouble.

With point-of-care testing in particular, you’re having nurses or techs draw the blood and put it in the cartridge. There are a lot of opportunities for mistakes. With in-house hospital testing, they do other things to check the blood. They look for platelet aggregation, or they look for a lipemic specimen or something. Point-of-care testing doesn’t have those options, and it’ll result even when something isn’t quite right. It doesn’t necessarily give an answer of “I don’t know.” It’ll give you an answer.

Kevin Pho: Like you said, you make thousands of decisions every emergency department shift. If you have a lack of confidence in reading lab tests, and you have that potential lab error in the back of your mind, how does that affect your decision-making during a busy shift?

Duane Penshorn: Well, I think as long as we keep in mind that it’s just a data point, that it’s not necessarily the truth, and that it has to go along with what we see in our patient. Hopefully we’ve spent enough time and had enough experience that we have suspicion for certain things. When you see something that doesn’t correlate, when something doesn’t make sense, you’ve got to take a minute to think, “OK, am I missing something, or is this a potential error?”

Some of my colleagues have gotten to the point where, if something is really, really unexpected, they just automatically order it again and repeat the study to make sure. If you get two that result the same, then that’s great. But there have been many, many times I’ve gotten discordant values because something happened, whether it was with the sample handling or whatever it was. On paper, point-of-care testing is very good. It’s very accurate. It’s almost as good as in-house hospital testing, but it’s so fast, and that’s why it’s implemented. And it can make mistakes.

Kevin Pho: When you bring up these issues with the lab itself, what types of quality control checks, what types of corrections, can they make to reduce these systemic sources of error?

Duane Penshorn: Well, typically with point-of-care testing, the nurses, techs, or whoever is using the machines go through training. They do daily checks to make sure everything’s calibrated correctly and that the reagents aren’t expired, that sort of thing. But at the end of the day, you don’t centrifuge samples to get serum or plasma levels. You’re using whole blood. You have variability in the temperature of the sample. You have the handling of the sample, whether there are little microclots in the pipette when it’s placed in the sample cartridge. When they’ve done studies of these, that’s where the errors seem to happen more often. The machines themselves are pretty reliable, but it’s the handling and the preparation of the sample that really makes the difference.

Kevin Pho: Do you feel like the labs themselves are, for lack of a better word, maximized in terms of what they can do to reduce error, and now we just have to rely on clinical suspicion, on things that just don’t quite look right, on unexpected lab values, and rely on those types of measures when it comes to interpreting our labs?

Duane Penshorn: Well, I think it comes down to when you see a big discordance, when the patient’s presentation doesn’t match your lab or your study or whatever. When I have a very high suspicion for appendicitis in someone and they get a normal CT scan result, I’ll call the radiologist and say, “Hey, listen. I’m worried about this, this, and this.” Usually they’re correct. Sometimes they’re not. It helps when you give them as much clinical information as you can when you order it in the first place, which always clues them in on what’s going on.

But you just have to question everything, especially when things don’t make sense. If you can’t see the picture, then you’re missing pieces of the puzzle, and you have to spend the extra time to tease that out and try to avoid unfortunate outcomes if you have a miss.

Kevin Pho: Can you give us a case study or example of something you’ve caught because of automation bias that made an appreciable difference in a patient’s clinical course?

Duane Penshorn: I’ve had a number of patients with low-risk chest pain, not many risk factors, getting serial troponins. The first or the second one might come back elevated, and it’s like, “That doesn’t make sense. This can’t be correct.” Then we repeat it, and it comes back normal. Usually it’s not severely elevated, but it’s enough that it would meet criteria for them getting admitted.

In my article, I wrote about the patient who had critical hypokalemia and hypocalcemia, and it was because that sample was diluted with saline. I actually admitted a child one time to a children’s hospital for thrombocytopenia, and when they rechecked the lab at the hospital, they said, “Oh, the platelets were clumped.” It was a mistake, and the child was admitted for no reason. The kid looked pretty good, but I’m not going to be able to put my hands on you and tell you what your platelet count is.

So it happens. There are times these things happen, and we always want to advocate for the patient and err on the side of caution. But it’s something we have to be aware of, because you can’t just look at the numbers, look for the red marks, say, “Nope, everything looks good, let’s go,” and ignore your clinical gestalt.

Kevin Pho: When I was in medical school, about twenty or thirty years ago, we were always taught that the diagnostic tests and labs only confirm what you initially clinically suspect. And of course, recently, people are just relying on these measures that we think are objective. Hearing this story and what you’re talking about goes back to what we all learned in medical school: These confirmatory tests should only be used to confirm what we clinically suspect.

Duane Penshorn: Right, exactly. As you get more experience, you know within a minute what’s wrong with the patient. You can tell. You put your hands on a surgical abdomen, and you know what the problem is. The vast majority of the time, probably 80 to 90 percent, you’re exactly correct: You’re ordering the lab or the study to confirm what you already suspect, or pretty much know to be true.

It’s the borderline cases that are dangerous. You can recognize the obvious ones, even with vital signs. If a screaming kid has a heart rate of 140 and feels hot to the touch, and they check the temperature and say, “Well, he’s 98.6,” you’d say, “Let’s redo that.” You can see those. It’s the borderline cases that are potentially dangerous, the ones you might miss.

Kevin Pho: Any suggestions you can make for hospitals and the administrators who may be listening now, on how to reduce these errors from automation bias? If you had a message for them, what would you say?

Duane Penshorn: Well, I think point-of-care testing is here to stay, especially in lower-volume places, and especially where throughput is important. A lab test that would run $3 or $4 in a hospital lab will run you $15 for point of care. It’s more expensive, but you have an answer back in ten minutes instead of an hour. The amount of time you save means more turnover. That’s more beds you can use, and more patients you can move through. That throughput is worth it economically at the end of the day. So they’re going to do what’s their fiduciary responsibility and make sure it makes sense economically.

I think there needs to be good, tight quality control and training for the staff in how they use it, and an awareness among physicians. When you have an abnormal value, some of my colleagues just repeat it. If I can have an answer in ten minutes, I can have another answer in ten minutes, and I feel good about that, especially if you’re going to do some kind of intervention. And if it’s something you’re really worried about in a patient, there’s no reason you can’t admit someone who doesn’t have anything abnormal. You just have that clinical gestalt, and a good hospitalist is going to believe what you say.

We’re very lucky in the system we work in here in San Antonio. We have a good team, and they really listen to us. When I tell somebody, “Hey, listen, everything’s coming back normal on this patient, but I just don’t feel good about it,” they’re really good about that. Empower the physicians to make those decisions, to make those calls. It’s OK to repeat a lab. It’s OK to say, “Hey, maybe we need to recheck this.”

Kevin Pho: We’re talking to Duane Penshorn. He’s an emergency physician and author. Today’s KevinMD article is “Automation bias is the new medical error.” Duane, let’s end with your take-home messages for the KevinMD audience.

Duane Penshorn: Spend the time with the patient, really listen to what’s going on, put your hands on them, and take a moment to formulate that differential. When you’re ordering your tests, if something comes back abnormal, really think, “Is that what I’m expecting to see?” If it’s not, don’t be afraid to repeat. Don’t be afraid to circle back with a patient. Make sure you’re not missing anything. Because the last thing you want is a hemolyzed specimen in somebody whose potassium is high with no reason for it. Then you start giving them something to bring their potassium down, they start breaking down, and you’ve caused harm.

Kevin Pho: Duane, thank you so much for sharing your story, time, and perspective, and thanks again for coming on the show.

Duane Penshorn: I appreciate it, Dr. Pho. So nice to meet you.

Listen and followApple PodcastsSpotifyYouTubeMore apps · Browse all episodes