The Impact of Diagnostic Errors

Bob Wachter has written a thought-provoking post on the hidden impact of diagnostic errors. The post discusses the current frenzy over preventing system-based errors (i.e. wrong drug, wrong side problems) which are made while providing patient care for an established diagnosis.  But what if the diagnosis is wrong?

It is the classic GIGO scenario: the result is only as good as the data inputs that created it.  As medical imaging becomes more sophisticated, genome mapping moves within reach, and competence in minimally-invasive surgical techniques increases, people naturally expect diagnostic accuracy to increase as well.  Why wouldn't it?

The problem with that logic is: Systems don't make diagnoses, people do. Specifically doctors (although I work with physicians assistants, nurse practitioners and certified nurse-midwives who also make diagnoses, although their training is limited to more straightforward diagnostic pathways.)

Dr. Wachter briefly reviews some of the AI work that has been done to mimic doctors' diagnostic thought processes.  These attempts have been unsuccessful, in part because diagnostic thought processes don't reduce entirely into algorithms, and--I suspect--that an incalculable element of gut instinct goes into every diagnosis we make.

Here's a recent example I diagrammed in my journal.  I was following a woman we'd admitted onto the hospitalist service with altered mental status.  I got a call from the RN that her heart rate was in the 180s:

Click here for a larger photo if you can't read my lousy handwriting.

Here's what I wrote about what was going through my mind when I was standing at the patient's bedside:

That moment when you are watching a rhythm strip without thinking about what to look for, but looking for it anyway. Then, when a nurse asks you what you're looking for, you can answer without searching for the answer, because you've been answering the same question for yourself. In other words, the awareness of problem solving is eliminated, and you are going through an algorithm without thinking about going through it.

Which was a blithering way of saying that, in the moment, I had no awareness of the decision-making process I was engaged in, but I was making decisions nonetheless.

I give you this example, not to repudiate the idea that diagnostic thought processes can be programmed to some extent, but to argue that there is an element of unconsciousness involved in the process.  As you can see from my feeble diagram, I was churning through a lot of data while I looked at the patient:

• Clinical appearance: stable or unstable?
• Telemetry interpretation: rate, rhythm, wide or narrow?
• Differential diagnosis: toxic/metabolic, ischemic, anatomic, EP
• End-organ effects: mentation, urine output, blood pressure
• Interventions: AV blockade, cardioversion, O2

Again, even though I wasn't fully conscious of my thought process, I was processing all of these questions and bits of information all at once.  I understand that non-linear processes such as this can be modeled by computers these days, but at the heart of this diagram is ME and the image I was forming of the patient.  I was looking at her.  I was processing not only the information above, but the following overall impression: She looks OK.  No emergency.  Don't know why the hell this is happening, but I'll tweak her meds.

I think I started a dilt drip, but I don't remember.  She got better.  She went to a SNF, but at least she left the hospital.

For me, the outcome mattered less than the process.  Here's another excerpt from my journal:

See how an algorithm, once learned, becames a series of parallel, mutually occurring feedback loops of reasoning. Or circular. Or big Medusa's web of questions/answers/possible solutions, each of which is ore or less ideal depending on the patient's changing status. So when the nurse says, "I guess it won't help to cardiovert if she keeps going in and out of flutter," I say, silently acknowledging that her statement is true:

"Also, she's not really falling into the unstable category right now." Answering her question (cardioversion less likely to help with paroxysmal SVT) and my own (unstable? no, stable).

I don't know why this fascinates me, except that I cannot for the life of me remember when I learned to think like this.

Oh dear, I do get a bit self-indulgent in my journal, but I don't think it's a bad idea to examine one's thought processes in this way.  Doing so helped me recognize the formal/teachable elements of my decision-making (algorithms, data points, all those arrows) as well as the instinctive element of looking at the patient--the element you can't feed into a computer.

Which gets me back to Dr. Wachter's post:  if, for the moment at least, a doctor's diagnostic acumen can't be fully programmed, we're back to doctor's as the origin of diagnostic errors.  What to do about these errors?  I have no solutions, only a few thoughts:

1. Although I've never been a fan of Groupthink, I do think it is helpful to have colleagues with whom I can discuss cases.  This was the biggest safety valve I had during residency--having other residents around to talk about patients.  Granted, we were all booger-nosed newbies, but we were grounded enough to look things up and ask for help when we felt we were in over our heads.  Unfortunately, in the real world I don't have immediate access to casual colleagues to run cases.  My hospital is small enough that only one hospitalist covers each shift.  Certainly, I can consult cardiologists, nephrologists, pulmonologists, neurologists, but even the most generous consultant I know doesn't want to get dragged into every single case, and professional liability makes them cautious about providing curbside opinions.  I'd like to see a culture shift towards informal collaborative problem-solving in medicine.

2. If, in the most basic form, diagnostic decision-making is a combination of algorithmic mastery and gut instinct, then both elements require a large pool of experiences to refine their accuracy.  Translation: Doctors should see a lot of patients with a lot of different problems in order to become better diagnosticians.  This must happen as part of medical education, which is one of the reasons I have written about my reservations about further reductions in resident work-hours.

So, how are we going to improve diagnostic accuracy? Let's start the informal collaborative problem-solving now--in your comments.

Addendum: Literally hours after completing this post, I started reading Jerome Groopman's How Doctors Think. Talk about timing! Here, Dr. Groopman cites Pat Croskerry, M.D., who explains the mental processes I described above as pattern recognition:

Pattern recognition, Croskerry told me, "reflects an immediacy of perception."  It occur within seconds, largely wihtout any conscious analysis; it draws most heavily on the doctor's visual appraisal of the patient.  And it does not occur by a linear, step-by-step combining of cues. The mind acts like a magnet, pulling in the cues from all directions.

Wow, there it is, in many fewer words that I wrote about it.  

Comments

[allison]

Recently, I took my toddler to the pediatrician's office because he was sick. Unfortunately, he got sick right around a holiday, the office was full, and I couldn't get him in to be seen for 4 days, so I settled for a triage nurse over the phone.

child had eye discharge, and as of that first day, no other sinus or ear infection signs. We were prescribed antibiotic drops.

Of course, the next day was Sat, and the other signs appeared. But it was Tuesday before the ped's office was open again, and again ,they were too full to see me. Child had all the signs of an ear infection now, plus a fever--fever above 102.5. I finally got in on Thursday to see a Nurse Practitioner.

Nurse heard me explain: child had eye discharge, now has runny nose, yellow to green snot, fever, cough, pulling on ears. No problems with eating or wet diapers, other than that kid feels terribly lousy. Father and mother had green snot pouring down their faces, too. Antibiotic eye drops had cleared up the eye issue immediately.

Nurse looked in ears, said "well, left one doesn't look normal, but doesn't look terribly infected". Wasn't sure about the sinuses. Then she asked about wet diapers, and said that she wanted to run more tests, including taking blood, putting in a catheter for a urine sample to check for a , and an XRay for his lungs in case it was pneumonia. when I asked why she suspected a UTI she said it was possible, even though there was no problem with wet diapers; when I asked why pneumonia, she said it was possible even though the lungs sounded good.

I allowed her to take blood but refused the rest. I didn't understand: her own explanations led to a differential diagnosis of sinus or ear infection
She wouldn't prescribe antibiotics, even though they had worked on the eyes. She wanted more tests even though they were HORRIBLY invasive.

why? because she didn't know what a slightly abnormal presentation looked like.

I called my MD (difficult to get through to his voice mail.. requires much guile) and told him about what I thought was a bad diagnostic experience, but by the time he called me back, child's fever was gone, and he was on the mend. So why the heck did I bother?

What would I have for a solution? I would have wanted this NP to have behave like a med student--and run all of this past some MD before asking for more tests, and hear the MD say "sounds like a normal ear/sinus infection to me."

They should discuss and triage all of their cases that day/week to talk about why they made the diagnosis they did. Or someone else should randomly pick their cases and ask them daily/weekly to present, until they all discuss it and get better at it.