0:00

So let's take a closer look at, uh, what's going on here.

0:04

Um, the great philosopher, uh, said, you know,

0:08

we only see what we know.

0:09

Um, but he also said, doubt grows with knowledge.

0:14

So, you know, knowing is not enough.

0:16

He also said that, um,

0:18

and the, uh, great mathematical physicist

0:20

and author, uh, Leonard m Lana said, you know,

0:23

perception requires imagination

0:25

because the data that people encounter in their lives is

0:28

never complete, and it's always equivocal.

0:31

And, you know, we operate in an environment in radiology

0:33

of extremely high uncertainty, you know,

0:35

and also high stakes.

0:36

And also, you know, to a degree

0:38

that I think people in other walks of life might find

0:40

to be fairly paralyzing,

0:42

but in action is rarely an option for us.

0:45

We have to, as the great Osler said, balance probabilities.

0:49

We have to make a decision. We have to take a stand,

0:51

and so we reliably are going to make mistakes.

0:55

So I developed this three part error model.

0:58

I think, you know, we can divide up the errors that we,

1:00

that we commonly see.

1:01

You know, certainly in my 20 year, uh, uh, practice in,

1:05

in quality and safety improvement.

1:07

Uh, there's errors of how we work,

1:09

you know, our systems errors.

1:10

We have bad, bad systems, bad plans, uh, in place

1:14

and, uh, in quality and safety strategies

1:17

and change management.

1:19

Those things really work very well on,

1:21

on those systems type errors.

1:23

There are errors of how we think, you know, cognitive errors

1:26

where we have cognitive biases, especially de-biasing, uh,

1:29

de-biasing strategies have been, uh, looked at as a strategy

1:32

to deal with those, and they're a little tougher.

1:35

But then there's errors of how we're made, which I think is

1:38

where perceptual errors and satisfaction of search come in.

1:42

I think they're not really all that well understood,

1:44

but it's probably a, a fundamental neurobiological process.

1:48

So errors of how we think, uh, I'm sorry, how, how we work,

1:51

uh, in quality and safety strategies.

1:53

Like you have bad processes, bad habits, you have policies

1:57

and procedures, um, culture

1:59

and leadership issues, you know, lots of workarounds.

2:02

You know, those are, are the kinds of things that we deal

2:06

with with systems errors.

2:07

And I think that those, you know, that the plan do,

2:09

check act cycle of Deming.

2:11

I think those things really, uh, work to, to deal with that

2:15

and to get rid of the, the, the tendencies, the,

2:18

the proneness to errors that our systems give us.

2:20

We can do things like root cause analysis and FMEA

2:23

and practice quality improvement projects.

2:25

Lean and six Sigma develop our culture of safety.

2:29

All these things are extremely worthwhile,

2:31

and that's been a lot of my work over the last 20 years.

2:34

So, getting back to that three-part error model,

2:37

we've talked now about errors of how we work,

2:39

and now I'd like to move into errors of how we think,

2:42

and those are the cognitive errors.

2:44

And the main problem that we have with that is, uh,

2:47

our cognitive biases and our mental shortcuts.

2:50

And, uh, the way to deal with those is

2:52

through de-biasing strategies and heuristics.

2:56

And you may have heard of the dual process

2:57

theory, which came to us

2:58

From Canman and D'S work,

3:00

and it was published in Kahneman's book.

3:02

Um, and, uh, Canman, of course won the Nobel Prize.

3:05

He died recently. Um,

3:06

and where they talked about these two types of thinking,

3:09

the, uh, the where you, the patient presents

3:12

and their, their, uh, abnormality is not recognized

3:16

and you sort of like, you know, think it through very slowly

3:20

and, uh, cautiously

3:21

and methodically until you come to a diagnosis.

3:24

But there's also this, uh, this type one process, uh,

3:28

where you just recognize it

3:29

and you immediately jump to the diagnosis.

3:31

And it's a nice mental shortcut, is quite a bit easier,

3:34

and it works great most of the time.

3:37

Uh, nice paper on it by cross carry there.

3:39

Um, uh, but it's not perfect

3:43

because this is where you really can have, um,

3:46

problems due to biases.

3:48

And that is, uh, preconceived notions that you may have, um,

3:52

that the change, the

3:54

or color the way that you, you know, jump to conclusions.

3:58

And, uh, there's a several of these cognitive biases

4:00

that have been recognized.

4:01

Maybe a hundred have been described,

4:03

but there's only a few that are really important

4:05

for medical diagnosis.

4:07

The first one, of course, is anchoring bias.

4:08

And that's why it's number one where you just get stuck on

4:11

the first thing that you think of.

4:13

Uh, and, um, and, you know, it's very hard sort

4:15

of push the wrong idea outta your head if it happens

4:18

to be the one you're anchored on.

4:20

I, I love the confirmation bias.

4:21

Look at this happy, smiling face.

4:24

All the ideas and things that he hears that confirm his,

4:27

his decision, you know, or happily stick

4:30

and everything else just passes right through.

4:32

Um, and, um, overconfidence, the great Osler said

4:36

that there's only two kinds of doctors

4:39

and, uh, the underconfident and the overconfident,

4:41

but, uh, having overconfidence

4:43

and a lack of humility is, is really a, a problem.

4:45

Um, and the outcome bias

4:47

where you judge a decision based on its outcome.

4:50

Uh, and, uh, the, the classic example is, you know,

4:53

you run a red light and you,

4:55

and you don't get into an

4:56

accident and you think, well, that was great.

4:57

I saved, you know, 10 minutes.

4:59

Um, when, um, if you got into an accident,

5:02

you would've said, oh, I shouldn't have run the red light.

5:04

Uh, well, you know, it was the same bad decision both times.

5:07

Uh, and the outcome doesn't really change the fact

5:10

that it was a bad decision.

5:11

Um, and, um, uh, placebo effect, uh, stereotyping,

5:16

survivorship bias, uh, salience bias, uh,

5:20

all these things have been, have been described.

5:22

But the, you know, the main ones that we really deal

5:24

with are anchoring bias and, you know, outcomes bias.

5:27

And, uh, also, um, you know, availability bias where we, we,

5:32

you know, we, we just had a case of pulmonary embolism,

5:34

so we, we think it's more likely than it is.

5:36

It kind of alters our pretest probability.

5:39

Um, and so how do you deal with these things?

5:41

Well, deep biasing strategies have been advocated.

5:44

And, and basically they all come down to just stopping

5:48

and forcing yourself out of your, uh, sort

5:50

of a cognitive rut by asking yourself,

5:53

what else could this be?

5:55

How else could this be happening?

5:57

Where else could I look for this?

5:59

And why else might this, uh, be occurring?

6:02

And if you ask yourself those questions, you can, uh, sort

6:05

of get yourself out of the cognitive bias,

6:07

and you can ask yourself, you know, am I experiencing one

6:09

of the cognitive biases right now?

6:11

Um, and, uh,

6:13

there's a thing called the cognitive autopsy you can do when

6:15

you find out you've made an error.

6:17

You know, you sort of think about it in a calm space, quiet,

6:20

just go through your mind.

6:21

How did you make the decision?

6:23

Uh, and, uh, don't discuss with anyone at first.

6:25

And, uh, you know, consider your decision processes sort

6:28

of proactively go through it again,

6:30

and, uh, not just retrospectively,

6:33

once you know where the outcome is.

6:35

Um, and, uh, ask yourself,

6:37

were there any cognitive biases involved?

6:39

And, you know, that is, uh, uh, a very good exercise to do.

6:44

So this brings us to the final category, which I call errors

6:47

of how we're made perceptual errors, satisfaction of search.

6:51

I think these things are not as well understood as,

6:54

as they could be, but we're beginning to, um,

6:56

you know, understand them better.

6:58

And I think they're not necessarily what they seem.

7:01

I believe that we're dealing with a human

7:02

neurocognitive mechanism.

7:04

The errors that we see appear to be essentially random

7:08

double reading has been really the only

7:11

effective strategy to deal with them.

7:13

Uh, so we're talking about errors

7:14

where you simply don't see something that's there,

7:17

and at least for now, uh, double reading seems to work.

7:20

And that was advocated by Dr. Garland back in 1949.

7:25

Uh, so, uh, we're hoping that artificial intelligence

7:29

with the, you know, neural network architecture can maybe be

7:32

that double reader and help to reduce these types of errors.

7:35

But remember, you know how 9,000, uh, was one

7:39

of the great artificial intelligences of,

7:42

of science fiction, and it was a homicidal maniac.

7:44

So, uh, I think that it's, it's a good idea

7:47

to be a little bit skeptical about this,

7:49

but hopefully, um,

7:51

hopefully artificial intelligence will be helpful for this.